XFS filesystem shutdown when deleting millions of files on RHEL
SOLUTION UNVERIFIED - 已更新 2019年三月8日20:43 -
環境
- Red Hat Enterprise Linux 7
問題
- XFS file system shut down when deleting 50 ~ 60 million small files (5 ~ 10 KB).
kernel: XFS (dm-3): xlog_write: reservation summary:
trans type = INACTIVE (3)
unit res = 78564 bytes
current res = -33148 bytes
total reg = 0 bytes (o/flow = 0 bytes)
ophdrs = 0 (ophdr space = 0 bytes)
ophdr + reg = 0 bytes
num regions = 0
kernel: XFS : xlog_write: reservation ran out. Need to up reservation
kernel: XFS : xfs_do_force_shutdown(0x2) called from line 2042 of file fs/xfs/xfs_log.c. Return address = 0xffffffffa033edb8
kernel: XFS : xfs_imap_to_bp: xfs_trans_read_buf() returned error 5.
kernel: XFS : Log I/O Error Detected. Shutting down filesystem
kernel: XFS : Please umount the filesystem and rectify the problem(s)
kernel: XFS : xfs_log_force: error 5 returned.
決議
This issue has been resolved by RHSA-2018:1062 please upgrade to kernel-3.10.0-862.el7 (7.5) or later for this fix.
Workaround
If you are unable to upgrade to the latest kernel some customers have reported success mounting the filesystem with the 'ikeep' mount option which prevents XFS from freeing unused inode clusters.
根源
This series refactors some of the transaction overrun reporting code in XFS to provide more useful error reports (patches 1-4) and refactors some of the log reservations associated with inode transactions to provide more accurate worst case reservations. This helps avoid transaction reservation overruns during particular workloads.
Below commits have been backported.
xfs: separate shutdown from ticket reservation print helper
xfs: refactor xlog_cil_insert_items() to facilitate transaction dump
xfs: dump transaction usage details on log reservation overrun
xfs: print transaction log reservation on overrun
xfs: include inobt buffers in ifree tx log reservation
xfs: fix up agi unlinked list reservations
xfs: truncate transaction does not modify the inobt
xfs: include an allocfree res for inobt modifications
xfs: refactor inode chunk alloc/free tx reservation
xfs: eliminate duplicate icreate tx reservation functions
診斷步驟
A metadump was taken of a filesystem exhibiting this issue for analysis.
# xfs_metadump -gow /dev/mapper/vg_data | bzip2 > reservation_failure.xfsmetadump.bz2
The image was restored to a local volume group for testing.
kdump was configured and the system set to panic on log reservation failures.
# sysctl fs.xfs.error_level=11 fs.xfs.panic_mask=2
fs.xfs.error_level = 11
fs.xfs.panic_mask = 2
# bzcat ~/xfs-metadump.img.gz | xfs_mdrestore - /dev/mapper/vg_metadump
# mount /dev/mapper/vg_metadump /mnt/metadump
# find /mnt/metadump -type f -delete
Initial analysis of the vmcore
XFS (dm-2): xlog_write: reservation summary:
XFS (dm-2): trans type = INACTIVE (3)
XFS (dm-2): unit res = 47832 bytes
XFS (dm-2): current res = -128 bytes
XFS (dm-2): total reg = 0 bytes (o/flow = 0 bytes)
XFS (dm-2): ophdrs = 0 (ophdr space = 0 bytes)
XFS (dm-2): ophdr + reg = 0 bytes
XFS (dm-2): num regions = 0
XFS (dm-2): Transforming an alert into a BUG.
XFS (dm-2): xlog_write: reservation ran out. Need to up reservation
------------[ cut here ]------------
kernel BUG at fs/xfs/xfs_message.c:97!
invalid opcode: 0000 [#1] SMP
Modules linked in: intel_powerclamp ... xfs ...
CPU: 0 PID: 29036 Comm: find Not tainted 3.10.0-514.el7.x86_64 #1
Hardware name: Dell Inc. PowerEdge R530/0CN7X8, BIOS 2.4.2 01/09/2017
task: ffff88085717af10 ti: ffff8806a1f50000 task.ti: ffff8806a1f50000
RIP: 0010:[<ffffffffa033c40c>] [<ffffffffa033c40c>] xfs_alert_tag+0xec/0x100 [xfs]
RSP: 0018:ffff8806a1f53c08 EFLAGS: 00010246
RAX: 0000000000000043 RBX: ffff880858970000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: ffff88085c60f838 RDI: ffff88085c60f838
RBP: ffff8806a1f53c78 R08: 0000000000000086 R09: 00000000000005d4
R10: 00000000000003ff R11: 0000000000000001 R12: ffff88066531cfd0
R13: ffff880858970000 R14: ffff880471dba108 R15: ffff880259574d68
FS: 00007f3fe617a800(0000) GS:ffff88085c600000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000840018 CR3: 0000000581fe9000 CR4: 00000000003407f0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Stack:
ffffffffa0372da8 ffffffffa0372da8 ffff8806a1f53c20 ffff880600000018
ffff8806a1f53c88 ffff8806a1f53c40 00000000668a0908 00000000668a0908
ffff8802902afcb8 ffff8801bc792f80 0000000000000000 0000000000000000
Call Trace:
[<ffffffffa0347f8f>] xlog_print_tic_res+0x14f/0x170 [xfs]
[<ffffffffa034a629>] xfs_log_commit_cil+0x499/0x4d0 [xfs]
[<ffffffffa034422d>] __xfs_trans_commit+0x12d/0x260 [xfs]
[<ffffffffa0344610>] xfs_trans_commit+0x10/0x20 [xfs]
[<ffffffffa0339bd4>] xfs_inactive_ifree+0x1d4/0x240 [xfs]
[<ffffffffa0339ccd>] xfs_inactive+0x8d/0x130 [xfs]
[<ffffffffa033f346>] xfs_fs_evict_inode+0xa6/0xe0 [xfs]
[<ffffffff8121a0e7>] evict+0xa7/0x170
[<ffffffff8121a985>] iput+0xf5/0x180
[<ffffffff8120f31e>] do_unlinkat+0x1ae/0x2b0
[<ffffffff812037f4>] ? SYSC_newfstatat+0x34/0x60
[<ffffffff812102eb>] SyS_unlinkat+0x1b/0x40
[<ffffffff816964c9>] system_call_fastpath+0x16/0x1b
crash> whatis xlog_print_tic_res
void xlog_print_tic_res(struct xfs_mount *, struct xlog_ticket *);
crash> dis -r ffffffffa0347f8f
0xffffffffa0347e40 <xlog_print_tic_res>: nopl 0x0(%rax,%rax,1) [FTRACE NOP]
0xffffffffa0347e45 <xlog_print_tic_res+0x5>: push %rbp
0xffffffffa0347e46 <xlog_print_tic_res+0x6>: mov %rsp,%rbp
0xffffffffa0347e49 <xlog_print_tic_res+0x9>: push %r13
0xffffffffa0347e4b <xlog_print_tic_res+0xb>: mov %rdi,%r13
0xffffffffa0347e4e <xlog_print_tic_res+0xe>: push %r12
0xffffffffa0347e50 <xlog_print_tic_res+0x10>: mov %rsi,%r12
... r12 and r13 aren't changed later
crash> struct xlog_ticket ffff88066531cfd0
struct xlog_ticket {
t_queue = {
next = 0xffff88066531cfd0,
prev = 0xffff88066531cfd0
},
t_task = 0xffff88085717af10,
t_tid = 0x8fb68988,
t_ref = {
counter = 0x1
},
t_curr_res = 0xffffff80, (-128)
t_unit_res = 0xbad8, (47832)
t_ocnt = 0x2,
t_cnt = 0x1,
t_clientid = 0x69,
t_flags = 0x3,
t_trans_type = 0x3,
t_res_num = 0x0,
t_res_num_ophdrs = 0x0,
t_res_arr_sum = 0x0,
t_res_o_flow = 0x0,
t_res_arr = {{
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}, {
r_len = 0x0,
r_type = 0x0
}}
}
- Related source code:
https://access.redhat.com/labs/psb/versions/kernel-3.10.0-229.el7/fs/xfs/xfs_log.c
void
xlog_print_tic_res(
struct xfs_mount *mp,
struct xlog_ticket *ticket)
{
uint i;
uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
/* match with XLOG_REG_TYPE_* in xfs_log.h */
static char *res_type_str[XLOG_REG_TYPE_MAX] = {
"bformat",
"bchunk",
"efi_format",
"efd_format",
"iformat",
"icore",
"iext",
"ibroot",
"ilocal",
"iattr_ext",
"iattr_broot",
"iattr_local",
"qformat",
"dquot",
"quotaoff",
"LR header",
"unmount",
"commit",
"trans header"
};
static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
"SETATTR_NOT_SIZE",
"SETATTR_SIZE",
"INACTIVE",
"CREATE",
"CREATE_TRUNC",
"TRUNCATE_FILE",
"REMOVE",
"LINK",
"RENAME",
"MKDIR",
"RMDIR",
"SYMLINK",
"SET_DMATTRS",
"GROWFS",
"STRAT_WRITE",
"DIOSTRAT",
"WRITE_SYNC",
"WRITEID",
"ADDAFORK",
"ATTRINVAL",
"ATRUNCATE",
"ATTR_SET",
"ATTR_RM",
"ATTR_FLAG",
"CLEAR_AGI_BUCKET",
"QM_SBCHANGE",
"DUMMY1",
"DUMMY2",
"QM_QUOTAOFF",
"QM_DQALLOC",
"QM_SETQLIM",
"QM_DQCLUSTER",
"QM_QINOCREATE",
"QM_QUOTAOFF_END",
"SB_UNIT",
"FSYNC_TS",
"GROWFSRT_ALLOC",
"GROWFSRT_ZERO",
"GROWFSRT_FREE",
"SWAPEXT"
};
xfs_warn(mp,
"xlog_write: reservation summary:\n"
" trans type = %s (%u)\n"
" unit res = %d bytes\n"
" current res = %d bytes\n"
" total reg = %u bytes (o/flow = %u bytes)\n"
" ophdrs = %u (ophdr space = %u bytes)\n"
" ophdr + reg = %u bytes\n"
" num regions = %u\n",
((ticket->t_trans_type <= 0 ||
ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
"bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
ticket->t_trans_type,
ticket->t_unit_res,
ticket->t_curr_res,
ticket->t_res_arr_sum, ticket->t_res_o_flow,
ticket->t_res_num_ophdrs, ophdr_spc,
ticket->t_res_arr_sum +
ticket->t_res_o_flow + ophdr_spc,
ticket->t_res_num);
for (i = 0; i < ticket->t_res_num; i++) {
uint r_type = ticket->t_res_arr[i].r_type;
xfs_warn(mp, "region[%u]: %s - %u bytes", i,
((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
"bad-rtype" : res_type_str[r_type-1]),
ticket->t_res_arr[i].r_len);
}
xfs_alert_tag(mp, XFS_PTAG_LOGRES,
"xlog_write: reservation ran out. Need to up reservation");
xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
}