Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 24 Oct 2022 17:41:10 +0200
From:      Mateusz Guzik <mjguzik@gmail.com>
To:        Andriy Gapon <avg@freebsd.org>
Cc:        freebsd-stable List <stable@freebsd.org>, freebsd-fs <fs@freebsd.org>
Subject:   Re: panic: condition seqc_in_modify not met, while replaying ZIL
Message-ID:  <CAGudoHH-_Mq23nzm2NQ8n2R7=pDQdfV%2Bk=AbbUrWH_XGoYTQzQ@mail.gmail.com>
In-Reply-To: <2f586acb-a8ee-fa17-bbbb-740f41dbfe82@FreeBSD.org>
References:  <93085f3c-4592-8f46-fadb-ba0ee2d2ffa9@FreeBSD.org> <CAGudoHH5iq2DmK5z0C8hAD=NF8_PAEQW4s43b%2B-G162xK7efrg@mail.gmail.com> <2f586acb-a8ee-fa17-bbbb-740f41dbfe82@FreeBSD.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On 10/24/22, Andriy Gapon <avg@freebsd.org> wrote:
> On 20/10/2022 17:32, Mateusz Guzik wrote:
>> This is a false-positive -- the seqc stuff is not modified on replay
>> and zfs internal asserts check for it. I think the easiest way forward
>> for this one is to appease the assert by entering it seqc.
>>
>> That said, could it be it is an invariant there are no nc entries at
>> this stage anyway and the call to cache_vop_rmdir can be avoided? I
>> can add cache_assert_no_entries or similar routine to be called
>> instead.
>
> I've checked the code and, indeed, the name cache is doubly disabled while
> ZIL
> is replayed.  ZFS sets z_use_namecache to false and z_replay to true before
> the
> replay.  And either is sufficient to disable ZFS name cache calls (which is
>
> redundant, but that's a different topic).
>
> So, I came up with this small proof-of-concept patch (and it helped with the
>
> problem at hand).
> It does not assert that the name cache for the filesystem being mounted is
> actually empty.  That would be a good addition, if you wish to add it.
>
> diff --git a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c
> b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c
> index f6bc9c0c6afb..87fa271dfd5c 100644
> --- a/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c
> +++ b/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_vnops_os.c
> @@ -1622,7 +1622,8 @@ zfs_rmdir_(vnode_t *dvp, vnode_t *vp, const char
> *name,
> cred_t *cr)
>
>   	dmu_tx_commit(tx);
>
> -	cache_vop_rmdir(dvp, vp);
> +	if (zfsvfs->z_use_namecache)
> +		cache_vop_rmdir(dvp, vp);
>   out:
>   	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
>   		zil_commit(zilog, 0);
>

Added the routine in
https://cgit.FreeBSD.org/src/commit/?id=d653aaec7a311a7d1455e9721904b2a341b32a65,
just call it for the else clause.

>
>
>> On 10/20/22, Andriy Gapon <avg@freebsd.org> wrote:
>>>
>>> This happens on stable/13, custom kernel compiled with DEBUG_VFS_LOCKS,
>>> from
>>>
>>> mid-September after an ungraceful reboot (unrelated crash).  As far as I
>>> can
>>> see
>>> in kgdb, both dvp and vp have v_seqc == 0.
>>>
>>> VNASSERT failed: ({ seqc_t __seqc = (_vp->v_seqc);
>>> __builtin_expect((__seqc
>>> &
>>> 1), 0); }) not true at
>>> /usr/home/avg/devel/freebsd-src-new/machines/trant/sys/kern/vfs_cache.c:2976
>>>
>>> (cache_vop_rmdir)
>>> 0xfffff8004bc0bd58: type VDIR
>>>       usecount 1, writecount 0, refcount 1 seqc users 0 mountedhere 0
>>>       hold count flags ()
>>>       flags ()
>>>       lock type zfs: UNLOCKED
>>> panic: condition seqc_in_modify(_vp->v_seqc) not met at
>>> /usr/home/avg/devel/freebsd-src-new/machines/trant/sys/kern/vfs_cache.c:2976
>>>
>>> (cache_vop_rmdir)
>>> cpuid = 5
>>> time = 1666250711
>>> KDB: stack backtrace:
>>> db_trace_self_wrapper() at 0xffffffff8061549b =
>>> db_trace_self_wrapper+0x2b/frame
>>> 0xfffffe01ce6c7090
>>> kdb_backtrace() at 0xffffffff80942927 = kdb_backtrace+0x37/frame
>>> 0xfffffe01ce6c7140
>>> vpanic() at 0xffffffff808f4fb4 = vpanic+0x184/frame 0xfffffe01ce6c71a0
>>> panic() at 0xffffffff808f4d63 = panic+0x43/frame 0xfffffe01ce6c7200
>>> cache_vop_rmdir() at 0xffffffff809b9b9f = cache_vop_rmdir+0xdf/frame
>>> 0xfffffe01ce6c7220
>>> zfs_rmdir_() at 0xffffffff80391cff = zfs_rmdir_+0x1df/frame
>>> 0xfffffe01ce6c7290
>>> zfs_rmdir() at 0xffffffff80391af8 = zfs_rmdir+0x48/frame
>>> 0xfffffe01ce6c7310
>>> zfs_replay_remove() at 0xffffffff804fc53b = zfs_replay_remove+0x7b/frame
>>> 0xfffffe01ce6c7340
>>> zil_replay_log_record() at 0xffffffff80508462 =
>>> zil_replay_log_record+0x212/frame 0xfffffe01ce6c7490
>>> zil_parse() at 0xffffffff80502f4b = zil_parse+0x5cb/frame
>>> 0xfffffe01ce6c76a0
>>> zil_replay() at 0xffffffff805081d8 = zil_replay+0xd8/frame
>>> 0xfffffe01ce6c7700
>>> zfsvfs_setup() at 0xffffffff8038f09d = zfsvfs_setup+0x24d/frame
>>> 0xfffffe01ce6c7930
>>> zfs_mount() at 0xffffffff8038c9b2 = zfs_mount+0x652/frame
>>> 0xfffffe01ce6c7ad0
>>> vfs_domount_first() at 0xffffffff809d0d66 =
>>> vfs_domount_first+0x216/frame
>>> 0xfffffe01ce6c7c00
>>> vfs_domount() at 0xffffffff809cdd43 = vfs_domount+0x2d3/frame
>>> 0xfffffe01ce6c7d30
>>> vfs_donmount() at 0xffffffff809ccb6f = vfs_donmount+0x81f/frame
>>> 0xfffffe01ce6c7dc0
>>> sys_nmount() at 0xffffffff809cc318 = sys_nmount+0x108/frame
>>> 0xfffffe01ce6c7df0
>>> amd64_syscall() at 0xffffffff80c31d16 = amd64_syscall+0x186/frame
>>> 0xfffffe01ce6c7f30
>>> fast_syscall_common() at 0xffffffff80c0889b =
>>> fast_syscall_common+0xf8/frame
>>>
>>> 0xfffffe01ce6c7f30
>>> --- syscall (378, FreeBSD ELF64, sys_nmount), rip = 0xb6aac9dd1a, rsp =
>>> 0xb6b538f468, rbp = 0xb6b538f4d0 ---
>>> Uptime: 37s
>>> Dumping 1426 out of 32644 MB: (CTRL-C to abort)
>>> ..2%..11%..21%..31%..41%..51%..61%..71%..81%..91%
>>>
>>> --
>>> Andriy Gapon
>>>
>>
>>
>
> --
> Andriy Gapon
>
>


-- 
Mateusz Guzik <mjguzik gmail.com>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAGudoHH-_Mq23nzm2NQ8n2R7=pDQdfV%2Bk=AbbUrWH_XGoYTQzQ>