Michal Privoznik [Mon, 28 Feb 2022 15:30:08 +0000 (16:30 +0100)]
lxcProcReadMeminfo: Fix case when @offset != 0
The idea behind lxcProcReadMeminfo() is that we read the host's
/proc/meminfo and copy it line by line producing the content for
container, changing only those lines we need. Thus, when a
process inside container opens the file and lseek()-s to a
different position (or reads the content in small chunks), we
mirror the seek in host's /proc/meminfo. But this doesn't work
really. We are not guaranteed to end up aligned on the beginning
of new line. It's better if we construct the new content and then
mimic seeking in it.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Mon, 28 Feb 2022 14:29:58 +0000 (15:29 +0100)]
lxcProcReadMeminfo: Drop @new_meminfo variable
In the lxcProcReadMeminfo() function we have @buffer variable
which is statically allocated and then @new_meminfo which is just
a pointer to the @buffer. This is needless, the @buffer can be
accessed directly.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Mon, 28 Feb 2022 14:21:21 +0000 (15:21 +0100)]
lxcProcReadMeminfo: Rename @fd to @fp
In lxcProcReadMeminfo() there's a variable named @fd which would
suggest it's type of int, but in fact it's type of FILE *. Rename
it to @fp to avoid confusion.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Tue, 22 Feb 2022 12:47:36 +0000 (13:47 +0100)]
lxcSetupFuse: Cleanup error paths
In the lxcSetupFuse() function there are multiple cleanup labels,
but with a bit of rewrite they can be joined into one 'error'
label. And while at it, set the @f argument only in the
successful path (currently is set in error case too).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Mon, 28 Feb 2022 11:53:39 +0000 (12:53 +0100)]
lxc_fuse: Prefer O_ACCMODE instead of & 3
In lxcProcOpen() we want to check whether the /proc/memfile is
being opened only for read. For that we check the fi->flags which
correspond to flags open() call. Instead of explicitly masking
the last two bits use O_ACCMODE constant, which is deemed to be
more portable.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Mon, 28 Feb 2022 11:53:25 +0000 (12:53 +0100)]
lxc_fuse: Drop some G_GNUC_UNUSED attributes
There are few arguments that are marked as G_GNUC_UNUSED even
though they are clearly used within their respective functions.
Drop the annotation in such cases.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Tue, 22 Feb 2022 11:51:32 +0000 (12:51 +0100)]
lxc_fuse.h: Don't include lxc_conf.h
Nothing in the lxc_fuse.h header file warrants inclusion of
lxc_conf.h. If anything, virconftypes.h must be included because
of virDomainDef required by lxcSetupFuse().
It's actually lxc_fuse.c that requires some macros from
lxc_fuse.h (e.g. LXC_STATE_DIR).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
When libc uses a define to rewrite stat64 to stat our mocks do not work if they
are chained because the symbol that we are looking up is being stringified and
therefore preventing the stat64->stat expansion per C-preprocessor rules. One
stringification macro is just enough to make it work.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
The returned packet can have less strict alignment (u_char) than the struct
(ether_header) we are casting it to, so to avoid alignment issues just copy the
header into the struct on the stack.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
syntax-check: Rework mock-noinline to get all files at once
The script can break if the number of files does not fit one invocation and
xargs has to split it. Instead pipe the list of files directly into the script
and in the script read them from stdin instead of the arguments.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
sockettest: Check for IPv4-in-IPv6 parsing and formatting
There are two standards how IPv4 address in IPv6 can be
expressed:
::10.1.2.3
::ffff:10.1.2.3
The former is obsolete and the latter should be used instead [1].
Add test cases to our sockettest to exercise parsing/formatting
of the valid address format.
Apparently clang was fixed as it no longer considers having
global variables static a problem. Make the variables static to
be sure they aren't used outside of the source file.
This effectively reverts v1.0.6-rc1~198 which started the trend.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
vircgroupmock: Be wiser about detecting fakerootdir change
The way that vircgroupmock works is that the vircgrouptest
creates a temporary directory and sets LIBVIRT_FAKE_ROOT_DIR env
variable which is then checked by the mock at the beginning of
basically every function it overrides (access(), stat in all its
flavours, mkdir(), etc.). The mock then creates a CGroup dir
structure. But the test is allowed to change the directory, to
accommodate environment for the particular test case. This is
done by changing the environment variable which is then detected
by the mock and the whole process repeats.
However, the way the mock detect changes is buggy. After it got
the environment variable it compares it to the last known value
(global variable @fakerootdir) and if they don't match the last
known value is set to point to the new value. Problem is that the
result of getenv() is assigned to the @fakerootdir directly.
Therefore, @fakerootdir points somewhere into the buffer of
environment variables. In turn, when the test sets new value (via
g_setenv()) it may be placed at the very same position in the env
var buffer and thus the mock fails to detect the change.
The solution is to keep our private copy of the value (by
g_strdup()) which makes the variable not rely on
getenv()/setenv() placing values at random positions.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peng Liang [Wed, 2 Mar 2022 09:22:05 +0000 (17:22 +0800)]
rpc: Fix memory leak of fds
In virSystemdActivationClaimFDs, the memory of ent->fds has been stolen
and stored in fds, but fds is never freed, which causes a memory leak.
Fix it by declaring fds as g_autofree.
Reported-by: Jie Tang <tangjie18@huawei.com> Signed-off-by: Peng Liang <liangpeng10@huawei.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Michal Privoznik [Thu, 17 Feb 2022 11:51:27 +0000 (12:51 +0100)]
libxl: Turn on user aliases
When I implemented user aliases I've invented this
virDomainDefFeatures flag so that individual drivers can signal
support for user provided aliases. The reasoning was that a
device alias might be part of guest ABI, or used in a different
way then in QEMU. Well, neither applies to the libxl driver, so
it's safe to allow user aliases there.
Resolves: https://gitlab.com/libvirt/libvirt/-/issues/231 Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Peter Krempa [Wed, 2 Mar 2022 12:47:09 +0000 (13:47 +0100)]
virsh: cmdDesc: Fix logic when '-edit' is used along with 'desc' argument
Historically the use of the '-desc' multiple argument parameter was not
forbidden toghether with '-edit', but use of both together has some
unexpected behaviour. Specifically the editor is filled with the
contents passed via '-desc' but if the user doesn't change the text in
any way virsh will claim that the description was not chaged even if it
differs from the currently set description. Similarly, when the user
would edit the description provided via 'desc' so that it's identical
with the one configured for the domain, virsh would claim that it was
updated:
# virsh desc cd
No description for domain: cd
# EDITOR=true virsh desc cd --edit "test desc"
Domain description not changed
After the fix:
# virsh desc cd
No description for domain: cd
# EDITOR=true virsh desc cd --edit "test desc"
Domain description updated successfully
# EDITOR=true virsh desc cd --edit "test desc"
Domain description not changed
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Mon, 28 Feb 2022 16:59:40 +0000 (17:59 +0100)]
virshGetOneDisplay: Refactor formatting of URI params
Unconditionally format the start of the query ('?') and make delimiters
('&') part of the arguments. At the end we can trim off 1 char from the
end of the buffer unconditionally.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
The code registering the event handlers in 'cmdEvent' had too many
blocks of code conditional on whether just one event is being listened
to or all events.
The code can be greatly simplified by uniting the code paths and having
only one branch when filling the list of events we want to listen for.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Mon, 21 Feb 2022 13:02:19 +0000 (14:02 +0100)]
virsh: Move 'cmdEvent' and all of its machinery to virsh-domain-event.c
'cmdEvent' along with all the helper functions it needs is ~950 LOC.
Move it out from virsh-domain.c to virsh-domain-event.c along with the
completer function so that the new module doesn't have to expose any new
types.
Semantically this creates a new category in 'virsh help' but all other
behaviour stays the same.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Mon, 21 Feb 2022 15:12:59 +0000 (16:12 +0100)]
virsh: Move 'virshDomainBlockJobToString' to virsh-util
The helper function is used in virshBlockJobInfo and also in the
callbacks of cmdEvent. Upcoming patch is going to move out the event
code into a helper so this needs to be in a shared place.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Add support for sending one FD from the client along with a monitor
command so that it's possible to use 'getfd' and 'add-fd' to use FDs
passed from the client with other QMP commands.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 17 Feb 2022 16:07:39 +0000 (17:07 +0100)]
virnetmessage: Introduce virNetMessageClearFDs
The helper splits out the clearing of the FDs transacted inside a
virNetMessage.
APIs transacting FDs both from and to the client at the same time will
need to clear the FDs stored in virNetMessage as the structure is
re-used for the reply and without clearing the list of FDs we'd return
the FDs sent by the client in addition to the new FDs sent by the API.t
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 23 Feb 2022 13:04:47 +0000 (14:04 +0100)]
qemu: block copy: Propagate 'detect_zeroes' properly into mirror definition
'qemuDomainPrepareDiskSourceData' propagates 'detect_zeroes' only for
the disk source image, but the mirror destination has the ambition to
replace the disk source when the job is finished, so we need to
propagate the 'detect_zeroes' setting also in that case.
Unfortunately it would become very hairy to either set 'disk->mirror'
sooner or propagate that we want this done into
'qemuDomainPrepareDiskSourceData', so the most straightforward solution
is to do the propagation inside 'qemuDomainBlockCopyCommon'.
Closes: https://gitlab.com/libvirt/libvirt/-/issues/277 Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
qemu: support firmware descriptor flash 'mode' for optional NVRAM
Currently the 'nvram_template' entry is mandatory when parsing the
firmware descriptor based on flash. QEMU is extending the firmware
descriptor spec to make the 'nvram_template' optional, depending
on the value of a new 'mode' field:
When undefining a VM, we must optionally delete any NVRAM that might
exist. When using firmware auto-select we always check the generated
path, ignoring any user specified path.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
tests: don't use auto-generated NVRAM path in tests
By using the auto-generated NVRAM path in test data files, we won't see
bugs where a user specified path gets accidentally overwritten by a
post-parse callback, or VM startup. For example, this caused us to miss
the bug fixed by:
Michal Privoznik [Wed, 23 Feb 2022 07:50:44 +0000 (08:50 +0100)]
qemu: Don't regenerate NVRAM path if parsed from domain XML
After v8.0.0-466-g08101bde5d we unconditionally regenerate per
domain NVRAM path even though it might have been parsed earlier
from domain XML. The way we do that leads to a memleak:
43 bytes in 1 blocks are definitely lost in loss record 330 of 682
at 0x483F7E5: malloc (vg_replace_malloc.c:381)
by 0x50D5B18: g_malloc (in /usr/lib64/libglib-2.0.so.0.7000.2)
by 0x50EFA4F: g_strdup (in /usr/lib64/libglib-2.0.so.0.7000.2)
by 0x49E774E: virXPathString (virxml.c:88)
by 0x4A3F0E4: virDomainDefParseBootLoaderOptions (domain_conf.c:18226)
by 0x4A3F49C: virDomainDefParseBootOptions (domain_conf.c:18298)
by 0x4A448C3: virDomainDefParseXML (domain_conf.c:19598)
by 0x4A487A1: virDomainDefParseNode (domain_conf.c:20404)
by 0x117FCF: testCompareXMLToArgv (qemuxml2argvtest.c:726)
by 0x142124: virTestRun (testutils.c:142)
by 0x1423D4: virTestRunLog (testutils.c:197)
by 0x140A76: mymain (qemuxml2argvtest.c:3406)
If we parsed NVRAM path from domain XML we must refrain from
generating new path.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Peter Krempa [Tue, 22 Feb 2022 16:34:46 +0000 (17:34 +0100)]
qemu: blockjob: Avoid spurious log errors when cancelling a shallow copy with reused images
In case when a user starts a block copy operation with
VIR_DOMAIN_BLOCK_COPY_SHALLOW and VIR_DOMAIN_BLOCK_COPY_REUSE_EXT and
both the reused image and the original disk have a backing image libvirt
specifically does not insert the backing image until after the job is
asked to be completed via virBlockJobAbort with
VIR_DOMAIN_BLOCK_JOB_ABORT_PIVOT.
This is so that management applications can copy the backing image on
the background.
Now when a user aborts the block job instead of cancelling it we'd
ignore the fact that we didn't insert the backing image yet and the
cancellation would result into a 'blockdev-del' of a invalid node name
and thus an 'error' severity entry in the log.
To solve this issue we use the same conditions when the backing image
addition is avoided to remove the internal state for them prior to the
call to unplug the mirror destination.
Reported-by: Kashyap Chamarthy <kchamart@redhat.com> Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Michal Privoznik [Tue, 22 Feb 2022 08:02:17 +0000 (09:02 +0100)]
qemu: Don't ignore failure when building default memory backend
When building the default memory backend (which has id='pc.ram')
and no guest NUMA is configured then
qemuBuildMemCommandLineMemoryDefaultBackend() is called. However,
its return value is ignored which means that on invalid
configuration (e.g. when non-existent hugepage size was
requested) an error is reported into the logs but QEMU is started
anyway. And while QEMU does error out its error message doesn't
give much clue what's going on:
qemu-system-x86_64: Memory backend 'pc.ram' not found
While at it, introduce a test case. While I could chose a nice
looking value (e.g. 4MiB) that's exactly what I wanted to avoid,
because while such value might not be possible on x84_64 it may
be possible on other arches (e.g. ppc is notoriously known for
supporting wide range of HP sizes). Let's stick with obviously
wrong value of 5MiB.
Reported-by: Charles Polisher <chas@chasmo.org> Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
qemu: inline code for filling in per-VM NVRAM path
Before creating a NVRAM path, the qemuDomainNVRAMPathGenerate
method checks whether the config is using the old style
firmware approach. This check is redundant in one of the two
callers. By inlining the check into the other caller, it makes
it clearer to understand that the NVRAM path filling is done
conditionally.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Since libvirt-guests script/service can operate on various URIs and we do
support both socket activation and traditional services, the ordering should be
specified for all the possible sockets and services.
Also remove the Wants= dependency since do not want to start any service. We
cannot know which one libvirt-guests is configured, so we'd have to start all
the daemons which would break if unused colliding services are not
masked (libvirtd.service in the modular case and all the modular daemon service
units in the monolithic scenario). Fortunately we can assume that the system is
configured properly to start services/sockets that are of interest to the user.
That also works with the setup described in https://libvirt.org/daemons.html .
To make it even more robust we add the daemon service into the machine units
created for individual domains as it was missing there.
Michal Privoznik [Mon, 21 Feb 2022 14:59:18 +0000 (15:59 +0100)]
qemu: Move some enums impl to qemu_monitor.c
There are some enums that are declared in qemu_monitor.h but
implemented in qemu_monitor_json.c. While from compiler and
linker POV it doesn't matter, the code is cleaner if an enum is
implemented in .c file that corresponds to .h file which declared
the enum.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Our last default template had a match of "node()" which incidentally matched
everything, including text nodes. Since this has the same priority according to
the XSLT spec, section 5.5:
runtime error: file ../docs/page.xsl line 223 element element
xsl:element: The effective name '' is not a valid QName.
because our last rule also matches text nodes and we are trying to extract the
node name out of them.
To fix this we change the match to "*" which only matches elements and not all
the nodes, and to avoid any possible errors with different XSLT processors we
also bump the priority of the match="text()" rule a little higher, just in case
someone needs to use an XSLT processor that chooses signalling the error instead
of the optional recovery.
https://bugs.gentoo.org/833586
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Michal Privoznik [Mon, 21 Feb 2022 13:04:43 +0000 (14:04 +0100)]
qemu: Use virDomainObjCheckActive() more
Using the following spatch, I've identified two places which
could be switched from explicit virDomainObjIsActive() +
virReportError() to virDomainObjCheckActive():
@@
expression dom;
@@
if (
- !virDomainObjIsActive(dom)
+ virDomainObjCheckActive(dom) < 0
) {
- virReportError(VIR_ERR_OPERATION_INVALID, "%s", _("domain is not running"));
...
}
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Hyman Huang(黄勇) [Sun, 20 Feb 2022 13:28:15 +0000 (21:28 +0800)]
qemu_driver: Add calc_mode for dirtyrate statistics
Add calc_mode for dirtyrate statistics retured by
virsh domstats --dirtyrate api, also add vcpu dirtyrate
if dirty-ring mode was used in last measurement.
Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn> Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Hyman Huang(黄勇) [Sun, 20 Feb 2022 13:28:14 +0000 (21:28 +0800)]
virsh: Add mode option to domdirtyrate-calc virsh api
Extend domdirtyrate-calc virsh api with mode option, either
of these three options "page-sampling,dirty-bitmap,dirty-ring"
can be specified when calculating dirty page rate.
Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn> Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Hyman Huang(黄勇) [Sun, 20 Feb 2022 13:28:13 +0000 (21:28 +0800)]
qemu_driver: Extend flags parameter of virDomainStartDirtyRateCalc
Extend flags parameter of virDomainStartDirtyRateCalc as a
superset of virDomainDirtyRateCalcFlags, parse the flags and
handle it correspondingly in qemuDomainStartDirtyRateCalc.
Signed-off-by: Hyman Huang(黄勇) <huangy81@chinatelecom.cn> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Jim Fehlig [Thu, 17 Feb 2022 18:48:13 +0000 (11:48 -0700)]
libxl: Fix libvirtd crash on domain restore
Commit cc2a3c2a94 missed one case in the libxl driver where virDomainDef
is returned from libxlDomainSaveImageOpen and a g_steal_pointer is needed.
Without it, the virDomainDef object is freed and the driver crashes later
in the restore process when accessing the object.
Signed-off-by: Jim Fehlig <jfehlig@suse.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 27 Jan 2022 16:14:32 +0000 (17:14 +0100)]
cmdStartGetFDs: Modernize
Calculate the length of the FD list beforehand to avoid multiple
expansions and mainly simplify the code and use automatic freeing to
remove the error code path.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
The systemd version in RHEL-7 lacked support for the LISTEN_FDNAMES env
variable with socket activation. Since we stopped targetting RHEL-7 we
can drop some considerable amount of compatibility code.
Reviewed-by: Ján Tomko <jtomko@redhat.com> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
projects / libvirt.git / log