On 04/01/2016 01:01 AM, Dave Chinner wrote:
On Thu, Mar 31, 2016 at 07:32:59PM +0300, Avi Kivity wrote:
Saw this nice gift this morning:
[2121372.825904] XFS (dm-10): Internal error xfs_trans_cancel at
line 1007 of file fs/xfs/xfs_trans.c. Caller xfs_create+0x40e/0x710
[xfs]
[2121372.827209] CPU: 0 PID: 32020 Comm: java Tainted: G W
------------ 3.10.0-327.10.1.el7.x86_64 #1
[2121372.828529] Hardware name: /DH77EB, BIOS
EBH7710H.86A.0099.2013.0125.1400 01/25/2013
[2121372.829873] ffff8807b2b11e80 00000000470753cc ffff88031058bb48
ffffffff816352cc
[2121372.831232] ffff88031058bb60 ffffffffa084be5b ffffffffa085b7ee
ffff88031058bb88
[2121372.832542] ffffffffa0866909 ffff88014a2f3b80 ffff8807f29a2800
0000000000000000
[2121372.833850] Call Trace:
[2121372.835125] [<ffffffff816352cc>] dump_stack+0x19/0x1b
[2121372.836397] [<ffffffffa084be5b>] xfs_error_report+0x3b/0x40 [xfs]
[2121372.837654] [<ffffffffa085b7ee>] ? xfs_create+0x40e/0x710 [xfs]
[2121372.838915] [<ffffffffa0866909>] xfs_trans_cancel+0xd9/0x100 [xfs]
[2121372.840178] [<ffffffffa085b7ee>] xfs_create+0x40e/0x710 [xfs]
[2121372.841444] [<ffffffffa0857d8b>] xfs_vn_mknod+0xbb/0x250 [xfs]
[2121372.842683] [<ffffffffa0857f53>] xfs_vn_create+0x13/0x20 [xfs]
[2121372.843887] [<ffffffff811eacdd>] vfs_create+0xcd/0x130
[2121372.845103] [<ffffffff811ec36f>] do_last+0xbef/0x1270
[2121372.846324] [<ffffffff811ee6d2>] path_openat+0xc2/0x490
[2121372.847538] [<ffffffff811efda2>] ? user_path_at_empty+0x72/0xc0
[2121372.848746] [<ffffffff811efe9b>] do_filp_open+0x4b/0xb0
[2121372.849917] [<ffffffff811fca27>] ? __alloc_fd+0xa7/0x130
[2121372.851090] [<ffffffff811dd843>] do_sys_open+0xf3/0x1f0
[2121372.852227] [<ffffffff811dd95e>] SyS_open+0x1e/0x20
[2121372.853356] [<ffffffff81645a49>] system_call_fastpath+0x16/0x1b
[2121372.854486] XFS (dm-10): xfs_do_force_shutdown(0x8) called from
line 1008 of file fs/xfs/xfs_trans.c. Return address =
0xffffffffa0866922
Filesystem appeared full,
ISTR there was a bug inthe inode allocation code that could lead to
multiple AGFs being dirtied (via AGFL fixups) and then not having
enough contiguous freee space to allocate a new inode chuck. I think
it was also a potential deadlock vector. Yeah:
e480a72 xfs: avoid AGI/AGF deadlock scenario for inode chunk allocation
Fixed in 3.15.
Apparently that was backported into the kernel I am using:
* Tue Mar 18 2014 Jarod Wilson <jarod@xxxxxxxxxx> [3.10.0-113.el7]
- [fs] xfs: avoid AGI/AGF deadlock scenario for inode chunk allocation
(Brian Foster) [1052789]
but after a reboot (critical server) it
went back down to 420GB free.
Lots of open unlinked (or O_TMPFILE) files, I'd guess.
The workload running on that machine makes it unlikely, but I cannot
rule it out.
_______________________________________________
xfs mailing list
xfs@xxxxxxxxxxx
http://oss.sgi.com/mailman/listinfo/xfs
