While the patch gives bdrv_graph_wrlock() as an example where the
issue can manifest, something similar can happen even when that is
disabled. Was able to reproduce the issue with
while true; do qm resize 115 scsi0 +4M; sleep 1; done
while running
fio --name=make-mirror-work --size=100M --direct=1 --rw=randwrite \
--bs=4k --ioengine=psync --numjobs=5 --runtime=1200 --time_based
in the VM.
Fix picked up from:
https://lists.nongnu.org/archive/html/qemu-devel/2023-12/msg01102.html
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
When using iothread, after commits
1665d9326f ("virtio-blk: implement BlockDevOps->drained_begin()")
766aa2de0f ("virtio-scsi: implement BlockDevOps->drained_begin()")
it can happen that polling gets stuck when draining. This would cause
IO in the guest to get completely stuck.
A workaround for users is stopping and resuming the vCPUs because that
would also stop and resume the dataplanes which would kick the host
notifiers.
This can happen with block jobs like backup and drive mirror as well
as with hotplug [2].
Reports in the community forum that might be about this issue[0][1]
and there is also one in the enterprise support channel.
As a workaround in the code, just re-enable notifications and kick the
virt queue after draining. Draining is already costly and rare, so no
need to worry about a performance penalty here. This was taken from
the following comment of a QEMU developer [3] (in my debugging,
I had already found re-enabling notification to work around the issue,
but also kicking the queue is more complete).
[0]: https://forum.proxmox.com/threads/137286/
[1]: https://forum.proxmox.com/threads/137536/
[2]: https://issues.redhat.com/browse/RHEL-3934
[3]: https://issues.redhat.com/browse/RHEL-3934?focusedId=23562096&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-23562096
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
The implementation of the helper is_path_tmpfs() is similar to the
existing qemu_fd_getfs() function in util/mmap-alloc.c, which
unfortunately only takes an existing fd.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Bigger notable changes:
* Commit 1a30b0f5d7 ("block: .bdrv_open is non-coroutine and
unlocked") broke the PVE backup patches, in particular setting up
the backup dump block driver, because bdrv_new_open_driver() cannot
be called from a coroutine. To fix it, bdrv_co_open() is used
instead, and while it's a much more involved function, the result
should be essentially the same. The only difference I noticed is
that the BDRV_O_ALLOW_RDWR flag is also set in the resulting bds
(block driver state), but that shouldn't hurt.
Smaller notable changes:
* aio_set_fd_handler() dropped its 'is_external' parameter stating
that all callers now pass false in 60f782b6b7 ("aio: remove
aio_disable_external() API"). The calls in the PVE patches also
passed false, so just drop the parameter too.
* global_state_store() does not have a return value anymore, so the
user in the PVE savevm-async patch was adapted. For context, see
c33f1829f8 ("migration: never fail in global_state_store()").
* Renames affecting the PVE savevm-async patch:
migrate_use_block() -> migrate_block() and ram_counters -> mig_stats
9d4b1e5f22 ("migration: Move migrate_use_block() to options.c")
aff3f6606d ("migration: Rename ram_counters to mig_stats")
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Since upstream QEMU 8.0, it's no longer possible to call
bdrv_img_create() from a coroutine anymore, meaning a backup with the
directory format would crash the QEMU instance.
The feature is only exposed via the monitor and was intended to be
experimental. There were no user reports about the breakage and it
only was noticed during the rebase for QEMU 8.1, because other parts
of the backup code needed adaptation and I decided to check the
BACKUP_FORMAT_DIR case too.
It should not stay in a broken state of course, but avoid the
maintenance cost and just make it a removed feature for Proxmox VE 8
retroactively.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
With the drive-backup QMP command, upstream QEMU uses a drained
section for the source drive when creating the backup job. Do the same
here to avoid subtle bugs.
There, the drained section extends until after the job is started, but
this cannot be done here for multi-disk backups (could at most start
the first job). The important thing is that the cbw
(copy-before-write) node is in place and the bcs (block-copy-state)
bitmap is initialized, which both happen during job creation (ensured
by the "block/backup: move bcs bitmap initialization to job creation"
PVE patch).
One such bug is one reported in the community forum [0], where using a
drive with iothread can lead to an overlapping block-copy request and
consequently an assertion failure. The block-copy code relies on the
bcs bitmap to determine if a request for a certain range can be
created. Each time a request is created, it resets the bcs bitmap at
that range to indicate that it's being handled.
The duplicate request can happen as follows:
Thread A attaches the cbw node
Thread B creates a request and resets the bitmap at that range
Thread A clears the bitmap and merges it with the PBS bitmap
The merging can lead to the bitmap being set again at the range of
the previous request, so the block-copy code thinks it's fine to
create a request there.
Thread B creates another requests at an overlapping range before the
other request is finished.
The drained section ensures that nothing else can interfere with the
bcs bitmap between attaching the copy-before-write block node and
initialization of the bitmap.
[0]: https://forum.proxmox.com/threads/133149/
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Add a filter to the "vma extract" command. A comma seperated list of
disk images that should be extracted can be passed with the "-d" option.
Example to extract an IDE drive and an SCSI drive from vzdump.vma:
vma extract vzdump.vma -d "drive-ide0,drive-scsi0" extractdir
Signed-off-by: Filip Schauer <f.schauer@proxmox.com>
Reported in the community forum [0]. By default, there can be large
amounts of memory left assigned to the QEMU process after backup.
Likely because of fragmentation, it's necessary to explicitly call
malloc_trim() to tell glibc that it shouldn't keep all that memory
resident for the process.
QEMU itself already does a malloc_trim() in the RCU thread, but that
code path might not be reached (or not for a long time) under usual
operation. The value of 4 MiB for the argument was also copied from
there.
Example with the following configuration:
> agent: 1
> boot: order=scsi0
> cores: 4
> cpu: x86-64-v2-AES
> ide2: none,media=cdrom
> memory: 1024
> name: backup-mem
> net0: virtio=DA:58:18:26:59:9F,bridge=vmbr0,firewall=1
> numa: 0
> ostype: l26
> scsi0: rbd:base-107-disk-0/vm-106-disk-1,size=4302M
> scsihw: virtio-scsi-pci
> smbios1: uuid=b2d4511e-8d01-44f1-afd6-9581b30c24a6
> sockets: 2
> startup: order=2
> virtio0: lvmthin:vm-106-disk-1,iothread=1,size=1G
> virtio1: lvmthin:vm-106-disk-2,iothread=1,size=1G
> virtio2: lvmthin:vm-106-disk-3,iothread=1,size=1G
> vmgenid: 0a1d8751-5e02-449d-977e-c0160e900231
Before the change:
> root@pve8a1 ~ # grep VmRSS /proc/$(cat /var/run/qemu-server/106.pid)/status
> VmRSS: 370948 kB
> root@pve8a1 ~ # vzdump 106 --storage pbs
> (...)
> INFO: Backup job finished successfully
> root@pve8a1 ~ # grep VmRSS /proc/$(cat /var/run/qemu-server/106.pid)/status
> VmRSS: 2114964 kB
After the change:
> root@pve8a1 ~ # grep VmRSS /proc/$(cat /var/run/qemu-server/106.pid)/status
> VmRSS: 398788 kB
> root@pve8a1 ~ # vzdump 106 --storage pbs
> (...)
> INFO: Backup job finished successfully
> root@pve8a1 ~ # grep VmRSS /proc/$(cat /var/run/qemu-server/106.pid)/status
> VmRSS: 424356 kB
[0]: https://forum.proxmox.com/threads/131339/
Co-diagnosed-by: Friedrich Weber <f.weber@proxmox.com>
Co-diagnosed-by: Dominik Csapak <d.csapak@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Acked-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Add format attributes to functions that take printf-like arguments. This
provides additional compile-time checking that the correct parameters
are passed to the functions.
This fixes compiler warnings generated by the -Wsuggest-attribute=format
flag.
Signed-off-by: Filip Schauer <f.schauer@proxmox.com>
Not difficult to run into, just have a drive with iothread, take a PBS
backup and then take a snapshot or hibernate. Resuming will fail with
> qemu: qemu_mutex_unlock_impl: Operation not permitted
because of not acquiring the correct AioContext first.
Migration is not affected, because it runs in coroutine context.
Reported in the community forum:
https://forum.proxmox.com/threads/129899/
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
The QAPI change for QEMU 8.0 dropped redundant has_foo parameters, but
in the blockdev_mirror_common() function (which is not part of the
QAPI itself but called from there) the argument pair was has_bitmap
and bitmap_name rather than has_bitmap and bitmap.
Reported-by: Aaron Lauterer <a.lauterer@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
where there is no good reason to keep them separate. It's a pain
during rebase if there are multiple patches changing the same code
over and over again. This was especially bad for the backup-related
patches. If the history of patches really is needed, it can be
extracted via git. Additionally, compilation with partial application
of patches was broken since a long time, because one of the master key
changes became part of an earlier patch during a past rebase.
If only the same files were changed by a subsequent patch and the
changes felt to belong together (obvious for later bug fixes, but also
done for features e.g. adding master key support for PBS), the patches
were squashed together.
The PBS namespace support patch was split into the individual parts
it changes, i.e. PBS block driver, pbs-restore binary and QMP backup
infrastructure, and squashed into the respective patches.
No code change is intended, git diff in the submodule should not show
any difference between applying all patches before this commit and
applying all patches after this commit.
The query-proxmox-support QMP function has been left as part of the
"PVE-Backup: Proxmox backup patches for QEMU" patch, because it's
currently only used there. If it ever is used elsewhere too, it can
be split out from there.
The recent alloc-track and BQL-related savevm-async changes have been
left separate for now, because it's not 100% clear they are the best
approach yet. This depends on what upstream decides about the BQL
stuff and whether and what kind of issues with the changes pop up.
The qemu-img dd snapshot patch has been re-ordered to after the other
qemu-img dd patches.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Namely, pvebackup_co_prepare() needs to call bdrv_co_open() rather
than bdrv_open(), because it is a coroutine itself.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Changes to other patches are all just metadata/context changes except
for pvebackup_co_prepare() needing to call bdrv_co_unref() rather than
bdrv_unref(), because it is a coroutine itself. This is documented in
d6ee2e324e ("block-coroutine-wrapper: Introduce no_co_wrapper"). The
change is necessary, because one of the stable fixes converts
bdrv_unref and blk_unref into no_co_wrappers (in preparation for a
second patch to fix a hang with the block resize QMP command).
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Many changes were necessary this time around:
* QAPI was changed to avoid redundant has_* variables, see commit
44ea9d9be3 ("qapi: Start to elide redundant has_FOO in generated C")
for details. This affected many QMP commands added by Proxmox too.
* Pending querying for migration got split into two functions, one to
estimate, one for exact value, see commit c8df4a7aef ("migration:
Split save_live_pending() into state_pending_*") for details. Relevant
for savevm-async and PBS dirty bitmap.
* Some block (driver) functions got converted to coroutines, so the
Proxmox block drivers needed to be adapted.
* Alloc track auto-detaching during PBS live restore got broken by
AioContext-related changes resulting in a deadlock. The current, hacky
method was replaced by a simpler one. Stefan apparently ran into a
problem with that when he wrote the driver, but there were
improvements in the stream job code since then and I didn't manage to
reproduce the issue. It's a separate patch "alloc-track: fix deadlock
during drop" for now, you can find the details there.
* Async snapshot-related changes:
- The pending querying got adapted to the above-mentioned split and
a patch is added to optimize it/make it more similar to what
upstream code does.
- Added initialization of the compression counters (for
future-proofing).
- It's necessary the hold the BQL (big QEMU lock = iothread mutex)
during the setup phase, because block layer functions are used there
and not doing so leads to racy, hard-to-debug crashes or hangs. It's
necessary to change some upstream code too for this, a version of
the patch "migration: for snapshots, hold the BQL during setup
callbacks" is intended to be upstreamed.
- Need to take the bdrv graph read lock before flushing.
* hmp_info_balloon was moved to a different file.
* Needed to include a new headers from time to time to still get the
correct functions.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
In qemu-server, we already allocate 2 * $mem_size + 500 MiB for driver
state (which was 32 MiB long ago according to git history). It seems
likely that the 30 MiB cutoff in the savevm-async implementation was
chosen based on that.
In bug #4476 [0], another issue caused the iteration to not make any
progress and the state file filled up all the way to the 30 MiB +
pending_size cutoff. Since the guest is not stopped immediately after
the check, it can still dirty some RAM and the current cutoff is not
enough for a reproducer VM (was done while bug #4476 still was not
fixed), dirtying memory with
> stress-ng -B 2 --bigheap-growth 64.0M'
After entering the final stage, savevm actually filled up the state
file completely, leading to an I/O error. It's probably the same
scenario as reported in the bug report, the error message was fixed in
commit a020815 ("savevm-async: fix function name in error message")
after the bug report.
If not for the bug, the cutoff will only be reached by a VM that's
dirtying RAM faster than can be written to the storage, so increase
the cutoff to 100 MiB to have a bigger chance to finish successfully,
while still trying to not increase downtime too much for
non-hibernation snapshots.
[0]: https://bugzilla.proxmox.com/show_bug.cgi?id=4476
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
when pend_postcopy is large. By definition, pend_postcopy won't
decrease when iterating, so a value larger than the cutoff of 400000
would lead to essentially empty iterations, filling up the state file
until only 30 MiB + pending_size remain and the second half of the
check would trigger.
Avoid this, by not considering pend_postcopy for the cutoff to enter
the final phase.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Avoids a patch and is required to compile when not all patches are
applied. No functional change is intended.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
User-facing breaking change:
The slirp submodule for user networking got removed. It would be
necessary to add the --enable-slirp option to the build and/or install
the appropriate library to continue building it. Since PVE is not
explicitly supporting it, it would require additionally installing the
libslirp0 package on all installations and there is *very* little
mention on the community forum when searching for "slirp" or
"netdev user", the plan is to only enable it again if there is some
real demand for it.
Notable changes:
* The big change for this release is the rework of job locking, using
a job mutex and introducing _locked() variants of job API functions
moving away from call-side AioContext locking. See (in the qemu
submodule) commit 6f592e5aca ("job.c: enable job lock/unlock and
remove Aiocontext locks") and previous commits for context.
Changes required for the backup patches:
* Use WITH_JOB_LOCK_GUARD() and call the _locked() variant of job
API functions where appropriate (many are only availalbe as
a _locked() variant).
* Remove acquiring/releasing AioContext around functions taking the
job mutex lock internally.
The patch introducing sequential transaction support for jobs needs
to temporarily unlock the job mutex to call job_start() when
starting the next job in the transaction.
* The zeroinit block driver now marks its child as primary.
The documentation in include/block/block-common.h states:
> Filter node has exactly one FILTERED|PRIMARY child, and may have
> other children which must not have these bits
Without this, an assert will trigger when copying to a zeroinit target
with qemu-img convert, because bdrv_child_cb_attach() expects any
non-PRIMARY child to be not FILTERED:
> qemu-img convert -n -p -f raw -O raw input.raw zeroinit:output.raw
> qemu-img: ../block.c:1476: bdrv_child_cb_attach: Assertion
> `!(child->role & BDRV_CHILD_FILTERED)' failed.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
The documentation in include/io/channel.h states that -1 or
QIO_CHANNEL_ERR_BLOCK should be returned upon error. Simply passing
along the return value from the blk-functions has the potential to
confuse the call sides. Non-blocking mode is not implemented
currently, so -1 it is.
The "return ret" was mistakenly left over from the previous
QEMUFileOps based implementation. Also, use error_setg_errno(), since
the blk(_co)_p{readv,writev} functions return errno codes.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
to be in-line with what other implementations in QEMU do. Commit
1d39c7098bbfa6862cb96066c4f8f6735ea397c5 mentions the EIO bit and
the function is expected to return 0 upon success (see other
implementations).
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Notable changes:
* The only big change is the switch to using a custom QIOChannel for
savevm-async, because the previously used QEMUFileOps was dropped.
Changes to the current implementation:
* Switch to vector based methods as required for an IO channel. For
short reads the passed-in IO vector is stuffed with zeroes at the
end, just to be sure.
* For reading: The documentation in include/io/channel.h states that
at least one byte should be read, so also error out when whe are
at the very end instead of returning 0.
* For reading: Fix off-by-one error when request goes beyond end.
The wrong code piece was:
if ((pos + size) > maxlen) {
size = maxlen - pos - 1;
}
Previously, the last byte would not be read. It's actually
possible to get a snapshot .raw file that has content all the way
up the final 512 byte (= BDRV_SECTOR_SIZE) boundary without any
trailing zero bytes (I wrote a script to do it).
Luckily, it didn't cause a real issue, because qemu_loadvm_state()
is not interested in the final (i.e. QEMU_VM_VMDESCRIPTION)
section. The buffer for reading it is simply freed up afterwards
and the function will assume that it read the whole section, even
if that's not the case.
* For writing: Make use of the generated blk_pwritev() wrapper
instead of manually wrapping the coroutine to simplify and save a
few lines.
* Adapt to changed interfaces for blk_{pread,pwrite}:
* a9262f551e ("block: Change blk_{pread,pwrite}() param order")
* 3b35d4542c ("block: Add a 'flags' param to blk_pread()")
* bf5b16fa40 ("block: Make blk_{pread,pwrite}() return 0 on success")
Those changes especially affected the qemu-img dd patches, because
the context also changed, but also some of our block drivers used
the functions.
* Drop qemu-common.h include: it got renamed after essentially
everything was moved to other headers. The only remaining user I
could find for things dropped from the header between 7.0 and 7.1
was qemu_get_vm_name() in the iscsi-initiatorname patch, but it
already includes the header to which the function was moved.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Without this change, it's necessary to send a second savevm-end QMP
command after aborting a snaphsot, before a new savevm-start QMP
command can succeed.
In process_savevm_finalize(), no longer set an error in the abort
scenario. If there already is another error, there's no need to
override it. If canceling was done intentionally, qmp_savevm_end()
is responsible for setting the state now.
Reported-by: Mira Limbeck <m.limbeck@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Reported in the community forum[0].
For 6.1.0, there were a few changes to the coroutine-sleep API, but
the adaptations in f376b2b ("update and rebase to QEMU v6.1.0") made
a mistake.
Currently, target_close_wait is NULL when passed to
qemu_co_sleep_ns_wakeable(), which further passes it to
qemu_co_sleep(), but there, it is dereferenced when trying to access
the 'to_wake' member:
> Thread 1 "kvm" received signal SIGSEGV, Segmentation fault.
> qemu_co_sleep (w=0x0) at ../util/qemu-coroutine-sleep.c:57
To fix it, create a proper struct and pass its address instead. Also
call qemu_co_sleep_wake unconditionally, because the NULL check (for
the 'to_wake' member) is done inside the function itself.
This patch is based on what the QEMU commits introducing the changes
to the coroutine-sleep API did to the callers in QEMU:
eaee072085 ("coroutine-sleep: allow qemu_co_sleep_wake that wakes nothing")
29a6ea24eb ("coroutine-sleep: replace QemuCoSleepState pointer with struct in the API")
[0]: https://forum.proxmox.com/threads/112130/
Tested-by: Mira Limbeck <m.limbeck@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Only very minor changes needed:
* Most patches in extra (or some version of them) are part of 7.0.0.
* aio_set_fd_handler got an extra parameter, but can just pass NULL
like we did for the related 'poll' parameter. See QEMU commit
826cc32423db2a99d184dbf4f507c737d7e7a4ae for more.
* Add include for qemu/memalign.h in vma.c and vma-writer.c.
* Add reverts for fixups of already reverted 0347a8fd4c ("block/rbd:
implement bdrv_co_block_status") that came in with 7.0.0. Those
fixups are not enough, see Proxmox bugzilla #4047.
* Two trivial context changes for bitmap-mirror patches.
* block_int.h got split up into multiple headers.
* Some context changes in configure and meson.build.
* Used the oppurtunity to squash fixup of bdrv_backuo_dump_create typo
in a later patch into the patch introducing the function (had to
move code to new header during rebase).
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Introduce a new map line for skipping a certain drive, of the form
skip=drive-scsi0
Since in PVE, most archives are compressed and piped to vma for
restore, it's not easily possible to skip reads.
For the reader, a new skip flag for VmaRestoreState is added and the
target is allowed to be NULL if skip is specified when registering.
If
the skip flag is set, no writes will be made as well as no check for
duplicate clusters. Therefore, the flag is not set for verify.
Originally-by: Fabian Ebner <f.ebner@proxmox.com>
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Acked-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
This is necessary for multi-disk backups where not all jobs are
immediately started after they are created. QEMU commit
06e0a9c16405c0a4c1eca33cf286cc04c42066a2 did already part of the work,
ensuring that new writes after job creation don't pass through to the
backup, but not yet for the MIRROR_SYNC_MODE_BITMAP case which is used
for PBS.
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
Notable changes:
* bdrv_co_p{discard,readv,writev,write_zeroes} function signatures
changed, to using int64_t for offsets/bytes and some still had int
rather than BrdvRequestFlags for the flags.
* job_cancel_sync now has a force parameter. Commit messages in
73895f3838cd7fdaf185cf1dbc47be58844a966f
4cfb3f05627ad82af473e7f7ae113c3884cd04e3
sound like using force=true makes more sense.
* Added 3 patches coming in via qemu-stable tag, most important one is
to work around a librbd issue.
* Added another 3 patches from qemu-devel to fix issue leading to
crash when live migrating with iothread.
* cluster_size calculation helper changed (see patch pve/0026).
* QAPI's if conditionals now use 'CONFIG_FOO' rather than
'defined(CONFIG_FOO)'
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
