This is a regression introduced by new RPC codes, previously
we advertise the service via ssh even if the daemon doesn't
listen on TLS port (TCP is not choosed). Now the service is
only advertised when it listens on TLS or TCP port. This breaks
upper layer apps which intends to discover the service, such
as virt-manager.
The libvirtd daemon had a few crude system tap probes. Some of
these were broken during the RPC rewrite. The new modular RPC
code is structured in a way that allows much more effective
tracing. Instead of trying to hook up the original probes,
define a new set of probes for the RPC and event code.
The master probes file is now src/probes.d. This contains
probes for virNetServerClientPtr, virNetClientPtr, virSocketPtr
virNetTLSContextPtr and virNetTLSSessionPtr modules. Also add
probes for the poll event loop.
The src/dtrace2systemtap.pl script can convert the probes.d
file into a libvirt_probes.stp file to make use from systemtap
much simpler.
The src/rpc/gensystemtap.pl script can generate a set of
systemtap functions for translating RPC enum values into
printable strings. This works for all RPC header enums (program,
type, status, procedure) and also the authentication enum
The PROBE macro will automatically generate a VIR_DEBUG
statement, so any place with a PROBE can remove any existing
manual DEBUG statements.
* daemon/libvirtd.stp, daemon/probes.d: Remove obsolete probing
* daemon/libvirtd.h: Remove probe macros
* daemon/Makefile.am: Remove all probe buildings/install
* daemon/remote.c: Update authentication probes
* src/dtrace2systemtap.pl, src/rpc/gensystemtap.pl: Scripts
to generate STP files
* src/internal.h: Add probe macros
* src/probes.d: Master list of probes
* src/rpc/virnetclient.c, src/rpc/virnetserverclient.c,
src/rpc/virnetsocket.c, src/rpc/virnettlscontext.c,
src/util/event_poll.c: Insert probe points, removing any
DEBUG statements that duplicate the info
To avoid static linking libvirtd to the RPC server code, which
then prevents sane introduction of DTrace probes, put it all
in the libvirt.so, and export it
* daemon/Makefile.am: Don't link to RPC libraries
* src/Makefile.am: Link all RPC libraries to libvirt.so
* src/libvirt_private.syms: Export all RPC functions
https://bugzilla.redhat.com/show_bug.cgi?id=740899 documents that
if qemu uses aio=native for its disks, then it consumes 128 aio
requests per disk. On a host with multiple guests, this can quickly
run out of kernel aio requests with the default aio-max-nr of
65536. Kernel developers have confirmed that there is no up-front
cost to raising this limit (a larger limit merely implies that more
aio requests can be issued in parallel, which in turn will result
in more kernel memory allocation, only if the system really does use
that many requests). Since the system default limit prevents 256
disks, which is well within libvirt's current scalability, this
patch installs a file to raise the limit and document it in case a
system administrator has further cause to tune the limit. The
install only works on platforms new enough to source /etc/sysctl.d/*
alongside /etc/sysctl.conf (F14 and RHEL 6).
* daemon/libvirtd.sysctl: New file.
* daemon/Makefile.am (EXTRA_DIST): Ship it.
(install-init, uninstall-init): Install it.
* libvirt.spec.in (%files): Include it in rpm.
Bug introduced in commit 675464b. On an OOM, this would try to
dereference a char* and free the contents as a pointer, which is
doomed to failure.
Adding a syntax check will prevent mistakes like this in the future.
* cfg.mk (sc_prohibit_internal_functions): New syntax check.
(exclude_file_name_regexp--sc_prohibit_internal_functions): Add
exemptions.
* daemon/remote.c (remoteRelayDomainEventIOError)
(remoteRelayDomainEventIOErrorReason)
(remoteRelayDomainEventGraphics, remoteRelayDomainEventBlockJob):
Use correct free function.
remoteRelayDomainEventBlockJob, remoteRelayDomainEventIOError,
remoteRelayDomainEventIOErrorReason and remoteRelayDomainEventGraphics
were using const string directly in rpc structure, before calling
remoteDispatchDomainEventSend(). But that routine now frees up all
the pointed allocated memory from the rpc structure and we end up
with a double free.
This now strdup() all the strings passed and provide mem_error goto
labels to be used when an allocation error occurs.
Note that the cleanup isn't completely finished because all relaying
function also call make_nonnull_domain() which also allocate a string
and never handle the error case. This patches doesn't try to address
this as this is only error correctness a priori and touches far more
functions in this module:
* daemon/remote.c: fix string allocations and memory error handling
for remoteRelayDomainEventBlockJob, remoteRelayDomainEventIOError,
remoteRelayDomainEventIOErrorReason and remoteRelayDomainEventGraphics
This is a bit painful for example when starting virt-manager
it tends to clutter libvirtd.log with invalid operation on cpu pinning
for defined but not running domains. A priori those kind of errors
don't indicate an error when executing the command but on a precondition
for running the API, and honnestly while the application should report
it, logging it as an error in libvirtd.log is not really useful,
Related bug: https://bugzilla.redhat.com/show_bug.cgi?id=590807
* daemon/libvirtd.c: extend daemonErrorLogFilter() to filter out
errors of type VIR_ERR_OPERATION_INVALID
This patch annotates APIs with low or high priority.
In low set MUST be all APIs which might eventually access monitor
(and thus block indefinitely). Other APIs may be marked as high
priority. However, some must be (e.g. domainDestroy).
For high priority calls (HPC), there are some high priority workers
(HPW) created in the pool. HPW can execute only HPC, although normal
worker can process any call regardless priority. Therefore, only those
APIs which are guaranteed to end in reasonable small amount of time
can be marked as HPC.
The size of this HPC pool is static, because HPC are expected to end
quickly, therefore jobs assigned to this pool will be served quickly.
It can be configured in libvirtd.conf via prio_workers variable.
Default is set to 5.
To mark API with low or high priority, append priority:{low|high} to
it's comment in src/remote/remote_protocol.x. This is similar to
autogen|skipgen. If not marked, the generator assumes low as default.
When libvirtd is running at non-root user, it won't create ${HOME}/.libvirt.
It will show error message:
17:44:16.838: 7035: error : virPidFileAcquirePath:322 : Failed to open pid file
Signed-off-by: Xu He Jie <xuhj@linux.vnet.ibm.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Commit 2c85644b0b attempted to
fix a problem with tracking RPC messages from streams by doing
- if (msg->header.type == VIR_NET_REPLY) {
+ if (msg->header.type == VIR_NET_REPLY ||
+ (msg->header.type == VIR_NET_STREAM &&
+ msg->header.status != VIR_NET_CONTINUE)) {
client->nrequests--;
In other words any stream packet, with status NET_OK or NET_ERROR
would cause nrequests to be decremented. This is great if the
packet from from a synchronous virStreamFinish or virStreamAbort
API call, but wildly wrong if from a server initiated abort.
The latter resulted in 'nrequests' being decremented below zero.
This then causes all I/O for that client to be stopped.
Instead of trying to infer whether we need to decrement the
nrequests field, from the message type/status, introduce an
explicit 'bool tracked' field to mark whether the virNetMessagePtr
object is subject to tracking.
Also add a virNetMessageClear function to allow a message
contents to be cleared out, without adversely impacting the
'tracked' field as a naive memset() would do
* src/rpc/virnetmessage.c, src/rpc/virnetmessage.h: Add
a 'bool tracked' field and virNetMessageClear() API
* daemon/remote.c, daemon/stream.c, src/rpc/virnetclientprogram.c,
src/rpc/virnetclientstream.c, src/rpc/virnetserverclient.c,
src/rpc/virnetserverprogram.c: Switch over to use
virNetMessageClear() and pass in the 'bool tracked' value
when creating messages.
When sending outbound stream RPC messages, a callback is
used to re-enable stream data transmission. If the stream
aborts while one of these messages is outstanding, the
stream may have been free'd by the time it is invoked. This
results in a use-after-free error
* daemon/stream.c: Ref-count streams to avoid use-after-free
When dispatching domain events we will create an XDR struct
containing the event info. Some of this data may be allocated
on the heap and so must be freed. The graphics event dispatcher
had a broken attempt to free one field, but missed others. All
the events have a dom->name string that needs freeing. The code
should have used the xdr_free() procedure for doing all this
* daemon/remote.c: Use xdr_free after dispatching events
Systemtap 1.2 <sys/sdt.h> tried to expand STAP_PROBE3 into an
initialization:
volatile __typeof__(arg) foo = arg;
but that fails if arg was declared as 'char arg[100]'.
Rather than make all callers to PROBE deal with the stupidity
of <sys/sdt.h>, we instead make PROBE cast away the problem.
Some of this preprocessor abuse copies ideas in src/libvirt.c.
* daemon/libvirtd.h (PROBE): Add casts to all arguments, using...
(VIR_ADD_CASTS, VIR_ADD_CAST, VIR_ADD_CAST2, VIR_ADD_CAST3)
(VIR_ADD_CAST_EXPAND, VIR_ADD_CAST_PASTE, VIR_COUNT_ARGS)
(VIR_ARG5, PROBE_EXPAND): New macros.
Reported by Wen Congyang.
When spice_tls is set but listen_tls is not, we don't initialize
GnuTLS library. So any later gnutls call (e.g. during migration,
where we initialize a certificate) will access uninitialized GnuTLS
internal structs and throws an error.
Although, we might now initialize GnuTLS twice, it is safe according
to the documentation:
This function can be called many times,
but will only do something the first time.
This patch creates 2 functions: virNetTLSInit and virNetTLSDeinit
with respect to written above.
The I/O event callback processes incoming packets first, and then
does outgoing packets. If the incoming packet caused the stream to
close, then the attempt to process outgoing data resulted in an
error. This caused libvirt to then send an error back to the client,
but the stream had already been stopped. This confused the client
since it sees 2 error events.
* daemon/stream.c: Don't attempt read if stream is closed
Every active stream results in a reference being held on the
virNetServerClientPtr object. This meant that if a client quit
with any streams active, although all I/O was stopped the
virNetServerClientPtr object would leak. This causes libvirtd
to leak any file handles associated with open streams when a
client quit
To fix this, when we call virNetServerClientClose there is a
callback invoked which lets the daemon release the streams
and thus the extra references
* daemon/remote.c: Add a hook to close all streams
* daemon/stream.c, daemon/stream.h: Add API for releasing
all streams
* src/rpc/virnetserverclient.c, src/rpc/virnetserverclient.h:
Allow registration of a hook to trigger when closing client
Early errors during start of libvirtd didn't have
an error reporting mechanism and caused libvirtd
to exit silently (only the return value indicated
an error).
Libvirt logging is initialized very early using
enviroment variables and the internal error reporting
API is used to report early errors.
v2 changes:
- print errors unconditionaly before logging starts
- fix message to US spelling
v2.5 changes:
- initialize logging from enviroment
- log all early errors using VIR_ERROR
v3 changes:
- move virSetLogFromEnv() after virInitialize()
fixes: https://bugzilla.redhat.com/show_bug.cgi?id=728654
This is introduced by commit df0b57a95a, which forgot to
add signal handler for SIGHUP.
A simple reproduce method:
1) Create a domain XML under /etc/libvirt/qemu
2) % kill -SIGHUP $(pidof libvirtd)
3) % virsh list --all (the new created domain XML is not listed)
Remove the current libvirtd pidfile handling code, in favour of
calling out to the new APIs. This ensures libvirtd's pidfile
handling is crashsafe
This also means that the non-root libvirtd instances (for handling
qemu:///session URIs) can now safely use pidfiles without racing
* daemon/libvirtd.c: Switch to use virPidFileAcquire and
virPidFileRelease
This patch introduces a internal RPC API "virNetServerClose", which
is standalone with "virNetServerFree". it closes all the socket fds,
and unlinks the unix socket paths, regardless of whether the socket
is still referenced or not.
This is to address regression bug:
https://bugzilla.redhat.com/show_bug.cgi?id=725702
Gettext annoyingly modifies CPPFLAGS in-place, putting
-I/usr/local/include into the search patch if libintl headers
must be used from that location. But since we must support
automake 1.9.6 which lacks AM_CPPFLAGS, and since CPPFLAGS is used
prior to INCLUDES, this means that the build picks up the _old_
installed libvirt.h in priority to the in-tree version, leading
to all sorts of weird build failures on FreeBSD.
Fix this by teaching configure to undo gettext's actions, but
to keep any changes required by gettext at the end of INCLUDES
after all in-tree locations are used first. Also requires
adding a wrapper Makefile.am and making gnulib-tool create
just gnulib.mk files during the bootstrap process.
Signed-off-by: Eric Blake <eblake@redhat.com>
I went with the shorter license notice used by src/libvirt.c,
rather than spelling out the full LGPLv2+ clause into each of
these files.
* configure.ac: Declare copyright.
* all Makefile.am: Likewise.
When libvirtd starts it it will sanity check its own certs,
and before libvirt clients connect to a remote server they
will sanity check their own certs. This patch allows such
sanity checking to be skipped. There is no strong reason to
need to do this, other than to bypass possible libvirt bugs
in sanity checking, or for testing purposes.
libvirt.conf gains tls_no_sanity_certificate parameter to
go along with tls_no_verify_certificate. The remote driver
client URIs gain a no_sanity URI parameter
* daemon/test_libvirtd.aug, daemon/libvirtd.conf,
daemon/libvirtd.c, daemon/libvirtd.aug: Add parameter to
allow cert sanity checks to be skipped
* src/remote/remote_driver.c: Add no_sanity parameter to
skip cert checks
* src/rpc/virnettlscontext.c, src/rpc/virnettlscontext.h:
Add new parameter for skipping sanity checks independantly
of skipping session cert validation checks
If the virStateInitialize call fails we must shutdown libvirtd
since drivers will not be available. Just free'ing the virNetServer
is not sufficient, we must send a SIGTERM to ourselves so that
we interrupt the event loop and trigger a orderly shutdown
* daemon/libvirtd.c: Kill ourselves if state init fails
* src/rpc/virnetserver.c: Add some debugging to event loop
When an operation started by virDomainBlockPull completes (either with
success or with failure), raise an event to indicate the final status.
This API allow users to avoid polling on virDomainGetBlockJobInfo if
they would prefer to use an event mechanism.
* daemon/remote.c: Dispatch events to client
* include/libvirt/libvirt.h.in: Define event ID and callback signature
* src/conf/domain_event.c, src/conf/domain_event.h,
src/libvirt_private.syms: Extend API to handle the new event
* src/qemu/qemu_driver.c: Connect to the QEMU monitor event
for block_stream completion and emit a libvirt block pull event
* src/remote/remote_driver.c: Receive and dispatch events to application
* src/remote/remote_protocol.x: Wire protocol definition for the event
* src/remote_protocol-structs: structure definitions for protocol verification
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c: Watch for BLOCK_STREAM_COMPLETED event
from QEMU monitor
The generator can handle everything except virDomainGetBlockJobInfo().
* src/remote/remote_protocol.x: provide defines for the new entry points
* src/remote/remote_driver.c daemon/remote.c: implement the client and
server side for virDomainGetBlockJobInfo.
* src/remote_protocol-structs: structure definitions for protocol verification
* src/rpc/gendispatch.pl: Permit some unsigned long parameters
The only 'void name(void)' style procedure in the protocol is 'close' that
is handled special, but also programming errors like a missing _args or
_ret suffix on the structs in the .x files can create such a situation by
accident. Making the generator aware of this avoids bogus errors from the
generator such as:
Use of uninitialized value in exists at ./rpc/gendispatch.pl line 967.
Also this allows to get rid of the -c option and the special case code for
the 'close' procedure, as the generator handles it now correctly.
Reported by Michal Privoznik
There were two API in driver.c that were silently masking flags
bits prior to calling out to the drivers, and several others
that were explicitly masking flags bits. This is not
forward-compatible - if we ever have that many flags in the
future, then talking to an old server that masks out the
flags would be indistinguishable from talking to a new server
that can honor the flag. In general, libvirt.c should forward
_all_ flags on to drivers, and only the drivers should reject
unknown flags.
In the case of virDrvSecretGetValue, the solution is to separate
the internal driver callback function to have two parameters
instead of one, with only one parameter affected by the public
API. In the case of virDomainGetXMLDesc, it turns out that
no one was ever mixing VIR_DOMAIN_XML_INTERNAL_STATUS with
the dumpxml path in the first place; that internal flag was
only used in saving and restoring state files, which happened
to be in functions internal to a single file, so there is no
mixing of the internal flag with a public flags argument.
Additionally, virDomainMemoryStats passed a flags argument
over RPC, but not to the driver.
* src/driver.h (VIR_DOMAIN_XML_FLAGS_MASK)
(VIR_SECRET_GET_VALUE_FLAGS_MASK): Delete.
(virDrvSecretGetValue): Separate out internal flags.
(virDrvDomainMemoryStats): Provide missing flags argument.
* src/driver.c (verify): Drop unused check.
* src/conf/domain_conf.h (virDomainObjParseFile): Delete
declaration.
(virDomainXMLInternalFlags): Move...
* src/conf/domain_conf.c: ...here. Delete redundant include.
(virDomainObjParseFile): Make static.
* src/libvirt.c (virDomainGetXMLDesc, virSecretGetValue): Update
clients.
(virDomainMemoryPeek, virInterfaceGetXMLDesc)
(virDomainMemoryStats, virDomainBlockPeek, virNetworkGetXMLDesc)
(virStoragePoolGetXMLDesc, virStorageVolGetXMLDesc)
(virNodeNumOfDevices, virNodeListDevices, virNWFilterGetXMLDesc):
Don't mask unknown flags.
* src/interface/netcf_driver.c (interfaceGetXMLDesc): Reject
unknown flags.
* src/secret/secret_driver.c (secretGetValue): Update clients.
* src/remote/remote_driver.c (remoteSecretGetValue)
(remoteDomainMemoryStats): Likewise.
* src/qemu/qemu_process.c (qemuProcessGetVolumeQcowPassphrase):
Likewise.
* src/qemu/qemu_driver.c (qemudDomainMemoryStats): Likewise.
* daemon/remote.c (remoteDispatchDomainMemoryStats): Likewise.
It is common for the $HOME/.libvirt/libvirtd.conf file to not
exist. Treat this situation as non-fatal since we can carry
on with our default settings just fine.
* daemon/libvirtd.c: Treat ENOENT as non-fatal when loading
config
The dispatch for the CLOSE RPC call was invoking the method
virNetServerClientClose(). This caused the client connection
to be immediately terminated. This meant the reply to the
final RPC message was never sent. Prior to the RPC rewrite
we merely flagged the connection for closing, and actually
closed it when the next RPC call dispatch had completed.
* daemon/remote.c: Flag connection for a delayed close
* daemon/stream.c: Update to use new API for closing
failed connection
* src/rpc/virnetserverclient.c, src/rpc/virnetserverclient.h:
Add support for a delayed connection close. Rename the
virNetServerClientMarkClose method to virNetServerClientImmediateClose
to clarify its semantics
If a client disconnects while it has a stream active, there is
a race condition which could see libvirtd crash. This is because
the client struct may be freed before the last stream event has
triggered. This is trivially solved by holding an extra reference
on the client for the stream callbak
* daemon/stream.c: Acquire reference on client when adding the
stream callback
