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
Pull the call to gnutls_x509_crt_get_dn up into a higher function
so that the 'dname' variable will be available for probe points
* src/rpc/virnettlscontext.c: Pull gnutls_x509_crt_get_dn up
one level
If we receive an error on the stream, set the EOF marker so
that any further (bogus) incoming data is dropped.
* src/rpc/virnetclientstream.c: Set EOF on stream
If we send back an unknown program error for async messages,
we will confuse the client because they only expect replies
for method calls. Just log & drop any invalid async messages
* src/rpc/virnetserver.c: Don't send error for async messages
Commit 597fe3cee6 accidentally
introduced a deadlock when reporting an unknown RPC program.
The virNetServerDispatchNewMessage method is called with
the client locked, and must therefore not attempt to send
any RPC messages back to the client. Only once the incoming
message is passed off to the virNetServerHandleJob worker
is it safe to start sending messages back
* src/rpc/virnetserver.c: Delay checking for unknown RPC
program until in worker thread
Do not crash if virStreamFinish is called after error.
==11000== Invalid read of size 4
==11000== at 0x373A8099A0: pthread_mutex_lock (pthread_mutex_lock.c:51)
==11000== by 0x4C7CADE: virMutexLock (threads-pthread.c:85)
==11000== by 0x4D57C31: virNetClientStreamRaiseError (virnetclientstream.c:203)
==11000== by 0x4D385E4: remoteStreamFinish (remote_driver.c:3541)
==11000== by 0x4D182F9: virStreamFinish (libvirt.c:14157)
==11000== by 0x40FDC4: cmdScreenshot (virsh.c:3075)
==11000== by 0x42BA40: vshCommandRun (virsh.c:14922)
==11000== by 0x42ECCA: main (virsh.c:16381)
==11000== Address 0x59b86c0 is 16 bytes inside a block of size 216 free'd
==11000== at 0x4A06928: free (vg_replace_malloc.c:427)
==11000== by 0x4C69E2B: virFree (memory.c:310)
==11000== by 0x4D57B56: virNetClientStreamFree (virnetclientstream.c:184)
==11000== by 0x4D3DB7A: remoteDomainScreenshot (remote_client_bodies.h:1812)
==11000== by 0x4CFD245: virDomainScreenshot (libvirt.c:2903)
==11000== by 0x40FB73: cmdScreenshot (virsh.c:3029)
==11000== by 0x42BA40: vshCommandRun (virsh.c:14922)
==11000== by 0x42ECCA: main (virsh.c:16381)
Mostly straight-forward, although this is the first API that
returns a new snapshot based on a snapshot rather than a domain.
* src/remote/remote_protocol.x
(REMOTE_PROC_DOMAIN_SNAPSHOT_GET_PARENT): New rpc.
(remote_domain_snapshot_get_parent_args)
(remote_domain_snapshot_get_parent_ret): New structs.
* src/rpc/gendispatch.pl: Adjust generator.
* src/remote/remote_driver.c (remote_driver): Use it.
* src/remote_protocol-structs: Update.
commit 984840a2c2 removed the
notification of waiting calls when VIR_NET_CONTINUE messages
arrive. This was to fix the case of a virStreamAbort() call
being prematurely notified of completion.
The problem is that sometimes there are dummy calls from a
virStreamRecv() call waiting that *do* need to be notified.
These dummy calls should have a status VIR_NET_CONTINUE. So
re-add the notification upon VIR_NET_CONTINUE, but only if
the waiter also has a status of VIR_NET_CONTINUE.
* src/rpc/virnetclient.c: Notify waiting call if stream data
arrives
* src/rpc/virnetclientstream.c: Mark dummy stream read packet
with status VIR_NET_CONTINUE
Libvirt special-cases a specific VIR_ERR_RPC from the remote driver
back into VIR_ERR_NO_SUPPORT on the client, so that clients can
handle missing rpc functions the same whether the hypervisor driver
is local or remote. However, commit c1b22644 introduced a regression:
VIR_FROM_THIS changed from VIR_FROM_REMOTE to VIR_FROM_RPC, so the
special casing no longer works if the server uses the newer error
domain.
* src/rpc/virnetclientprogram.c
(virNetClientProgramDispatchError): Also cater to 0.9.3 and newer.
* src/rpc/virnettlscontext.c: fix memory leak on
virNetTLSContextValidCertificate.
* Detected in valgrind run:
==25667==
==25667== 6,085 (44 direct, 6,041 indirect) bytes in 1 blocks are definitely
lost in loss record 326 of 351
==25667== at 0x4005447: calloc (vg_replace_malloc.c:467)
==25667== by 0x4F2791F3: _asn1_add_node_only (structure.c:53)
==25667== by 0x4F27997A: _asn1_copy_structure3 (structure.c:421)
==25667== by 0x4F276A50: _asn1_append_sequence_set (element.c:144)
==25667== by 0x4F2743FF: asn1_der_decoding (decoding.c:1194)
==25667== by 0x4F22B9CC: gnutls_x509_crt_import (x509.c:229)
==25667== by 0x805274B: virNetTLSContextCheckCertificate
(virnettlscontext.c:1009)
==25667== by 0x804DE32: testTLSSessionInit (virnettlscontexttest.c:693)
==25667== by 0x804F14D: virtTestRun (testutils.c:140)
==25667==
==25667== 23,188 (88 direct, 23,100 indirect) bytes in 11 blocks are definitely
lost in loss record 346 of 351
==25667== at 0x4005447: calloc (vg_replace_malloc.c:467)
==25667== by 0x4F22B841: gnutls_x509_crt_init (x509.c:50)
==25667== by 0x805272B: virNetTLSContextCheckCertificate
(virnettlscontext.c:1003)
==25667== by 0x804DDD1: testTLSSessionInit (virnettlscontexttest.c:673)
==25667== by 0x804F14D: virtTestRun (testutils.c:140)
* How to reproduce?
% cd libvirt && ./configure && make && make -C tests valgrind
or
% valgrind -v --leak-check=full ./tests/virnettlscontexttest
Signed-off-by: Alex Jia <ajia@redhat.com>
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.
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.
The bufferOffset has been initialized to zero in virNetMessageEncodePayloadRaw(),
so, we use bufferLength to represent the length of message which is going to be
sent to client side.
From: Matthias Bolte <matthias.bolte@googlemail.com>
Tested-by: Jim Fehlig <jfehlig@novell.com>
Matthias provided this patch to fix an issue I encountered in the
generator with APIs containing call-by-ref long type, e.g.
int virDomainMigrateGetMaxSpeed(virDomainPtr domain,
unsigned long *bandwidth,
unsigned int flags);
In case we add a new program in the future (we did that in the past and
we are going to do it again soon) current daemon will behave badly with
new client that wants to use the new program. Before the RPC rewrite we
used to just send an error reply to any request with unknown program.
With the RPC rewrite in 0.9.3 the daemon just closes the connection
through which such request was sent. This patch fixes this regression.
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.
If a client had initiated a stream abort, it will have a call
waiting for a reply in the queue. If more data continues to
arrive on the stream, the abort command could mistakenly get
signalled as complete. Remove the code from async data processing
that looked for waiting calls. Add a sanity check to ensure no
async call can ever be marked as needing a reply
* src/rpc/virnetclient.c: Ensure async data packets can't
trigger a reply
If a stream gets a server initiated abort, the client may still
send an abort request before it receives the server side abort.
This causes the server to send back another abort for the
stream. Since the protocol defines that abort is the last thing
to be sent, the client gets confused by this second abort from
the server. If the stream is already shutdown, just drop any
client requested abort, rather than sending back another message.
This fixes the regression from previous versions.
Tested as follows
In one virsh session
virsh # start foo
virsh # console foo
In other virsh session
virsh # destroy foo
The first virsh session should be able to continue issuing
commands without error. Prior to this patch it saw
virsh # list
error: Failed to list active domains
error: An error occurred, but the cause is unknown
virsh # list
error: Failed to list active domains
error: no call waiting for reply with prog 536903814 vers 1 serial 9
* src/rpc/virnetserverprogram.c: Drop abort requests
for streams which no longer exist
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
When trying to use any SASL authentication for TCP sockets by
setting auth_tls = "sasl" in libvirtd.conf on server side, the
client will hang because of the sasl session relocking other than
dropping the lock when exiting virNetSASLSessionExtKeySize()
* src/rpc/virnetsaslcontext.c: virNetSASLSessionExtKeySize drop the
lock on exit
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
In virNetServerNew, Coverity didn't realize that srv->mdsnGroupName
can only be non-NULL if mdsnGroupName was non-NULL.
In virNetServerRun, Coverity didn't realize that the array is non-NULL
if the array count is non-zero.
* src/rpc/virnetserver.c (virNetServerNew): Use alternate pointer.
(virNetServerRun): Give coverity a hint.
Detected by Coverity. Freeing the wrong variable results in both
a memory leak and the likelihood of the caller dereferencing through
a freed pointer.
* src/rpc/virnettlscontext.c (virNetTLSSessionNew): Free correct
variable.
Detected by Coverity. We want to compare the result of fnmatch 'rv',
not our pre-set return value 'ret'.
* src/rpc/virnetsaslcontext.c (virNetSASLContextCheckIdentity):
Check correct variable.
Spotted by Coverity. Gnutls documents that buffer must be NULL
if gnutls_x509_crt_get_key_purpose_oid is to be used to determine
the correct size needed for allocating a buffer.
* src/rpc/virnettlscontext.c
(virNetTLSContextCheckCertKeyPurpose): Initialize buffer.
Spotted by coverity. If pipe2 fails, then we attempt to close
uninitialized fds, which may result in a double-close.
* src/rpc/virnetserver.c (virNetServerSignalSetup): Initialize fds.
Steps to reproduce this problem (vm1 is not running):
for i in `seq 50`; do virsh managedsave vm1& done; killall virsh
Pre-patch, virNetServerClientClose could end up setting client->sock
to NULL prior to other cleanup functions trying to use client->sock.
This fixes things by checking for NULL in more places, and by deferring
the cleanup until after all queued messages have been served.
* src/rpc/virnetserverclient.c (virNetServerClientRegisterEvent)
(virNetServerClientGetFD, virNetServerClientIsSecure)
(virNetServerClientLocalAddrString)
(virNetServerClientRemoteAddrString): Check for closed socket.
(virNetServerClientClose): Rearrange close sequence.
Analysis from Wen Congyang.
Without this, cygwin failed to compile:
In file included from ../src/rpc/virnetmessage.h:24,
from ../src/rpc/virnetclient.h:27,
from remote/remote_driver.c:31:
../src/rpc/virnetprotocol.h:9:21: error: rpc/rpc.h: No such file or directory
With that fixed, compilation warned:
rpc/virnetsocket.c: In function 'virNetSocketNewListenUNIX':
rpc/virnetsocket.c:347: warning: format '%d' expects type 'int', but argument 8 has type 'gid_t' [-Wformat]
rpc/virnetsocket.c: In function 'virNetSocketGetLocalIdentity':
rpc/virnetsocket.c:743: warning: pointer targets in passing argument 5 of 'getsockopt' differ in signedness
* src/Makefile.am (libvirt_driver_remote_la_CFLAGS)
(libvirt_net_rpc_client_la_CFLAGS)
(libvirt_net_rpc_server_la_CFLAGS): Include XDR_CFLAGS, for rpc
headers on cygwin.
* src/rpc/virnetsocket.c (virNetSocketNewListenUNIX)
(virNetSocketGetLocalIdentity): Avoid compiler warnings.
On RHEL 5, with gcc 4.1.2:
rpc/virnetsaslcontext.c: In function 'virNetSASLSessionUpdateBufSize':
rpc/virnetsaslcontext.c:396: warning: dereferencing type-punned pointer will break strict-aliasing rules [-Wstrict-aliasing]
* src/rpc/virnetsaslcontext.c (virNetSASLSessionUpdateBufSize):
Use a union to work around gcc warning.
When an incoming RPC message is ready for processing,
virNetServerClientDispatchRead()
will invoke the 'dispatchFunc' callback. This is set to
virNetServerDispatchNewMessage
This function puts the message + client in a queue for processing by the thread
pool. The thread pool worker function is
virNetServerHandleJob
The first thing this does is acquire an extra reference on the 'client'.
Unfortunately, between the time the message+client are put on the thread pool
queue, and the time the worker runs, the client object may have had its last
reference removed.
We clearly need to add the reference to the client object before putting the
client on the processing queue
* src/rpc/virnetserverclient.c: Add a reference to the client when
invoking the dispatch function
* src/rpc/virnetserver.c: Don't acquire a reference to the client
when in the worker thread
The virNetSASLContext, virNetSASLSession, virNetTLSContext and
virNetTLSSession classes previously relied in their owners
(virNetClient / virNetServer / virNetServerClient) to provide
locking protection for concurrent usage. When virNetSocket
gained its own locking code, this invalidated the implicit
safety the SASL/TLS modules relied on. Thus we need to give
them all explicit locking of their own via new mutexes.
* src/rpc/virnetsaslcontext.c, src/rpc/virnettlscontext.c: Add
a mutex per object
When setting up a server socket, we must skip EADDRINUSE errors
from bind, since the IPv6 socket bind may have already bound to
the IPv4 socket too. If we don't manage to bind to any sockets
at all though, we should then report the EADDRINUSE error as
normal.
This fixes the case where libvirtd would not exit if some other
program was listening on its TCP/TLS ports.
* src/rpc/virnetsocket.c: Report EADDRINUSE
