Peter Krempa [Tue, 19 Mar 2024 14:19:26 +0000 (15:19 +0100)]
virshtest: Reimplement 'virsh-output' test
Adapt the 'tests/virsh-output-commands' file from 'virsh-output' test as
a source. Apart from expanding the bash function to each command, I've
also had to drop the negative tests for argument population, as a
command parsing error aborts the execution of the script right away
rather than just reporting the error.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Mon, 18 Mar 2024 14:38:39 +0000 (15:38 +0100)]
virshtest: Adapt 'event' option arg handling tests from 'virsh-optparse'
Move the argument parsing tests excercising 'virsh event' options
from 'virsh-optparse' to 'virshtest'.
As the test invokes 'virsh event' with a timeout and thus waits for one
second pointlessly the patch also adds infrastructure to mark individual
cases as expensive and is skipped normally.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 20 Mar 2024 16:34:45 +0000 (17:34 +0100)]
virshtest: Replace list and nodeinfo tests by equivalent VIR_TEST_SCRIPT variant
Add the basic list and info commands into a script and run it via
VIR_TEST_RUN_SCRIPT to simplify the code and save up on 'virsh'
instances exec'd for the test.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Tue, 12 Mar 2024 12:38:24 +0000 (13:38 +0100)]
virshtest: Prepare for simpler testing - echo tests
Embedding the expected output in a C source code makes it very hard to
extend tests. In order to be able to test the outputs against data in
files on disk we need better naming of the tests themselves.
Use virTestCounterNext/Reset with appropriate tags to give reasonable
names to the 'virsh echo' tests' and prepare the 'DO_TEST' macro for
wider use.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 20 Mar 2024 16:45:56 +0000 (17:45 +0100)]
virshtest: Filter multiple occurences of string to drop in testFilterLine
Upcoming patches will require that possibly multiple occurences of the
string to drop are present in the output string thus we need to adapt
testFilterLine to handle them.
Additionally we drop the unused return value.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Tue, 12 Mar 2024 16:15:39 +0000 (17:15 +0100)]
virshtest: Allow to test failure of commands
Modify the test code so that if virsh fails both 'stdout' and 'stderr'
are captured and compared against the output and also the return value
is checked by appending it to the output.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Tue, 5 Mar 2024 21:37:43 +0000 (22:37 +0100)]
vsh: Make positional parsing of arguments opt-in
Switch the command parser from using the VSH_OFLAG_REQ_OPT flag
opting out from positional parsing of arguments to a combination of the
'positional' flags for truly positional arguments and
'unwanted_positional' preserving semantics for the existing arguments
where the parser did it due to bad design.
This patch retires VSH_OFLAG_REQ_OPT along with the infrastructure that
was needed to refactor all uses properly.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Historically the command parser in virsh parses/fills even optional
arguments with values as if they were positional unless opted out using
VSH_OFLAG_REQ_OPT. This creates unexpected situations when commands can
break in this unwanted semantics:
To prevent any further addition annotate the rest of the arguments with
the 'unwanted_positional' flag, so that the parser can keep parsing them
as such but any further optional argument will not have this behaviour.
Certain arguments where it makes sense are annotated as 'positional' too
in this patch.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 14 Mar 2024 16:17:19 +0000 (17:17 +0100)]
virsh: Annodate 'unwanted_positional' arguments
Historically the command parser in virsh parses/fills even optional
arguments with values as if they were positional unless opted out using
VSH_OFLAG_REQ_OPT. This creates unexpected situations when commands can
break in this unwanted semantics:
To prevent any further addition annotate the rest of the arguments with
the 'unwanted_positional' flag, so that the parser can keep parsing them
as such but any further optional argument will not have this behaviour.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 14 Mar 2024 14:34:03 +0000 (15:34 +0100)]
virsh: volume: Mark optional 'pool' argument as 'positional'
Annotate arguments which can be unintentionally parsed positionally.
(See previous commits for explanation.)
The pool name is optional but in all cases it can be promoted to an
optional positional argument so that it can be properly aligned with the
expectations of the parser.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 22:01:18 +0000 (23:01 +0100)]
virsh: snapshot: Make 'snapshotname' argument positional
The 'snapshotname' argument is optional as by default "current" snapshot
is considered. Regardless of that we should treat it as positional as
it's the common usage. This is now possible as we can have one optional
positional argument.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 14 Mar 2024 15:58:40 +0000 (16:58 +0100)]
vsh: Make the only argument of 'connect', 'cd', and 'help' commands positional
The intended use of those commands is to use the argument directly
without the flag. Since the argument is optional in all cases we
couldn't declare them as positional until now.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 14 Mar 2024 15:56:16 +0000 (16:56 +0100)]
virsh-checkpoint: Make 'checkpointname' positional and required
The argument was being parsed positionally due to the command parser
quirk as we didn't opt out of it.
Since the code in virshLookupCheckpoint requires that the checkpointname
is present we can mark all the options as positional and required and
remove the redundant check from virshLookupCheckpoint.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
While making virsh require the flags is technically a breaking change,
there were multiple instances where arguments were added to the argument
list thus changing the order the positional arguments would be
interpreted as. Examples are commits: 7e157858b4b, bc5a8090afa, ca21d75d25. As of such there are multiple breaks of compatibility for
the positional arguments.
As of such, require the option flag for all optional arguments with
value for 'virsh attach-disk'.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 21:52:12 +0000 (22:52 +0100)]
virsh: Fix "positional" argument annotations for 'migrate' command
Annotate arguments which can be unintentionally parsed positionally.
(See previous commits for explanation.)
Annotate '--migrateuri', '--graphicsuri', '--listen-address', '-dname',
'--timeout', '--xml', '--migrate-disks' and '--disks port' as
'unwanted_positional'. These were declared in chronological order per
git history.
All others are annotated with VSH_OFLAG_REQ_OPT which makes the parser
require the '--optionname'. This is due to the fact that '--disks-uri'
was introduced later and put in front of others declared earlier
breaking the order they would be accepted, thus changing the behaviour
between versions.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 21:50:02 +0000 (22:50 +0100)]
virsh: Require option flags for 'blkdeviotune' arguments
Make all of the tunable parameter flags require the option name (don't
parse them positionally).
While techically this would be a breaking change if anyone were to
specify the tunable values positionally this is not the case as the
first two tunables are not compatible with each other:
$ virsh blkdeviotune cd vda 4 5
error: Unable to change block I/O throttle
error: invalid argument: total and read/write of bytes_sec cannot be set at the same time
The above is produced by all implementations of the API (qemu and test
drivers). It is true that the first tunable can be specified
positionally (--total-bytes-sec) but it is misleading and shoud not be
allowed either.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 15:58:52 +0000 (16:58 +0100)]
vsh: Fix option formatting for 'VHS_OT_ARGV' options
While previous fixes kept the help output unchanged as base for the
refactors it turns out that the formatting of help for argv options is
wrong.
Specifically in SYNOPSIS the non-positional _ARGV would have the option
name in square brackets (which in other cases means that given thing is
optional) despite being required.
Similarly in the DESCRIPTION section positional versions would not show
the optional argument name and also didn't use the three dots to signal
that it can be used multiple times.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 21:44:44 +0000 (22:44 +0100)]
virsh: Annotate '--diskspec' _ARGV options as unwanted positional
Our documentation in most places explicitly mentions --diskspec and it
was never meant to be positional, although we can't change the parser
any more. Annotate them as 'unwanted_positional'.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 20:13:44 +0000 (21:13 +0100)]
vsh: Introduce tool to find unwanted positional arguments to 'self-test'
While the virsh option definitions specify (either explicitly after
recent refactors, or implicitly before) whether an argument is
positional or not, the actual parser is way more lax and actually and
allows also arguments which were considered/documented as non-positional
to be filled positionally unless VSH_OFLAG_REQ_OPT is used in the flags.
This creates situations such as 'snapshot-create-as' which has the
following docs:
In the above example e.g. '--memspec' is not populated.
This disconnect makes it impossible to refactor the parser itself and
allows users to write buggy interactions with virsh.
In order to address this we'll be annotating every single of these
unwanted positional options as such so that this doesn't happen in the
future, while still preserving the quirk in the parser.
This patch introduces a tool which outputs list of options which are not
marked as positional but are lacking the VSH_OFLAG_REQ_OPT flag.
This tool will be removed once all the offenders found by it will be
addressed.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Thu, 14 Mar 2024 09:31:17 +0000 (10:31 +0100)]
vshCmddefCheckInternals: Improve some checks
- move the check that completer_flags are 0 if no completer is set
into a common place and remove duplication
- add check that _BOOL arguments are not positional
- add missing checks to _ALIAS
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 13 Mar 2024 14:54:47 +0000 (15:54 +0100)]
vshCmddefHelp: Drop empty line at the end
All virsh commands in non-quiet mode append another separator line thus
having two is unnecessary and in quiet mode it still has a trailing
blank line. Remove it.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Pavel Hrdina [Wed, 21 Feb 2024 16:45:50 +0000 (17:45 +0100)]
qemu_snapshot: correctly update metadata when deleting external snapshot with multiple branches
XML metadata for snapshot contains only single list of disk overlays
from the moment when the snapshot was taken. When user creates multiple
branches of snapshots the parent snapshot will still list only the
original disk overlays. This may cause an issue in a specific scenario:
s1
|
+- s2
+- s3 (active)
For this snapshot topology when we delete s2 metadata for s1 are not
updated. Now when we delete s1 the code operated with incorrect
overlays from s1 metadata in order to update s3 metadata resulting in no
changes to s3 metadata.
Now when user tries to delete s3 it fails with following error:
error: Failed to delete snapshot s3
error: operation failed: snapshot VM disk source and parent disk source are not the same
For the actual deletion there is a code to figure out the correct disk
source but it was not used to update metadata as well. Due to reasons
how block commit in libvirt works we need to create a copy of that disk
source in order to have it available when updating metadata as the
original source will be freed at that point.
Resolves: https://issues.redhat.com/browse/RHEL-26276 Signed-off-by: Pavel Hrdina <phrdina@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Pavel Hrdina [Wed, 21 Feb 2024 16:45:08 +0000 (17:45 +0100)]
qemu_snapshot: call qemuSnapshotDeleteUpdateDisks only for external snapshots
Calling this function when deleting internal snapshot isn't required
because with internal snapshots all changes are done within the file
itself so there is no file deletion and no need to update snapshot
metadata.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Peter Krempa [Tue, 26 Mar 2024 14:06:58 +0000 (15:06 +0100)]
pci: Remove error reporting from PCI VPD parsing
The PCI VPD (Vital Product Data) may be missing or the kernel can report
presence but not actually have the data. Also the data is specified by
the device vendor and thus may be invalid in some cases.
To avoid log spamming, since the only usage in the node device driver is
ignoring errors, remove all error reporting.
Closes: https://gitlab.com/libvirt/libvirt/-/issues/607 Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Wed, 27 Mar 2024 12:52:54 +0000 (13:52 +0100)]
virpcivpd: Revert error reporting from PCI VPD parser
The VPD parsing is fragile and depends on hardware vendor's adherance to
standards. Since libvirt only ever uses this data to report it in the
nodedev XML which ignores any errors there's no much point in having
error reporting which I've added recently.
Turn the errors into VIR_DEBUG statements in preparation for upcoming
patch which completely removes the expectation to report errors.
Adam Julis [Fri, 22 Mar 2024 08:12:26 +0000 (09:12 +0100)]
qemuDomainChangeNet: Error when boot index changes in live XML
If the original code detected a missing or null boot index in the
new XML, it automatically added the current value. This
autocompletion was incorrect because it was impossible to
distinguish between user intent and user error - changing the
boot order itself is forbidden and should always be an error.
Resolves: https://issues.redhat.com/browse/RHEL-23416 Fixes: aa3e07caec6179dfa6479deab14a21a493637d53 Signed-off-by: Adam Julis <ajulis@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
remote: check for negative array lengths before allocation
While the C API entry points will validate non-negative lengths
for various parameters, the RPC server de-serialization code
will need to allocate memory for arrays before entering the C
API. These allocations will thus happen before the non-negative
length check is performed.
Passing a negative length to the g_new0 function will usually
result in a crash due to the negative length being treated as
a huge positive number.
This was found and diagnosed by ALT Linux Team with AFLplusplus.
CVE-2024-2494 Reviewed-by: Michal Privoznik <mprivozn@redhat.com> Found-by: Alexandr Shashkin <dutyrok@altlinux.org> Co-developed-by: Alexander Kuznetsov <kuznetsovam@altlinux.org> Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Andrea Bolognani [Tue, 28 Nov 2023 15:09:43 +0000 (16:09 +0100)]
qemu: Tweak augeas schema
Current entries should always be listed before obsolete ones.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Andrea Bolognani [Tue, 21 Nov 2023 17:20:32 +0000 (18:20 +0100)]
security: Drop virSecurity(DAC|SELinux)SetImageLabelRelative()
The single caller for each function passes the same value
for @src and @parent, which means that we don't really need
the additional API.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Andrea Bolognani [Mon, 20 Nov 2023 18:17:02 +0000 (19:17 +0100)]
security: Drop virSecurity(DAC|SELinux)RestoreImageLabelSingle()
Each one only has a single, trivial caller.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Andrea Bolognani [Tue, 21 Nov 2023 16:36:56 +0000 (17:36 +0100)]
security: Fix name for _virSecurityDACChardevCallbackData
It was clearly copied over from the SELinux driver without
updating its name in the process.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Andrea Bolognani [Tue, 21 Nov 2023 16:17:10 +0000 (17:17 +0100)]
security: Fix alignment
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Stefan Berger <stefanb@linux.ibm.com>
Following callbacks have been implemented
* domainRestore
* domainRestoreFlags
The path parameter to these callbacks has to be of the directory where
libvirt has performed save. Additionally, call restore in `domainCreate`
if the domain has managedsave.
Signed-off-by: Purna Pavan Chandra Aekkaladevi <paekkaladevi@linux.microsoft.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Following callbacks have been implemented
* domainSaveImageGetXMLDesc
* domainManagedSaveRemove
* domainManagedSaveGetXMLDesc
* domainHasManagedSaveImage
Signed-off-by: Purna Pavan Chandra Aekkaladevi <paekkaladevi@linux.microsoft.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Implemented save callbacks. CH's vmm.snapshot API is called to save the
domain state. The path passed to these callbacks has to be of directory
as CH takes dir as input to snapshot and saves multiple files under it.
Signed-off-by: Purna Pavan Chandra Aekkaladevi <paekkaladevi@linux.microsoft.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Andrea Bolognani [Tue, 27 Feb 2024 14:58:27 +0000 (15:58 +0100)]
meson: Check for sched_get_priority_min()
virProcessSetScheduler() uses not just sched_setscheduler() but
also sched_get_priority_{min,max}(). Currently we assume that
the former being available implies that the latter are as well,
but that's not the case for at least GNU/Hurd.
Make sure all functions are actually available before
attempting to use them.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Andrea Bolognani [Tue, 27 Feb 2024 14:52:15 +0000 (15:52 +0100)]
meson: Restore check for sched_getaffinity()
Commit c07cf0a68693 replaced this check with one for the
presence of cpu_set_t.
The idea at the time was that only sched_{get,set}affinity()
were visible by default, while making cpu_set_t visible required
defining _WITH_CPU_SET_T. So libvirt would detect the function
and attempt to use it, but the code would not compile because
the necessary data type had not been made accessible.
The commit in question brought three FreeBSD commits as evidence
of this. While [1] and [2] do indeed seem to support this
explanation, [3] from just a few days later made it so that not
just cpu_set_t, but also the functions, required user action to
be visible. This arguably would have made the change unnecessary.
However, [4] from roughly a month later changed things once
again: it completely removed _WITH_CPU_SET_T, making both the
functions and the data type visible by default.
This is the status quo that seems to have persisted until
today. If one were to check any recent FreeBSD build job
performed as part of our CI pipeline, for example [5] and [6]
for FreeBSD 13 and 14 respectively, they would be able to
confirm that in both cases cpu_set_t is detected as available.
Since there is no longer a difference between the availability
of the functions and that of the data type, go back to what we
had before.
This has the interesting side-effect of fixing a bug
introduced by the commit in question.
When detection was changed from the function to the data type,
most uses of WITH_SCHED_GETAFFINITY were replaced with uses of
WITH_DECL_CPU_SET_T, but not all of them: specifically, those
that decided whether qemuProcessInitCpuAffinity() would be
actually implemented or replaced with a no-op stub were not
updated, which means that we've been running the stub version
everywhere except on FreeBSD ever since.
The code has been copied to the Cloud Hypervisor driver in
the meantime, which is similarly affected. Now that we're
building the actual implementation, we need to add virnuma.h
to the includes.
As a nice bonus this also makes things work correctly on
GNU/Hurd, where cpu_set_t is available but
sched_{get,set}affinity() are non-working stubs.
Andrea Bolognani [Fri, 23 Feb 2024 00:29:28 +0000 (01:29 +0100)]
util: Accept TIDs for virProcess{Get,Set}Affinity() on BSD
Depending on the situation, the IDs that we pass to these
functions can be either referring to processes or threads.
Linux doesn't have separate interfaces for one or the other,
but FreeBSD does and we're explicitly telling it that the ID
refers to a process. When it refers to a thread instead, the
call will fail, and the VM will not be able to start.
Luckily, another possible choice is CPU_WHICH_TIDPID, which
makes things behave the same as Linux.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Rayhan Faizel [Tue, 19 Mar 2024 15:16:30 +0000 (16:16 +0100)]
qemu_command: Generate command line for MTP filesystem
The source tag sets the rootdir property of the device, which is
the directory exposed to the guest via the MTP device. The target
tag sets the desc property. This device supports read-only mode
as well. Like virtiofs, it does not support additional access
modes.
Signed-off-by: Rayhan Faizel <rayhan.faizel@gmail.com> Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>