[pve-devel] [RFC qemu] fix #1420: fix stop mode backup with virtio-blk

[Fabian Grünbichler]

Signed-off-by: Fabian Grünbichler <f.gruenbichler at proxmox.com>
---
this seems in line with our other VMA modifications, and I get the following
stacktrace without:

Thread 1 (Thread 0x7fa27cffd700 (LWP 29427)):
 #0  blk_bs (blk=0x0) at block/block-backend.c:1655
 #1  blk_set_aio_context (blk=0x0, new_context=0x7fa27e16a8c0) at block/block-backend.c:1656
 #2  0x000055d7d03c68c8 in block_job_attached_aio_context (new_context=<optimized out>, opaque=0x7fa1768f6d80) at blockjob.c:115
         job = 0x7fa1768f6d80
 #3  0x000055d7d03c2a8b in bdrv_attach_aio_context (bs=bs at entry=0x7fa27e1c6000, new_context=new_context at entry=0x7fa27e16a8c0) at block.c:4400
         ban = <optimized out>
         ban_tmp = 0x0
         child = <optimized out>
         __PRETTY_FUNCTION__ = "bdrv_attach_aio_context"
 #4  0x000055d7d03c2b29 in bdrv_set_aio_context (bs=bs at entry=0x7fa27e1c6000, new_context=new_context at entry=0x7fa27e16a8c0) at block.c:4424
 #5  0x000055d7d03fe6b6 in blk_set_aio_context (blk=0x7fa27e031000, new_context=0x7fa27e16a8c0) at block/block-backend.c:1662
 #6  0x000055d7d014571c in virtio_blk_data_plane_start (vdev=<optimized out>) at hw/block/dataplane/virtio-blk.c:202
         vblk = 0x7fa177039510
         __func__ = "virtio_blk_data_plane_start"
         s = 0x7fa177091f00
         qbus = <optimized out>
         k = <optimized out>
         i = <optimized out>
         nvqs = <optimized out>
         r = <optimized out>

with the inner-most two having a blk parameter of 0x0 leading to a segfault.

blockjob.c:115 does the following:
job->driver->attached_aio_context(job, new_context);

and job / job->driver in frame #2 are:
$1 = {driver = 0x55d7d0a261a0 <backup_job_driver>, blk = 0x7fa27e033800, id =
0x7fa17703c4e0 "drive-virtio0", co = 0x7fa27e16c440, cancelled = false,
pause_count = 2, user_paused = false, busy = false, paused = true, ready =
false, deferred_to_main_loop = false, job_list = {le_next = 0x0, le_prev =
0x55d7d0adb198 <block_jobs>}, iostatus = BLOCK_DEVICE_IO_STATUS_OK, offset =
285147136, len = 34359738368, speed = 0, cb = 0x55d7d01f8800
<pvebackup_complete_cb>, blocker = 0x7fa176910f40, nodes = 0x7fa27e02b750,
opaque = 0x7fa27e1cc000, refcnt = 1, completed = false, ret = 0, txn = 0x0,
txn_list = {le_next = 0x0, le_prev = 0x0}}

$3 = {common = {driver = 0x55d7d0a261a0 <backup_job_driver>, blk =
0x7fa27e033800, id = 0x7fa17703c4e0 "drive-virtio0", co = 0x7fa27e16c440,
cancelled = false, pause_count = 2, user_paused = false, busy = false, paused =
true, ready = false, deferred_to_main_loop = false, job_list = {le_next = 0x0,
le_prev = 0x55d7d0adb198 <block_jobs>}, iostatus = BLOCK_DEVICE_IO_STATUS_OK,
offset = 285147136, len = 34359738368, speed = 0, cb = 0x55d7d01f8800
<pvebackup_complete_cb>, blocker = 0x7fa176910f40, nodes = 0x7fa27e02b750,
opaque = 0x7fa27e1cc000, refcnt = 1, completed = false, ret = 0, txn = 0x0,
txn_list = { le_next = 0x0, le_prev = 0x0}}, target = 0x0, sync_bitmap = 0x0,
sync_mode = MIRROR_SYNC_MODE_FULL, limit = {slice_start_time = 0,
slice_end_time = 0, slice_quota = 0, slice_ns = 0, dispatched = 0}, dump_cb =
0x55d7d01f83c0 <pvebackup_dump_cb>, on_source_error = BLOCKDEV_ON_ERROR_REPORT,
on_target_error = BLOCKDEV_ON_ERROR_REPORT, flush_rwlock = {pending_writer = 0,
reader = 0, mutex = { locked = 0, ctx = 0x0, from_push = {slh_first = 0x0},
to_pop = {slh_first = 0x0}, handoff = 0, sequence = 0, holder = 0x0}, queue =
{entries = {sqh_first = 0x0, sqh_last = 0x7fa1768f6eb0}}}, sectors_read =
556928, done_bitmap = 0x7fa179c85000, cluster_size = 65536, compress = false,
before_write = {notify = 0x55d7d0416ed0 <backup_before_write_notify>, node =
{le_next = 0x0, le_prev = 0x7fa27e1c91c8}}, inflight_reqs = {lh_first = 0x0}}

this is IMHO the place to fix it (without a target, there is nothing further to
set the context of). review of someone more familiar with the qemu and VMA code
bases very welcome ;)

 debian/patches/pve/0028-adding-old-vma-files.patch | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/debian/patches/pve/0028-adding-old-vma-files.patch b/debian/patches/pve/0028-adding-old-vma-files.patch
index 5e87ed7..85927a3 100644
--- a/debian/patches/pve/0028-adding-old-vma-files.patch
+++ b/debian/patches/pve/0028-adding-old-vma-files.patch
@@ -105,7 +105,17 @@ index 1ede70c061..51b5ba6eda 100644
      assert(s->target);
      blk_unref(s->target);
      s->target = NULL;
-@@ -330,9 +344,11 @@ static BlockErrorAction backup_error_action(BackupBlockJob *job,
+@@ -255,7 +269,8 @@ static void backup_attached_aio_context(BlockJob *job, AioContext *aio_context)
+ {
+     BackupBlockJob *s = container_of(job, BackupBlockJob, common);
+ 
+-    blk_set_aio_context(s->target, aio_context);
++    if (s->target)
++        blk_set_aio_context(s->target, aio_context);
+ }
+ 
+ void backup_do_checkpoint(BlockJob *job, Error **errp)
+@@ -330,9 +345,11 @@ static BlockErrorAction backup_error_action(BackupBlockJob *job,
      if (read) {
          return block_job_error_action(&job->common, job->on_source_error,
                                        true, error);
-- 
2.11.0