On 2/5/21 5:10 AM, Vladimir Sementsov-Ogievskiy wrote: > I reproduced the following crash fast enough: > > 0 raise () at /lib64/libc.so.6 > 1 abort () at /lib64/libc.so.6 > 2 _nl_load_domain.cold () at /lib64/libc.so.6 > 3 annobin_assert.c_end () at /lib64/libc.so.6 > 4 bdrv_reopen_multiple (bs_queue=0x55de75fa9b70, errp=0x0) > at ../block.c:3820 > 5 bdrv_reopen_set_read_only (bs=0x55de760fc020, read_only=true, > errp=0x0) at ../block.c:3870 > 6 stream_clean (job=0x55de75fa9410) at ../block/stream.c:99
> Then I had 38 successful iterations and another crash: > 0 bdrv_check_update_perm (bs=0x5631ac97bc50, q=0x0, new_used_perm=1, > new_shared_perm=31, ignore_children=0x0, errp=0x7ffd9d477cf8) at > ../block.c:2197 > 1 bdrv_root_attach_child > (child_bs=0x5631ac97bc50, child_name=0x5631aaf6b1f9 "backing", > child_class=0x5631ab280ca0 <child_of_bds>, child_role=8, > ctx=0x5631ab757300, perm=1, shared_perm=31, opaque=0x5631abb8c020, > errp=0x7ffd9d477cf8) > at ../block.c:2642 > 2 bdrv_attach_child (parent_bs=0x5631abb8c020, > child_bs=0x5631ac97bc50, child_name=0x5631aaf6b1f9 "backing", > child_class=0x5631ab280ca0 <child_of_bds>, child_role=8, > errp=0x7ffd9d477cf8) > at ../block.c:2719 > 3 bdrv_set_backing_hd (bs=0x5631abb8c020, backing_hd=0x5631ac97bc50, > errp=0x7ffd9d477cf8) at ../block.c:2854 > 4 stream_prepare (job=0x5631ac751eb0) at ../block/stream.c:74 So we definitely have a race that can show a wide variety of knock-on effects. > > and trying to reproduce it on top of > "block: update graph permissions update" I had 634 successful > iterations > and then the following crash (which looks much better): This part of the commit message is odd - if we check it in to git as written, you're pointing to a future commit, while still stating that it is not a perfect commit. But maybe by the time that commit gets in we'll have figured out this last crash and corrected it as well. Sticking to just the first two logs is fine by me. > > So it seems reasonable to drop test from auto group at least until we > merge "block: update graph permissions update" > > Signed-off-by: Vladimir Sementsov-Ogievskiy <vsement...@virtuozzo.com> > --- > > Note: be free to shorten commit message if needed :) Indeed. But as to the patch itself, I agree, and may Peter wants to apply it directly to master instead of waiting for it to come through on of the block maintainers? Reviewed-by: Eric Blake <ebl...@redhat.com> -- Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
