Given an LXC guest with a root filesystem path of
/export/lxc/roots/helloworld/root
During startup, we will pivot the root filesystem to end up
at
/.oldroot/export/lxc/roots/helloworld/root
We then try to open
/.oldroot/export/lxc/roots/helloworld/root/dev/pts
Now consider if '/export/lxc' is an absolute symlink pointing
to '/media/lxc'. The kernel will try to open
/media/lxc/roots/helloworld/root/dev/pts
whereas it should be trying to open
/.oldroot//media/lxc/roots/helloworld/root/dev/pts
To deal with the fact that the root filesystem can be moved,
we need to resolve symlinks in *any* part of the filesystem
source path.
* src/libvirt_private.syms, src/util/util.c,
src/util/util.h: Add virFileResolveAllLinks to resolve
all symlinks in a path
* src/lxc/lxc_container.c: Resolve all symlinks in filesystem
paths during startup
Currently the LXC controller attempts to deal with EOF on a
tty by spawning a thread to do an edge triggered epoll_wait().
This avoids the normal event loop spinning on POLLHUP. There
is a subtle mistake though - even after seeing POLLHUP on a
master PTY, it is still perfectly possible & valid to write
data to the PTY. There is a buffer that can be filled with
data, even when no client is present.
The second mistake is that the epoll_wait() thread was not
looking for the EPOLLOUT condition, so when a new client
connects to the LXC console, it had to explicitly send a
character before any queued output would appear.
Finally, there was in fact no need to spawn a new thread to
deal with epoll_wait(). The epoll file descriptor itself
can be poll()'d on normally.
This patch attempts to deal with all these problems.
- The blocking epoll_wait() thread is replaced by a poll
on the epoll file descriptor which then does a non-blocking
epoll_wait() to handle events
- Even if POLLHUP is seen, we continue trying to write
any pending output until getting EAGAIN from write.
- Once write returns EAGAIN, we modify the epoll event
mask to also look for EPOLLOUT
* src/lxc/lxc_controller.c: Avoid stalled I/O upon
connected to an LXC console
The lifetime of the virDomainEventState object is tied to
the lifetime of the driver, which in stateless drivers is
tied to the lifetime of the virConnectPtr.
If we add & remove a timer when allocating/freeing the
virDomainEventState object, we can get a situation where
the timer still triggers once after virDomainEventState
has been freed. The timeout callback can't keep a ref
on the event state though, since that would be a circular
reference.
The trick is to only register the timer when a callback
is registered with the event state & remove the timer
when the callback is unregistered.
The demo for the bug is to run
while true ; do date ; ../tools/virsh -q -c test:///default 'shutdown test; undefine test; dominfo test' ; done
prior to this fix, it will frequently hang and / or
crash, or corrupt memory
Currently all drivers using domain events need to provide a callback
for handling a timer to dispatch events in a clean stack. There is
no technical reason for dispatch to go via driver specific code. It
could trivially be dispatched directly from the domain event code,
thus removing tedious boilerplate code from all drivers
Also fix the libxl & xen drivers to pass 'true' when creating the
virDomainEventState, since they run inside the daemon & thus always
expect events to be present.
* src/conf/domain_event.c, src/conf/domain_event.h: Internalize
dispatch of events from timer callback
* src/libxl/libxl_driver.c, src/lxc/lxc_driver.c,
src/qemu/qemu_domain.c, src/qemu/qemu_driver.c,
src/remote/remote_driver.c, src/test/test_driver.c,
src/uml/uml_driver.c, src/vbox/vbox_tmpl.c,
src/xen/xen_driver.c: Remove all timer dispatch functions
When registering a callback for a particular event some callers
need to know how many callbacks already exist for that event.
While it is possible to ask for a count, this is not free from
race conditions when threaded. Thus the API for registering
callbacks should return the count of callbacks. Also rename
virDomainEventStateDeregisterAny to virDomainEventStateDeregisterID
* src/conf/domain_event.c, src/conf/domain_event.h,
src/libvirt_private.syms: Return count of callbacks when
registering callbacks
* src/libxl/libxl_driver.c, src/libxl/libxl_driver.c,
src/qemu/qemu_driver.c, src/remote/remote_driver.c,
src/remote/remote_driver.c, src/uml/uml_driver.c,
src/vbox/vbox_tmpl.c, src/xen/xen_driver.c: Update
for change in APIs
A preparatory patch for DHCP snooping where we want to be able to
differentiate between a VM's interface using the tuple of
<VM UUID, Interface MAC address>. We assume that MAC addresses could
possibly be re-used between different networks (VLANs) thus do not only
want to rely on the MAC address to identify an interface.
At the current 'final destination' in virNWFilterInstantiate I am leaving
the vmuuid parameter as ATTRIBUTE_UNUSED until the DHCP snooping patches arrive.
(we may not post the DHCP snooping patches for 0.9.9, though)
Mostly this is a pretty trivial patch. On the lowest layers, in lxc_driver
and uml_conf, I am passing the virDomainDefPtr around until I am passing
only the VM's uuid into the NWFilter calls.
The virTimestamp and virTimeMs functions in src/util/util.h
duplicate functionality from virtime.h, in a non-async signal
safe manner. Remove them, and convert all code over to the new
APIs.
* src/util/util.c, src/util/util.h: Delete virTimeMs and virTimestamp
* src/lxc/lxc_driver.c, src/qemu/qemu_domain.c,
src/qemu/qemu_driver.c, src/qemu/qemu_migration.c,
src/qemu/qemu_process.c, src/util/event_poll.c: Convert to use
virtime APIs
After the previous patch, there are now some redundant checks.
* src/qemu/qemu_driver.c (qemudDomainGetVcpuPinInfo)
(qemuGetSchedulerParametersFlags): Drop checks now guaranteed by
libvirt.c.
* src/lxc/lxc_driver.c (lxcGetSchedulerParametersFlags):
Likewise.
Add the core functions that implement the functionality of the API.
Suspend is done by using an asynchronous mechanism so that we can return
the status to the caller before the host gets suspended. This asynchronous
operation is achieved by suspending the host in a separate thread of
execution. However, returning the status to the caller is only best-effort,
but not guaranteed.
To resume the host, an RTC alarm is set up (based on how long we want to
suspend) before suspending the host. When this alarm fires, the host
gets woken up.
Suspend-to-RAM operation on a host running Linux can take upto more than 20
seconds, depending on the load of the system. (Freezing of tasks, an operation
preceding any suspend operation, is given up after a 20 second timeout).
And Suspend-to-Disk can take even more time, considering the time required
for compaction, creating the memory image and writing it to disk etc.
So, we do not allow the user to specify a suspend duration of less than 60
seconds, to be on the safer side, since we don't want to prematurely declare
failure when we only had to wait for some more time.
To make lxcSetContainerResources smaller, pull the mem tune
and device ACL setup code out into separate methods
* src/lxc/lxc_controller.c: Introduce lxcSetContainerMemTune
and lxcSetContainerDeviceACL
While LXC does not have the concept of VCPUS, so we can't do
per-VCPU pCPU placement, we can support the VM level CPU
placement. Todo this simply set the CPU affinity of the LXC
controller at startup. All child processes will inherit this
affinity.
* src/lxc/lxc_controller.c: Set process affinity
Support creation of macvlan devices for LXC containers. Do not
allow setting of bandwidth controls or vport profiles due to the
complication that there is no host side visible device to work
with.
* src/lxc/lxc_driver.c: Support type=direct interfaces
The current lxcSetupInterfaces() method directly performs setup
of the bridge devices. Since it will shortly need to also create
macvlan devices, move the bridge related code into a separate
method
* src/lxc/lxc_driver.c: Split lxcSetupInterfaces() to create a
new lxcSetupInterfaceBridge()
Move the virNetDevSetName and virNetDevSetNamespace APIs out
of LXC's veth.c and into virnetdev.c.
Move the remaining content of the file to src/util/virnetdevveth.c
* src/lxc/veth.c: Rename to src/util/virnetdevveth.c
* src/lxc/veth.h: Rename to src/util/virnetdevveth.h
* src/util/virnetdev.c, src/util/virnetdev.h: Add
virNetDevSetName and virNetDevSetNamespace
* src/lxc/lxc_container.c, src/lxc/lxc_controller.c,
src/lxc/lxc_driver.c: Update include paths
The src/lxc/veth.c file contains APIs for managing veth devices,
but some of the APIs duplicate stuff from src/util/virnetdev.h.
Delete thed duplicate APIs and rename the remaining ones to
follow virNetDevVethXXXX
* src/lxc/veth.c, src/lxc/veth.h: Rename APIs & delete duplicates
* src/lxc/lxc_container.c, src/lxc/lxc_controller.c,
src/lxc/lxc_driver.c: Update for API renaming
Following the renaming of the bridge management APIs, we can now
split the source file into 3 corresponding pieces
* src/util/virnetdev.c: APIs for any type of network interface
* src/util/virnetdevbridge.c: APIs for bridge interfaces
* src/util/virnetdevtap.c: APIs for TAP interfaces
* src/util/virnetdev.c, src/util/virnetdev.h,
src/util/virnetdevbridge.c, src/util/virnetdevbridge.h,
src/util/virnetdevtap.c, src/util/virnetdevtap.h: Copied
from bridge.{c,h}
* src/util/bridge.c, src/util/bridge.h: Split into 3 pieces
* src/lxc/lxc_driver.c, src/network/bridge_driver.c,
src/openvz/openvz_driver.c, src/qemu/qemu_command.c,
src/qemu/qemu_conf.h, src/uml/uml_conf.c, src/uml/uml_conf.h,
src/uml/uml_driver.c: Update #include directives
The existing brXXX APIs in src/util/bridge.h are renamed to
follow one of three different conventions
- virNetDevXXX - operations for any type of interface
- virNetDevBridgeXXX - operations for bridge interfaces
- virNetDevTapXXX - operations for tap interfaces
* src/util/bridge.h, src/util/bridge.c: Rename all APIs
* src/lxc/lxc_driver.c, src/network/bridge_driver.c,
src/qemu/qemu_command.c, src/uml/uml_conf.c,
src/uml/uml_driver.c: Update for API renaming
Currently every caller of the brXXX APIs has to store the returned
errno value and then raise an error message. This results in
inconsistent error messages across drivers, additional burden on
the callers and makes the error reporting inaccurate since it is
hard to distinguish different scenarios from 1 errno value.
* src/util/bridge.c: Raise errors instead of returning errnos
* src/lxc/lxc_driver.c, src/network/bridge_driver.c,
src/qemu/qemu_command.c, src/uml/uml_conf.c,
src/uml/uml_driver.c: Remove error reporting code
The bridge management APIs in src/util/bridge.c require a brControl
object to be passed around. This holds the file descriptor for the
control socket. This extra object complicates use of the API for
only a minor efficiency gain, which is in turn entirely offset by
the need to fork/exec the brctl command for STP configuration.
This patch removes the 'brControl' object entirely, instead opening
the control socket & closing it again within the scope of each method.
The parameter names for the APIs are also made to consistently use
'brname' for bridge device name, and 'ifname' for an interface
device name. Finally annotations are added for non-NULL parameters
and return check validation
* src/util/bridge.c, src/util/bridge.h: Remove brControl object
and update API parameter names & annotations.
* src/lxc/lxc_driver.c, src/network/bridge_driver.c,
src/uml/uml_conf.h, src/uml/uml_conf.c, src/uml/uml_driver.c,
src/qemu/qemu_command.c, src/qemu/qemu_conf.h,
src/qemu/qemu_driver.c: Remove reference to 'brControl' object
I got this weird failure:
error: Failed to start domain simple
error: internal error cannot mix caller fds with blocking execution
and tracked it down to a use-after-free - virCommandSetOutputFD
was storing the address of a stack-local variable, which then
went out of scope before the virCommandRun that dereferenced it.
Bug introduced in commit 451cfd05 (0.9.2).
* src/lxc/lxc_driver.c (lxcBuildControllerCmd): Move log fd
registration...
(lxcVmStart): ...to caller.
Based on a Coverity report - the return value of waitpid() should
always be checked, to avoid problems with leaking resources.
* src/lxc/lxc_controller.c (lxcControllerRun): Use simpler virPidAbort.
The default console type may vary based on the OS type. ie a Xen
paravirt guests wants a 'xen' console, while a fullvirt guests
wants a 'serial' console.
A plain integer default console type in the capabilities does
not suffice. Instead introduce a callback that is passed the
OS type.
* src/conf/capabilities.h: Use a callback for default console
type
* src/conf/domain_conf.c, src/conf/domain_conf.h: Use callback
for default console type. Add missing LXC/OpenVZ console types.
* src/esx/esx_driver.c, src/libxl/libxl_conf.c,
src/lxc/lxc_conf.c, src/openvz/openvz_conf.c,
src/phyp/phyp_driver.c, src/qemu/qemu_capabilities.c,
src/uml/uml_conf.c, src/vbox/vbox_tmpl.c,
src/vmware/vmware_conf.c, src/xen/xen_hypervisor.c,
src/xenapi/xenapi_driver.c: Set default console type callback
To allow virDomainOpenConsole to access non-primary consoles,
device aliases are required to be set. Until now only the QEMU
driver has done this. Update LXC & UML to set aliases for any
console devices
* src/lxc/lxc_driver.c, src/uml/uml_driver.c: Set aliases
for console devices
Currently the LXC controller only supports setup of a single
text console. This is wired up to the container init's stdio,
as well as /dev/console and /dev/tty1. Extending support for
multiple consoles, means wiring up additional PTYs to /dev/tty2,
/dev/tty3, etc, etc. The LXC controller is passed multiple open
file handles, one for each console requested.
* src/lxc/lxc_container.c, src/lxc/lxc_container.h: Wire up
all the /dev/ttyN links required to symlink to /dev/pts/NN
* src/lxc/lxc_container.h: Open more container side /dev/pts/NN
devices, and adapt event loop to handle I/O from all consoles
* src/lxc/lxc_driver.c: Setup multiple host side PTYs
The current I/O code for LXC uses a hand crafted event loop
to forward I/O between the container & host app, based on
epoll to handle EOF on PTYs. This event loop is not easily
extensible to add more consoles, or monitor other types of
file descriptors.
Remove the custom event loop and replace it with a normal
libvirt event loop. When detecting EOF on a PTY, disable
the event watch on that FD, and fork off a background thread
that does a edge-triggered epoll() on the FD. When the FD
finally shows new incoming data, the thread re-enables the
watch on the FD and exits.
When getting EOF from a read() on the PTY, the existing code
would do waitpid(WNOHANG) to see if the container had exited.
Unfortunately there is a race condition, because even though
the process has closed its stdio handles, it might still
exist.
To deal with this the new event loop uses a SIG_CHILD handler
to perform the waitpid only when the container is known to
have actually exited.
* src/lxc/lxc_controller.c: Rewrite the event loop to use
the standard APIs.
While Xen only has a single paravirt console, UML, and
QEMU both support multiple paravirt consoles. The LXC
driver can also be trivially made to support multiple
consoles. This patch extends the XML to allow multiple
<console> elements in the XML. It also makes the UML
and QEMU drivers support this config.
* src/conf/domain_conf.c, src/conf/domain_conf.h: Allow
multiple <console> devices
* src/lxc/lxc_driver.c, src/xen/xen_driver.c,
src/xenxs/xen_sxpr.c, src/xenxs/xen_xm.c: Update for
internal API changes
* src/security/security_selinux.c, src/security/virt-aa-helper.c:
Only label consoles that aren't a copy of the serial device
* src/qemu/qemu_command.c, src/qemu/qemu_driver.c,
src/qemu/qemu_process.c, src/uml/uml_conf.c,
src/uml/uml_driver.c: Support multiple console devices
* tests/qemuxml2xmltest.c, tests/qemuxml2argvtest.c: Extra
tests for multiple virtio consoles. Set QEMU_CAPS_CHARDEV
for all console /channel tests
* tests/qemuxml2argvdata/qemuxml2argv-channel-virtio-auto.args,
tests/qemuxml2argvdata/qemuxml2argv-channel-virtio.args
tests/qemuxml2argvdata/qemuxml2argv-console-virtio.args: Update
for correct chardev syntax
* tests/qemuxml2argvdata/qemuxml2argv-console-virtio-many.args,
tests/qemuxml2argvdata/qemuxml2argv-console-virtio-many.xml: New
test file
Allow the user to call with nparams too small, per API documentation.
Also, libvirt.c filters out nparams of 0 for scheduler parameters.
* src/lxc/lxc_driver.c (lxcDomainGetMemoryParameters): Allow fewer
than max.
(lxcGetSchedulerParametersFlags): Drop redundant check.
Document the parameter names that will be used by
virDomain{Get,Set}SchedulerParameters{,Flags}, rather than
hard-coding those names in each driver, to match what is
done with memory, blkio, and blockstats parameters.
* include/libvirt/libvirt.h.in (VIR_DOMAIN_SCHEDULER_CPU_SHARES)
(VIR_DOMAIN_SCHEDULER_VCPU_PERIOD)
(VIR_DOMAIN_SCHEDULER_VCPU_QUOTA, VIR_DOMAIN_SCHEDULER_WEIGHT)
(VIR_DOMAIN_SCHEDULER_CAP, VIR_DOMAIN_SCHEDULER_RESERVATION)
(VIR_DOMAIN_SCHEDULER_LIMIT, VIR_DOMAIN_SCHEDULER_SHARES): New
field name macros.
* src/qemu/qemu_driver.c (qemuSetSchedulerParametersFlags)
(qemuGetSchedulerParametersFlags): Use new defines.
* src/test/test_driver.c (testDomainGetSchedulerParamsFlags)
(testDomainSetSchedulerParamsFlags): Likewise.
* src/xen/xen_hypervisor.c (xenHypervisorGetSchedulerParameters)
(xenHypervisorSetSchedulerParameters): Likewise.
* src/xen/xend_internal.c (xenDaemonGetSchedulerParameters)
(xenDaemonSetSchedulerParameters): Likewise.
* src/lxc/lxc_driver.c (lxcSetSchedulerParametersFlags)
(lxcGetSchedulerParametersFlags): Likewise.
* src/esx/esx_driver.c (esxDomainGetSchedulerParametersFlags)
(esxDomainSetSchedulerParametersFlags): Likewise.
* src/libxl/libxl_driver.c (libxlDomainGetSchedulerParametersFlags)
(libxlDomainSetSchedulerParametersFlags): Likewise.
If an LXC VM fails to start, quite a few cleanup paths will
result in the original error message being overwritten. Some
other cleanup paths also forgot to actually terminate the VM.
* src/lxc/lxc_driver.c: Ensure VM is terminated on startup
failure and preserve original error
The LXC code for mounting container filesystems from block devices
tries all filesystems in /etc/filesystems and possibly those in
/proc/filesystems. The regular mount binary, however, first tries
using libblkid to detect the format. Add support for doing the same
in libvirt, since Fedora's /etc/filesystems is missing many formats,
most notably ext4 which is the default filesystem Fedora uses!
* src/Makefile.am: Link libvirt_lxc to libblkid
* src/lxc/lxc_container.c: Probe filesystem format with libblkid
If we looped through /etc/filesystems trying to mount with each
type and failed all options, we forget to actually raise an
error message.
* src/lxc/lxc_container.c: Raise error if unable to detect
the filesystems. Also fix existing error message
The kernel automounter is mostly broken wrt to containers. Most
notably if you start a new filesystem namespace and then attempt
to unmount any autofs filesystem, it will typically fail with a
weird error message like
Failed to unmount '/.oldroot/sys/kernel/security':Too many levels of symbolic links
Attempting to detach the autofs mount using umount2(MNT_DETACH)
will also fail with the same error. Therefore if we get any error on
unmount()ing a filesystem from the old root FS when starting a
container, we must immediately break out and detach the entire
old root filesystem (ignoring any mounts below it).
This has the effect of making the old root filesystem inaccessible
to anything inside the container, but at the cost that the mounts
live on in the kernel until the container exits. Given that SystemD
uses autofs by default, we need LXC to be robust this scenario and
thus this tradeoff is worthwhile.
* src/lxc/lxc_container.c: Detach root filesystem if any umount
operation fails.
The /etc/filesystems file can contain a '*' on the last line to
indicate that /proc/filessystems should be tried next. We have
a check that this '*' only occurs on the last line. Unfortunately
when we then start reading /proc/filesystems, we mistakenly think
we've seen '*' in /proc/filesystems and fail
* src/lxc/lxc_container.c: Skip '*' validation when we're reading
/proc/filesystems
Only some of the return paths of lxcContainerWaitForContinue will
have set errno. In other paths we need to set it manually to avoid
the caller getting a random stale errno value
* src/lxc/lxc_container.c: Set errno in lxcContainerWaitForContinue
Previous commit clears number of items alocated in lxcSetupLoopDevices
if VIR_REALLOC_N fails. In that case, the pointer is not NULL, and
causes leaking FDs that have been allocated.
* src/lxc/lxc_controller.c: revert zeroing array size
If the function lxcSetupLoopDevices(def, &nloopDevs, &loopDevs) failed,
the variable loopDevs will keep a initial NULL value, however, the
function VIR_FORCE_CLOSE(loopDevs[i]) will directly deref it.
This patch also fixes returning a bogous number of devices from
lxcSetupLoopDevices on an error path.
* rc/lxc/lxc_controller.c: fixed a null pointer dereference.
Signed-off-by: Alex Jia <ajia@redhat.com>
Cppcheck detected a syntaxError on lxcDomainInterfaceStats.
* src/lxc/lxc_driver.c: fixed missing '{' in the function lxcDomainInterfaceStats.
Signed-off-by: Alex Jia <ajia@redhat.com>
Based on a report by Coverity. waitpid() can leak resources if it
fails with EINTR, so it should never be used without checking return
status. But we already have a helper function that does that, so
use it in more places.
* src/lxc/lxc_container.c (lxcContainerAvailable): Use safer
virWaitPid.
* daemon/libvirtd.c (daemonForkIntoBackground): Likewise.
* tests/testutils.c (virtTestCaptureProgramOutput, virtTestMain):
Likewise.
* src/libvirt.c (virConnectAuthGainPolkit): Simplify with virCommand.
