1
0
mirror of https://passt.top/passt synced 2024-06-29 22:42:46 +00:00
passt/udp.c

1364 lines
36 KiB
C
Raw Normal View History

passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
// SPDX-License-Identifier: AGPL-3.0-or-later
/* PASST - Plug A Simple Socket Transport
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* for qemu/UNIX domain socket mode
*
* PASTA - Pack A Subtle Tap Abstraction
* for network namespace/tap device mode
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*
* udp.c - UDP L2-L4 translation routines
*
* Copyright (c) 2020-2021 Red Hat GmbH
* Author: Stefano Brivio <sbrivio@redhat.com>
*/
/**
* DOC: Theory of Operation
*
*
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* For UDP, a reduced version of port-based connection tracking is implemented
* with two purposes:
* - binding ephemeral ports when they're used as source port by the guest, so
* that replies on those ports can be forwarded back to the guest, with a
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* fixed timeout for this binding
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* - packets received from the local host get their source changed to a local
* address (gateway address) so that they can be forwarded to the guest, and
* packets sent as replies by the guest need their destination address to
* be changed back to the address of the local host. This is dynamic to allow
* connections from the gateway as well, and uses the same fixed 180s timeout
*
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* Sockets for bound ports are created at initialisation time, one set for IPv4
* and one for IPv6.
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*
* Packets are forwarded back and forth, by prepending and stripping UDP headers
* in the obvious way, with no port translation.
*
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* In PASTA mode, the L2-L4 translation is skipped for connections to ports
* bound between namespaces using the loopback interface, messages are directly
* transferred between L4 sockets instead. These are called spliced connections
* for consistency with the TCP implementation, but the splice() syscall isn't
* actually used as it wouldn't make sense for datagram-based connections: a
* pair of recvmmsg() and sendmmsg() deals with this case.
*
* The connection tracking for PASTA mode is slightly complicated by the absence
* of actual connections, see struct udp_splice_port, and these examples:
*
* - from init to namespace:
*
* - forward direction: 127.0.0.1:5000 -> 127.0.0.1:80 in init from bound
* socket s, with epoll reference: index = 80, splice = UDP_TO_NS
* - if udp_splice_map[V4][5000].ns_conn_sock:
* - send packet to udp4_splice_map[5000].ns_conn_sock
* - otherwise:
* - create new socket udp_splice_map[V4][5000].ns_conn_sock
* - connect in namespace to 127.0.0.1:80 (note: this destination port
* might be remapped to another port instead)
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* - get source port of new connected socket (10000) with getsockname()
* - add to epoll with reference: index = 10000, splice: UDP_BACK_TO_INIT
* - set udp_splice_map[V4][10000].init_bound_sock to s
* - set udp_splice_map[V4][10000].init_dst_port to 5000
* - update udp_splice_map[V4][5000].ns_conn_ts with current time
*
* - reverse direction: 127.0.0.1:80 -> 127.0.0.1:10000 in namespace from
* connected socket s, having epoll reference: index = 10000,
* splice = UDP_BACK_TO_INIT
* - if udp_splice_map[V4][10000].init_bound_sock:
* - send to udp_splice_map[V4][10000].init_bound_sock, with destination
* port udp_splice_map[V4][10000].init_dst_port (5000)
* - otherwise, discard
*
* - from namespace to init:
*
* - forward direction: 127.0.0.1:2000 -> 127.0.0.1:22 in namespace from bound
* socket s, with epoll reference: index = 22, splice = UDP_TO_INIT
* - if udp4_splice_map[V4][2000].init_conn_sock:
* - send packet to udp4_splice_map[2000].init_conn_sock
* - otherwise:
* - create new socket udp_splice_map[V4][2000].init_conn_sock
* - connect in init to 127.0.0.1:22 (note: this destination port
* might be remapped to another port instead)
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* - get source port of new connected socket (4000) with getsockname()
* - add to epoll with reference: index = 4000, splice = UDP_BACK_TO_NS
* - set udp_splice_map[V4][4000].ns_bound_sock to s
* - set udp_splice_map[V4][4000].ns_dst_port to 2000
* - update udp_splice_map[V4][4000].init_conn_ts with current time
*
* - reverse direction: 127.0.0.1:22 -> 127.0.0.1:4000 in init from connected
* socket s, having epoll reference: index = 4000, splice = UDP_BACK_TO_NS
* - if udp_splice_map[V4][4000].ns_bound_sock:
* - send to udp_splice_map[V4][4000].ns_bound_sock, with destination port
* udp_splice_map[4000].ns_dst_port (2000)
* - otherwise, discard
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*/
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
#include <sched.h>
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#include <stdio.h>
#include <errno.h>
#include <limits.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/udp.h>
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#include <stdint.h>
#include <stddef.h>
#include <string.h>
#include <sys/epoll.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/uio.h>
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#include <unistd.h>
#include <time.h>
#include <assert.h>
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#include "checksum.h"
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
#include "util.h"
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
#include "passt.h"
#include "tap.h"
#include "pcap.h"
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
#define UDP_CONN_TIMEOUT 180 /* s, timeout for ephemeral or local bind */
#define UDP_SPLICE_FRAMES 32
#define UDP_TAP_FRAMES_MEM 32
#define UDP_TAP_FRAMES (c->mode == MODE_PASST ? UDP_TAP_FRAMES_MEM : 1)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
/**
* struct udp_tap_port - Port tracking based on tap-facing source port
* @sock: Socket bound to source port used as index
* @flags: Flags for local bind, loopback address/unicast address as source
* @ts: Activity timestamp from tap, used for socket aging
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
*/
struct udp_tap_port {
int sock;
uint8_t flags;
#define PORT_LOCAL BIT(0)
#define PORT_LOOPBACK BIT(1)
#define PORT_GUA BIT(2)
time_t ts;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
};
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
/**
* struct udp_splice_port - Source port tracking for traffic between namespaces
* @ns_conn_sock: Socket connected in namespace for init source port
* @init_conn_sock: Socket connected in init for namespace source port
* @ns_conn_ts: Timestamp of activity for socket connected in namespace
* @init_conn_ts: Timestamp of activity for socket connceted in init
* @ns_dst_port: Destination port in namespace for init source port
* @init_dst_port: Destination port in init for namespace source port
* @ns_bound_sock: Bound socket in namespace for this source port in init
* @init_bound_sock: Bound socket in init for this source port in namespace
*/
struct udp_splice_port {
int ns_conn_sock;
int init_conn_sock;
time_t ns_conn_ts;
time_t init_conn_ts;
in_port_t ns_dst_port;
in_port_t init_dst_port;
int ns_bound_sock;
int init_bound_sock;
};
/* Port tracking, arrays indexed by packet source port (host order) */
static struct udp_tap_port udp_tap_map [IP_VERSIONS][NUM_PORTS];
static struct udp_splice_port udp_splice_map [IP_VERSIONS][NUM_PORTS];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
enum udp_act_type {
UDP_ACT_TAP,
UDP_ACT_NS_CONN,
UDP_ACT_INIT_CONN,
UDP_ACT_TYPE_MAX,
};
/* Activity-based aging for bindings */
static uint8_t udp_act[IP_VERSIONS][UDP_ACT_TYPE_MAX][DIV_ROUND_UP(NUM_PORTS, 8)];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
/* Static buffers */
/**
* udp4_l2_buf_t - Pre-cooked IPv4 packet buffers for tap connections
* @s_in: Source socket address, filled in by recvmmsg()
* @psum: Partial IP header checksum (excluding tot_len and saddr)
* @vnet_len: 4-byte qemu vnet buffer length descriptor, only for passt mode
* @eh: Pre-filled Ethernet header
* @iph: Pre-filled IP header (except for tot_len and saddr)
* @uh: Headroom for UDP header
* @data: Storage for UDP payload
*/
static struct udp4_l2_buf_t {
struct sockaddr_in s_in;
uint32_t psum;
uint32_t vnet_len;
struct ethhdr eh;
struct iphdr iph;
struct udphdr uh;
uint8_t data[USHRT_MAX -
(sizeof(struct iphdr) + sizeof(struct udphdr))];
} __attribute__ ((packed, aligned(__alignof__(unsigned int))))
udp4_l2_buf[UDP_TAP_FRAMES_MEM];
/**
* udp6_l2_buf_t - Pre-cooked IPv6 packet buffers for tap connections
* @s_in6: Source socket address, filled in by recvmmsg()
* @vnet_len: 4-byte qemu vnet buffer length descriptor, only for passt mode
* @eh: Pre-filled Ethernet header
* @ip6h: Pre-filled IP header (except for payload_len and addresses)
* @uh: Headroom for UDP header
* @data: Storage for UDP payload
*/
struct udp6_l2_buf_t {
struct sockaddr_in6 s_in6;
#ifdef __AVX2__
/* Align ip6h to 32-byte boundary. */
uint8_t pad[64 - (sizeof(struct sockaddr_in6) + sizeof(struct ethhdr) +
sizeof(uint32_t))];
#endif
uint32_t vnet_len;
struct ethhdr eh;
struct ipv6hdr ip6h;
struct udphdr uh;
uint8_t data[USHRT_MAX -
(sizeof(struct ipv6hdr) + sizeof(struct udphdr))];
#ifdef __AVX2__
} __attribute__ ((packed, aligned(32)))
#else
} __attribute__ ((packed, aligned(__alignof__(unsigned int))))
#endif
udp6_l2_buf[UDP_TAP_FRAMES_MEM];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
static struct sockaddr_storage udp_splice_namebuf;
static uint8_t udp_splice_buf[UDP_SPLICE_FRAMES][USHRT_MAX];
/* recvmmsg()/sendmmsg() data for tap */
static struct iovec udp4_l2_iov_sock [UDP_TAP_FRAMES_MEM];
static struct iovec udp6_l2_iov_sock [UDP_TAP_FRAMES_MEM];
static struct iovec udp4_l2_iov_tap [UDP_TAP_FRAMES_MEM];
static struct iovec udp6_l2_iov_tap [UDP_TAP_FRAMES_MEM];
static struct mmsghdr udp4_l2_mh_sock [UDP_TAP_FRAMES_MEM];
static struct mmsghdr udp6_l2_mh_sock [UDP_TAP_FRAMES_MEM];
static struct mmsghdr udp4_l2_mh_tap [UDP_TAP_FRAMES_MEM];
static struct mmsghdr udp6_l2_mh_tap [UDP_TAP_FRAMES_MEM];
/* recvmmsg()/sendmmsg() data for "spliced" connections */
static struct iovec udp_iov_recv [UDP_SPLICE_FRAMES];
static struct mmsghdr udp_mmh_recv [UDP_SPLICE_FRAMES];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
static struct iovec udp_iov_send [UDP_SPLICE_FRAMES];
static struct mmsghdr udp_mmh_send [UDP_SPLICE_FRAMES];
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
static struct iovec udp_iov_sendto [UDP_SPLICE_FRAMES];
static struct mmsghdr udp_mmh_sendto [UDP_SPLICE_FRAMES];
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
/**
* udp_invert_portmap() - Compute reverse port translations for return packets
* @fwd: Port forwarding configuration to compute reverse map for
*/
static void udp_invert_portmap(struct udp_port_fwd *fwd)
{
int i;
assert(ARRAY_SIZE(fwd->f.delta) == ARRAY_SIZE(fwd->rdelta));
for (i = 0; i < ARRAY_SIZE(fwd->f.delta); i++) {
in_port_t delta = fwd->f.delta[i];
if (delta)
fwd->rdelta[(in_port_t)i + delta] = NUM_PORTS - delta;
}
}
/**
* udp_update_check4() - Update checksum with variable parts from stored one
* @buf: L2 packet buffer with final IPv4 header
*/
static void udp_update_check4(struct udp4_l2_buf_t *buf)
{
uint32_t sum = buf->psum;
sum += buf->iph.tot_len;
sum += (buf->iph.saddr >> 16) & 0xffff;
sum += buf->iph.saddr & 0xffff;
buf->iph.check = (uint16_t)~csum_fold(sum);
}
/**
* udp_update_l2_buf() - Update L2 buffers with Ethernet and IPv4 addresses
* @eth_d: Ethernet destination address, NULL if unchanged
* @eth_s: Ethernet source address, NULL if unchanged
* @ip_da: Pointer to IPv4 destination address, NULL if unchanged
*/
void udp_update_l2_buf(const unsigned char *eth_d, const unsigned char *eth_s,
const uint32_t *ip_da)
{
int i;
for (i = 0; i < UDP_TAP_FRAMES_MEM; i++) {
struct udp4_l2_buf_t *b4 = &udp4_l2_buf[i];
struct udp6_l2_buf_t *b6 = &udp6_l2_buf[i];
if (eth_d) {
memcpy(b4->eh.h_dest, eth_d, ETH_ALEN);
memcpy(b6->eh.h_dest, eth_d, ETH_ALEN);
}
if (eth_s) {
memcpy(b4->eh.h_source, eth_s, ETH_ALEN);
memcpy(b6->eh.h_source, eth_s, ETH_ALEN);
}
if (ip_da) {
b4->iph.daddr = *ip_da;
if (!i) {
b4->iph.saddr = 0;
b4->iph.tot_len = 0;
b4->iph.check = 0;
b4->psum = sum_16b(&b4->iph, 20);
} else {
b4->psum = udp4_l2_buf[0].psum;
}
}
}
}
/**
* udp_sock4_iov_init() - Initialise scatter-gather L2 buffers for IPv4 sockets
*/
static void udp_sock4_iov_init(void)
{
struct mmsghdr *h;
int i;
for (i = 0; i < ARRAY_SIZE(udp4_l2_buf); i++) {
udp4_l2_buf[i] = (struct udp4_l2_buf_t) {
{ 0 }, 0, 0,
L2_BUF_ETH_IP4_INIT, L2_BUF_IP4_INIT(IPPROTO_UDP),
{{{ 0 }}}, { 0 },
};
}
for (i = 0, h = udp4_l2_mh_sock; i < UDP_TAP_FRAMES_MEM; i++, h++) {
struct msghdr *mh = &h->msg_hdr;
mh->msg_name = &udp4_l2_buf[i].s_in;
mh->msg_namelen = sizeof(udp4_l2_buf[i].s_in);
udp4_l2_iov_sock[i].iov_base = udp4_l2_buf[i].data;
udp4_l2_iov_sock[i].iov_len = sizeof(udp4_l2_buf[i].data);
mh->msg_iov = &udp4_l2_iov_sock[i];
mh->msg_iovlen = 1;
}
for (i = 0, h = udp4_l2_mh_tap; i < UDP_TAP_FRAMES_MEM; i++, h++) {
struct msghdr *mh = &h->msg_hdr;
udp4_l2_iov_tap[i].iov_base = &udp4_l2_buf[i].vnet_len;
mh->msg_iov = &udp4_l2_iov_tap[i];
mh->msg_iovlen = 1;
}
}
/**
* udp_sock6_iov_init() - Initialise scatter-gather L2 buffers for IPv6 sockets
*/
static void udp_sock6_iov_init(void)
{
struct mmsghdr *h;
int i;
for (i = 0; i < ARRAY_SIZE(udp6_l2_buf); i++) {
udp6_l2_buf[i] = (struct udp6_l2_buf_t) {
{ 0 },
#ifdef __AVX2__
{ 0 },
#endif
0, L2_BUF_ETH_IP6_INIT, L2_BUF_IP6_INIT(IPPROTO_UDP),
{{{ 0 }}}, { 0 },
};
}
for (i = 0, h = udp6_l2_mh_sock; i < UDP_TAP_FRAMES_MEM; i++, h++) {
struct msghdr *mh = &h->msg_hdr;
mh->msg_name = &udp6_l2_buf[i].s_in6;
mh->msg_namelen = sizeof(struct sockaddr_in6);
udp6_l2_iov_sock[i].iov_base = udp6_l2_buf[i].data;
udp6_l2_iov_sock[i].iov_len = sizeof(udp6_l2_buf[i].data);
mh->msg_iov = &udp6_l2_iov_sock[i];
mh->msg_iovlen = 1;
}
for (i = 0, h = udp6_l2_mh_tap; i < UDP_TAP_FRAMES_MEM; i++, h++) {
struct msghdr *mh = &h->msg_hdr;
udp6_l2_iov_tap[i].iov_base = &udp6_l2_buf[i].vnet_len;
mh->msg_iov = &udp6_l2_iov_tap[i];
mh->msg_iovlen = 1;
}
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
/**
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* udp_splice_connect() - Create and connect socket for "spliced" binding
* @c: Execution context
* @v6: Set for IPv6 connections
* @bound_sock: Originating bound socket
* @src: Source port of original connection, host order
* @dst: Destination port of original connection, host order
* @splice: UDP_BACK_TO_INIT from init, UDP_BACK_TO_NS from namespace
*
* Return: connected socket, negative error code on failure
*
* #syscalls:pasta getsockname
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
*/
int udp_splice_connect(const struct ctx *c, int v6, int bound_sock,
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
in_port_t src, in_port_t dst, int splice)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
{
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct epoll_event ev = { .events = EPOLLIN | EPOLLRDHUP | EPOLLHUP };
union epoll_ref ref = { .r.proto = IPPROTO_UDP,
.r.p.udp.udp = { .splice = splice, .v6 = v6 }
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
};
struct sockaddr_storage sa;
struct udp_splice_port *sp;
socklen_t sl = sizeof(sa);
int s;
s = socket(v6 ? AF_INET6 : AF_INET, SOCK_DGRAM | SOCK_NONBLOCK,
IPPROTO_UDP);
if (s > SOCKET_MAX) {
close(s);
return -EIO;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (s < 0)
return s;
ref.r.s = s;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (v6) {
struct sockaddr_in6 addr6 = {
.sin6_family = AF_INET6,
.sin6_port = htons(dst),
.sin6_addr = IN6ADDR_LOOPBACK_INIT,
};
if (connect(s, (struct sockaddr *)&addr6, sizeof(addr6)))
goto fail;
} else {
struct sockaddr_in addr4 = {
.sin_family = AF_INET,
.sin_port = htons(dst),
.sin_addr = { .s_addr = htonl(INADDR_LOOPBACK) },
};
if (connect(s, (struct sockaddr *)&addr4, sizeof(addr4)))
goto fail;
}
if (getsockname(s, (struct sockaddr *)&sa, &sl))
goto fail;
if (v6) {
struct sockaddr_in6 sa6;
memcpy(&sa6, &sa, sizeof(sa6));
ref.r.p.udp.udp.port = ntohs(sa6.sin6_port);
} else {
struct sockaddr_in sa4;
memcpy(&sa4, &sa, sizeof(sa4));
ref.r.p.udp.udp.port = ntohs(sa4.sin_port);
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
sp = &udp_splice_map[v6 ? V6 : V4][ref.r.p.udp.udp.port];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (splice == UDP_BACK_TO_INIT) {
sp->init_bound_sock = bound_sock;
sp->init_dst_port = src;
udp_splice_map[v6 ? V6 : V4][src].ns_conn_sock = s;
bitmap_set(udp_act[v6 ? V6 : V4][UDP_ACT_NS_CONN], src);
} else if (splice == UDP_BACK_TO_NS) {
sp->ns_bound_sock = bound_sock;
sp->ns_dst_port = src;
udp_splice_map[v6 ? V6 : V4][src].init_conn_sock = s;
bitmap_set(udp_act[v6 ? V6 : V4][UDP_ACT_INIT_CONN], src);
}
ev.data.u64 = ref.u64;
epoll_ctl(c->epollfd, EPOLL_CTL_ADD, s, &ev);
return s;
fail:
close(s);
return -1;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
/**
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* struct udp_splice_connect_ns_arg - Arguments for udp_splice_connect_ns()
* @c: Execution context
* @v6: Set for inbound IPv6 connection
* @bound_sock: Originating bound socket
* @src: Source port of original connection, host order
* @dst: Destination port of original connection, host order
* @s: Newly created socket or negative error code
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
*/
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct udp_splice_connect_ns_arg {
const struct ctx *c;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
int v6;
int bound_sock;
in_port_t src;
in_port_t dst;
int s;
};
/**
* udp_splice_connect_ns() - Enter namespace and call udp_splice_connect()
* @arg: See struct udp_splice_connect_ns_arg
*
* Return: 0
*/
static int udp_splice_connect_ns(void *arg)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
{
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct udp_splice_connect_ns_arg *a;
a = (struct udp_splice_connect_ns_arg *)arg;
passt, pasta: Namespace-based sandboxing, defer seccomp policy application To reach (at least) a conceptually equivalent security level as implemented by --enable-sandbox in slirp4netns, we need to create a new mount namespace and pivot_root() into a new (empty) mountpoint, so that passt and pasta can't access any filesystem resource after initialisation. While at it, also detach IPC, PID (only for passt, to prevent vulnerabilities based on the knowledge of a target PID), and UTS namespaces. With this approach, if we apply the seccomp filters right after the configuration step, the number of allowed syscalls grows further. To prevent this, defer the application of seccomp policies after the initialisation phase, before the main loop, that's where we expect bad things to happen, potentially. This way, we get back to 22 allowed syscalls for passt and 34 for pasta, on x86_64. While at it, move #syscalls notes to specific code paths wherever it conceptually makes sense. We have to open all the file handles we'll ever need before sandboxing: - the packet capture file can only be opened once, drop instance numbers from the default path and use the (pre-sandbox) PID instead - /proc/net/tcp{,v6} and /proc/net/udp{,v6}, for automatic detection of bound ports in pasta mode, are now opened only once, before sandboxing, and their handles are stored in the execution context - the UNIX domain socket for passt is also bound only once, before sandboxing: to reject clients after the first one, instead of closing the listening socket, keep it open, accept and immediately discard new connection if we already have a valid one Clarify the (unchanged) behaviour for --netns-only in the man page. To actually make passt and pasta processes run in a separate PID namespace, we need to unshare(CLONE_NEWPID) before forking to background (if configured to do so). Introduce a small daemon() implementation, __daemon(), that additionally saves the PID file before forking. While running in foreground, the process itself can't move to a new PID namespace (a process can't change the notion of its own PID): mention that in the man page. For some reason, fork() in a detached PID namespace causes SIGTERM and SIGQUIT to be ignored, even if the handler is still reported as SIG_DFL: add a signal handler that just exits. We can now drop most of the pasta_child_handler() implementation, that took care of terminating all processes running in the same namespace, if pasta started a shell: the shell itself is now the init process in that namespace, and all children will terminate once the init process exits. Issuing 'echo $$' in a detached PID namespace won't return the actual namespace PID as seen from the init namespace: adapt demo and test setup scripts to reflect that. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2022-02-07 20:11:37 +00:00
if (ns_enter(a->c))
return 0;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
a->s = udp_splice_connect(a->c, a->v6, a->bound_sock, a->src, a->dst,
UDP_BACK_TO_INIT);
return 0;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
/**
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* udp_sock_handler_splice() - Handler for socket mapped to "spliced" connection
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @c: Execution context
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* @ref: epoll reference
* @events: epoll events bitmap
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @now: Current timestamp
*/
static void udp_sock_handler_splice(const struct ctx *c, union epoll_ref ref,
uint32_t events, const struct timespec *now)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
{
in_port_t src, dst = ref.r.p.udp.udp.port, send_dst = 0;
struct msghdr *mh = &udp_mmh_recv[0].msg_hdr;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct sockaddr_storage *sa_s = mh->msg_name;
int s, v6 = ref.r.p.udp.udp.v6, n, i;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (!(events & EPOLLIN))
return;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
n = recvmmsg(ref.r.s, udp_mmh_recv, UDP_SPLICE_FRAMES, 0, NULL);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (n <= 0)
return;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (v6) {
struct sockaddr_in6 *sa = (struct sockaddr_in6 *)sa_s;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
src = htons(sa->sin6_port);
} else {
struct sockaddr_in *sa = (struct sockaddr_in *)sa_s;
src = ntohs(sa->sin_port);
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
switch (ref.r.p.udp.udp.splice) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
case UDP_TO_NS:
src += c->udp.fwd_out.rdelta[src];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (!(s = udp_splice_map[v6][src].ns_conn_sock)) {
struct udp_splice_connect_ns_arg arg = {
c, v6, ref.r.s, src, dst, -1,
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
};
NS_CALL(udp_splice_connect_ns, &arg);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if ((s = arg.s) < 0)
return;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_splice_map[v6][src].ns_conn_ts = now->tv_sec;
break;
case UDP_BACK_TO_INIT:
if (!(s = udp_splice_map[v6][dst].init_bound_sock))
return;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
send_dst = udp_splice_map[v6][dst].init_dst_port;
break;
case UDP_TO_INIT:
src += c->udp.fwd_in.rdelta[src];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (!(s = udp_splice_map[v6][src].init_conn_sock)) {
s = udp_splice_connect(c, v6, ref.r.s, src, dst,
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
UDP_BACK_TO_NS);
if (s < 0)
return;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_splice_map[v6][src].init_conn_ts = now->tv_sec;
break;
case UDP_BACK_TO_NS:
if (!(s = udp_splice_map[v6][dst].ns_bound_sock))
return;
send_dst = udp_splice_map[v6][dst].ns_dst_port;
break;
default:
return;
}
if (ref.r.p.udp.udp.splice == UDP_TO_NS ||
ref.r.p.udp.udp.splice == UDP_TO_INIT) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
for (i = 0; i < n; i++) {
struct msghdr *mh_s = &udp_mmh_send[i].msg_hdr;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
mh_s->msg_iov->iov_len = udp_mmh_recv[i].msg_len;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
sendmmsg(s, udp_mmh_send, n, MSG_NOSIGNAL);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
return;
}
for (i = 0; i < n; i++) {
struct msghdr *mh_s = &udp_mmh_sendto[i].msg_hdr;
mh_s->msg_iov->iov_len = udp_mmh_recv[i].msg_len;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (v6) {
*((struct sockaddr_in6 *)&udp_splice_namebuf) =
((struct sockaddr_in6) {
.sin6_family = AF_INET6,
.sin6_addr = IN6ADDR_LOOPBACK_INIT,
.sin6_port = htons(send_dst),
.sin6_scope_id = 0,
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
});
} else {
*((struct sockaddr_in *)&udp_splice_namebuf) =
((struct sockaddr_in) {
.sin_family = AF_INET,
.sin_addr = { .s_addr = htonl(INADDR_LOOPBACK) },
.sin_port = htons(send_dst),
.sin_zero = { 0 },
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
});
}
sendmmsg(s, udp_mmh_sendto, n, MSG_NOSIGNAL);
}
/**
* udp_sock_fill_data_v4() - Fill and queue one buffer. In pasta mode, write it
* @c: Execution context
* @n: Index of buffer in udp4_l2_buf pool
* @ref: epoll reference from socket
* @msg_idx: Index within message being prepared (spans multiple buffers)
* @msg_len: Length of current message being prepared for sending
* @now: Current timestamp
*/
static void udp_sock_fill_data_v4(const struct ctx *c, int n,
union epoll_ref ref,
int *msg_idx, int *msg_bufs, ssize_t *msg_len,
const struct timespec *now)
{
struct msghdr *mh = &udp6_l2_mh_tap[*msg_idx].msg_hdr;
struct udp4_l2_buf_t *b = &udp4_l2_buf[n];
size_t ip_len, buf_len;
in_port_t src_port;
in_addr_t src;
ip_len = udp4_l2_mh_sock[n].msg_len + sizeof(b->iph) + sizeof(b->uh);
b->iph.tot_len = htons(ip_len);
src = ntohl(b->s_in.sin_addr.s_addr);
src_port = htons(b->s_in.sin_port);
if (src >> IN_CLASSA_NSHIFT == IN_LOOPBACKNET ||
src == INADDR_ANY || src == ntohl(c->ip4.addr_seen)) {
b->iph.saddr = c->ip4.gw;
udp_tap_map[V4][src_port].ts = now->tv_sec;
udp_tap_map[V4][src_port].flags |= PORT_LOCAL;
if (b->s_in.sin_addr.s_addr == c->ip4.addr_seen)
udp_tap_map[V4][src_port].flags &= ~PORT_LOOPBACK;
else
udp_tap_map[V4][src_port].flags |= PORT_LOOPBACK;
bitmap_set(udp_act[V4][UDP_ACT_TAP], src_port);
} else if (c->ip4.dns_fwd &&
src == ntohl(c->ip4.dns[0]) && ntohs(src_port) == 53) {
b->iph.saddr = c->ip4.dns_fwd;
} else {
b->iph.saddr = b->s_in.sin_addr.s_addr;
}
udp_update_check4(b);
b->uh.source = b->s_in.sin_port;
b->uh.dest = htons(ref.r.p.udp.udp.port);
b->uh.len = htons(udp4_l2_mh_sock[n].msg_len + sizeof(b->uh));
if (c->mode == MODE_PASTA) {
PRAGMA_STRINGOP_OVERREAD_IGNORE
if (write(c->fd_tap, &b->eh, sizeof(b->eh) + ip_len) < 0)
debug("tap write: %s", strerror(errno));
PRAGMA_STRINGOP_OVERREAD_IGNORE_POP
pcap((char *)&b->eh, sizeof(b->eh) + ip_len);
return;
}
b->vnet_len = htonl(ip_len + sizeof(struct ethhdr));
buf_len = sizeof(uint32_t) + sizeof(struct ethhdr) + ip_len;
udp4_l2_iov_tap[n].iov_len = buf_len;
/* With bigger messages, qemu closes the connection. */
if (*msg_bufs && *msg_len + buf_len > SHRT_MAX) {
mh->msg_iovlen = *msg_bufs;
(*msg_idx)++;
udp4_l2_mh_tap[*msg_idx].msg_hdr.msg_iov = &udp4_l2_iov_tap[n];
*msg_len = *msg_bufs = 0;
}
*msg_len += buf_len;
(*msg_bufs)++;
}
/**
* udp_sock_fill_data_v4() - Fill and queue one buffer. In pasta mode, write it
* @c: Execution context
* @n: Index of buffer in udp4_l2_buf pool
* @ref: epoll reference from socket
* @msg_idx: Index within message being prepared (spans multiple buffers)
* @msg_len: Length of current message being prepared for sending
* @now: Current timestamp
*/
static void udp_sock_fill_data_v6(const struct ctx *c, int n,
union epoll_ref ref,
int *msg_idx, int *msg_bufs, ssize_t *msg_len,
const struct timespec *now)
{
struct msghdr *mh = &udp6_l2_mh_tap[*msg_idx].msg_hdr;
struct udp6_l2_buf_t *b = &udp6_l2_buf[n];
size_t ip_len, buf_len;
struct in6_addr *src;
in_port_t src_port;
src = &b->s_in6.sin6_addr;
src_port = ntohs(b->s_in6.sin6_port);
ip_len = udp6_l2_mh_sock[n].msg_len + sizeof(b->ip6h) + sizeof(b->uh);
b->ip6h.payload_len = htons(udp6_l2_mh_sock[n].msg_len + sizeof(b->uh));
if (IN6_IS_ADDR_LINKLOCAL(src)) {
b->ip6h.daddr = c->ip6.addr_ll_seen;
b->ip6h.saddr = b->s_in6.sin6_addr;
} else if (IN6_IS_ADDR_LOOPBACK(src) ||
IN6_ARE_ADDR_EQUAL(src, &c->ip6.addr_seen) ||
IN6_ARE_ADDR_EQUAL(src, &c->ip6.addr)) {
b->ip6h.daddr = c->ip6.addr_ll_seen;
if (IN6_IS_ADDR_LINKLOCAL(&c->ip6.gw))
b->ip6h.saddr = c->ip6.gw;
else
b->ip6h.saddr = c->ip6.addr_ll;
udp_tap_map[V6][src_port].ts = now->tv_sec;
udp_tap_map[V6][src_port].flags |= PORT_LOCAL;
if (IN6_IS_ADDR_LOOPBACK(src))
udp_tap_map[V6][src_port].flags |= PORT_LOOPBACK;
else
udp_tap_map[V6][src_port].flags &= ~PORT_LOOPBACK;
if (IN6_ARE_ADDR_EQUAL(src, &c->ip6.addr))
udp_tap_map[V6][src_port].flags |= PORT_GUA;
else
udp_tap_map[V6][src_port].flags &= ~PORT_GUA;
bitmap_set(udp_act[V6][UDP_ACT_TAP], src_port);
} else if (!IN6_IS_ADDR_UNSPECIFIED(&c->ip6.dns_fwd) &&
IN6_ARE_ADDR_EQUAL(src, &c->ip6.dns_fwd) && src_port == 53) {
b->ip6h.daddr = c->ip6.addr_seen;
b->ip6h.saddr = c->ip6.dns_fwd;
} else {
b->ip6h.daddr = c->ip6.addr_seen;
b->ip6h.saddr = b->s_in6.sin6_addr;
}
b->uh.source = b->s_in6.sin6_port;
b->uh.dest = htons(ref.r.p.udp.udp.port);
b->uh.len = b->ip6h.payload_len;
b->ip6h.hop_limit = IPPROTO_UDP;
b->ip6h.version = b->ip6h.nexthdr = b->uh.check = 0;
b->uh.check = csum(&b->ip6h, ip_len, 0);
b->ip6h.version = 6;
b->ip6h.nexthdr = IPPROTO_UDP;
b->ip6h.hop_limit = 255;
if (c->mode == MODE_PASTA) {
PRAGMA_STRINGOP_OVERREAD_IGNORE
if (write(c->fd_tap, &b->eh, sizeof(b->eh) + ip_len) < 0)
debug("tap write: %s", strerror(errno));
PRAGMA_STRINGOP_OVERREAD_IGNORE_POP
pcap((char *)&b->eh, sizeof(b->eh) + ip_len);
return;
}
b->vnet_len = htonl(ip_len + sizeof(struct ethhdr));
buf_len = sizeof(uint32_t) + sizeof(struct ethhdr) + ip_len;
udp6_l2_iov_tap[n].iov_len = buf_len;
/* With bigger messages, qemu closes the connection. */
if (*msg_bufs && *msg_len + buf_len > SHRT_MAX) {
mh->msg_iovlen = *msg_bufs;
(*msg_idx)++;
udp6_l2_mh_tap[*msg_idx].msg_hdr.msg_iov = &udp6_l2_iov_tap[n];
*msg_len = *msg_bufs = 0;
}
*msg_len += buf_len;
(*msg_bufs)++;
}
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
/**
* udp_sock_handler() - Handle new data from socket
* @c: Execution context
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* @ref: epoll reference
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
* @events: epoll events bitmap
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @now: Current timestamp
*
* #syscalls recvmmsg
* #syscalls:passt sendmmsg sendmsg
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*/
void udp_sock_handler(const struct ctx *c, union epoll_ref ref, uint32_t events,
const struct timespec *now)
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
{
ssize_t n, msg_len = 0, missing = 0;
int msg_bufs = 0, msg_i = 0, ret;
struct mmsghdr *tap_mmh;
struct msghdr *last_mh;
unsigned int i;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
if (events == EPOLLERR)
return;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
if (ref.r.p.udp.udp.splice) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_sock_handler_splice(c, ref, events, now);
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
return;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
if (ref.r.p.udp.udp.v6) {
n = recvmmsg(ref.r.s, udp6_l2_mh_sock, UDP_TAP_FRAMES, 0, NULL);
if (n <= 0)
return;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
udp6_l2_mh_tap[0].msg_hdr.msg_iov = &udp6_l2_iov_tap[0];
for (i = 0; i < (unsigned)n; i++) {
udp_sock_fill_data_v6(c, i, ref,
&msg_i, &msg_bufs, &msg_len, now);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
udp6_l2_mh_tap[msg_i].msg_hdr.msg_iovlen = msg_bufs;
tap_mmh = udp6_l2_mh_tap;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
} else {
n = recvmmsg(ref.r.s, udp4_l2_mh_sock, UDP_TAP_FRAMES, 0, NULL);
if (n <= 0)
return;
udp6_l2_mh_tap[0].msg_hdr.msg_iov = &udp6_l2_iov_tap[0];
for (i = 0; i < (unsigned)n; i++) {
udp_sock_fill_data_v4(c, i, ref,
&msg_i, &msg_bufs, &msg_len, now);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
udp4_l2_mh_tap[msg_i].msg_hdr.msg_iovlen = msg_bufs;
tap_mmh = udp4_l2_mh_tap;
}
if (c->mode == MODE_PASTA)
return;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
ret = sendmmsg(c->fd_tap, tap_mmh, msg_i + 1,
MSG_NOSIGNAL | MSG_DONTWAIT);
if (ret <= 0)
return;
/* If we lose some messages to sendmmsg() here, fine, it's UDP. However,
* the last message needs to be delivered completely, otherwise qemu
* will fail to reassemble the next message and close the connection. Go
* through headers from the last sent message, counting bytes, and, if
* and as soon as we see more bytes than sendmmsg() sent, re-send the
* rest with a blocking call.
*
* In pictures, given this example:
*
* iov #0 iov #1 iov #2 iov #3
* tap_mmh[ret - 1].msg_hdr: .... ...... ..... ......
* tap_mmh[ret - 1].msg_len: 7 .... ...
*
* when 'msglen' reaches: 10 ^
* and 'missing' below is: 3 ---
*
* re-send everything from here: ^-- ----- ------
*/
last_mh = &tap_mmh[ret - 1].msg_hdr;
for (i = 0, msg_len = 0; i < last_mh->msg_iovlen; i++) {
if (missing <= 0) {
msg_len += last_mh->msg_iov[i].iov_len;
missing = msg_len - tap_mmh[ret - 1].msg_len;
}
if (missing > 0) {
uint8_t **iov_base;
int first_offset;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
iov_base = (uint8_t **)&last_mh->msg_iov[i].iov_base;
first_offset = last_mh->msg_iov[i].iov_len - missing;
*iov_base += first_offset;
last_mh->msg_iov[i].iov_len = missing;
last_mh->msg_iov = &last_mh->msg_iov[i];
if (sendmsg(c->fd_tap, last_mh, MSG_NOSIGNAL) < 0)
debug("UDP: %li bytes to tap missing", missing);
*iov_base -= first_offset;
break;
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
pcapmm(tap_mmh, ret);
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
/**
* udp_tap_handler() - Handle packets from tap
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
* @c: Execution context
* @af: Address family, AF_INET or AF_INET6
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* @addr: Destination address
* @p: Pool of UDP packets, with UDP headers
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @now: Current timestamp
*
* Return: count of consumed packets
*
* #syscalls sendmmsg
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*/
int udp_tap_handler(struct ctx *c, int af, const void *addr,
const struct pool *p, const struct timespec *now)
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
{
struct mmsghdr mm[UIO_MAXIOV];
struct iovec m[UIO_MAXIOV];
struct sockaddr_in6 s_in6;
struct sockaddr_in s_in;
struct sockaddr *sa;
int i, s, count = 0;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
in_port_t src, dst;
struct udphdr *uh;
socklen_t sl;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
(void)c;
uh = packet_get(p, 0, 0, sizeof(*uh), NULL);
if (!uh)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
return 1;
/* The caller already checks that all the messages have the same source
* and destination, so we can just take those from the first message.
*/
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
src = ntohs(uh->source);
dst = ntohs(uh->dest);
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
if (af == AF_INET) {
s_in = (struct sockaddr_in) {
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
.sin_family = AF_INET,
.sin_port = uh->dest,
.sin_addr = *(struct in_addr *)addr,
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
};
sa = (struct sockaddr *)&s_in;
sl = sizeof(s_in);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (!(s = udp_tap_map[V4][src].sock)) {
union udp_epoll_ref uref = { .udp.bound = 1,
.udp.port = src };
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
s = sock_l4(c, AF_INET, IPPROTO_UDP, NULL, src,
uref.u32);
if (s < 0)
return p->count;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_tap_map[V4][src].sock = s;
bitmap_set(udp_act[V4][UDP_ACT_TAP], src);
}
udp_tap_map[V4][src].ts = now->tv_sec;
if (s_in.sin_addr.s_addr == c->ip4.gw && !c->no_map_gw) {
if (!(udp_tap_map[V4][dst].flags & PORT_LOCAL) ||
(udp_tap_map[V4][dst].flags & PORT_LOOPBACK))
s_in.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
else
s_in.sin_addr.s_addr = c->ip4.addr_seen;
} else if (s_in.sin_addr.s_addr == c->ip4.dns_fwd &&
ntohs(s_in.sin_port) == 53) {
s_in.sin_addr.s_addr = c->ip4.dns[0];
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
} else {
s_in6 = (struct sockaddr_in6) {
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
.sin6_family = AF_INET6,
.sin6_port = uh->dest,
.sin6_addr = *(struct in6_addr *)addr,
};
const void *bind_addr = &in6addr_any;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
sa = (struct sockaddr *)&s_in6;
sl = sizeof(s_in6);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (IN6_ARE_ADDR_EQUAL(addr, &c->ip6.gw) && !c->no_map_gw) {
if (!(udp_tap_map[V6][dst].flags & PORT_LOCAL) ||
(udp_tap_map[V6][dst].flags & PORT_LOOPBACK))
s_in6.sin6_addr = in6addr_loopback;
else if (udp_tap_map[V6][dst].flags & PORT_GUA)
s_in6.sin6_addr = c->ip6.addr;
else
s_in6.sin6_addr = c->ip6.addr_seen;
} else if (IN6_ARE_ADDR_EQUAL(addr, &c->ip6.dns_fwd) &&
ntohs(s_in6.sin6_port) == 53) {
s_in6.sin6_addr = c->ip6.dns[0];
} else if (IN6_IS_ADDR_LINKLOCAL(&s_in6.sin6_addr)) {
bind_addr = &c->ip6.addr_ll;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (!(s = udp_tap_map[V6][src].sock)) {
union udp_epoll_ref uref = { .udp.bound = 1,
.udp.v6 = 1,
.udp.port = src };
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
s = sock_l4(c, AF_INET6, IPPROTO_UDP, bind_addr, src,
uref.u32);
if (s < 0)
return p->count;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_tap_map[V6][src].sock = s;
bitmap_set(udp_act[V6][UDP_ACT_TAP], src);
}
udp_tap_map[V6][src].ts = now->tv_sec;
}
for (i = 0; i < (int)p->count; i++) {
struct udphdr *uh_send;
size_t len;
uh_send = packet_get(p, i, 0, sizeof(*uh), &len);
if (!uh_send)
return p->count;
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
mm[i].msg_hdr.msg_name = sa;
mm[i].msg_hdr.msg_namelen = sl;
if (len) {
m[i].iov_base = (char *)(uh_send + 1);
m[i].iov_len = len;
mm[i].msg_hdr.msg_iov = m + i;
mm[i].msg_hdr.msg_iovlen = 1;
} else {
mm[i].msg_hdr.msg_iov = NULL;
mm[i].msg_hdr.msg_iovlen = 0;
}
mm[i].msg_hdr.msg_control = NULL;
mm[i].msg_hdr.msg_controllen = 0;
mm[i].msg_hdr.msg_flags = 0;
count++;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
count = sendmmsg(s, mm, count, MSG_NOSIGNAL);
if (count < 0)
return 1;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
return count;
}
/**
* udp_sock_init() - Initialise listening sockets for a given port
* @c: Execution context
* @ns: In pasta mode, if set, bind with loopback address in namespace
* @af: Address family to select a specific IP version, or AF_UNSPEC
* @addr: Pointer to address for binding, NULL if not configured
* @port: Port, host order
*/
void udp_sock_init(const struct ctx *c, int ns, sa_family_t af,
const void *addr, in_port_t port)
{
union udp_epoll_ref uref = { .udp.bound = 1 };
const void *bind_addr;
int s;
if (ns) {
uref.udp.port = (in_port_t)(port +
c->udp.fwd_out.f.delta[port]);
} else {
uref.udp.port = (in_port_t)(port +
c->udp.fwd_in.f.delta[port]);
}
if (af == AF_INET || af == AF_UNSPEC) {
if (!addr && c->mode == MODE_PASTA)
bind_addr = &c->ip4.addr;
else
bind_addr = addr;
uref.udp.v6 = 0;
if (!ns) {
uref.udp.splice = 0;
s = sock_l4(c, AF_INET, IPPROTO_UDP, bind_addr, port,
uref.u32);
udp_tap_map[V4][uref.udp.port].sock = s;
}
if (c->mode == MODE_PASTA) {
bind_addr = &(uint32_t){ htonl(INADDR_LOOPBACK) };
uref.udp.splice = UDP_TO_NS;
sock_l4(c, AF_INET, IPPROTO_UDP, bind_addr, port,
uref.u32);
}
if (ns) {
bind_addr = &(uint32_t){ htonl(INADDR_LOOPBACK) };
uref.udp.splice = UDP_TO_INIT;
sock_l4(c, AF_INET, IPPROTO_UDP, bind_addr, port,
uref.u32);
}
}
if (af == AF_INET6 || af == AF_UNSPEC) {
if (!addr && c->mode == MODE_PASTA)
bind_addr = &c->ip6.addr;
else
bind_addr = addr;
uref.udp.v6 = 1;
if (!ns) {
uref.udp.splice = 0;
s = sock_l4(c, AF_INET6, IPPROTO_UDP, bind_addr, port,
uref.u32);
udp_tap_map[V6][uref.udp.port].sock = s;
}
if (c->mode == MODE_PASTA) {
bind_addr = &in6addr_loopback;
uref.udp.splice = UDP_TO_NS;
sock_l4(c, AF_INET6, IPPROTO_UDP, bind_addr, port,
uref.u32);
}
if (ns) {
bind_addr = &in6addr_loopback;
uref.udp.splice = UDP_TO_INIT;
sock_l4(c, AF_INET6, IPPROTO_UDP, bind_addr, port,
uref.u32);
}
}
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
/**
* udp_sock_init_ns() - Bind sockets in namespace for inbound connections
* @arg: Execution context
*
* Return: 0
*/
int udp_sock_init_ns(void *arg)
{
struct ctx *c = (struct ctx *)arg;
unsigned dst;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
passt, pasta: Namespace-based sandboxing, defer seccomp policy application To reach (at least) a conceptually equivalent security level as implemented by --enable-sandbox in slirp4netns, we need to create a new mount namespace and pivot_root() into a new (empty) mountpoint, so that passt and pasta can't access any filesystem resource after initialisation. While at it, also detach IPC, PID (only for passt, to prevent vulnerabilities based on the knowledge of a target PID), and UTS namespaces. With this approach, if we apply the seccomp filters right after the configuration step, the number of allowed syscalls grows further. To prevent this, defer the application of seccomp policies after the initialisation phase, before the main loop, that's where we expect bad things to happen, potentially. This way, we get back to 22 allowed syscalls for passt and 34 for pasta, on x86_64. While at it, move #syscalls notes to specific code paths wherever it conceptually makes sense. We have to open all the file handles we'll ever need before sandboxing: - the packet capture file can only be opened once, drop instance numbers from the default path and use the (pre-sandbox) PID instead - /proc/net/tcp{,v6} and /proc/net/udp{,v6}, for automatic detection of bound ports in pasta mode, are now opened only once, before sandboxing, and their handles are stored in the execution context - the UNIX domain socket for passt is also bound only once, before sandboxing: to reject clients after the first one, instead of closing the listening socket, keep it open, accept and immediately discard new connection if we already have a valid one Clarify the (unchanged) behaviour for --netns-only in the man page. To actually make passt and pasta processes run in a separate PID namespace, we need to unshare(CLONE_NEWPID) before forking to background (if configured to do so). Introduce a small daemon() implementation, __daemon(), that additionally saves the PID file before forking. While running in foreground, the process itself can't move to a new PID namespace (a process can't change the notion of its own PID): mention that in the man page. For some reason, fork() in a detached PID namespace causes SIGTERM and SIGQUIT to be ignored, even if the handler is still reported as SIG_DFL: add a signal handler that just exits. We can now drop most of the pasta_child_handler() implementation, that took care of terminating all processes running in the same namespace, if pasta started a shell: the shell itself is now the init process in that namespace, and all children will terminate once the init process exits. Issuing 'echo $$' in a detached PID namespace won't return the actual namespace PID as seen from the init namespace: adapt demo and test setup scripts to reflect that. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2022-02-07 20:11:37 +00:00
if (ns_enter(c))
return 0;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
for (dst = 0; dst < NUM_PORTS; dst++) {
if (!bitmap_isset(c->udp.fwd_out.f.map, dst))
continue;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
udp_sock_init(c, 1, AF_UNSPEC, NULL, dst);
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
return 0;
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
/**
* udp_splice_iov_init() - Set up buffers and descriptors for recvmmsg/sendmmsg
*/
static void udp_splice_iov_init(void)
{
struct mmsghdr *h;
struct iovec *iov;
int i;
for (i = 0, h = udp_mmh_recv; i < UDP_SPLICE_FRAMES; i++, h++) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct msghdr *mh = &h->msg_hdr;
if (!i) {
mh->msg_name = &udp_splice_namebuf;
mh->msg_namelen = sizeof(udp_splice_namebuf);
}
mh->msg_iov = &udp_iov_recv[i];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
mh->msg_iovlen = 1;
}
for (i = 0, iov = udp_iov_recv; i < UDP_SPLICE_FRAMES; i++, iov++) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
iov->iov_base = udp_splice_buf[i];
iov->iov_len = sizeof(udp_splice_buf[i]);
}
for (i = 0, h = udp_mmh_send; i < UDP_SPLICE_FRAMES; i++, h++) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct msghdr *mh = &h->msg_hdr;
mh->msg_iov = &udp_iov_send[i];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
mh->msg_iovlen = 1;
}
for (i = 0, iov = udp_iov_send; i < UDP_SPLICE_FRAMES; i++, iov++)
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
iov->iov_base = udp_splice_buf[i];
for (i = 0, h = udp_mmh_sendto; i < UDP_SPLICE_FRAMES; i++, h++) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct msghdr *mh = &h->msg_hdr;
mh->msg_name = &udp_splice_namebuf;
mh->msg_namelen = sizeof(udp_splice_namebuf);
mh->msg_iov = &udp_iov_sendto[i];
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
mh->msg_iovlen = 1;
}
for (i = 0, iov = udp_iov_sendto; i < UDP_SPLICE_FRAMES; i++, iov++)
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
iov->iov_base = udp_splice_buf[i];
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
}
/**
* udp_init() - Initialise per-socket data, and sockets in namespace
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
* @c: Execution context
*
* Return: 0
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
*/
int udp_init(struct ctx *c)
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
{
if (c->ifi4)
udp_sock4_iov_init();
if (c->ifi6)
udp_sock6_iov_init();
udp_invert_portmap(&c->udp.fwd_in);
udp_invert_portmap(&c->udp.fwd_out);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
if (c->mode == MODE_PASTA) {
udp_splice_iov_init();
NS_CALL(udp_sock_init_ns, c);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
passt: New design and implementation with native Layer 4 sockets This is a reimplementation, partially building on the earlier draft, that uses L4 sockets (SOCK_DGRAM, SOCK_STREAM) instead of SOCK_RAW, providing L4-L2 translation functionality without requiring any security capability. Conceptually, this follows the design presented at: https://gitlab.com/abologna/kubevirt-and-kvm/-/blob/master/Networking.md The most significant novelty here comes from TCP and UDP translation layers. In particular, the TCP state and translation logic follows the intent of being minimalistic, without reimplementing a full TCP stack in either direction, and synchronising as much as possible the TCP dynamic and flows between guest and host kernel. Another important introduction concerns addressing, port translation and forwarding. The Layer 4 implementations now attempt to bind on all unbound ports, in order to forward connections in a transparent way. While at it: - the qemu 'tap' back-end can't be used as-is by qrap anymore, because of explicit checks now introduced in qemu to ensure that the corresponding file descriptor is actually a tap device. For this reason, qrap now operates on a 'socket' back-end type, accounting for and building the additional header reporting frame length - provide a demo script that sets up namespaces, addresses and routes, and starts the daemon. A virtual machine started in the network namespace, wrapped by qrap, will now directly interface with passt and communicate using Layer 4 sockets provided by the host kernel. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-02-16 06:25:09 +00:00
return 0;
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
/**
* udp_timer_one() - Handler for timed events on one port
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* @c: Execution context
* @v6: Set for IPv6 connections
* @type: Socket type
* @port: Port number, host order
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @ts: Timestamp from caller
*/
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
static void udp_timer_one(struct ctx *c, int v6, enum udp_act_type type,
in_port_t port, const struct timespec *ts)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
{
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
struct udp_splice_port *sp;
struct udp_tap_port *tp;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
int s = -1;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
switch (type) {
case UDP_ACT_TAP:
tp = &udp_tap_map[v6 ? V6 : V4][port];
if (ts->tv_sec - tp->ts > UDP_CONN_TIMEOUT) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
s = tp->sock;
tp->flags = 0;
}
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
break;
case UDP_ACT_INIT_CONN:
sp = &udp_splice_map[v6 ? V6 : V4][port];
if (ts->tv_sec - sp->init_conn_ts > UDP_CONN_TIMEOUT)
s = sp->init_conn_sock;
break;
case UDP_ACT_NS_CONN:
sp = &udp_splice_map[v6 ? V6 : V4][port];
if (ts->tv_sec - sp->ns_conn_ts > UDP_CONN_TIMEOUT)
s = sp->ns_conn_sock;
break;
default:
return;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
if (s > 0) {
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
epoll_ctl(c->epollfd, EPOLL_CTL_DEL, s, NULL);
close(s);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
bitmap_clear(udp_act[v6 ? V6 : V4][type], port);
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
}
/**
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
* udp_timer() - Scan activity bitmaps for ports with associated timed events
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
* @c: Execution context
* @ts: Timestamp from caller
*/
void udp_timer(struct ctx *c, const struct timespec *ts)
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
{
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
int n, t, v6 = 0;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
unsigned int i;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
long *word, tmp;
if (!c->ifi4)
v6 = 1;
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
v6:
for (t = 0; t < UDP_ACT_TYPE_MAX; t++) {
word = (long *)udp_act[v6 ? V6 : V4][t];
for (i = 0; i < ARRAY_SIZE(udp_act[0][0]);
i += sizeof(long), word++) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
tmp = *word;
while ((n = ffsl(tmp))) {
tmp &= ~(1UL << (n - 1));
udp_timer_one(c, v6, t, i * 8 + n - 1, ts);
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
}
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
}
if (!v6 && c->ifi6) {
passt: Add PASTA mode, major rework PASTA (Pack A Subtle Tap Abstraction) provides quasi-native host connectivity to an otherwise disconnected, unprivileged network and user namespace, similarly to slirp4netns. Given that the implementation is largely overlapping with PASST, no separate binary is built: 'pasta' (and 'passt4netns' for clarity) both link to 'passt', and the mode of operation is selected depending on how the binary is invoked. Usage example: $ unshare -rUn # echo $$ 1871759 $ ./pasta 1871759 # From another terminal # udhcpc -i pasta0 2>/dev/null # ping -c1 pasta.pizza PING pasta.pizza (64.190.62.111) 56(84) bytes of data. 64 bytes from 64.190.62.111 (64.190.62.111): icmp_seq=1 ttl=255 time=34.6 ms --- pasta.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 34.575/34.575/34.575/0.000 ms # ping -c1 spaghetti.pizza PING spaghetti.pizza(2606:4700:3034::6815:147a (2606:4700:3034::6815:147a)) 56 data bytes 64 bytes from 2606:4700:3034::6815:147a (2606:4700:3034::6815:147a): icmp_seq=1 ttl=255 time=29.0 ms --- spaghetti.pizza ping statistics --- 1 packets transmitted, 1 received, 0% packet loss, time 0ms rtt min/avg/max/mdev = 28.967/28.967/28.967/0.000 ms This entails a major rework, especially with regard to the storage of tracked connections and to the semantics of epoll(7) references. Indexing TCP and UDP bindings merely by socket proved to be inflexible and unsuitable to handle different connection flows: pasta also provides Layer-2 to Layer-2 socket mapping between init and a separate namespace for local connections, using a pair of splice() system calls for TCP, and a recvmmsg()/sendmmsg() pair for UDP local bindings. For instance, building on the previous example: # ip link set dev lo up # iperf3 -s $ iperf3 -c ::1 -Z -w 32M -l 1024k -P2 | tail -n4 [SUM] 0.00-10.00 sec 52.3 GBytes 44.9 Gbits/sec 283 sender [SUM] 0.00-10.43 sec 52.3 GBytes 43.1 Gbits/sec receiver iperf Done. epoll(7) references now include a generic part in order to demultiplex data to the relevant protocol handler, using 24 bits for the socket number, and an opaque portion reserved for usage by the single protocol handlers, in order to track sockets back to corresponding connections and bindings. A number of fixes pertaining to TCP state machine and congestion window handling are also included here. Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-07-17 06:34:53 +00:00
v6 = 1;
goto v6;
udp: Connection tracking for ephemeral, local ports, and related fixes As we support UDP forwarding for packets that are sent to local ports, we actually need some kind of connection tracking for UDP. While at it, this commit introduces a number of vaguely related fixes for issues observed while trying this out. In detail: - implement an explicit, albeit minimalistic, connection tracking for UDP, to allow usage of ephemeral ports by the guest and by the host at the same time, by binding them dynamically as needed, and to allow mapping address changes for packets with a loopback address as destination - set the guest MAC address whenever we receive a packet from tap instead of waiting for an ARP request, and set it to broadcast on start, otherwise DHCPv6 might not work if all DHCPv6 requests time out before the guest starts talking IPv4 - split context IPv6 address into address we assign, global or site address seen on tap, and link-local address seen on tap, and make sure we use the addresses we've seen as destination (link-local choice depends on source address). Similarly, for IPv4, split into address we assign and address we observe, and use the address we observe as destination - introduce a clock_gettime() syscall right after epoll_wait() wakes up, so that we can remove all the other ones and pass the current timestamp to tap and socket handlers -- this is additionally needed by UDP to time out bindings to ephemeral ports and mappings between loopback address and a local address - rename sock_l4_add() to sock_l4(), no semantic changes intended - include <arpa/inet.h> in passt.c before kernel headers so that we can use <netinet/in.h> macros to check IPv6 address types, and remove a duplicate <linux/ip.h> inclusion Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
2021-04-29 14:59:20 +00:00
}
}