The tests in pasta/tcp and pasta/udp for inbound transfers have the server
listening within the namespace explicitly bound to 127.0.0.1 or ::1. This
only works because of the behaviour of inbound splice connections, which
always appear with both source and destination addresses as loopback in
the namespace. That's not an inherent property for "spliced" connections
and arguably an undesirable one. Also update the test names to make it
clearer that these tests are expecting to exercise the "splice" path.
Interestingly this was already correct for the equivalent passt_in_ns/*,
although we also update the test names for clarity there.
Note that there are similar issues in some of the podman tests, addressed
in https://github.com/containers/podman/pull/24064
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This section didn't mention the effect of the --map-host-loopback option
which now alters this behaviour. Update it accordingly.
It used "local addresses" to mean specifically 127.0.0.0/8 and ::1.
However, "local" could also refer to link-local addresses or to addresses
of any scope which happen to be configured on the host. Use "loopback
address" to be more precise about this.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
The description of this option says that it's deprecated, but unlike
--no-copy-addrs and --no-copy-routes it doesn't have a clear label. Add
one to make it easier to spot.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
After running dhclient -6 we expect the DHCPv6 assigned address to be
immediately usable. That's true with the Fedora dhclient-script (and the
upstream ISC DHCP one), however it's not true with the Debian
dhclient-script. The Debian script can complete with the address still
in "tentative" state, and the address won't be usable until Duplicate
Address Detection (DAD) completes. That's arguably a bug in Debian (see
link below), but for the time being we need to work around it anyway.
We usually get away with this, because by the time we do anything where the
address matters, DAD has completed. However, it's not robust, so we should
explicitly wait for DAD to complete when we get an DHCPv6 address.
Link: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1085231
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Getting a SLAAC address takes a little while because the kernel must
complete Duplicate Address Detection (DAD) before marking the address as
ready. In several places we have an explicit 'sleep 2' to wait for that
to complete.
Fixed length delays are never a great idea, although this one is pretty
solid. Still, it would be better to explicitly wait for DAD to complete
in case of long delays (which might happen on slow emulated hosts, or with
heavy load), and to speed the tests up if DAD completes quicker.
Replace the fixed sleeps with a loop waiting for DAD to complete. We do
this by looping waiting for all tentative addresses to disappear.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This fixes a number of harmless but slightly ugly warts in the ARP
resolution code:
* Use in4addr_any to represent 0.0.0.0 rather than hand constructing an
example.
* When comparing am->sip against 0.0.0.0 use sizeof(am->sip) instead of
sizeof(am->tip) (same value, but makes more logical sense)
* Described the guest's assigned address as such, rather than as "our
address" - that's not usually what we mean by "our address" these days
* Remove "we might have the same IP address" comment which I can't make
sense of in context (possibly it's relating to the statement below,
which already has its own comment?)
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Starting from commit 9178a9e346 ("tcp: Always send an ACK segment
once the handshake is completed"), we always send an ACK segment,
without any payload, to complete the three-way handshake while
establishing a connection started from a socket.
We queue that segment after checking if we already have data to send
to the tap, which means that its sequence number is higher than any
segment with data we're sending in the same iteration, if any data is
available on the socket.
However, in tcp_defer_handler(), we first flush "flags" buffers, that
is, we send out segments without any data first, and then segments
with data, which means that our "empty" ACK is sent before the ACK
segment with data (if any), which has a lower sequence number.
This appears to be harmless as the guest or container will generally
reorder segments, but it looks rather weird and we can't exclude it's
actually causing problems.
Queue the empty ACK first, so that it gets a lower sequence number,
before checking for any data from the socket.
Reported-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Just like we do for PCAP, DEBUG and KERNEL. Otherwise, running tests
with TRACE=1 will not actually enable tracing output.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
...instead of echo: otherwise, bash won't handle escape sequences we
use to colour messages (and 'echo -e' is not specified by POSIX).
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
When redirecting DNS queries with the --dns-forward option, passt/pasta
needs a host side nameserver to redirect the queries to. This is
controlled by the c->ip[46].dns_host variables. This is set to the first
first nameserver listed in the host's /etc/resolv.conf, and there isn't
currently a way to override it from the command line.
Prior to 0b25cac9 ("conf: Treat --dns addresses as guest visible
addresses") it was possible to alter this with the -D/--dns option.
However, doing so was confusing and had some nonsensical edge cases because
-D generally takes guest side addresses, rather than host side addresses.
Add a new --dns-host option to restore this functionality in a more
sensible way.
Link: https://bugs.passt.top/show_bug.cgi?id=102
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
In a couple of recent reports, we've seen that it can be useful for pasta
to forward ports from addresses which are not currently configured on the
host, but might be in future. That can be done with the sysctl
net.ipv4.ip_nonlocal_bind, but that does require CAP_NET_ADMIN to set in
the first place. We can allow the same thing on a per-socket basis with
the IP_FREEBIND (or IPV6_FREEBIND) socket option.
Add a --freebind command line argument to enable this socket option on
all listening sockets.
Link: https://bugs.passt.top/show_bug.cgi?id=101
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
As for tcp_update_check_tcp4()/tcp_update_check_tcp6(),
change csum_udp4() and csum_udp6() to use an iovec array.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
TCP header and payload are supposed to be in the same buffer,
and tcp_update_check_tcp4()/tcp_update_check_tcp6() compute
the checksum from the base address of the header using the
length of the IP payload.
In the future (for vhost-user) we need to dispatch the TCP header and
the TCP payload through several buffers. To be able to manage that, we
provide an iovec array that points to the data of the TCP frame.
We provide also an offset to be able to provide an array that contains
the TCP frame embedded in an lower level frame, and this offset points
to the TCP header inside the iovec array.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
The offset allows any headers that are not part of the data
to checksum to be skipped.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
The offset is passed directly to pcap_frame() and allows
any headers that are not part of the frame to
capture to be skipped.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
As tcp_update_check_tcp4() and tcp_update_check_tcp6() compute the
checksum using the TCP header and the TCP payload, it is clearer
to use a pointer to tcp_payload_t that includes tcphdr and payload
rather than a pointer to tcphdr (and guessing TCP header is
followed by the payload).
Move tcp_payload_t and tcp_flags_t to tcp_internal.h.
(They will be used also by vhost-user).
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This is quite useful at least for myself as I'm usually running tests
using a guest kernel that's not the same as the one on the host.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
We already have an inany_ntop() function to format inany addresses into
text. Add inany_pton() to parse them from text, and use it in
conf_ports().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
tcp_sock_init() and udp_sock_init() take an address to bind to as an
address family and void * pair. Use an inany instead. Formerly AF_UNSPEC
was used to indicate that we want to listen on both 0.0.0.0 and ::, now use
a NULL inany to indicate that.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
The sock_l4() function is very convenient for creating sockets bound to
a given address, but its interface has some problems.
Most importantly, the address and port alone aren't enough in some cases.
For link-local addresses (at least) we also need the pif in order to
properly construct a socket adddress. This case doesn't yet arise, but
it might cause us trouble in future.
Additionally, sock_l4() can take AF_UNSPEC with the special meaning that it
should attempt to create a "dual stack" socket which will respond to both
IPv4 and IPv6 traffic. This only makes sense if there is no specific
address given. We verify this at runtime, but it would be nicer if we
could enforce it structurally.
For sockets associated specifically with a single flow we already replaced
sock_l4() with flowside_sock_l4() which avoids those problems. Now,
replace all the remaining users with a new pif_sock_l4() which also takes
an explicit pif.
The new function takes the address as an inany *, with NULL indicating the
dual stack case. This does add some complexity in some of the callers,
however future planned cleanups should make this go away again.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
To save some kernel memory we try to use "dual stack" sockets (that listen
to both IPv4 and IPv6 traffic) when possible. However udp_sock_init()
attempts to do this in some cases that can't work. Specifically we can
only do this when listening on any address. That's never true for the
ns (splicing) case, because we always listen on loopback. For the !ns
case and AF_UNSPEC case, addr should always be NULL, but add an assert to
verify.
This is harmless: if addr is non-NULL, sock_l4() will just fail and we'll
fall back to the other path. But, it's messy and makes some upcoming
changes harder, so avoid attempting this in cases we know can't work.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
We can need not to set TCP checksum. Add a parameter to
tcp_fill_headers4() and tcp_fill_headers6() to disable it.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
We can need not to set the UDP checksum. Add a parameter to
udp_update_hdr4() and udp_update_hdr6() to disable it.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
write_remainder() steps through the buffers in an IO vector writing out
everything past a certain byte offset. However, on each iteration it
rescans the buffer from the beginning to find out where we're up to. With
an unfortunate set of write sizes this could lead to quadratic behaviour.
In an even less likely set of circumstances (total vector length > maximum
size_t) the 'skip' variable could overflow. This is one factor in a
longstanding Coverity error we've seen (although I still can't figure out
the remainder of its complaint).
Rework write_remainder() to always work out our new position in the vector
relative to our old/current position, rather than starting from the
beginning each time. As a bonus this seems to fix the Coverity error.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
write(2) might not write all the data it is given. Add a write_all_buf()
helper to keep calling it until all the given data is written, or we get an
error.
Currently we use write_remainder() to do this operation in pcap_frame().
That's a little awkward since it requires constructing an iovec, and future
changes we want to make to write_remainder() will be easier in terms of
this single buffer helper.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This parameter is already treated as a boolean internally. Make it a
'bool' type for clarity.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This function has a block conditional on !snd_wnd_cap shortly before an
snd_wnd_cap is statically false).
Therefore, simplify this down to a single conditional with an else branch.
While we're there, fix some improperly indented closing braces.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
When available, we want to retrieve our socket peer's advertised window and
forward that to the guest. That information has been available from the
kernel via the TCP_INFO getsockopt() since kernel commit 8f7baad7f035.
Currently our probing for this is a bit odd. The HAS_SND_WND define
determines if our headers include the tcp_snd_wnd field, but that doesn't
necessarily mean the running kernel supports it. Currently we start by
assuming it's _not_ available, but mark it as available if we ever see
a non-zero value in the field. This is a bit hit and miss in two ways:
* Zero is perfectly possible window the peer could report, so we can
get false negatives
* We're reading TCP_INFO into a local variable, which might not be zero
initialised, so if the kernel _doesn't_ write it it could have non-zero
garbage, giving us false positives.
We can use a more direct way of probing for this: getsockopt() reports the
length of the information retreived. So, check whether that's long enough
to include the field. This lets us probe the availability of the field
once and for all during initialisation. That in turn allows ctx to become
a const pointer to tcp_prepare_flags() which cascades through many other
functions.
We also move the flag for the probe result from the ctx structure to a
global, to match peek_offset_cap.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
tcp_send_flag() and tcp_probe_peek_offset_cap() are not used outside tcp.c,
and have no prototype in a header. Make them static.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
When handling the DUP_ACK flag, we copy all the buffers making up the ack
frame. However, all our frames share the same buffer for the Ethernet
header (tcp4_eth_src or tcp6_eth_src), so copying the TCP_IOV_ETH will
result in a (perfectly) overlapping memcpy(). This seems to have been
harmless so far, but overlapping ranges to memcpy() is undefined behaviour,
so we really should avoid it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
This initialisation for IPv4 flags buffers is redundant with the very next
line which sets both iov_base and iov_len.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Since commit eedc81b6ef ("fwd, conf: Probe host's ephemeral ports"),
we might need to read from /proc/sys/net/ipv4/ip_local_port_range in
both passt and pasta.
While pasta was already allowed to open and write /proc/sys/net
entries, read access was missing in SELinux's type enforcement: add
that.
In passt, instead, this is the first time we need to access an entry
there: add everything we need.
Fixes: eedc81b6ef ("fwd, conf: Probe host's ephemeral ports")
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
tap_pasta_input() keeps reading frames from the tap device until the
buffer is full. However, this has an ugly edge case, when we get close
to buffer full, we will provide just the remaining space as a read()
buffer. If this is shorter than the next frame to read, the tap device
will truncate the frame and discard the remainder.
Adjust the code to make sure we always have room for a maximum size frame.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
tap_pasta_input() has a rather confusing structure, using two gotos.
Remove these by restructuring the function to have the main loop condition
based on filling our buffer space, with errors or running out of data
treated as the exception, rather than the other way around. This allows
us to handle the EINTR which triggered the 'restart' goto with a continue.
The outer 'redo' was triggered if we completely filled our buffer, to flush
it and do another pass. This one is unnecessary since we don't (yet) use
EPOLLET on the tap device: if there's still more data we'll get another
event and re-enter the loop.
Along the way handle a couple of extra edge cases:
- Check for EWOULDBLOCK as well as EAGAIN for the benefit of any future
ports where those might not have the same value
- Detect EOF on the tap device and exit in that case
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
When tap_passt_input() gets an error from recv() it (correctly) does not
print any error message for EINTR, EAGAIN or EWOULDBLOCK. However in all
three cases it returns from the function. That makes sense for EAGAIN and
EWOULDBLOCK, since we then want to wait for the next EPOLLIN event before
trying again. For EINTR, however, it makes more sense to retry immediately
- as it stands we're likely to get a renewer EPOLLIN event immediately in
that case, since we're using level triggered signalling.
So, handle EINTR separately by immediately retrying until we succeed or
get a different type of error.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Currently, tap_handler_pas{st,ta}() check for EPOLLRDHUP, EPOLLHUP and
EPOLLERR events, then assume anything left is EPOLLIN. We have some future
cases that may want to also handle EPOLLOUT, so in preparation explicitly
handle EPOLLIN, moving the logic to a subfunction.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
If the nanoseconds of the minuend timestamp are less than the
nanoseconds of the subtrahend timestamp, we need to carry one second
in the subtraction.
I subtracted this second from the minuend, but didn't actually carry
it in the subtraction of nanoseconds, and logged timestamps would jump
back whenever we switched to the first branch of timespec_diff_us()
from the second one.
Most likely, the reason why I didn't carry the second is that I
instinctively thought that swapping the operands would have the same
effect. But it doesn't, in general: that only happens with arithmetic
in modulo powers of 2. Undo the swap as well.
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
cppcheck-2.15.0 has apparently broadened when it throws a warning about
redundant initialization to include some cases where we have an initializer
for some fields, but then set other fields in the function body.
This is arguably a false positive: although we are technically overwriting
the zero-initialization the compiler supplies for fields not explicitly
initialized, this sort of construct makes sense when there are some fields
we know at the top of the function where the initializer is, but others
that require more complex calculation.
That said, in the two places this shows up, it's pretty easy to work
around. The results are arguably slightly clearer than what we had, since
they move the parts of the initialization closer together.
So do that rather than having ugly suppressions or dealing with the
tedious process of reporting a cppcheck false positive.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Currently, for not established connections, we monitor sockets with
edge-triggered events (EPOLLET) if we are in the TAP_SYN_RCVD state
(outbound connection being established) but not in the
TAP_SYN_ACK_SENT case of it (socket is connected, and we sent SYN,ACK
to the container/guest).
While debugging https://bugs.passt.top/show_bug.cgi?id=94, I spotted
another possibility for a short EPOLLRDHUP storm (10 seconds), which
doesn't seem to happen in actual use cases, but I could reproduce it:
start a connection from a container, while dropping (using netfilter)
ACK segments coming out of the container itself.
On the server side, outside the container, accept the connection and
shutdown the writing side of it immediately.
At this point, we're in the TAP_SYN_ACK_SENT case (not just a mere
TAP_SYN_RCVD state), we get EPOLLRDHUP from the socket, but we don't
have any reasonable way to handle it other than waiting for the tap
side to complete the three-way handshake. So we'll just keep getting
this EPOLLRDHUP until the SYN_TIMEOUT kicks in.
Always enable EPOLLET when EPOLLRDHUP is the only epoll event we
subscribe to: in this case, getting multiple EPOLLRDHUP reports is
totally useless.
In the only remaining non-established state, SOCK_ACCEPTED, for
inbound connections, we're anyway discarding EPOLLRDHUP events until
we established the conection, because we don't know what to do with
them until we get an answer from the tap side, so it's safe to enable
EPOLLET also in that case.
Link: https://bugs.passt.top/show_bug.cgi?id=94
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
udp_sock_errs() reads out everything in the socket error queue. However
we've seen some cases[0] where an EPOLLERR event is active, but there isn't
anything in the queue.
One possibility is that the error is reported instead by the SO_ERROR
sockopt. Check for that case and report it as best we can. If we still
get an EPOLLERR without visible error, we have no way to clear the error
state, so treat it as an unrecoverable error.
[0] https://github.com/containers/podman/issues/23686#issuecomment-2324945010
Link: https://bugs.passt.top/show_bug.cgi?id=95
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
We can get network errors, usually transient, reported via the socket error
queue. However, at least theoretically, we could get errors trying to
read the queue itself. Since we have no idea how to clear an error
condition in that case, treat it as unrecoverable.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Currently udp_sock_recv() both attempts to clear socket errors and read
a batch of datagrams for forwarding. That made sense initially, since
both listening and reply sockets need to do this. However, we have certain
error cases which will add additional complexity to the error processing.
Furthermore, if we ever wanted to more thoroughly handle errors received
here - e.g. by synthesising ICMP messages on the tap device - it will
likely require different handling for the listening and reply socket cases.
So, split handling of error events into its own udp_sock_errs() function.
While we're there, allow it to report "unrecoverable errors". We don't
have any of these so far, but some cases we're working on might require it.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
The details of a flow - endpoints, interfaces etc. - can be pretty
important for debugging. We log this on flow state transitions, but it can
also be useful to log this when we report specific conditions. Add some
helper functions and macros to make it easy to do that.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Unlike TCP, UDP has no in-band signalling for the end of a flow. So the
only way we remove flows is on a timer if they have no activity for 180s.
However, we've started to investigate some error conditions in which we
want to prematurely abort / abandon a UDP flow. We can call
udp_flow_close(), which will make the flow inert (sockets closed, no epoll
events, can't be looked up in hash). However it will still wait 3 minutes
to clear away the stale entry.
Clean this up by adding an explicit 'closed' flag which will cause a flow
to be more promptly cleaned up. We also publish udp_flow_close() so it
can be called from other places to abort UDP flows().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Our flow hash table uses linear probing in which we step backwards through
clusters of adjacent hash entries when we have near collisions. Usually
that's implemented by flow_hash_probe(). However, due to some details we
need a second implementation in flowside_lookup(). An embarrassing
oversight in rebasing from earlier versions has mean that version is
incorrect, trying to step forward through clusters rather than backward.
In situations with the right sorts of has near-collisions this can lead to
us not associating an ACK from the tap device with the right flow, leaving
it in a not-quite-established state. If the remote peer does a shutdown()
at the right time, this can lead to a storm of EPOLLRDHUP events causing
high CPU load.
Fixes: acca4235c4 ("flow, tcp: Generalise TCP hash table to general flow hash table")
Link: https://bugs.passt.top/show_bug.cgi?id=94
Suggested-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
In fecb1b65b1 ("log: Don't prefix message with timestamp on --debug
if it's a continuation"), I fixed this for --debug on standard error,
but not for log files: if messages are continuations, they shouldn't
be prefixed by timestamp and severity.
Otherwise, we'll print stuff like this:
0.0028: ERROR: Receive error on guest connection, reset0.0028: ERROR: : Bad file descriptor
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
On some systems source fortification is enabled whenever code
optimization is enabled (e.g. with -O2). Since code fortification
is explicitly enabled too (with possibly different value than the
system wants, there are three levels [1]), distros are required
to patch our Makefile, e.g. [2].
Detect whether fortification is not already enabled and enable it
explicitly only if really needed.
1: https://www.gnu.org/software/libc/manual/html_node/Source-Fortification.html
2: edfeb8763a
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
When we forward "all" ports (-t all or -u all), or use an exclude-only
range, we don't actually forward *all* ports - that wouln't leave local
ports to use for outgoing connections. Rather we forward all non-ephemeral
ports - those that won't be used for outgoing connections or datagrams.
Currently we assume the range of ephemeral ports is that recommended by
RFC 6335, 49152-65535. However, that's not the range used by default on
Linux, 32768-60999 but configurable with the net.ipv4.ip_local_port_range
sysctl.
We can't really know what range the guest will consider ephemeral, but if
it differs too much from the host it's likely to cause problems we can't
avoid anyway. So, using the host's ephemeral range is a better guess than
using the RFC 6335 range.
Therefore, add logic to probe the host's ephemeral range, falling back to
the RFC 6335 range if that fails. This has the bonus advantage of
reducing the number of ports bound by -t all -u all on most Linux machines
thereby reducing kernel memory usage. Specifically this reduces kernel
memory usage with -t all -u all from ~380MiB to ~289MiB.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
When using -t all, -u all or exclude-only ranges, we'll attempt to forward
all non-ephemeral port numbers, including port 0. However, this won't work
as intended: bind() treats a zero port not as literal port 0, but as
"pick a port for me". Because of the special meaning of port 0, we mostly
outright exclude it in our handling.
Do the same for setting up forwards, not attempting to forward for port 0.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
