My latest patch for RPC rework (a2c304f687) introduced a memory leak.
virNetMessageEncodeHeader() is calling VIR_ALLOC_N(msg->buffer, ...)
despite fact, that msg->buffer isn't VIR_FREE()'d on all paths calling
the function. Therefore, rather than injecting free statement switch to
VIR_REALLOC_N().
Previous commit
commit 32a9aac2e0
Author: William Jon McCann <william.jon.mccann@gmail.com>
Date: Thu May 3 12:36:27 2012 -0400
Use XDG Base Directories instead of storing in home directory
Accidentally changed the umask when creating /var/run/libvirt
to 077. This prevents /var/run/libvirt being readable by non-root,
which is required for non-root to connect to libvirtd. Fix the
code so that umask 077 is only used for the non-privileged libvirtd
instance.
Only the non-privileged libvirtd instance uses $HOME. So avoid
running the code for migrating to XDG directories unless using
a non-privileged libvirtd
Commits 51082301, 16d7b39, and 521cc447 introduced support for
'virsh snapshot-list --from' when talking to a server older than
0.9.5, but broke support for plain 'virsh snapshot-list' for the
same old server in the process. Because the code is not properly
gated, we end up with a SIGSEGV during a strcmp with a NULL argument.
* tools/virsh.c (cmdSnapshotList): Don't waste time on fallbacks
when --from is not present.
when do remount,the source and target should be the same
values specified in the initial mount() call.
So change fs->dst to src.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
This fixes the build on 32bit systems which otherwise fails with:
virnetmessagetest.c: In function 'testMessageHeaderEncode':
virnetmessagetest.c:75:9: error: format '%zu' expects argument of type 'size_t', but argument 7 has type 'long unsigned int' [-Werror=format]
virDomainSnapshotPtr has a refcount member, but no one was able
to use it. Furthermore, all of our other vir*Ptr objects have
a *Ref method to match their *Free method. Thankfully, this is
client-side only, so we can use this new function regardless of
how old the server side is! (I have future patches to virsh
that want to use it.)
* include/libvirt/libvirt.h.in (virDomainSnapshotRef): Declare.
* src/libvirt.c (virDomainSnapshotRef): Implement it.
* src/libvirt_public.syms (LIBVIRT_0.9.13): Export it.
When libvirtd forks off a new child, the child then calls virLogReset(),
which ends up closing file descriptors used as log outputs. However, we
recently started logging closed file descriptors, which means we need to
lock logging mutex which was already locked by virLogReset(). We don't
really want to log anything when we are in the process of closing log
outputs.
For pseries guest, spapr-vlan and spapr-vty is based
on spapr-vio address. According to model of network
device, the address type should be assigned automatically.
For serial device, serial pty device is recognized as
spapr-vty device, which is also on spapr-vio.
So this patch is to correct the address type of
spapr-vlan and spapr-vty, and build correct
command line of spapr-vty.
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
Reviewed-by: Michael Ellerman<michaele@au1.ibm.com>
While libvirt intentionally avoids -Wundef (after all, C99
guarantees sane semantics of treating undefined macros as 0),
the glibc insanity of #warning on _FORTIFY_SOURCE coupled with
what some people feel is the black magic of autoconf means
that other projects are likely to copy our snippet verbatim.
We can be nicer to other projects by making it easier to
integrate into projects that use -Wundef.
Suggested by Christophe Fergeau.
* m4/virt-compile-warnings.m4 (LIBVIRT_COMPILE_WARNINGS): Be nice
to other projects using -Wundef.
There is a theoretical problem of an extreme bug where we can get
into deadlock due to command handshaking. Thanks to a pair of pipes,
we have a situation where the parent thinks the child reported an
error and is waiting for a message from the child to explain the
error; but at the same time the child thinks it reported success
and is waiting for the parent to acknowledge the success; so both
processes are now blocked.
Thankfully, I don't think this deadlock is possible without at
least one other bug in the code, but I did see exactly that sort
of situation prior to commit da831af - I saw a backtrace where a
double close bug in the parent caused the parent to read from the
wrong fd and assume the child failed, even though the child really
sent success.
This potential deadlock is not quite like commit 858c247 (a deadlock
due to multiple readers on one pipe preventing a write from completing),
although the solution is similar - always close unused pipe fds before
blocking, rather than after.
* src/util/command.c (virCommandHandshakeWait): Close unused fds
sooner.
When libvirtd is started and there is an unusable/not-connectable
leftover from earlier started machine, it's more reasonable to say
that the machine "crashed" if we know it was started with
"-no-shutdown".
This patch fixes that and also changes the other result (when machine
was started without "-no-shutdown") to "unknown", because the previous
"failed" reason means (according to include/libvirt/libvirt.h.in:174),
that the machine failed to start.
If you compile without NLS support, where _() is a no-op macro,
then we end up passing a string literal to a char*, provoking:
In file included from virsh.c:3639:0:
virsh-edit.c: In function ‘cmdSaveImageEdit’:
virsh-edit.c:97:13: error: assignment discards ‘const’ qualifier from pointer target type [-Werror]
virsh-edit.c:106:13: error: assignment discards ‘const’ qualifier from pointer target type [-Werror]
* tools/virsh-edit.c: Be const-safe.
Commit 7bff56a worked in an incremental build, but fails for a
fresh clone; apparently, if make sees both an actual file
spelling and an inference rule, only the exact spelling is used.
CCLD libvirt_driver_test.la
CC libvirt_driver_remote_la-remote_driver.lo
remote/remote_driver.c:4707:34: fatal error: remote_client_bodies.h: No such file or directory
compilation terminated.
BUILT_SOURCES to the rescue, instead of trying to mess with .lo
dependencies directly.
* src/Makefile.am (REMOTE_DRIVER_PREREQS, %remote_driver.lo): Drop...
(BUILT_SOURCES): ...and add here instead.
Commit 1c275e9a accidentally dropped the storage driver from
libvirtd, because it depended on a C preprocessor macro that
was not defined. Furthermore, if you do './configure
--without-storage-dir --with-storage-disk' or any other combination
where you explicitly build a subset of storage backends excluding
the dir backend, then the build is broken.
Based on analysis by Osier Yang.
* configure.ac (WITH_STORAGE): Define top-level conditional.
* src/Makefile.am (mod_LTLIBRARIES): Build driver even when
storage_dir is disabled.
* daemon/libvirtd.c: Pick up storage driver for any backend, not
just dir.
* daemon/Makefile.am (libvirtd_LDADD): Likewise.
Since we are allocating RPC buffer dynamically, we can increase limits
for max. size of RPC message and RPC string. This is needed to cover
some corner cases where libvirt is run on such huge machines that their
capabilities XML is 4 times bigger than our current limit. This leaves
users with inability to even connect.
Currently, we are allocating buffer for RPC messages statically.
This is not such pain when RPC limits are small. However, if we want
ever to increase those limits, we need to allocate buffer dynamically,
based on RPC message len (= the first 4 bytes). Therefore we will
decrease our mem usage in most cases and still be flexible enough in
corner cases.
We had a distributed file (remote_protocol.h, which in turn was
a prereq to remote_driver.c) depending on a generated file
(libvirt_probes.h), which is a no-no for a VPATH build from a
read-only source tree (no wonder 'make distcheck' tests precisely
that situation):
File `libvirt_driver_remote.la' does not exist.
File `libvirt_driver_remote_la-remote_driver.lo' does not exist.
Prerequisite `libvirt_probes.h' is newer than target `../../src/remote/remote_protocol.h'.
Must remake target `../../src/remote/remote_protocol.h'.
Invoking recipe from Makefile:7464 to update target `../../src/remote/remote_protocol.h'.
make[3]: Entering directory `/home/remote/eblake/libvirt-tmp2/build/libvirt-0.9.12/_build/src'
GEN ../../src/remote/remote_protocol.h
cannot create ../../src/remote/remote_protocol.h: Permission denied at ../../src/rpc/genprotocol.pl line 31.
make[3]: *** [../../src/remote/remote_protocol.h] Error 13
Rather than making distributed .c files depend on generated files, we
really want to ensure that compilation into .lo files is not attempted
until the generated files are present, done by this patch. Since there
were two different sets of conditionally generated files that both
feed the .lo file, I had to introduce a new variable REMOTE_DRIVER_PREREQS
to keep automake happy.
After that fix, the next issue was that make treats './foo' and 'foo'
differently in determining whether an implicit %foo rule is applicable,
with the result that locking/qemu-sanlock.conf wasn't properly being
built at the right times. Also, the output for using the .aug test
files was a bit verbose.
After fixing the src directory, the next error is related to the docs
directory, where the tarball is missing a stamp file and thus tries to
regenerate files that are already present:
GEN ../../docs/apibuild.py.stamp
Traceback (most recent call last):
File "../../docs/apibuild.py", line 2511, in <module>
rebuild("libvirt")
File "../../docs/apibuild.py", line 2495, in rebuild
builder.serialize()
File "../../docs/apibuild.py", line 2424, in serialize
output = open(filename, "w")
IOError: [Errno 13] Permission denied: '../../docs/libvirt-api.xml'
make[5]: *** [../../docs/apibuild.py.stamp] Error 1
and fixing that exposed another case of a distributed file (generated
html) depending on a built file (libvirt.h), but only when doing an
in-tree build, because of a file glob.
* src/Makefile.am ($(srcdir)/remote/remote_driver.c): Change...
(libvirt_driver_remote_la-remote_driver.lo): ...to the real
dependency.
($(builddir)/locking/%-sanlock.conf): Drop $(builddir), so that
rule gets run in time for test_libvirt_sanlock.aug.
(test_libvir*.aug): Cater to silent build.
(conf_DATA): Don't ship qemu-sanlock.conf in the tarball, since it
is trivial to regenerate.
* docs/Makefile.am (EXTRA_DIST): Ship our stamp file.
($(APIBUILD_STAMP)): Don't depend on generated file.
QEMU 1.1.0 has been officially released. With 1.1.0 QEMU went back to
three-digits version even for the initial release and I renamed the data
files to match this fact. They were generated with
qemu-system-x86_64 -help >tests/qemuhelpdata/qemu-1.1.0
qemu-system-x86_64 \
-device ? \
-device pci-assign,? \
-device virtio-blk-pci,? \
-device virtio-net-pci,? \
-device scsi-disk,? 2>tests/qemuhelpdata/qemu-1.1.0-device
I came across a bug that the command line generated for passthrough
of the host parallel port /dev/parport0 by libvirt for QEMU is incorrect.
It currently produces:
-chardev tty,id=charparallel0,path=/dev/parport0
-device isa-parallel,chardev=charparallel0,id=parallel0
The first parameter is "tty". It sould be "parport".
If I launch qemu with -chardev parport,... it works as expected.
I have already filled a bug report (
https://bugzilla.redhat.com/show_bug.cgi?id=823879 ), the topic was
already on the list some months ago:
https://www.redhat.com/archives/libvirt-users/2011-September/msg00095.html
Signed-off-by: Eric Blake <eblake@redhat.com>
It is possible to deadlock libvirt by having a domain with XML
longer than PIPE_BUF, and by writing a hook script that closes
stdin early. This is because libvirt was keeping a copy of the
child's stdin read fd open, which means the write fd in the
parent will never see EPIPE (remember, libvirt should always be
run with SIGPIPE ignored, so we should never get a SIGPIPE signal).
Since there is no error, libvirt blocks waiting for a write to
complete, even though the only reader is also libvirt. The
solution is to ensure that only the child can act as a reader
before the parent does any writes; and then dealing with the
fallout of dealing with EPIPE.
Thankfully, this is not a security hole - since the only way to
trigger the deadlock is to install a custom hook script, anyone
that already has privileges to install a hook script already has
privileges to do any number of other equally disruptive things
to libvirt; it would only be a security hole if an unprivileged
user could install a hook script to DoS a privileged user.
* src/util/command.c (virCommandRun): Close parent's copy of child
read fd earlier.
(virCommandProcessIO): Don't let EPIPE be fatal; the child may
be done parsing input.
* tests/commandhelper.c (main): Set up a SIGPIPE situation.
* tests/commandtest.c (test20): Trigger it.
* tests/commanddata/test20.log: New file.
Although src/util/viratomic.h has been added to the repo, up until now
it hasn't been used. Stefan Berger is using it in his proposed dhcp
snooping patches, and an rpm build with those patches failed due to
viratomic.h not being packed up with the rest of the sources.
glibc 2.15 (on Fedora 17) coupled with explicit disabling of
optimization during development dies a painful death:
In file included from /usr/include/limits.h:27:0,
from /usr/lib/gcc/x86_64-redhat-linux/4.7.0/include/limits.h:169,
from /usr/lib/gcc/x86_64-redhat-linux/4.7.0/include/syslimits.h:7,
from /usr/lib/gcc/x86_64-redhat-linux/4.7.0/include/limits.h:34,
from util/bitmap.c:26:
/usr/include/features.h:314:4: error: #warning _FORTIFY_SOURCE requires compiling with optimization (-O) [-Werror=cpp]
cc1: all warnings being treated as errors
Work around this by only conditionally defining _FORTIFY_SOURCE,
in the case where glibc can actually use it. The trick is using
AH_VERBATIM instead of AC_DEFINE.
* m4/virt-compile-warnings.m4 (LIBVIRT_COMPILE_WARNINGS): Squelch
_FORTIFY_SOURCE when needed to avoid glibc #warnings.
I noticed this during 'make syntax-check':
prohibit_readlink
grep: Unmatched ( or \(
* cfg.mk (exclude_file_name_regexp--sc_prohibit_readlink): Fix
mismatched '('.
EBADF errors are logged as warnings as they normally indicate a double
close bug. This patch also provides VIR_MASS_CLOSE helper to be user in
the only case of mass close after fork when EBADF should rather be
ignored.
If users *-edit but make a mistake in XML all changes are
permanently lost. However, if virsh is not running within
a script we can ask user if he wants to re-edit the file
and correct the mistakes.
Currently, we either generate some cmd*Edit commands (cmdPoolEdit
and cmdNetworkEdit) via sed script or copy the body of cmdEdit
(e.g. cmdInterfaceEdit, cmdNWFilterEdit, etc.). This fact makes
it harder to implement any new feature to our editing system.
Therefore switch to new implementation - define macros to:
- dump XML (EDIT_GET_XML)
- take an action if XML wasn't changed,
usually just vshPrint() (EDIT_NOT_CHANGED)
- define new object (EDIT_DEFINE) - the edited XML is in @doc_edited
- free object defined by EDIT_DEFINE (EDIT_FREE)
and #include "virsh-edit.c"
With support for multiple IP addresses per interface in place, this patch
now adds support for multiple IP addresses per interface for the DHCP
snooping code.
Testing:
Since the infrastructure I tested this with does not provide multiple IP
addresses per MAC address (anymore), I either had to plug the VM's interface
from the virtual bride connected directly to the infrastructure to virbr0
to get a 2nd IP address from dnsmasq (kill and run dhclient inside the VM)
or changed the lease file (/var/run/libvirt/network/nwfilter.leases) and
restart libvirtd to have a 2nd IP address on an existing interface.
Note that dnsmasq can take a lease timeout parameter as part of the --dhcp-range
command line parameter, so that timeouts can be tested that way
(--dhcp-range 192.168.122.2,192.168.122.254,120). So, terminating and restarting
dnsmasq with that parameter is another choice to watch an IP address disappear
after 120 seconds.
Regards,
Stefan
The goal of this patch is to prepare for support for multiple IP
addresses per interface in the DHCP snooping code.
Move the code for the IP address map that maps interface names to
IP addresses into their own file. Rename the functions on the way
but otherwise leave the code as-is. Initialize this new layer
separately before dependent layers (iplearning, dhcpsnooping)
and shut it down after them.
This patch adds DHCP snooping support to libvirt. The learning method for
IP addresses is specified by setting the "CTRL_IP_LEARNING" variable to one of
"any" [default] (existing IP learning code), "none" (static only addresses)
or "dhcp" (DHCP snooping).
Active leases are saved in a lease file and reloaded on restart or HUP.
The following interface XML activates and uses the DHCP snooping:
<interface type='bridge'>
<source bridge='virbr0'/>
<filterref filter='clean-traffic'>
<parameter name='CTRL_IP_LEARNING' value='dhcp'/>
</filterref>
</interface>
All filters containing the variable 'IP' are automatically adjusted when
the VM receives an IP address via DHCP. However, multiple IP addresses per
interface are silently ignored in this patch, thus only supporting one IP
address per interface. Multiple IP address support is added in a later
patch in this series.
Signed-off-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com>
Currently, monitoring QEMU virtual machines with standard Unix
sysadmin tools is harder than it has to be. The QEMU command line is
often miles long and mostly redundant, it's hard to tell which process
is which.
This patch reorders the QEMU -name argument to be the first, so it's
immediately visible in "ps x", htop and "atop -c" output.
This patch resolves:
https://bugzilla.redhat.com/show_bug.cgi?id=827519
The problem is that an interface with type='hostdev' will have an
alias of the form "hostdev%d", while the function that looks through
existing netdevs to determine the name to use for a new addition will
fail if there's an existing entry that does not match the form
"net%d".
This is another of the handful of places that need an exception due to
the hybrid nature of <interface type='hostdev'> (which is not exactly
an <interface> or a <hostdev>, but is both at the same time).
If we migrate to fd, spec->fwdType is not MIGRATION_FWD_DIRECT,
we will close spec->dest.fd.local in qemuMigrationRun(). So we
should set spec->dest.fd.local to -1 in qemuMigrationRun().
Bug present since 0.9.5 (commit 326176179).
Wen Congyang reported that we have a double-close bug if we fail
virFDStreamOpenInternal, since childfd duplicated one of the fds[]
array contents. In truth, since we always transfer both members
of fds to other variables, we should close the fds through those
other names, and just use fds[] for pipe().
Bug present since 0.9.0 (commit e886237a).
* src/fdstream.c (virFDStreamOpenFileInternal): Swap scope of
childfd and fds[], to avoid a double close.
KAMEZAWA Hiroyuki reported a nasty double-free bug when virCommand
is used to convert a string into input to a child command. The
problem is that the poll() loop of virCommandProcessIO would close()
the write end of the pipe in order to let the child see EOF, then
the caller virCommandRun() would also close the same fd number, with
the second close possibly nuking an fd opened by some other thread
in the meantime. This in turn can have all sorts of bad effects.
The bug has been present since the introduction of virCommand in
commit f16ad06f.
This is based on his first attempt at a patch, at
https://bugzilla.redhat.com/show_bug.cgi?id=823716
* src/util/command.c (_virCommand): Drop inpipe member.
(virCommandProcessIO): Add argument, to avoid closing caller's fd
without informing caller.
(virCommandRun, virCommandNewArgs): Adjust clients.
