QEMU domains were marked as having managed save image even if they were
saved using the regular save. With this patch, domains are marked so
only when using managed save API.
QEMU (and librbd) flush the cache on the source before the
destination starts, and the destination does not read any
changeable data before that, so live migration with rbd caching
is safe.
This makes 'virsh migrate' work with rbd and caching without the
--unsafe flag.
Reported-by: Vladimir Bashkirtsev <vladimir@bashkirtsev.com>
Signed-off-by: Josh Durgin <josh.durgin@inktank.com>
virDomainBlockStatsFlags can't collect total_time_ns for read/write/flush
because of key typo when retriveing from qemu cmd result
Signed-off-by: lvroyce <lvroyce@linux.vnet.ibm.com>
Reported by Jason Helfman as a build-breaker on FreeBSD.
* src/conf/domain_conf.c (virDomainFSDefParseXML): Use POSIX
spelling.
* src/openvz/openvz_conf.c (openvzReadFSConf): Likewise.
Below patch fixes this coverity report:
/libvirt/src/conf/nwfilter_conf.c:382:
leaked_storage: Variable "varAccess" going out of scope leaks the storage it points to.
Since we are mounting a new /dev in the container, we must
remove any sub-mounts like /dev/shm, /dev/mqueue, etc,
otherwise they'll be recorded in /proc/mounts, but not be
accessible to applications.
Hello,
This is a patch to fix vm's outbound traffic control problem.
Currently, vm's outbound traffic control by libvirt doesn't go well.
This problem was previously discussed at libvir-list ML, however
it seems that there isn't still any answer to the problem.
http://www.redhat.com/archives/libvir-list/2011-August/msg00333.html
I measured Guest(with virtio-net) to Host TCP throughput with the
command "netperf -H".
Here are the outbound QoS parameters and the results.
outbound average rate[kilobytes/s] : Guest to Host throughput[Mbit/s]
======================================================================
1024 (8Mbit/s) : 4.56
2048 (16Mbit/s) : 3.29
4096 (32Mbit/s) : 3.35
8192 (64Mbit/s) : 3.95
16384 (128Mbit/s) : 4.08
32768 (256Mbit/s) : 3.94
65536 (512Mbit/s) : 3.23
The outbound traffic goes down unreasonably and is even not controled.
The cause of this problem is too large mtu value in "tc filter" command run by
libvirt. The command uses burst value to set mtu and the burst is equal to
average rate value if it's not set. This value is too large. For example
if the average rate is set to 1024 kilobytes/s, the mtu value is set to 1024
kilobytes. That's too large compared to the size of network packets.
Here libvirt applies tc ingress filter to Host's vnet(tun) device.
Tc ingress filter is implemented with TBF(Token Buckets Filter) algorithm. TBF
uses mtu value to calculate the amount of token consumed by each packet. With too
large mtu value, the token consumption rate is set too large. This leads to
token starvation and deterioration of TCP throughput.
Then, should we use the default mtu value 2 kilobytes?
The anser is No, because Guest with virtio-net device uses 65536 bytes
as mtu to transmit packets to Host, and the tc filter with the default mtu
value 2k drops packets whose size is larger than 2k. So, the most packets
is droped and again leads to deterioration of TCP throughput.
The appropriate mtu value is 65536 bytes which is equal to the maximum value
of network interface device defined in <linux/netdevice.h>. The value is
not so large that it causes token starvation and not so small that it
drops most packets.
Therefore this patch set the mtu value to 64kb(== 65535 bytes).
Again, here are the outbound QoS parameters and the TCP throughput with
the libvirt patched.
outbound average rate[kilobytes/s] : Guest to Host throughput[Mbit/s]
======================================================================
1024 (8Mbit/s) : 8.22
2048 (16Mbit/s) : 16.42
4096 (32Mbit/s) : 32.93
8192 (64Mbit/s) : 66.85
16384 (128Mbit/s) : 133.88
32768 (256Mbit/s) : 271.01
65536 (512Mbit/s) : 547.32
The outbound traffic conforms to the given limit.
Thank you,
Signed-off-by: Eiichi Tsukata <eiichi.tsukata.xh@hitachi.com>
If the user specified invalid protocol type in a network's SRV record
the error path ended up in freeing uninitialized pointers causing a
daemon crash.
*network_conf.c: virNetworkDNSSrvDefParseXML(): initialize local
variables
VirtualBox doesn't use the common virDomainObj implementation so this
patch adds a separate implementation using the VirtualBox API.
This driver implementation supports all currently defined flags. As
VirtualBox does not support transient guests, managed save images and
autostarting we assume all guests are persistent, don't have a managed
save image and are not autostarted. Filtering for existence of those
properities results in empty list.
mnt_fsname can not be the same, as we check the duplicate pool
sources earlier before, means it can't be the same pool, moreover,
a pool can't be started if it's already active anyway. So no reason
to act as success.
virConnectDomainEventRegisterAny() takes a domain as an argument.
So it should be possible to register the same event (be it
VIR_DOMAIN_EVENT_ID_LIFECYCLE for example) for two different domains.
That is, we need to take domain into account when searching for
duplicate event being already registered.
In order to retrieve some sysinfo data we need to parse /proc/sysinfo and
/proc/cpuinfo.
Signed-off-by: Thang Pham <thang.pham@us.ibm.com>
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
For the s390x architecture the sysfs core_id alone is not unique. As a
result it can happen that libvirt thinks there are less host CPUs available
than really present.
Currently, a logical CPU is equivalent to a core for s390x. We therefore
produce a fake core id from the CPU number.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Minimal CPU "parser" for s390 to avoid compile time warning.
Signed-off-by: Thang Pham <thang.pham@us.ibm.com>
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Adding CPU encoder/decoder for s390 to avoid runtime error messages.
Signed-off-by: Thang Pham <thang.pham@us.ibm.com>
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Starting a KVM guest on s390 fails immediately. This is because
"qemu --help" reports -no-acpi even for the s390(x) architecture but
-no-acpi isn't supported there.
Workaround is to remove QEMU_CAPS_NO_ACPI from the capability set
after the version/capability extraction.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
We used to prefix 'rbd:' to volume names, this is not necessary.
Qemu takes RBD devices in this way, like: qemu -drive rbd:pool/image
When attaching a network disk like RBD to a guest we however do not use this prefix.
Currently you can't map a RBD volume name directly to a domain without removing the prefix.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
If no 'listen' attribute or <listen> element is set in the
guest XML, the default driver configured listen address is
used. There is no way to client applications to determine
what this address is though. When starting the guest, we
should update the live XML to include this default listen
address
Storage is one of the last domains in libvirt where we don't fully
utilize inactive and live XML. Okay, it might be because we don't
have support for that. So implement such support. However, we need
to fallback when talking to old daemon which doesn't support this
new flag called VIR_STORAGE_XML_INACTIVE.
Currently, we share the idea of old & new def with domains. Users can
*-edit an object (domain, pool) which spawns a new internal
representation for them. This is referenced via
{domainObj,poolObj}->newDef [compared to ->def]. However, for pool we
were never overwriting def with newDef. This must be done on
pool-destroy (like we do analogically in domain detroy).
Currently libvirt-lxc checks to see if the destination exists and is a
directory. If it is not a directory then the mount fails. Since
libvirt-lxc can bind mount files on an inode, this patch is needed to
allow us to bind mount files on files. Currently we want to bind mount
on top of /etc/machine-id, and /etc/adjtime
If the destination of the mount point does not exists, it checks if the
src is a directory and then attempts to create a directory, otherwise it
creates an empty file for the destination. The code will then bind mount
over the destination.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Some GNULIB headers (eg unistd.h) will often need to include
winsock2.h for various symbols. There is a rule that winsock2.h
must be included before windows.h. This means that any file
which does
#ifdef WIN32
#include <windows.h>
#endif
#include <unistd.h>
is potentially broken. A simple rule is that /all/ includes of
windows.h must be matched with a preceding include of winsock2.h
regardless of whether unistd.h is used currently
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
A sanlock lease can be marked as shared (rather
than exclusive) using SANLK_RES_SHARED flag. This
adds support for that flag and ensures that in auto
disk mode, any shared disks use shared leases. This
also makes any read-only disks be completely
ignored.
These changes remove the need for the option
ignore_readonly_and_shared_disks
so that is removed
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Currently you can configure LXC to bind a host directory to
a guest directory, but not to bind a guest directory to a
guest directory. While the guest container init could do
this itself, allowing it in the libvirt XML means a stricter
SELinux policy can be written
Introduce a new syntax for filesystems to allow use of a RAM
filesystem
<filesystem type='ram'>
<source usage='10' units='MiB'/>
<target dir='/mnt'/>
</filesystem>
The usage units default to KiB to limit consumption of host memory.
* docs/formatdomain.html.in: Document new syntax
* docs/schemas/domaincommon.rng: Add new attributes
* src/conf/domain_conf.c: Parsing/formatting of RAM filesystems
* src/lxc/lxc_container.c: Mounting of RAM filesystems
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The test of ref count is not protected by lock, which is unsafe because
the ref count may have been changed by other threads during the test.
This patch fixes this.
When shutting down libvirtd, the virNetServer shutdown can deadlock
if there are in-flight jobs being handled by virNetServerHandleJob().
virNetServerFree() will acquire the virNetServer lock and call
virThreadPoolFree() to terminate the workers, waiting for the workers
to finish. But in-flight workers will attempt to acquire the
virNetServer lock, resulting in deadlock.
Fix the deadlock by unlocking the virNetServer lock before calling
virThreadPoolFree(). This is safe since the virNetServerPtr object
is ref-counted and only decrementing the ref count needs to be
protected. Additionally, there is no need to re-acquire the lock
after virThreadPoolFree() completes as all the workers have
terminated.
The lxc contoller eventually makes use of virRandomBits(), which was
segfaulting since virRandomInitialize() is never invoked.
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff554d560 in random_r () from /lib64/libc.so.6
(gdb) bt
0 0x00007ffff554d560 in random_r () from /lib64/libc.so.6
1 0x0000000000469eaa in virRandomBits (nbits=32) at util/virrandom.c:80
2 0x000000000045bf69 in virHashCreateFull (size=256,
dataFree=0x4aa2a2 <hashDataFree>, keyCode=0x45bd40 <virHashStrCode>,
keyEqual=0x45bdad <virHashStrEqual>, keyCopy=0x45bdfa <virHashStrCopy>,
keyFree=0x45be37 <virHashStrFree>) at util/virhash.c:134
3 0x000000000045c069 in virHashCreate (size=0, dataFree=0x4aa2a2 <hashDataFree>)
at util/virhash.c:164
4 0x00000000004aa562 in virNWFilterHashTableCreate (n=0)
at conf/nwfilter_params.c:686
5 0x00000000004aa95b in virNWFilterParseParamAttributes (cur=0x711d30)
at conf/nwfilter_params.c:793
6 0x0000000000481a7f in virDomainNetDefParseXML (caps=0x702c90, node=0x7116b0,
ctxt=0x7101b0, bootMap=0x0, flags=0) at conf/domain_conf.c:4589
7 0x000000000048cc36 in virDomainDefParseXML (caps=0x702c90, xml=0x710040,
root=0x7103b0, ctxt=0x7101b0, expectedVirtTypes=16, flags=0)
at conf/domain_conf.c:8658
8 0x000000000048f011 in virDomainDefParseNode (caps=0x702c90, xml=0x710040,
root=0x7103b0, expectedVirtTypes=16, flags=0) at conf/domain_conf.c:9360
9 0x000000000048ee30 in virDomainDefParse (xmlStr=0x0,
filename=0x702ae0 "/var/run/libvirt/lxc/x.xml", caps=0x702c90,
expectedVirtTypes=16, flags=0) at conf/domain_conf.c:9310
10 0x000000000048ef00 in virDomainDefParseFile (caps=0x702c90,
filename=0x702ae0 "/var/run/libvirt/lxc/x.xml", expectedVirtTypes=16, flags=0)
at conf/domain_conf.c:9332
11 0x0000000000425053 in main (argc=5, argv=0x7fffffffe2b8)
at lxc/lxc_controller.c:1773
The comment says:
/* Now create the final dir in the path with the uid/gid/mode
* requested in the config. If the dir already exists, just set
* the perms.
*/
However, virDirCreate is only invoked if the target path doesn't
exist yet (which is opposite with the comment), or the uid from
the config is not -1 (I don't understand why, think it's just
another mistake). And the result is the perms of the pool won't
be changed if one tries to build the pool with different perms
again.
Besides these logic error fix, if no uid and gid are specified in
the config, the practical used uid, gid are reflected.
The two new APIs are rather trivial; based on bits and pieces of
other existing APIs. But rather than blindly return 0 or 1 for
HasMetadata, I chose to first validate that the snapshot in
question in fact exists.
* src/esx/esx_driver.c (esxDomainSnapshotIsCurrent)
(esxDomainSnapshotHasMetadata): New functions.
* src/vbox/vbox_tmpl.c (vboxDomainSnapshotIsCurrent)
(vboxDomainSnapshotHasMetadata): Likewise.
Blindly returning success is misleading if the object no longer
exists; it is a bit better to check for existence up front before
returning information about that object. This pattern matches the
fact that most of our other APIs check for existence as a side
effect prior to getting at the real piece of information being
queried.
* src/esx/esx_driver.c (esxDomainIsUpdated, esxDomainIsPersistent):
Add existence checks.
* src/vbox/vbox_tmpl.c (vboxDomainIsPersistent)
(vboxDomainIsUpdated): Likewise.
This patch adds support for listing all domains into drivers that use
the common virDomainObj implementation: libxl, lxc, openvz, qemu, test,
uml, vmware.
For drivers that don't support managed save images the guests are
treated as if they had none, so filtering guests that do have such an
image on this driver succeeds and produces 0 results.
The two new functions are very similar to the existing functions;
just a matter of different arguments and a call to a different
helper function.
* src/qemu/qemu_driver.c (qemuDomainSnapshotListNames)
(qemuDomainSnapshotNum, qemuDomainSnapshotListChildrenNames)
(qemuDomainSnapshotNumChildren): Support new flags.
(qemuDomainListAllSnapshots): New functions.
Wraps the conversion from 'char *name' to virDomainSnapshotPtr in
a reusable manner.
* src/conf/virdomainlist.h (virDomainListSnapshots): New declaration.
* src/conf/virdomainlist.c (virDomainListSnapshots): Implement it.
* src/libvirt_private.syms (virdomainlist.h): Export it.
The generator doesn't handle lists of virDomainSnapshotPtr, so
this commit requires a bit more work than some RPC additions.
* src/remote/remote_protocol.x
(REMOTE_PROC_DOMAIN_LIST_ALL_SNAPSHOTS)
(REMOTE_PROC_DOMAIN_SNAPSHOT_LIST_ALL_CHILDREN): New RPC calls,
with corresponding structs.
* daemon/remote.c (remoteDispatchDomainListAllSnapshots)
(remoteDispatchDomainSnapshotListAllChildren): New functions.
* src/remote/remote_driver.c (remoteDomainListAllSnapshots)
(remoteDomainSnapshotListAllChildren): Likewise.
* src/remote_protocol-structs: Regenerate.
There was an inherent race between virDomainSnapshotNum() and
virDomainSnapshotListNames(), where an additional snapshot could
be created in the meantime, or where a snapshot could be deleted
before converting the name back to a virDomainSnapshotPtr. It
was also an awkward name: the function operates on domains, not
domain snapshots. virDomainSnapshotListChildrenNames() suffered
from the same inherent race, although its naming was nicer.
This patch makes things nicer by grabbing a snapshot list
atomically, in the format most useful to the user.
* include/libvirt/libvirt.h.in (virDomainListAllSnapshots)
(virDomainSnapshotListAllChildren): New declarations.
* src/libvirt.c (virDomainSnapshotListNames)
(virDomainSnapshotListChildrenNames): Add cross-references.
(virDomainListAllSnapshots, virDomainSnapshotListAllChildren):
New functions.
* src/libvirt_public.syms (LIBVIRT_0.9.13): Export them.
* src/driver.h (virDrvDomainListAllSnapshots)
(virDrvDomainSnapshotListAllChildren): New callbacks.
* python/generator.py (skip_function): Prepare for later
hand-written versions.
It turns out that one-bit filtering makes it hard to select the inverse
set, so it is easier to provide filtering groups. For back-compat,
omitting all bits within a group means the group is not used for
filtering, and by definition of a group (each snapshot matches exactly
one bit within the group, and the set of bits in the group covers all
snapshots), selecting all bits also makes the group useless.
Unfortunately, virDomainSnapshotListChildren defined the bit
VIR_DOMAIN_SNAPSHOT_LIST_DESCENDANTS as an expansion rather than a
filter, so we cannot make it part of a filter group, so that bit
(and its counterpart VIR_DOMAIN_SNAPSHOT_LIST_ROOTS for
virDomainSnapshotList) remains a single control bit.
* include/libvirt/libvirt.h.in (virDomainSnapshotListFlags): Add a
couple more flags.
* src/libvirt.c (virDomainSnapshotNum)
(virDomainSnapshotNumChildren): Document them.
(virDomainSnapshotListNames, virDomainSnapshotListChildrenNames):
Likewise, and add thread-safety caveats.
* src/conf/virdomainlist.h (VIR_DOMAIN_SNAPSHOT_FILTERS_*): New
convenience macros.
* src/conf/domain_conf.c (virDomainSnapshotObjListCopyNames)
(virDomainSnapshotObjListCount): Support the new flags.
Until now, it was possible to crash libvirtd when defining domain with
channel device with missing source element.
When creating new virDomainChrDef, target.port is set to -1, but
unfortunately it is an union with addresses that virDomainChrDefFree
tries to free in case the deviceType is channel. Having the port set
to -1 is intended, however the cleanest way to get around the problems
with the crash seems to be renumbering the VIR_DOMAIN_CHR_CHANNEL_
target types to cover new NONE type (with value 0) being the default
(no target type yet).
Macro virCheckNullArgGoto is supposed to check for NULL argument but
checks non-NULL instead.
Macro virCheckNonNullArgReturn reports error as if the argument should
be NULL when it shouldn't.
when lxcContainerIdentifyCGroups failed, the memory it allocated
has been freed, so we should not free this memory again in
lxcContainerSetupPivortRoot and lxcContainerSetupExtraMounts.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Another case where we can do the same amount of work with fewer
lines of redundant code, which will make adding new filters easier.
* src/conf/domain_conf.c (virDomainSnapshotNameData): Adjust
struct.
(virDomainSnapshotObjListCount): Delete, now taken care of...
(virDomainSnapshotObjListCopyNames): ...here.
(virDomainSnapshotObjListGetNames): Adjust caller to handle
counting.
(virDomainSnapshotObjListNum): Simplify.
Now that domain listing is a thin wrapper around child listing,
it's easier to have a common entry point. This restores the
hashForEach optimization lost in the previous patch when there
are no snapshots being filtered out of the entire list.
* src/conf/domain_conf.h (virDomainSnapshotObjListGetNames)
(virDomainSnapshotObjListNum): Add parameter.
(virDomainSnapshotObjListGetNamesFrom)
(virDomainSnapshotObjListNumFrom): Delete.
* src/libvirt_private.syms (domain_conf.h): Drop deleted functions.
* src/conf/domain_conf.c (virDomainSnapshotObjListGetNames):
Merge, and (re)add an optimization.
* src/qemu/qemu_driver.c (qemuDomainUndefineFlags)
(qemuDomainSnapshotListNames, qemuDomainSnapshotNum)
(qemuDomainSnapshotListChildrenNames)
(qemuDomainSnapshotNumChildren): Update callers.
* src/qemu/qemu_migration.c (qemuMigrationIsAllowed): Likewise.
* src/conf/virdomainlist.c (virDomainListPopulate): Likewise.
This idea was first suggested by Daniel Veillard here:
https://www.redhat.com/archives/libvir-list/2011-October/msg00353.html
Now that I am about to add more complexity to snapshot listing, it
makes sense to avoid code duplication and special casing for domain
listing (all snapshots) vs. snapshot listing (descendants); adding
a metaroot reduces the number of code lines by having the domain
listing turn into a descendant listing of the metaroot.
Note that this has one minor pessimization - if we are going to list
ALL snapshots without filtering, then virHashForeach is more efficient
than recursing through the child relationships; restoring that minor
optimization will occur in the next patch.
* src/conf/domain_conf.h (_virDomainSnapshotObj)
(_virDomainSnapshotObjList): Repurpose some fields.
(virDomainSnapshotDropParent): Drop unused parameter.
* src/conf/domain_conf.c (virDomainSnapshotObjListGetNames)
(virDomainSnapshotObjListCount): Simplify.
(virDomainSnapshotFindByName, virDomainSnapshotSetRelations)
(virDomainSnapshotDropParent): Match new field semantics.
* src/qemu/qemu_driver.c (qemuDomainSnapshotCreateXML)
(qemuDomainSnapshotReparentChildren, qemuDomainSnapshotDelete):
Adjust clients.
This patch adds common code to list domains in fashion used by
virListAllDomains with all currently supported flags. The header file
also contains macros that group filters together that are used to
shorten filter conditions.
This patch stores existence of the image in the object. At start of the
daemon the state is checked and then updated in key moments in domain
lifecycle.
This patch wires up the RPC protocol handlers for
virConnectListAllDomains(). The RPC generator has no support for the way
how virConnectListAllDomains() returns the results so the handler code
had to be done manually.
The new api is handled by REMOTE_PROC_CONNECT_LIST_ALL_DOMAINS, with
number 273 and marked with high priority.
This patch adds a new public api that lists domains. The new approach is
different from those used before. There are key points to this:
1) The list is acquired atomically and contains both active and inactive
domains (guests). This eliminates the need to call two different list
APIs, where the state might change in between the calls.
2) The returned list consists of virDomainPtrs instead of names or ID's
that have to be converted to virDomainPtrs anyways using separate calls
for each one of them. This is more convenient and saves hypervisor calls.
3) The returned list is auto-allocated. This saves a lot of hassle for
the users.
4) Built in support for filtering. The API call supports various
filtering flags that modify the output list according to user needs.
Available filter groups:
Domain status:
VIR_CONNECT_LIST_DOMAINS_ACTIVE, VIR_CONNECT_LIST_DOMAINS_INACTIVE
Domain persistence:
VIR_CONNECT_LIST_DOMAINS_PERSISTENT,
VIR_CONNECT_LIST_DOMAINS_TRANSIENT
Domain state:
VIR_CONNECT_LIST_DOMAINS_RUNNING, VIR_CONNECT_LIST_DOMAINS_PAUSED,
VIR_CONNECT_LIST_DOMAINS_SHUTOFF, VIR_CONNECT_LIST_DOMAINS_OTHER
Existence of managed save image:
VIR_CONNECT_LIST_DOMAINS_MANAGEDSAVE,
VIR_CONNECT_LIST_DOMAINS_NO_MANAGEDSAVE
Auto-start option:
VIR_CONNECT_LIST_DOMAINS_AUTOSTART,
VIR_CONNECT_LIST_DOMAINS_NO_AUTOSTART
Existence of snapshot:
VIR_CONNECT_LIST_DOMAINS_HAS_SNAPSHOT,
VIR_CONNECT_LIST_DOMAINS_NO_SNAPSHOT
5) The python binding returns a list of domain objects that is very neat
to work with.
The only problem with this approach is no support from code generators
so both RPC code and python bindings had to be written manually.
*include/libvirt/libvirt.h.in: - add API prototype
- clean up whitespace mistakes nearby
*python/generator.py: - inhibit generation of the bindings for the new
api
*src/driver.h: - add driver prototype
- clean up some whitespace mistakes nearby
*src/libvirt.c: - add public implementation
*src/libvirt_public.syms: - export the new symbol
when libvirt_lxc trigger oom error in lxcContainerGetSubtree
we should free the alloced memory for mounts.
so when lxcContainerGetSubtree failed,we should do some
memory cleanup in lxcContainerUnmountSubtree.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
we alloc the memory for format in lxcContainerMountDetectFilesystem
but without free it in lxcContainerMountFSBlockHelper.
this patch just call VIR_FREE to free it.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
With latest changes to qemu-ga success on some commands is not reported
anymore, e.g. guest-shutdown or guest-suspend-*. However, errors are
still being reported. Therefore, we need to find different source of
indication if operation was successful. Events.
Commit 6e769eba made it a runtime error if libvirt was compiled
without yajl support but targets a new enough qemu. But enough
users are hitting this on self-compiled libvirt that it is worth
erroring out at compilation time, rather than an obscure failure
when trying to use the built executable.
* configure.ac: If qemu is requested and -version works, require
yajl when qemu version is new enough.
* src/qemu/qemu_capabilities.c (qemuCapsComputeCmdFlags): Add
comment.
When libpcap is not available, the NWFilter driver provides a
no-op stub for the DHCP snooping initialization. This was
mistakenly returning '-1' instead of '0', so the entire driver
initialization failed
dump-guest-memory is a new dump mechanism, and it can work when the
guest uses host devices. This patch adds a API to use this new
monitor command.
We will always use json mode if qemu's version is >= 0.15, so I
don't implement the API for text mode.
This reverts
commit c16b4c43fc
Author: Daniel P. Berrange <berrange@redhat.com>
Date: Fri May 11 15:09:27 2012 +0100
Avoid LXC pivot root in the root source is still /
This commit broke setup of /dev, because the code which
deals with setting up a private /dev and /dev/pts only
works if you do a pivotroot.
The original intent of avoiding the pivot root was to
try and ensure the new root has a minimumal mount
tree. The better way todo this is to just unmount the
bits we don't want (ie old /proc & /sys subtrees.
So apply the logic from
commit c529b47a75
Author: Daniel P. Berrange <berrange@redhat.com>
Date: Fri May 11 11:35:28 2012 +0100
Trim /proc & /sys subtrees before mounting new instances
to the pivot_root codepath as well
Libvirt updates the configuration of SPICE server only when something
changes. This is unfortunate when the user wants to disconnect a
existing spice session when the connected attribute is already
"disconnect".
This patch modifies the conditions for calling the password updater to
be called when nothing changes, but the connected attribute is already
"disconnect".
While unescaping the commands the commands passed through to the monitor
function qemuMonitorUnescapeArg() initialized lenght of the input string
to strlen()+1 which is fine for alloc but not for iteration of the
string.
This patch fixes the off-by-one error and drops the pointless check for
a single trailing slash that is automaticaly handled by the default
branch of switch.
commit 52d064f42d added
VIR_NETWORK_XML_INACTIVE in order to allow suppressing the
auto-generated list of VFs in network definitions, and a --inactive
flag to virsh net-dumpxml to take advantage of the flag. However, it
missed out on two opportunities:
1) Use INACTIVE to get the current config of the network as it
exists on disk, rather than the currently active config.
2) Add INACTIVE to the flags used for the virsh net-edit command, so
that it won't include the forward-pool interfaces that were
autogenerated, and so that a re-edit of the network prior to
restarting it will show any other edits made since the last restart
of the network. (prior to this patch, if you edited a network a 2nd
time without restarting, all of the previous edits would magically
disappear).
In order to fit with the new #define-based generic edit function in
virsh.c, a new function vshNetworkGetXMLDesc() was added. This
function first tries to call virNetworkGetXMLDesc with the INACTIVE
flag added, then retries without if the first attempt fails (in the
manner expected when the server doesn't support it).
A core use case of the hook scripts is to be able to do things
to a guest's network configuration. It is possible to hook into
the 'start' operation for a QEMU guest which runs just before
the guest is started. The TAP devices will exist at this point,
but the QEMU process will not. It can be desirable to have a
'started' hook too, which runs once QEMU has started.
If libvirtd is restarted it will re-populate firewall rules,
but there is no QEMU hook to trigger for existing domains.
This is solved with a 'reconnect' hook.
Finally, if attaching to an external QEMU process there needs
to be an 'attach' hook script.
This all also applies to the LXC driver
* docs/hooks.html.in: Document new operations
* src/util/hooks.c, src/util/hooks.c: Add 'started', 'reconnect'
and 'attach' operations for QEMU. Add 'prepare', 'started',
'release' and 'reconnect' operations for LXC
* src/lxc/lxc_driver.c: Add hooks for 'prepare', 'started',
'release' and 'reconnect' operations
* src/qemu/qemu_process.c: Add hooks for 'started', 'reconnect'
and 'reconnect' operations
First 'poll' can't return EWOULDBLOCK, and second, we're checking errno
so far away from the poll() call that we've probably already trashed the
original errno value.
In addition to keepalive responses, we also need to send keepalive
requests from client IO loop to properly detect dead connection in case
a libvirt API is called from the main loop, which prevents any timers to
be called.
The previous commit removed the only usage of ``all'' parameter in
virKeepAliveStopInternal, which was actually the only reason for having
virKeepAliveStopInternal. This effectively reverts most of commit
6446a9e20c.
When a libvirt API is called from the main event loop (which seems to be
common in event-based glib apps), the client IO loop would properly
handle keepalive requests sent by a server but will not actually send
them because the main event loop is blocked with the API. This patch
gets rid of response timer and the thread which is processing keepalive
requests is also responsible for queueing responses for delivery.
Add virKeepAliveTimeout and virKeepAliveTrigger APIs that can be used to
set poll timeouts and trigger keepalive timer. virKeepAliveTrigger
checks if it is called to early and does nothing in that case.
The code that needs to be run every keepalive interval of inactivity was
only called from a timer and thus from the main event loop. We will need
to call the code directly from another place.
As non-blocking calls are no longer dropped, we don't really need to
care that much about their fate and wait for the thread with the buck
to process them. If another thread has the buck, we can just push a
non-blocking call to the queue and be done with it.
So far, we were dropping non-blocking calls whenever sending them would
block. In case a client is sending lots of stream calls (which are not
supposed to generate any reply), the assumption that having other calls
in a queue is sufficient to get a reply from the server doesn't work. I
tried to fix this in b1e374a7ac but
failed and reverted that commit.
With this patch, non-blocking calls are never dropped (unless the
connection is being closed) and will always be sent.
Normally, when every call has a thread associated with it, the thread
may get the buck and be in charge of sending all calls until its own
call is done. When we introduced non-blocking calls, we had to add
special handling of new non-blocking calls. This patch uses event loop
to send data if there is no thread to get the buck so that any
non-blocking calls left in the queue are properly sent without having to
handle them specially. It also avoids adding even more cruft to client
IO loop in the following patches.
With this change in, non-blocking calls may see unpredictable delays in
delivery when the client has no event loop registered. However, the only
non-blocking calls we have are keepalives and we already require event
loop for them, which makes this a non-issue until someone introduces new
non-blocking calls.
'make dist' was depending on *protocol-structs files, which are
stored in git but in turn depended on generated files. We still
want to ship the protocol-structs files, but by renaming the
tests to something not matching a file name, we separate 'make
check' (which depends on the generated file) from 'make dist'
(which only depends on the git files). After all, the tarball
should never depend on a generated file not stored in git.
I found one more case of a git file depending on a generated
file, in a bogus virkeycode.c listing; but at least this one
had no associated rules so it never broke 'make dist'.
Reported by Wen Congyang. Latent bug has been present since
commit 62dee6f, but only recently exposed by commit 7bff56a.
* src/Makefile.am ($(srcdir)/util/virkeycode.c): Drop useless
dependency.
(BUILT_SOURCES): ...and build virkeymaps.h sooner.
(PROTOCOL_STRUCTS): Rather than depend on the struct file...
(check-local): ...convert things into a phony target of...
(check-protocol): ...a new check.
($(srcdir)/remote_protocol-struct): Rename to isolate the distributed
file from the conditional test.
(PDWTAGS): Deal with rename. Swap to compare 'expected actual'.
Currently, if qemuProcessStart fail at some point, e.g. because
domain being started wants a PCI/USB device already assigned to
a different domain, we jump to cleanup label where qemuProcessStop
is performed. This unconditionally calls virSecurityManagerRestoreAllLabel
which is wrong because the other domain is still using those devices.
However, once we successfully label all devices/paths in
qemuProcessStart() from that point on, we have to perform a rollback
on failure - that is - we have to virSecurityManagerRestoreAllLabel.
The two APIs are rather trivial; based on bits and pieces of other
existing APIs. It leaves the door open for future extension to
qemu to report snapshots without metadata based on reading qcow2
internal snapshot names.
* src/qemu/qemu_driver.c (qemuDomainSnapshotIsCurrent)
(qemuDomainSnapshotHasMetadata): New functions.
Right now, starting from just a virDomainSnapshotPtr, and wanting to
know if it is the current snapshot for its respective domain, you have
to use virDomainSnapshotGetDomain(), then virDomainSnapshotCurrent(),
then compare the two names returned by virDomainSnapshotGetName().
It is a bit easier if we can directly query this information from the
snapshot itself.
Right now, it is possible to filter a snapshot listing based on
whether snapshots have metadata that would prevent domain deletion,
but the only way to learn if an individual snapshot has metadata is
to see if that snapshot appears in the list returned by a listing.
Additionally, I hope to expand the qemu driver in a future patch to
use qemu-img to reconstruct snapshot XML corresponding to internal
qcow2 snapshot names not otherwise tracked by libvirt (in part, so
that libvirt can guarantee that new snapshots are not created with
a name that would silently corrupt the existing portion of the qcow2
file); if I ever get that in, then it would no longer be an all-or-none
decision on whether snapshots have metadata, and becomes all the more
important to be able to directly determine that information from a
particular snapshot.
Other query functions (such as virDomainIsActive) do not have a flags
argument, but since virDomainHasCurrentSnapshot takes a flags argument,
I figured it was safer to provide a flags argument here as well.
* include/libvirt/libvirt.h.in (virDomainSnapshotIsCurrent)
(virDomainSnapshotHasMetadata): New declarations.
* src/libvirt.c (virDomainSnapshotIsCurrent)
(virDomainSnapshotHasMetadata): New functions.
* src/libvirt_public.syms (LIBVIRT_0.9.13): Export them.
* src/driver.h (virDrvDomainSnapshotIsCurrent)
(virDrvDomainSnapshotHasMetadata): New driver callbacks.
Right now, the only way to get at the contents of a virBuffer is
to destroy it. But there are cases in my upcoming patches where
peeking at the contents makes life easier. I suppose this does
open up the potential for bad code to dereference a stale pointer,
by disregarding the docs that the return value is invalid on the
next virBuf operation, but such is life.
* src/util/buf.h (virBufferCurrentContent): New declaration.
* src/util/buf.c (virBufferCurrentContent): Implement it.
* src/libvirt_private.syms (buf.h): Export it.
* tests/virbuftest.c (testBufAutoIndent): Test it.
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().
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>
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>
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.
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.
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.
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.
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.
Apart from the non-sanlock check build, there is also a little fix for
qemu (EXTRA_DIST had qemu.conf and others inside even if the build was
supposed to be without qemu).
Some of our rules used $(PERL), while others used 'perl'. Always
using the variable allows a developer to point to a different (often
better) perl than the default one found on $PATH.
* daemon/Makefile.am ($(srcdir)/remote_dispatch.h): s/perl/$(PERL).
* src/Makefile.am ($(srcdir)/remote/remote_client_bodies.h)
(PDWTAGS, %protocol.c, %_probes.stp): Likewise.
Without this fix, a VPATH build (such as used by ./autobuild.sh)
fails with messages like:
make[3]: Entering directory `/home/remote/eblake/libvirt-tmp2/build/daemon'
../../build-aux/augeas-gentest.pl libvirtd.conf ../../daemon/test_libvirtd.aug.in test_libvirtd.aug
cannot read libvirtd.conf: No such file or directory at ../../build-aux/augeas-gentest.pl line 38.
Since the test files are not part of the tarball, we can generate
them into the build dir, but rather than create a subdirectory
just for the test file, it is easier to test them directly in
libvirt.git/src.
* daemon/Makefile.am (AUG_GENTEST): Factor out definition.
(test_libvirtd.aug): Look for correct file.
* src/Makefile.am (AUG_GENTEST): Use $(PERL).
(qemu/test_libvirtd_qemu.aug, lxc/test_libvirtd_lxc.aug)
(locking/test_libvirt_sanlock.aug): Rename to avoid subdirectories.
(check-augeas-qemu, check-augeas-lxc, check-augeas-sanlock): Reflect
location of built tests.
* configure.ac (PERL): Substitute perl.
Currently, we are logging only one side of pipes we
create in virCommandRequireHandshake(); This is enough
in cases where pipe2() returns two consecutive FDs. However,
it is not guaranteed and it may return any FDs.
Therefore, it's wise to log the other ends as well.
When getting number of CPUs the host has assigned, there was always
number "1" returned. Even though all lxc domains with no pinning
launched by libvirt run on all pCPUs (by default, no matter what's the
number), we should at least return the same number as the user
specified when creating the domain.
I added libvirt_qemu_probes.h into BUILT_SOURCES. That makes it
generated, but most probably it is not the clearest way how to do
that, but it fixes the build.
The previous patch fixed an incremental build, but missed that on
a fresh checkout, we now have nothing left that stops make from
nuking libvirt_qemu_probes.o.
* src/Makefile.am ($(libvirt_driver_qemu_la_SOURCES)): Delete,
since this variable is empty.
(.PRECIOUS): Add %_probes.o, so they don't get nuked as an
intermediate by-product after creating %_probes.lo.
The moment you specify a _DEPENDENCIES, older automake (stupidly)
assumes that you will specify _all_ dependencies for that target.
This stupidity has been fixed in automake 1.12, but we cannot rely on
newer automake everywhere. For libvirt_la_DEPENDENCIES, we took
care of providing the full list, but for libvirt_qemu_la_DEPENDENCIES,
we were missing the dependency on libvirt_qemu_impl.la, which resulted
in a failed build:
make[3]: Entering directory `/home/ajia/Workspace/libvirt/src'
CCLD libvirt_driver_qemu.la
libtool: link: `libvirt_qemu_probes.lo' is not a valid libtool object
* src/Makefile.am (libvirt_driver_qemu_la_DEPENDENCIES): Delete;
automake does a better job if it does the entire job.
==3240== 23 bytes in 1 blocks are definitely lost in loss record 242 of 744
==3240== at 0x4C2A4CD: malloc (vg_replace_malloc.c:236)
==3240== by 0x8077537: __vasprintf_chk (vasprintf_chk.c:82)
==3240== by 0x509C677: virVasprintf (stdio2.h:199)
==3240== by 0x509C733: virAsprintf (util.c:1912)
==3240== by 0x1906583A: qemudStartup (qemu_driver.c:679)
==3240== by 0x511991D: virStateInitialize (libvirt.c:809)
==3240== by 0x40CD84: daemonRunStateInit (libvirtd.c:751)
==3240== by 0x5098745: virThreadHelper (threads-pthread.c:161)
==3240== by 0x7953D8F: start_thread (pthread_create.c:309)
==3240== by 0x805FF5C: clone (clone.S:115)
To ensure consistent error reporting of invalid arguments,
provide a number of predefined helper methods & macros.
- An arg which must not be NULL:
virCheckNonNullArgReturn(argname, retvalue)
virCheckNonNullArgGoto(argname, label)
- An arg which must be NULL
virCheckNullArgGoto(argname, label)
- An arg which must be positive (ie 1 or greater)
virCheckPositiveArgGoto(argname, label)
- An arg which must not be 0
virCheckNonZeroArgGoto(argname, label)
- An arg which must be zero
virCheckZeroArgGoto(argname, label)
- An arg which must not be negative (ie 0 or greater)
virCheckNonNegativeArgGoto(argname, label)
* src/libvirt.c, src/libvirt-qemu.c,
src/nodeinfo.c, src/datatypes.c: Update to use
virCheckXXXX macros
* po/POTFILES.in: Add libvirt-qemu.c and virterror_internal.h
* src/internal.h: Define macros for checking invalid args
* src/util/virterror_internal.h: Define macros for reporting
invalid args
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Libtool is picky about linking against a module library (aka a .so);
giving lots of warnings like this in the tests directory:
CCLD networkxml2argvtest
*** Warning: Linking the executable networkxml2argvtest against the loadable module
*** libvirt_driver_network.so is not portable!
Fix that by splitting things into a convenience library which can
be used directly by the tests, and making the real .so just wrap
the convenience library.
Based on a suggestion by Daniel P. Berrange.
* configure.ac (--with-driver-modules): Fix help test.
* src/Makefile.am (libvirt_driver_xen.la, libvirt_driver_libxl.la)
(libvirt_driver_qemu.la, libvirt_driver_lxc.la)
(libvirt_driver_uml.la): Factor into new convenience libraries.
* tests/Makefile.am (xen_LDADDS, qemu_LDADDS, lxc_LDADDS)
(networkxml2argvtest_LDADD): Link to convenience libraries, not
shared libraries.
There was no rule forcing libvirt_qemu_probes.o to be built
before libvirt_qemu_probes.lo was used. Also libvirtd was
still referencing the .o file, rather than the .lo file.
Both the .lo and .o file must be listed as DEPENDENCIES,
otherwise libtool will unhelpfully delete the .o file
once the .lo file is created.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The CoTaskMemFree function requires the ole32 DLL to be
linked against. Currently this is only done for the
VirtualBox driver. Also add it to libvirt_util.la
* configure.ac: Unconditionally add ole32 DLL to Win32
* src/Makefile.am: Link old32 to libvirt_util.la
When adding new config file parameters, the corresponding
additions to the augeas lens' are constantly forgotten.
Also there are augeas test cases, these don't catch the
error, since they too are never updated.
To address this, the augeas test cases need to be auto-generated
from the example config files.
* build-aux/augeas-gentest.pl: Helper to generate an
augeas test file, substituting in elements from the
example config files
* src/Makefile.am, daemon/Makefile.am: Switch to
auto-generated augeas test cases
* daemon/test_libvirtd.aug, daemon/test_libvirtd.aug.in,
src/locking/test_libvirt_sanlock.aug,
src/locking/test_libvirt_sanlock.aug.in,
src/lxc/test_libvirtd_lxc.aug,
src/lxc/test_libvirtd_lxc.aug.in,
src/qemu/test_libvirtd_qemu.aug,
src/qemu/test_libvirtd_qemu.aug.in: Remove example
config file data, replacing with a ::CONFIG:: placeholder
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Currently all the config options are listed under a 'vnc_entry'
group. Create a bunch of new groups & move options to the
right place
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Add nmissing 'host_uuid' entry to libvirtd.conf lens and
rename spice_passwd to spice_password in qemu.conf lens
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Instead of doing
# example_config
use
#example_config
so it is possible to programatically uncomment example config
options, as distinct from their comment/descriptions
Also delete rogue trailing comma not allowed by lens
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Add an impl of +virGetUserRuntimeDirectory, virGetUserCacheDirectory
virGetUserConfigDirectory and virGetUserDirectory for Win32 platform.
Also create stubs for non-Win32 platforms which lack getpwuid_r()
In adding these two helpers were added virFileIsAbsPath and
virFileSkipRoot, along with some macros VIR_FILE_DIR_SEPARATOR,
VIR_FILE_DIR_SEPARATOR_S, VIR_FILE_IS_DIR_SEPARATOR,
VIR_FILE_PATH_SEPARATOR, VIR_FILE_PATH_SEPARATOR_S
All this code was adapted from GLib2 under terms of LGPLv2+ license.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Remove the uid param from virGetUserConfigDirectory,
virGetUserCacheDirectory, virGetUserRuntimeDirectory,
and virGetUserDirectory
These functions were universally called with the
results of getuid() or geteuid(). To make it practical
to port to Win32, remove the uid parameter and hardcode
geteuid()
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When you try to connect to a socket in the abstract namespace,
the error will be ECONNREFUSED for a non-listening daemon. With
the non-abstract namespace though, you instead get ENOENT. Add
a check for this extra errno when auto-spawning the daemon
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Remove a number of pointless checks against PATH_MAX and
add a syntax-check rule to prevent its use in future
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Libtool supports linking directly against .o files on some platforms
(such as Linux), which happens to be the only place where we are
actually doing that (for the dtrace-generated probes.o files). However,
it raises a big stink about the non-portability, even though we don't
attempt it on platforms where it would actually fail:
CCLD libvirt_driver_qemu.la
*** Warning: Linking the shared library libvirt_driver_qemu.la against
the non-libtool
*** objects libvirt_qemu_probes.o is not portable!
This shuts libtool up by creating a proper .lo file that matches
what libtool normally expects.
* src/Makefile.am (%_probes.lo): New rule.
(libvirt_probes.stp, libvirt_qemu_probes.stp): Simplify into...
(%_probes.stp): ...shorter rule.
(CLEANFILES): Clean new .lo files.
(libvirt_la_BUILT_LIBADD, libvirt_driver_qemu_la_LIBADD)
(libvirt_lxc_LDADD, virt_aa_helper_LDADD): Link against .lo file.
* tests/Makefile.am (PROBES_O, qemu_LDADDS): Likewise.
If vdsm is installed and configured in Fedora 17, we add the following
items into qemu.conf:
spice_tls=1
spice_tls_x509_cert_dir="/etc/pki/vdsm/libvirt-spice"
However, after this changes, augtool cannot identify qemu.conf anymore.
Add a VIR_ERR_DOMAIN_LAST sentinel for virErrorDomain and
replace the virErrorDomainName function by a VIR_ENUM_IMPL
In the process the naming of error domains is sanitized
* src/util/virterror.c: Use VIR_ENUM_IMPL for converting
error domains to strings
* include/libvirt/virterror.h: Add VIR_ERR_DOMAIN_LAST
The libvirt_private.syms file exports virNetlinkEventServiceLocalPid
so there needs to be a no-op stub for Win32 to avoid linker errors
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
* daemon/libvirtd.c: Set custom driver module dir if the current
binary name is 'lt-libvirtd' (indicating execution directly
from GIT checkout)
* src/driver.c, src/driver.h, src/libvirt_driver_modules.syms: Add
virDriverModuleInitialize to allow driver module location to
be changed
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When building as driver modules, it is not possible for the QEMU
driver module to reference the DTrace/SystemTAP probes linked into
the main libvirt.so. Thus we need to move the QEMU probes into a
separate file 'libvirt_qemu_probes.d'. Also rename the existing
file from 'probes.d' to 'libvirt_probes.d' while we're at it
* daemon/Makefile.am, src/internal.h: Include libvirt_probes.h
instead of probes.h
* src/Makefile.am: Add rules for libvirt_qemu_probes.d
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor_json.c,
src/qemu/qemu_monitor_text.c: Include libvirt_qemu_probes.h
* src/libvirt_probes.d: Rename from probes.d
* src/libvirt_qemu_probes.d: QEMU specific probes formerly
in probes.d
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Since we have drivers which depend on each other (ie QEMU/LXC
depend on the network driver APIs), we need to use RTLD_GLOBAL
instead of RTLD_LOCAL. While this pollutes the calling binary
with many more symbols, this is no worse than if we directly
link to the drivers, and this only applies to libvirtd
* src/driver.c: s/RTLD_LOCAL/RTLD_GLOBAL/
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Only libvirt_driver_storage.la links to libblkid currently. If
we are running in a scenario with driver modules, LXC must
directly link to it, since it can't assume the storage driver
is present
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The libvirt_test.la library was introduced to allow test suites
to reference internal-only symbols. These days, nearly every
symbol we care about is in src/libvirt_private.syms, so there
is no need for libvirt_test.la to continue to exist
* src/Makefile.am: Delete libvirt_test.la & add new .syms files
* src/libvirt_private.syms: Export symbols needed by test suite
* tests/Makefile.am: Link to libvirt_test.la. Ensure LXC tests link
to network_driver.la
* src/libvirt_esx.syms, src/libvirt_openvz.syms: Add exports needed
by test suite
libvirt_driver_nodedev.la should not link against either
libvirt_util.la or gnulib.la, since libvirt.so brings
in those deps.
* src/Makefile.am: Fix broken linkage of libvirt_driver_nodedev.la
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The driver modules all use symbols which are defined in libvirt.so.
Thus for loading of modules to work, the binary that libvirt.so
is linked to must export its symbols back to modules. If the
libvirt.so itself is dlopen()d then the RTLD_GLOBAL flag must
be set. Unfortunately few, if any, programming languages use
the RTLD_GLOBAL flag when loading modules :-( This means is it
not practical to use driver modules for any libvirt client side
drivers (OpenVZ, VMWare, Hyper-V, Remote client, test).
This patch changes the build process so only server side drivers
are built as modules (Xen, QEMU, LXC, UML)
* daemon/libvirtd.c: Add missing load of 'interface' driver
* src/Makefile.am: Only build server side drivers as modules
* src/libvirt.c: Don't load any driver modules
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
and use it for virDomainParseMemory. This allows to parse arbitrary
scaled value, not only memory related values as needed for the
filesystem limits code following later in this series.
This reverts commit b1e374a7ac, which was
rather bad since I failed to consider all sides of the issue. The main
things I didn't consider properly are:
- a thread which sends a non-blocking call waits for the thread with
the buck to process the call
- the code doesn't expect non-blocking calls to remain in the queue
unless they were already partially sent
Thus, the reverted patch actually breaks more than what it fixes and
clients (which may even be libvirtd during p2p migrations) will likely
end up in a deadlock.
The pciDevice structure corresponding to the device being hot-unplugged
was freed after it was "stolen" from activeList. The pointer was still
used for eg-inactive list. This patch removes the free of the structure
and frees it only if reset fails on the device.
I'm tired of writing:
bool sep = false;
while (...) {
if (sep)
virBufferAddChar(buf, ',');
sep = true;
virBufferAdd(buf, str);
}
This makes it easier, allowing one to write:
while (...)
virBufferAsprintf(buf, "%s,", str);
virBufferTrim(buf, ",", -1);
to trim any remaining comma.
* src/util/buf.h (virBufferTrim): Declare.
* src/util/buf.c (virBufferTrim): New function.
* tests/virbuftest.c (testBufTrim): Test it.
This patch adds support for a new storage backend with RBD support.
RBD is the RADOS Block Device and is part of the Ceph distributed storage
system.
It comes in two flavours: Qemu-RBD and Kernel RBD, this storage backend only
supports Qemu-RBD, thus limiting the use of this storage driver to Qemu only.
To function this backend relies on librbd and librados being present on the
local system.
The backend also supports Cephx authentication for safe authentication with
the Ceph cluster.
For storing credentials it uses the built-in secret mechanism of libvirt.
Signed-off-by: Wido den Hollander <wido@widodh.nl>
When the last reference to a virConnectPtr is released by
libvirtd, it was possible for a deadlock to occur in the
virDomainEventState functions. The virDomainEventStatePtr
holds a reference on virConnectPtr for each registered
callback. When removing a callback, the virUnrefConnect
function is run. If this causes the last reference on the
virConnectPtr to be released, then virReleaseConnect can
be run, which in turns calls qemudClose. This function has
a call to virDomainEventStateDeregisterConn which is intended
to remove all callbacks associated with the virConnectPtr
instance. This will try to grab a lock on virDomainEventState
but this lock is already held. Deadlock ensues
Thread 1 (Thread 0x7fcbb526a840 (LWP 23185)):
Since each callback associated with a virConnectPtr holds a
reference on virConnectPtr, it is impossible for the qemudClose
method to be invoked while any callbacks are still registered.
Thus the call to virDomainEventStateDeregisterConn must in fact
be a no-op. Thus it is possible to just remove all trace of
virDomainEventStateDeregisterConn and avoid the deadlock.
* src/conf/domain_event.c, src/conf/domain_event.h,
src/libvirt_private.syms: Delete virDomainEventStateDeregisterConn
* src/libxl/libxl_driver.c, src/lxc/lxc_driver.c,
src/qemu/qemu_driver.c, src/uml/uml_driver.c: Remove
calls to virDomainEventStateDeregisterConn
This patch adds support for the recent ipset iptables extension
to libvirt's nwfilter subsystem. Ipset allows to maintain 'sets'
of IP addresses, ports and other packet parameters and allows for
faster lookup (in the order of O(1) vs. O(n)) and rule evaluation
to achieve higher throughput than what can be achieved with
individual iptables rules.
On the command line iptables supports ipset using
iptables ... -m set --match-set <ipset name> <flags> -j ...
where 'ipset name' is the name of a previously created ipset and
flags is a comma-separated list of up to 6 flags. Flags use 'src' and 'dst'
for selecting IP addresses, ports etc. from the source or
destination part of a packet. So a concrete example may look like this:
iptables -A INPUT -m set --match-set test src,src -j ACCEPT
Since ipset management is quite complex, the idea was to leave ipset
management outside of libvirt but still allow users to reference an ipset.
The user would have to make sure the ipset is available once the VM is
started so that the iptables rule(s) referencing the ipset can be created.
Using XML to describe an ipset in an nwfilter rule would then look as
follows:
<rule action='accept' direction='in'>
<all ipset='test' ipsetflags='src,src'/>
</rule>
The two parameters on the command line are also the two distinct XML attributes
'ipset' and 'ipsetflags'.
FYI: Here is the man page for ipset:
https://ipset.netfilter.org/ipset.man.html
Regards,
Stefan
We were being lazy - virnetlink.c was getting uint32_t as a
side-effect from glibc 2.14's <unistd.h>, but older glibc 2.11
does not provide uint32_t from <unistd.h>. In fact, POSIX states
that <unistd.h> need only provide intptr_t, not all of <stdint.h>,
so the bug really is ours. Reported by Jonathan Alescio.
* src/util/virnetlink.h: Include <stdint.h>.
This involves setting the cpuacct cgroup to a per-vcpu granularity,
as well as summing the each vcpu accounting into a common array.
Now that we are reading more than one cgroup file, we double-check
that cpus weren't hot-plugged between reads to invalidate our
summing.
Signed-off-by: Eric Blake <eblake@redhat.com>
If qemuPrepareHostdevUSBDevices fail it will roll back devices added
to the driver list of used devices. However, if it may fail because
the device is being used already. But then again - with roll back.
Therefore don't try to remove a usb device manually if the function
fail. Although, we want to remove the device if any operation
performed afterwards fail.
Allow the logging APIs to be called with a va_list for format
args, instead of requiring var-args usage.
* src/util/logging.h, src/util/logging.c: Add virLogVMessage
The use of readlink() in lxc_container.c is intentional; we don't
want an absolute pathname there.
* src/util/cgroup.h (VIR_CGROUP_SYSFS_MOUNT): Indent properly.
* cfg.mk (exclude_file_name_regexp--sc_prohibit_readlink): Add
exemption.
One of our latest USB device handling patches
05abd1507d introduced a regression.
That is, we first create a temporary list of all USB devices that
are to be used by domain just starting up. Then we iterate over and
check if a device from the list is in the global list of currently
assigned devices (activeUsbHostdevs). If not, we add it there and
continue with next iteration then. But if a device from temporary
list is either taken already or adding to the activeUsbHostdevs fails,
we remove all devices in temp list from the activeUsbHostdevs list.
Therefore, if a device is already taken we remove it from
activeUsbHostdevs even if we should not. Thus, next time we allow
the device to be assigned to another domain.
Most versions of libselinux do not contain the function
selinux_lxc_contexts_path() that the security driver
recently started using for LXC. We must add a conditional
check for it in configure and then disable the LXC security
driver for builds where libselinux lacks this function.
* configure.ac: Check for selinux_lxc_contexts_path
* src/security/security_selinux.c: Disable LXC security
if selinux_lxc_contexts_path() is missing
Normal practice is for cgroups controllers to be mounted at
/sys/fs/cgroup. When setting up a container, /sys is mounted
with a new sysfs instance, thus we must re-mount all the
cgroups controllers. The complexity is that we must mount
them in the same layout as the host OS. ie if 'cpu' and 'cpuacct'
were mounted at the same location in the host we must preserve
this in the container. Also if any controllers are co-located
we must setup symlinks from the individual controller name to
the co-located mount-point
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Both /proc and /sys may have sub-mounts in them from the host
OS. We must explicitly unmount them all before mounting the
new instance over that location. If we don't then /proc/mounts
will show the sub-mounts as existing, even though nothing will
be able to access them, due to the over-mount.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If the LXC config has a filesystem
<filesystem>
<source dir='/'/>
<target dir='/'/>
</filesystem>
then there is no need to go down the pivot root codepath.
We can simply use the existing root as needed.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
