]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/xfs/xfs_log.c
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_error.h"
29 #include "xfs_log_priv.h"
30 #include "xfs_buf_item.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_alloc_btree.h"
33 #include "xfs_ialloc_btree.h"
34 #include "xfs_log_recover.h"
35 #include "xfs_trans_priv.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
39 #include "xfs_trace.h"
41 kmem_zone_t
*xfs_log_ticket_zone
;
43 /* Local miscellaneous function prototypes */
44 STATIC
int xlog_commit_record(struct log
*log
, struct xlog_ticket
*ticket
,
45 xlog_in_core_t
**, xfs_lsn_t
*);
46 STATIC xlog_t
* xlog_alloc_log(xfs_mount_t
*mp
,
47 xfs_buftarg_t
*log_target
,
48 xfs_daddr_t blk_offset
,
50 STATIC
int xlog_space_left(struct log
*log
, atomic64_t
*head
);
51 STATIC
int xlog_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
52 STATIC
void xlog_dealloc_log(xlog_t
*log
);
54 /* local state machine functions */
55 STATIC
void xlog_state_done_syncing(xlog_in_core_t
*iclog
, int);
56 STATIC
void xlog_state_do_callback(xlog_t
*log
,int aborted
, xlog_in_core_t
*iclog
);
57 STATIC
int xlog_state_get_iclog_space(xlog_t
*log
,
59 xlog_in_core_t
**iclog
,
60 xlog_ticket_t
*ticket
,
63 STATIC
int xlog_state_release_iclog(xlog_t
*log
,
64 xlog_in_core_t
*iclog
);
65 STATIC
void xlog_state_switch_iclogs(xlog_t
*log
,
66 xlog_in_core_t
*iclog
,
68 STATIC
void xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
70 /* local functions to manipulate grant head */
71 STATIC
int xlog_grant_log_space(xlog_t
*log
,
73 STATIC
void xlog_grant_push_ail(struct log
*log
,
75 STATIC
void xlog_regrant_reserve_log_space(xlog_t
*log
,
76 xlog_ticket_t
*ticket
);
77 STATIC
int xlog_regrant_write_log_space(xlog_t
*log
,
78 xlog_ticket_t
*ticket
);
79 STATIC
void xlog_ungrant_log_space(xlog_t
*log
,
80 xlog_ticket_t
*ticket
);
83 STATIC
void xlog_verify_dest_ptr(xlog_t
*log
, char *ptr
);
84 STATIC
void xlog_verify_grant_tail(struct log
*log
);
85 STATIC
void xlog_verify_iclog(xlog_t
*log
, xlog_in_core_t
*iclog
,
86 int count
, boolean_t syncing
);
87 STATIC
void xlog_verify_tail_lsn(xlog_t
*log
, xlog_in_core_t
*iclog
,
90 #define xlog_verify_dest_ptr(a,b)
91 #define xlog_verify_grant_tail(a)
92 #define xlog_verify_iclog(a,b,c,d)
93 #define xlog_verify_tail_lsn(a,b,c)
96 STATIC
int xlog_iclogs_empty(xlog_t
*log
);
104 int64_t head_val
= atomic64_read(head
);
110 xlog_crack_grant_head_val(head_val
, &cycle
, &space
);
114 space
+= log
->l_logsize
;
119 new = xlog_assign_grant_head_val(cycle
, space
);
120 head_val
= atomic64_cmpxchg(head
, old
, new);
121 } while (head_val
!= old
);
125 xlog_grant_add_space(
130 int64_t head_val
= atomic64_read(head
);
137 xlog_crack_grant_head_val(head_val
, &cycle
, &space
);
139 tmp
= log
->l_logsize
- space
;
148 new = xlog_assign_grant_head_val(cycle
, space
);
149 head_val
= atomic64_cmpxchg(head
, old
, new);
150 } while (head_val
!= old
);
154 xlog_tic_reset_res(xlog_ticket_t
*tic
)
157 tic
->t_res_arr_sum
= 0;
158 tic
->t_res_num_ophdrs
= 0;
162 xlog_tic_add_region(xlog_ticket_t
*tic
, uint len
, uint type
)
164 if (tic
->t_res_num
== XLOG_TIC_LEN_MAX
) {
165 /* add to overflow and start again */
166 tic
->t_res_o_flow
+= tic
->t_res_arr_sum
;
168 tic
->t_res_arr_sum
= 0;
171 tic
->t_res_arr
[tic
->t_res_num
].r_len
= len
;
172 tic
->t_res_arr
[tic
->t_res_num
].r_type
= type
;
173 tic
->t_res_arr_sum
+= len
;
180 * 1. currblock field gets updated at startup and after in-core logs
181 * marked as with WANT_SYNC.
185 * This routine is called when a user of a log manager ticket is done with
186 * the reservation. If the ticket was ever used, then a commit record for
187 * the associated transaction is written out as a log operation header with
188 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
189 * a given ticket. If the ticket was one with a permanent reservation, then
190 * a few operations are done differently. Permanent reservation tickets by
191 * default don't release the reservation. They just commit the current
192 * transaction with the belief that the reservation is still needed. A flag
193 * must be passed in before permanent reservations are actually released.
194 * When these type of tickets are not released, they need to be set into
195 * the inited state again. By doing this, a start record will be written
196 * out when the next write occurs.
200 struct xfs_mount
*mp
,
201 struct xlog_ticket
*ticket
,
202 struct xlog_in_core
**iclog
,
205 struct log
*log
= mp
->m_log
;
208 if (XLOG_FORCED_SHUTDOWN(log
) ||
210 * If nothing was ever written, don't write out commit record.
211 * If we get an error, just continue and give back the log ticket.
213 (((ticket
->t_flags
& XLOG_TIC_INITED
) == 0) &&
214 (xlog_commit_record(log
, ticket
, iclog
, &lsn
)))) {
215 lsn
= (xfs_lsn_t
) -1;
216 if (ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) {
217 flags
|= XFS_LOG_REL_PERM_RESERV
;
222 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) == 0 ||
223 (flags
& XFS_LOG_REL_PERM_RESERV
)) {
224 trace_xfs_log_done_nonperm(log
, ticket
);
227 * Release ticket if not permanent reservation or a specific
228 * request has been made to release a permanent reservation.
230 xlog_ungrant_log_space(log
, ticket
);
231 xfs_log_ticket_put(ticket
);
233 trace_xfs_log_done_perm(log
, ticket
);
235 xlog_regrant_reserve_log_space(log
, ticket
);
236 /* If this ticket was a permanent reservation and we aren't
237 * trying to release it, reset the inited flags; so next time
238 * we write, a start record will be written out.
240 ticket
->t_flags
|= XLOG_TIC_INITED
;
247 * Attaches a new iclog I/O completion callback routine during
248 * transaction commit. If the log is in error state, a non-zero
249 * return code is handed back and the caller is responsible for
250 * executing the callback at an appropriate time.
254 struct xfs_mount
*mp
,
255 struct xlog_in_core
*iclog
,
256 xfs_log_callback_t
*cb
)
260 spin_lock(&iclog
->ic_callback_lock
);
261 abortflg
= (iclog
->ic_state
& XLOG_STATE_IOERROR
);
263 ASSERT_ALWAYS((iclog
->ic_state
== XLOG_STATE_ACTIVE
) ||
264 (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
));
266 *(iclog
->ic_callback_tail
) = cb
;
267 iclog
->ic_callback_tail
= &(cb
->cb_next
);
269 spin_unlock(&iclog
->ic_callback_lock
);
274 xfs_log_release_iclog(
275 struct xfs_mount
*mp
,
276 struct xlog_in_core
*iclog
)
278 if (xlog_state_release_iclog(mp
->m_log
, iclog
)) {
279 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
287 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
288 * to the reservation.
289 * 2. Potentially, push buffers at tail of log to disk.
291 * Each reservation is going to reserve extra space for a log record header.
292 * When writes happen to the on-disk log, we don't subtract the length of the
293 * log record header from any reservation. By wasting space in each
294 * reservation, we prevent over allocation problems.
298 struct xfs_mount
*mp
,
301 struct xlog_ticket
**ticket
,
306 struct log
*log
= mp
->m_log
;
307 struct xlog_ticket
*internal_ticket
;
310 ASSERT(client
== XFS_TRANSACTION
|| client
== XFS_LOG
);
312 if (XLOG_FORCED_SHUTDOWN(log
))
313 return XFS_ERROR(EIO
);
315 XFS_STATS_INC(xs_try_logspace
);
318 if (*ticket
!= NULL
) {
319 ASSERT(flags
& XFS_LOG_PERM_RESERV
);
320 internal_ticket
= *ticket
;
323 * this is a new transaction on the ticket, so we need to
324 * change the transaction ID so that the next transaction has a
325 * different TID in the log. Just add one to the existing tid
326 * so that we can see chains of rolling transactions in the log
329 internal_ticket
->t_tid
++;
331 trace_xfs_log_reserve(log
, internal_ticket
);
333 xlog_grant_push_ail(log
, internal_ticket
->t_unit_res
);
334 retval
= xlog_regrant_write_log_space(log
, internal_ticket
);
336 /* may sleep if need to allocate more tickets */
337 internal_ticket
= xlog_ticket_alloc(log
, unit_bytes
, cnt
,
339 KM_SLEEP
|KM_MAYFAIL
);
340 if (!internal_ticket
)
341 return XFS_ERROR(ENOMEM
);
342 internal_ticket
->t_trans_type
= t_type
;
343 *ticket
= internal_ticket
;
345 trace_xfs_log_reserve(log
, internal_ticket
);
347 xlog_grant_push_ail(log
,
348 (internal_ticket
->t_unit_res
*
349 internal_ticket
->t_cnt
));
350 retval
= xlog_grant_log_space(log
, internal_ticket
);
354 } /* xfs_log_reserve */
358 * Mount a log filesystem
360 * mp - ubiquitous xfs mount point structure
361 * log_target - buftarg of on-disk log device
362 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
363 * num_bblocks - Number of BBSIZE blocks in on-disk log
365 * Return error or zero.
370 xfs_buftarg_t
*log_target
,
371 xfs_daddr_t blk_offset
,
376 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
377 xfs_notice(mp
, "Mounting Filesystem");
380 "Mounting filesystem in no-recovery mode. Filesystem will be inconsistent.");
381 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
384 mp
->m_log
= xlog_alloc_log(mp
, log_target
, blk_offset
, num_bblks
);
385 if (IS_ERR(mp
->m_log
)) {
386 error
= -PTR_ERR(mp
->m_log
);
391 * Initialize the AIL now we have a log.
393 error
= xfs_trans_ail_init(mp
);
395 xfs_warn(mp
, "AIL initialisation failed: error %d", error
);
398 mp
->m_log
->l_ailp
= mp
->m_ail
;
401 * skip log recovery on a norecovery mount. pretend it all
404 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
)) {
405 int readonly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
408 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
410 error
= xlog_recover(mp
->m_log
);
413 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
415 xfs_warn(mp
, "log mount/recovery failed: error %d",
417 goto out_destroy_ail
;
421 /* Normal transactions can now occur */
422 mp
->m_log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
425 * Now the log has been fully initialised and we know were our
426 * space grant counters are, we can initialise the permanent ticket
427 * needed for delayed logging to work.
429 xlog_cil_init_post_recovery(mp
->m_log
);
434 xfs_trans_ail_destroy(mp
);
436 xlog_dealloc_log(mp
->m_log
);
442 * Finish the recovery of the file system. This is separate from
443 * the xfs_log_mount() call, because it depends on the code in
444 * xfs_mountfs() to read in the root and real-time bitmap inodes
445 * between calling xfs_log_mount() and here.
447 * mp - ubiquitous xfs mount point structure
450 xfs_log_mount_finish(xfs_mount_t
*mp
)
454 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
455 error
= xlog_recover_finish(mp
->m_log
);
458 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
465 * Final log writes as part of unmount.
467 * Mark the filesystem clean as unmount happens. Note that during relocation
468 * this routine needs to be executed as part of source-bag while the
469 * deallocation must not be done until source-end.
473 * Unmount record used to have a string "Unmount filesystem--" in the
474 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
475 * We just write the magic number now since that particular field isn't
476 * currently architecture converted and "nUmount" is a bit foo.
477 * As far as I know, there weren't any dependencies on the old behaviour.
481 xfs_log_unmount_write(xfs_mount_t
*mp
)
483 xlog_t
*log
= mp
->m_log
;
484 xlog_in_core_t
*iclog
;
486 xlog_in_core_t
*first_iclog
;
488 xlog_ticket_t
*tic
= NULL
;
493 * Don't write out unmount record on read-only mounts.
494 * Or, if we are doing a forced umount (typically because of IO errors).
496 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
499 error
= _xfs_log_force(mp
, XFS_LOG_SYNC
, NULL
);
500 ASSERT(error
|| !(XLOG_FORCED_SHUTDOWN(log
)));
503 first_iclog
= iclog
= log
->l_iclog
;
505 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
506 ASSERT(iclog
->ic_state
& XLOG_STATE_ACTIVE
);
507 ASSERT(iclog
->ic_offset
== 0);
509 iclog
= iclog
->ic_next
;
510 } while (iclog
!= first_iclog
);
512 if (! (XLOG_FORCED_SHUTDOWN(log
))) {
513 error
= xfs_log_reserve(mp
, 600, 1, &tic
,
514 XFS_LOG
, 0, XLOG_UNMOUNT_REC_TYPE
);
516 /* the data section must be 32 bit size aligned */
520 __uint32_t pad2
; /* may as well make it 64 bits */
522 .magic
= XLOG_UNMOUNT_TYPE
,
524 struct xfs_log_iovec reg
= {
526 .i_len
= sizeof(magic
),
527 .i_type
= XLOG_REG_TYPE_UNMOUNT
,
529 struct xfs_log_vec vec
= {
534 /* remove inited flag */
536 error
= xlog_write(log
, &vec
, tic
, &lsn
,
537 NULL
, XLOG_UNMOUNT_TRANS
);
539 * At this point, we're umounting anyway,
540 * so there's no point in transitioning log state
541 * to IOERROR. Just continue...
546 xfs_alert(mp
, "%s: unmount record failed", __func__
);
549 spin_lock(&log
->l_icloglock
);
550 iclog
= log
->l_iclog
;
551 atomic_inc(&iclog
->ic_refcnt
);
552 xlog_state_want_sync(log
, iclog
);
553 spin_unlock(&log
->l_icloglock
);
554 error
= xlog_state_release_iclog(log
, iclog
);
556 spin_lock(&log
->l_icloglock
);
557 if (!(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
558 iclog
->ic_state
== XLOG_STATE_DIRTY
)) {
559 if (!XLOG_FORCED_SHUTDOWN(log
)) {
560 xlog_wait(&iclog
->ic_force_wait
,
563 spin_unlock(&log
->l_icloglock
);
566 spin_unlock(&log
->l_icloglock
);
569 trace_xfs_log_umount_write(log
, tic
);
570 xlog_ungrant_log_space(log
, tic
);
571 xfs_log_ticket_put(tic
);
575 * We're already in forced_shutdown mode, couldn't
576 * even attempt to write out the unmount transaction.
578 * Go through the motions of sync'ing and releasing
579 * the iclog, even though no I/O will actually happen,
580 * we need to wait for other log I/Os that may already
581 * be in progress. Do this as a separate section of
582 * code so we'll know if we ever get stuck here that
583 * we're in this odd situation of trying to unmount
584 * a file system that went into forced_shutdown as
585 * the result of an unmount..
587 spin_lock(&log
->l_icloglock
);
588 iclog
= log
->l_iclog
;
589 atomic_inc(&iclog
->ic_refcnt
);
591 xlog_state_want_sync(log
, iclog
);
592 spin_unlock(&log
->l_icloglock
);
593 error
= xlog_state_release_iclog(log
, iclog
);
595 spin_lock(&log
->l_icloglock
);
597 if ( ! ( iclog
->ic_state
== XLOG_STATE_ACTIVE
598 || iclog
->ic_state
== XLOG_STATE_DIRTY
599 || iclog
->ic_state
== XLOG_STATE_IOERROR
) ) {
601 xlog_wait(&iclog
->ic_force_wait
,
604 spin_unlock(&log
->l_icloglock
);
609 } /* xfs_log_unmount_write */
612 * Deallocate log structures for unmount/relocation.
614 * We need to stop the aild from running before we destroy
615 * and deallocate the log as the aild references the log.
618 xfs_log_unmount(xfs_mount_t
*mp
)
620 xfs_trans_ail_destroy(mp
);
621 xlog_dealloc_log(mp
->m_log
);
626 struct xfs_mount
*mp
,
627 struct xfs_log_item
*item
,
629 struct xfs_item_ops
*ops
)
631 item
->li_mountp
= mp
;
632 item
->li_ailp
= mp
->m_ail
;
633 item
->li_type
= type
;
637 INIT_LIST_HEAD(&item
->li_ail
);
638 INIT_LIST_HEAD(&item
->li_cil
);
642 * Write region vectors to log. The write happens using the space reservation
643 * of the ticket (tic). It is not a requirement that all writes for a given
644 * transaction occur with one call to xfs_log_write(). However, it is important
645 * to note that the transaction reservation code makes an assumption about the
646 * number of log headers a transaction requires that may be violated if you
647 * don't pass all the transaction vectors in one call....
651 struct xfs_mount
*mp
,
652 struct xfs_log_iovec reg
[],
654 struct xlog_ticket
*tic
,
655 xfs_lsn_t
*start_lsn
)
657 struct log
*log
= mp
->m_log
;
659 struct xfs_log_vec vec
= {
660 .lv_niovecs
= nentries
,
664 if (XLOG_FORCED_SHUTDOWN(log
))
665 return XFS_ERROR(EIO
);
667 error
= xlog_write(log
, &vec
, tic
, start_lsn
, NULL
, 0);
669 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
674 xfs_log_move_tail(xfs_mount_t
*mp
,
678 xlog_t
*log
= mp
->m_log
;
679 int need_bytes
, free_bytes
;
681 if (XLOG_FORCED_SHUTDOWN(log
))
685 tail_lsn
= atomic64_read(&log
->l_last_sync_lsn
);
687 /* tail_lsn == 1 implies that we weren't passed a valid value. */
689 atomic64_set(&log
->l_tail_lsn
, tail_lsn
);
691 if (!list_empty_careful(&log
->l_writeq
)) {
693 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
694 panic("Recovery problem");
696 spin_lock(&log
->l_grant_write_lock
);
697 free_bytes
= xlog_space_left(log
, &log
->l_grant_write_head
);
698 list_for_each_entry(tic
, &log
->l_writeq
, t_queue
) {
699 ASSERT(tic
->t_flags
& XLOG_TIC_PERM_RESERV
);
701 if (free_bytes
< tic
->t_unit_res
&& tail_lsn
!= 1)
704 free_bytes
-= tic
->t_unit_res
;
705 trace_xfs_log_regrant_write_wake_up(log
, tic
);
706 wake_up(&tic
->t_wait
);
708 spin_unlock(&log
->l_grant_write_lock
);
711 if (!list_empty_careful(&log
->l_reserveq
)) {
713 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
714 panic("Recovery problem");
716 spin_lock(&log
->l_grant_reserve_lock
);
717 free_bytes
= xlog_space_left(log
, &log
->l_grant_reserve_head
);
718 list_for_each_entry(tic
, &log
->l_reserveq
, t_queue
) {
719 if (tic
->t_flags
& XLOG_TIC_PERM_RESERV
)
720 need_bytes
= tic
->t_unit_res
*tic
->t_cnt
;
722 need_bytes
= tic
->t_unit_res
;
723 if (free_bytes
< need_bytes
&& tail_lsn
!= 1)
726 free_bytes
-= need_bytes
;
727 trace_xfs_log_grant_wake_up(log
, tic
);
728 wake_up(&tic
->t_wait
);
730 spin_unlock(&log
->l_grant_reserve_lock
);
735 * Determine if we have a transaction that has gone to disk
736 * that needs to be covered. To begin the transition to the idle state
737 * firstly the log needs to be idle (no AIL and nothing in the iclogs).
738 * If we are then in a state where covering is needed, the caller is informed
739 * that dummy transactions are required to move the log into the idle state.
741 * Because this is called as part of the sync process, we should also indicate
742 * that dummy transactions should be issued in anything but the covered or
743 * idle states. This ensures that the log tail is accurately reflected in
744 * the log at the end of the sync, hence if a crash occurrs avoids replay
745 * of transactions where the metadata is already on disk.
748 xfs_log_need_covered(xfs_mount_t
*mp
)
751 xlog_t
*log
= mp
->m_log
;
753 if (!xfs_fs_writable(mp
))
756 spin_lock(&log
->l_icloglock
);
757 switch (log
->l_covered_state
) {
758 case XLOG_STATE_COVER_DONE
:
759 case XLOG_STATE_COVER_DONE2
:
760 case XLOG_STATE_COVER_IDLE
:
762 case XLOG_STATE_COVER_NEED
:
763 case XLOG_STATE_COVER_NEED2
:
764 if (!xfs_ail_min_lsn(log
->l_ailp
) &&
765 xlog_iclogs_empty(log
)) {
766 if (log
->l_covered_state
== XLOG_STATE_COVER_NEED
)
767 log
->l_covered_state
= XLOG_STATE_COVER_DONE
;
769 log
->l_covered_state
= XLOG_STATE_COVER_DONE2
;
776 spin_unlock(&log
->l_icloglock
);
780 /******************************************************************************
784 ******************************************************************************
787 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
788 * The log manager must keep track of the last LR which was committed
789 * to disk. The lsn of this LR will become the new tail_lsn whenever
790 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
791 * the situation where stuff could be written into the log but nothing
792 * was ever in the AIL when asked. Eventually, we panic since the
793 * tail hits the head.
795 * We may be holding the log iclog lock upon entering this routine.
798 xlog_assign_tail_lsn(
799 struct xfs_mount
*mp
)
802 struct log
*log
= mp
->m_log
;
804 tail_lsn
= xfs_ail_min_lsn(mp
->m_ail
);
806 tail_lsn
= atomic64_read(&log
->l_last_sync_lsn
);
808 atomic64_set(&log
->l_tail_lsn
, tail_lsn
);
813 * Return the space in the log between the tail and the head. The head
814 * is passed in the cycle/bytes formal parms. In the special case where
815 * the reserve head has wrapped passed the tail, this calculation is no
816 * longer valid. In this case, just return 0 which means there is no space
817 * in the log. This works for all places where this function is called
818 * with the reserve head. Of course, if the write head were to ever
819 * wrap the tail, we should blow up. Rather than catch this case here,
820 * we depend on other ASSERTions in other parts of the code. XXXmiken
822 * This code also handles the case where the reservation head is behind
823 * the tail. The details of this case are described below, but the end
824 * result is that we return the size of the log as the amount of space left.
837 xlog_crack_grant_head(head
, &head_cycle
, &head_bytes
);
838 xlog_crack_atomic_lsn(&log
->l_tail_lsn
, &tail_cycle
, &tail_bytes
);
839 tail_bytes
= BBTOB(tail_bytes
);
840 if (tail_cycle
== head_cycle
&& head_bytes
>= tail_bytes
)
841 free_bytes
= log
->l_logsize
- (head_bytes
- tail_bytes
);
842 else if (tail_cycle
+ 1 < head_cycle
)
844 else if (tail_cycle
< head_cycle
) {
845 ASSERT(tail_cycle
== (head_cycle
- 1));
846 free_bytes
= tail_bytes
- head_bytes
;
849 * The reservation head is behind the tail.
850 * In this case we just want to return the size of the
851 * log as the amount of space left.
854 "xlog_space_left: head behind tail\n"
855 " tail_cycle = %d, tail_bytes = %d\n"
856 " GH cycle = %d, GH bytes = %d",
857 tail_cycle
, tail_bytes
, head_cycle
, head_bytes
);
859 free_bytes
= log
->l_logsize
;
866 * Log function which is called when an io completes.
868 * The log manager needs its own routine, in order to control what
869 * happens with the buffer after the write completes.
872 xlog_iodone(xfs_buf_t
*bp
)
874 xlog_in_core_t
*iclog
= bp
->b_fspriv
;
875 xlog_t
*l
= iclog
->ic_log
;
879 * Race to shutdown the filesystem if we see an error.
881 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp
)), l
->l_mp
,
882 XFS_ERRTAG_IODONE_IOERR
, XFS_RANDOM_IODONE_IOERR
)) {
883 xfs_ioerror_alert("xlog_iodone", l
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
885 xfs_force_shutdown(l
->l_mp
, SHUTDOWN_LOG_IO_ERROR
);
887 * This flag will be propagated to the trans-committed
888 * callback routines to let them know that the log-commit
891 aborted
= XFS_LI_ABORTED
;
892 } else if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
893 aborted
= XFS_LI_ABORTED
;
896 /* log I/O is always issued ASYNC */
897 ASSERT(XFS_BUF_ISASYNC(bp
));
898 xlog_state_done_syncing(iclog
, aborted
);
900 * do not reference the buffer (bp) here as we could race
901 * with it being freed after writing the unmount record to the
908 * Return size of each in-core log record buffer.
910 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
912 * If the filesystem blocksize is too large, we may need to choose a
913 * larger size since the directory code currently logs entire blocks.
917 xlog_get_iclog_buffer_size(xfs_mount_t
*mp
,
923 if (mp
->m_logbufs
<= 0)
924 log
->l_iclog_bufs
= XLOG_MAX_ICLOGS
;
926 log
->l_iclog_bufs
= mp
->m_logbufs
;
929 * Buffer size passed in from mount system call.
931 if (mp
->m_logbsize
> 0) {
932 size
= log
->l_iclog_size
= mp
->m_logbsize
;
933 log
->l_iclog_size_log
= 0;
935 log
->l_iclog_size_log
++;
939 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
940 /* # headers = size / 32k
941 * one header holds cycles from 32k of data
944 xhdrs
= mp
->m_logbsize
/ XLOG_HEADER_CYCLE_SIZE
;
945 if (mp
->m_logbsize
% XLOG_HEADER_CYCLE_SIZE
)
947 log
->l_iclog_hsize
= xhdrs
<< BBSHIFT
;
948 log
->l_iclog_heads
= xhdrs
;
950 ASSERT(mp
->m_logbsize
<= XLOG_BIG_RECORD_BSIZE
);
951 log
->l_iclog_hsize
= BBSIZE
;
952 log
->l_iclog_heads
= 1;
957 /* All machines use 32kB buffers by default. */
958 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
959 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
961 /* the default log size is 16k or 32k which is one header sector */
962 log
->l_iclog_hsize
= BBSIZE
;
963 log
->l_iclog_heads
= 1;
966 /* are we being asked to make the sizes selected above visible? */
967 if (mp
->m_logbufs
== 0)
968 mp
->m_logbufs
= log
->l_iclog_bufs
;
969 if (mp
->m_logbsize
== 0)
970 mp
->m_logbsize
= log
->l_iclog_size
;
971 } /* xlog_get_iclog_buffer_size */
975 * This routine initializes some of the log structure for a given mount point.
976 * Its primary purpose is to fill in enough, so recovery can occur. However,
977 * some other stuff may be filled in too.
980 xlog_alloc_log(xfs_mount_t
*mp
,
981 xfs_buftarg_t
*log_target
,
982 xfs_daddr_t blk_offset
,
986 xlog_rec_header_t
*head
;
987 xlog_in_core_t
**iclogp
;
988 xlog_in_core_t
*iclog
, *prev_iclog
=NULL
;
994 log
= kmem_zalloc(sizeof(xlog_t
), KM_MAYFAIL
);
996 xfs_warn(mp
, "Log allocation failed: No memory!");
1001 log
->l_targ
= log_target
;
1002 log
->l_logsize
= BBTOB(num_bblks
);
1003 log
->l_logBBstart
= blk_offset
;
1004 log
->l_logBBsize
= num_bblks
;
1005 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1006 log
->l_flags
|= XLOG_ACTIVE_RECOVERY
;
1008 log
->l_prev_block
= -1;
1009 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1010 xlog_assign_atomic_lsn(&log
->l_tail_lsn
, 1, 0);
1011 xlog_assign_atomic_lsn(&log
->l_last_sync_lsn
, 1, 0);
1012 log
->l_curr_cycle
= 1; /* 0 is bad since this is initial value */
1013 xlog_assign_grant_head(&log
->l_grant_reserve_head
, 1, 0);
1014 xlog_assign_grant_head(&log
->l_grant_write_head
, 1, 0);
1015 INIT_LIST_HEAD(&log
->l_reserveq
);
1016 INIT_LIST_HEAD(&log
->l_writeq
);
1017 spin_lock_init(&log
->l_grant_reserve_lock
);
1018 spin_lock_init(&log
->l_grant_write_lock
);
1020 error
= EFSCORRUPTED
;
1021 if (xfs_sb_version_hassector(&mp
->m_sb
)) {
1022 log2_size
= mp
->m_sb
.sb_logsectlog
;
1023 if (log2_size
< BBSHIFT
) {
1024 xfs_warn(mp
, "Log sector size too small (0x%x < 0x%x)",
1025 log2_size
, BBSHIFT
);
1029 log2_size
-= BBSHIFT
;
1030 if (log2_size
> mp
->m_sectbb_log
) {
1031 xfs_warn(mp
, "Log sector size too large (0x%x > 0x%x)",
1032 log2_size
, mp
->m_sectbb_log
);
1036 /* for larger sector sizes, must have v2 or external log */
1037 if (log2_size
&& log
->l_logBBstart
> 0 &&
1038 !xfs_sb_version_haslogv2(&mp
->m_sb
)) {
1040 "log sector size (0x%x) invalid for configuration.",
1045 log
->l_sectBBsize
= 1 << log2_size
;
1047 xlog_get_iclog_buffer_size(mp
, log
);
1050 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1053 bp
->b_iodone
= xlog_iodone
;
1054 ASSERT(XFS_BUF_ISBUSY(bp
));
1055 ASSERT(xfs_buf_islocked(bp
));
1058 spin_lock_init(&log
->l_icloglock
);
1059 init_waitqueue_head(&log
->l_flush_wait
);
1061 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1062 ASSERT((XFS_BUF_SIZE(bp
) & BBMASK
) == 0);
1064 iclogp
= &log
->l_iclog
;
1066 * The amount of memory to allocate for the iclog structure is
1067 * rather funky due to the way the structure is defined. It is
1068 * done this way so that we can use different sizes for machines
1069 * with different amounts of memory. See the definition of
1070 * xlog_in_core_t in xfs_log_priv.h for details.
1072 ASSERT(log
->l_iclog_size
>= 4096);
1073 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
1074 *iclogp
= kmem_zalloc(sizeof(xlog_in_core_t
), KM_MAYFAIL
);
1076 goto out_free_iclog
;
1079 iclog
->ic_prev
= prev_iclog
;
1082 bp
= xfs_buf_get_uncached(mp
->m_logdev_targp
,
1083 log
->l_iclog_size
, 0);
1085 goto out_free_iclog
;
1087 bp
->b_iodone
= xlog_iodone
;
1089 iclog
->ic_data
= bp
->b_addr
;
1091 log
->l_iclog_bak
[i
] = (xfs_caddr_t
)&(iclog
->ic_header
);
1093 head
= &iclog
->ic_header
;
1094 memset(head
, 0, sizeof(xlog_rec_header_t
));
1095 head
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1096 head
->h_version
= cpu_to_be32(
1097 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1098 head
->h_size
= cpu_to_be32(log
->l_iclog_size
);
1100 head
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1101 memcpy(&head
->h_fs_uuid
, &mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1103 iclog
->ic_size
= XFS_BUF_SIZE(bp
) - log
->l_iclog_hsize
;
1104 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1105 iclog
->ic_log
= log
;
1106 atomic_set(&iclog
->ic_refcnt
, 0);
1107 spin_lock_init(&iclog
->ic_callback_lock
);
1108 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
1109 iclog
->ic_datap
= (char *)iclog
->ic_data
+ log
->l_iclog_hsize
;
1111 ASSERT(XFS_BUF_ISBUSY(iclog
->ic_bp
));
1112 ASSERT(xfs_buf_islocked(iclog
->ic_bp
));
1113 init_waitqueue_head(&iclog
->ic_force_wait
);
1114 init_waitqueue_head(&iclog
->ic_write_wait
);
1116 iclogp
= &iclog
->ic_next
;
1118 *iclogp
= log
->l_iclog
; /* complete ring */
1119 log
->l_iclog
->ic_prev
= prev_iclog
; /* re-write 1st prev ptr */
1121 error
= xlog_cil_init(log
);
1123 goto out_free_iclog
;
1127 for (iclog
= log
->l_iclog
; iclog
; iclog
= prev_iclog
) {
1128 prev_iclog
= iclog
->ic_next
;
1130 xfs_buf_free(iclog
->ic_bp
);
1133 spinlock_destroy(&log
->l_icloglock
);
1134 xfs_buf_free(log
->l_xbuf
);
1138 return ERR_PTR(-error
);
1139 } /* xlog_alloc_log */
1143 * Write out the commit record of a transaction associated with the given
1144 * ticket. Return the lsn of the commit record.
1149 struct xlog_ticket
*ticket
,
1150 struct xlog_in_core
**iclog
,
1151 xfs_lsn_t
*commitlsnp
)
1153 struct xfs_mount
*mp
= log
->l_mp
;
1155 struct xfs_log_iovec reg
= {
1158 .i_type
= XLOG_REG_TYPE_COMMIT
,
1160 struct xfs_log_vec vec
= {
1165 ASSERT_ALWAYS(iclog
);
1166 error
= xlog_write(log
, &vec
, ticket
, commitlsnp
, iclog
,
1169 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
1174 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1175 * log space. This code pushes on the lsn which would supposedly free up
1176 * the 25% which we want to leave free. We may need to adopt a policy which
1177 * pushes on an lsn which is further along in the log once we reach the high
1178 * water mark. In this manner, we would be creating a low water mark.
1181 xlog_grant_push_ail(
1185 xfs_lsn_t threshold_lsn
= 0;
1186 xfs_lsn_t last_sync_lsn
;
1189 int threshold_block
;
1190 int threshold_cycle
;
1193 ASSERT(BTOBB(need_bytes
) < log
->l_logBBsize
);
1195 free_bytes
= xlog_space_left(log
, &log
->l_grant_reserve_head
);
1196 free_blocks
= BTOBBT(free_bytes
);
1199 * Set the threshold for the minimum number of free blocks in the
1200 * log to the maximum of what the caller needs, one quarter of the
1201 * log, and 256 blocks.
1203 free_threshold
= BTOBB(need_bytes
);
1204 free_threshold
= MAX(free_threshold
, (log
->l_logBBsize
>> 2));
1205 free_threshold
= MAX(free_threshold
, 256);
1206 if (free_blocks
>= free_threshold
)
1209 xlog_crack_atomic_lsn(&log
->l_tail_lsn
, &threshold_cycle
,
1211 threshold_block
+= free_threshold
;
1212 if (threshold_block
>= log
->l_logBBsize
) {
1213 threshold_block
-= log
->l_logBBsize
;
1214 threshold_cycle
+= 1;
1216 threshold_lsn
= xlog_assign_lsn(threshold_cycle
,
1219 * Don't pass in an lsn greater than the lsn of the last
1220 * log record known to be on disk. Use a snapshot of the last sync lsn
1221 * so that it doesn't change between the compare and the set.
1223 last_sync_lsn
= atomic64_read(&log
->l_last_sync_lsn
);
1224 if (XFS_LSN_CMP(threshold_lsn
, last_sync_lsn
) > 0)
1225 threshold_lsn
= last_sync_lsn
;
1228 * Get the transaction layer to kick the dirty buffers out to
1229 * disk asynchronously. No point in trying to do this if
1230 * the filesystem is shutting down.
1232 if (!XLOG_FORCED_SHUTDOWN(log
))
1233 xfs_ail_push(log
->l_ailp
, threshold_lsn
);
1237 * The bdstrat callback function for log bufs. This gives us a central
1238 * place to trap bufs in case we get hit by a log I/O error and need to
1239 * shutdown. Actually, in practice, even when we didn't get a log error,
1240 * we transition the iclogs to IOERROR state *after* flushing all existing
1241 * iclogs to disk. This is because we don't want anymore new transactions to be
1242 * started or completed afterwards.
1248 struct xlog_in_core
*iclog
= bp
->b_fspriv
;
1250 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
1251 XFS_BUF_ERROR(bp
, EIO
);
1253 xfs_buf_ioend(bp
, 0);
1255 * It would seem logical to return EIO here, but we rely on
1256 * the log state machine to propagate I/O errors instead of
1262 xfs_buf_iorequest(bp
);
1267 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1268 * fashion. Previously, we should have moved the current iclog
1269 * ptr in the log to point to the next available iclog. This allows further
1270 * write to continue while this code syncs out an iclog ready to go.
1271 * Before an in-core log can be written out, the data section must be scanned
1272 * to save away the 1st word of each BBSIZE block into the header. We replace
1273 * it with the current cycle count. Each BBSIZE block is tagged with the
1274 * cycle count because there in an implicit assumption that drives will
1275 * guarantee that entire 512 byte blocks get written at once. In other words,
1276 * we can't have part of a 512 byte block written and part not written. By
1277 * tagging each block, we will know which blocks are valid when recovering
1278 * after an unclean shutdown.
1280 * This routine is single threaded on the iclog. No other thread can be in
1281 * this routine with the same iclog. Changing contents of iclog can there-
1282 * fore be done without grabbing the state machine lock. Updating the global
1283 * log will require grabbing the lock though.
1285 * The entire log manager uses a logical block numbering scheme. Only
1286 * log_sync (and then only bwrite()) know about the fact that the log may
1287 * not start with block zero on a given device. The log block start offset
1288 * is added immediately before calling bwrite().
1292 xlog_sync(xlog_t
*log
,
1293 xlog_in_core_t
*iclog
)
1295 xfs_caddr_t dptr
; /* pointer to byte sized element */
1298 uint count
; /* byte count of bwrite */
1299 uint count_init
; /* initial count before roundup */
1300 int roundoff
; /* roundoff to BB or stripe */
1301 int split
= 0; /* split write into two regions */
1303 int v2
= xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
);
1305 XFS_STATS_INC(xs_log_writes
);
1306 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
1308 /* Add for LR header */
1309 count_init
= log
->l_iclog_hsize
+ iclog
->ic_offset
;
1311 /* Round out the log write size */
1312 if (v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1) {
1313 /* we have a v2 stripe unit to use */
1314 count
= XLOG_LSUNITTOB(log
, XLOG_BTOLSUNIT(log
, count_init
));
1316 count
= BBTOB(BTOBB(count_init
));
1318 roundoff
= count
- count_init
;
1319 ASSERT(roundoff
>= 0);
1320 ASSERT((v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1 &&
1321 roundoff
< log
->l_mp
->m_sb
.sb_logsunit
)
1323 (log
->l_mp
->m_sb
.sb_logsunit
<= 1 &&
1324 roundoff
< BBTOB(1)));
1326 /* move grant heads by roundoff in sync */
1327 xlog_grant_add_space(log
, &log
->l_grant_reserve_head
, roundoff
);
1328 xlog_grant_add_space(log
, &log
->l_grant_write_head
, roundoff
);
1330 /* put cycle number in every block */
1331 xlog_pack_data(log
, iclog
, roundoff
);
1333 /* real byte length */
1335 iclog
->ic_header
.h_len
=
1336 cpu_to_be32(iclog
->ic_offset
+ roundoff
);
1338 iclog
->ic_header
.h_len
=
1339 cpu_to_be32(iclog
->ic_offset
);
1343 XFS_BUF_SET_ADDR(bp
, BLOCK_LSN(be64_to_cpu(iclog
->ic_header
.h_lsn
)));
1345 XFS_STATS_ADD(xs_log_blocks
, BTOBB(count
));
1347 /* Do we need to split this write into 2 parts? */
1348 if (XFS_BUF_ADDR(bp
) + BTOBB(count
) > log
->l_logBBsize
) {
1349 split
= count
- (BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
)));
1350 count
= BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
));
1351 iclog
->ic_bwritecnt
= 2; /* split into 2 writes */
1353 iclog
->ic_bwritecnt
= 1;
1355 XFS_BUF_SET_COUNT(bp
, count
);
1356 bp
->b_fspriv
= iclog
;
1357 XFS_BUF_ZEROFLAGS(bp
);
1360 bp
->b_flags
|= XBF_SYNCIO
;
1362 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
) {
1363 bp
->b_flags
|= XBF_FUA
;
1366 * Flush the data device before flushing the log to make
1367 * sure all meta data written back from the AIL actually made
1368 * it to disk before stamping the new log tail LSN into the
1369 * log buffer. For an external log we need to issue the
1370 * flush explicitly, and unfortunately synchronously here;
1371 * for an internal log we can simply use the block layer
1372 * state machine for preflushes.
1374 if (log
->l_mp
->m_logdev_targp
!= log
->l_mp
->m_ddev_targp
)
1375 xfs_blkdev_issue_flush(log
->l_mp
->m_ddev_targp
);
1377 bp
->b_flags
|= XBF_FLUSH
;
1380 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1381 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1383 xlog_verify_iclog(log
, iclog
, count
, B_TRUE
);
1385 /* account for log which doesn't start at block #0 */
1386 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1388 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1393 if ((error
= xlog_bdstrat(bp
))) {
1394 xfs_ioerror_alert("xlog_sync", log
->l_mp
, bp
,
1399 bp
= iclog
->ic_log
->l_xbuf
;
1400 XFS_BUF_SET_ADDR(bp
, 0); /* logical 0 */
1401 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
)((__psint_t
)&(iclog
->ic_header
)+
1402 (__psint_t
)count
), split
);
1403 bp
->b_fspriv
= iclog
;
1404 XFS_BUF_ZEROFLAGS(bp
);
1407 bp
->b_flags
|= XBF_SYNCIO
;
1408 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1409 bp
->b_flags
|= XBF_FUA
;
1410 dptr
= XFS_BUF_PTR(bp
);
1412 * Bump the cycle numbers at the start of each block
1413 * since this part of the buffer is at the start of
1414 * a new cycle. Watch out for the header magic number
1417 for (i
= 0; i
< split
; i
+= BBSIZE
) {
1418 be32_add_cpu((__be32
*)dptr
, 1);
1419 if (be32_to_cpu(*(__be32
*)dptr
) == XLOG_HEADER_MAGIC_NUM
)
1420 be32_add_cpu((__be32
*)dptr
, 1);
1424 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1425 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1427 /* account for internal log which doesn't start at block #0 */
1428 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1430 if ((error
= xlog_bdstrat(bp
))) {
1431 xfs_ioerror_alert("xlog_sync (split)", log
->l_mp
,
1432 bp
, XFS_BUF_ADDR(bp
));
1441 * Deallocate a log structure
1444 xlog_dealloc_log(xlog_t
*log
)
1446 xlog_in_core_t
*iclog
, *next_iclog
;
1449 xlog_cil_destroy(log
);
1452 * always need to ensure that the extra buffer does not point to memory
1453 * owned by another log buffer before we free it.
1455 xfs_buf_set_empty(log
->l_xbuf
, log
->l_iclog_size
);
1456 xfs_buf_free(log
->l_xbuf
);
1458 iclog
= log
->l_iclog
;
1459 for (i
=0; i
<log
->l_iclog_bufs
; i
++) {
1460 xfs_buf_free(iclog
->ic_bp
);
1461 next_iclog
= iclog
->ic_next
;
1465 spinlock_destroy(&log
->l_icloglock
);
1467 log
->l_mp
->m_log
= NULL
;
1469 } /* xlog_dealloc_log */
1472 * Update counters atomically now that memcpy is done.
1476 xlog_state_finish_copy(xlog_t
*log
,
1477 xlog_in_core_t
*iclog
,
1481 spin_lock(&log
->l_icloglock
);
1483 be32_add_cpu(&iclog
->ic_header
.h_num_logops
, record_cnt
);
1484 iclog
->ic_offset
+= copy_bytes
;
1486 spin_unlock(&log
->l_icloglock
);
1487 } /* xlog_state_finish_copy */
1493 * print out info relating to regions written which consume
1498 struct xfs_mount
*mp
,
1499 struct xlog_ticket
*ticket
)
1502 uint ophdr_spc
= ticket
->t_res_num_ophdrs
* (uint
)sizeof(xlog_op_header_t
);
1504 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1505 static char *res_type_str
[XLOG_REG_TYPE_MAX
] = {
1526 static char *trans_type_str
[XFS_TRANS_TYPE_MAX
] = {
1570 "xfs_log_write: reservation summary:\n"
1571 " trans type = %s (%u)\n"
1572 " unit res = %d bytes\n"
1573 " current res = %d bytes\n"
1574 " total reg = %u bytes (o/flow = %u bytes)\n"
1575 " ophdrs = %u (ophdr space = %u bytes)\n"
1576 " ophdr + reg = %u bytes\n"
1577 " num regions = %u\n",
1578 ((ticket
->t_trans_type
<= 0 ||
1579 ticket
->t_trans_type
> XFS_TRANS_TYPE_MAX
) ?
1580 "bad-trans-type" : trans_type_str
[ticket
->t_trans_type
-1]),
1581 ticket
->t_trans_type
,
1584 ticket
->t_res_arr_sum
, ticket
->t_res_o_flow
,
1585 ticket
->t_res_num_ophdrs
, ophdr_spc
,
1586 ticket
->t_res_arr_sum
+
1587 ticket
->t_res_o_flow
+ ophdr_spc
,
1590 for (i
= 0; i
< ticket
->t_res_num
; i
++) {
1591 uint r_type
= ticket
->t_res_arr
[i
].r_type
;
1592 xfs_warn(mp
, "region[%u]: %s - %u bytes\n", i
,
1593 ((r_type
<= 0 || r_type
> XLOG_REG_TYPE_MAX
) ?
1594 "bad-rtype" : res_type_str
[r_type
-1]),
1595 ticket
->t_res_arr
[i
].r_len
);
1598 xfs_alert_tag(mp
, XFS_PTAG_LOGRES
,
1599 "xfs_log_write: reservation ran out. Need to up reservation");
1600 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
1604 * Calculate the potential space needed by the log vector. Each region gets
1605 * its own xlog_op_header_t and may need to be double word aligned.
1608 xlog_write_calc_vec_length(
1609 struct xlog_ticket
*ticket
,
1610 struct xfs_log_vec
*log_vector
)
1612 struct xfs_log_vec
*lv
;
1617 /* acct for start rec of xact */
1618 if (ticket
->t_flags
& XLOG_TIC_INITED
)
1621 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
1622 headers
+= lv
->lv_niovecs
;
1624 for (i
= 0; i
< lv
->lv_niovecs
; i
++) {
1625 struct xfs_log_iovec
*vecp
= &lv
->lv_iovecp
[i
];
1628 xlog_tic_add_region(ticket
, vecp
->i_len
, vecp
->i_type
);
1632 ticket
->t_res_num_ophdrs
+= headers
;
1633 len
+= headers
* sizeof(struct xlog_op_header
);
1639 * If first write for transaction, insert start record We can't be trying to
1640 * commit if we are inited. We can't have any "partial_copy" if we are inited.
1643 xlog_write_start_rec(
1644 struct xlog_op_header
*ophdr
,
1645 struct xlog_ticket
*ticket
)
1647 if (!(ticket
->t_flags
& XLOG_TIC_INITED
))
1650 ophdr
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1651 ophdr
->oh_clientid
= ticket
->t_clientid
;
1653 ophdr
->oh_flags
= XLOG_START_TRANS
;
1656 ticket
->t_flags
&= ~XLOG_TIC_INITED
;
1658 return sizeof(struct xlog_op_header
);
1661 static xlog_op_header_t
*
1662 xlog_write_setup_ophdr(
1664 struct xlog_op_header
*ophdr
,
1665 struct xlog_ticket
*ticket
,
1668 ophdr
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1669 ophdr
->oh_clientid
= ticket
->t_clientid
;
1672 /* are we copying a commit or unmount record? */
1673 ophdr
->oh_flags
= flags
;
1676 * We've seen logs corrupted with bad transaction client ids. This
1677 * makes sure that XFS doesn't generate them on. Turn this into an EIO
1678 * and shut down the filesystem.
1680 switch (ophdr
->oh_clientid
) {
1681 case XFS_TRANSACTION
:
1687 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1688 ophdr
->oh_clientid
, ticket
);
1696 * Set up the parameters of the region copy into the log. This has
1697 * to handle region write split across multiple log buffers - this
1698 * state is kept external to this function so that this code can
1699 * can be written in an obvious, self documenting manner.
1702 xlog_write_setup_copy(
1703 struct xlog_ticket
*ticket
,
1704 struct xlog_op_header
*ophdr
,
1705 int space_available
,
1709 int *last_was_partial_copy
,
1710 int *bytes_consumed
)
1714 still_to_copy
= space_required
- *bytes_consumed
;
1715 *copy_off
= *bytes_consumed
;
1717 if (still_to_copy
<= space_available
) {
1718 /* write of region completes here */
1719 *copy_len
= still_to_copy
;
1720 ophdr
->oh_len
= cpu_to_be32(*copy_len
);
1721 if (*last_was_partial_copy
)
1722 ophdr
->oh_flags
|= (XLOG_END_TRANS
|XLOG_WAS_CONT_TRANS
);
1723 *last_was_partial_copy
= 0;
1724 *bytes_consumed
= 0;
1728 /* partial write of region, needs extra log op header reservation */
1729 *copy_len
= space_available
;
1730 ophdr
->oh_len
= cpu_to_be32(*copy_len
);
1731 ophdr
->oh_flags
|= XLOG_CONTINUE_TRANS
;
1732 if (*last_was_partial_copy
)
1733 ophdr
->oh_flags
|= XLOG_WAS_CONT_TRANS
;
1734 *bytes_consumed
+= *copy_len
;
1735 (*last_was_partial_copy
)++;
1737 /* account for new log op header */
1738 ticket
->t_curr_res
-= sizeof(struct xlog_op_header
);
1739 ticket
->t_res_num_ophdrs
++;
1741 return sizeof(struct xlog_op_header
);
1745 xlog_write_copy_finish(
1747 struct xlog_in_core
*iclog
,
1752 int *partial_copy_len
,
1754 struct xlog_in_core
**commit_iclog
)
1756 if (*partial_copy
) {
1758 * This iclog has already been marked WANT_SYNC by
1759 * xlog_state_get_iclog_space.
1761 xlog_state_finish_copy(log
, iclog
, *record_cnt
, *data_cnt
);
1764 return xlog_state_release_iclog(log
, iclog
);
1768 *partial_copy_len
= 0;
1770 if (iclog
->ic_size
- log_offset
<= sizeof(xlog_op_header_t
)) {
1771 /* no more space in this iclog - push it. */
1772 xlog_state_finish_copy(log
, iclog
, *record_cnt
, *data_cnt
);
1776 spin_lock(&log
->l_icloglock
);
1777 xlog_state_want_sync(log
, iclog
);
1778 spin_unlock(&log
->l_icloglock
);
1781 return xlog_state_release_iclog(log
, iclog
);
1782 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1783 *commit_iclog
= iclog
;
1790 * Write some region out to in-core log
1792 * This will be called when writing externally provided regions or when
1793 * writing out a commit record for a given transaction.
1795 * General algorithm:
1796 * 1. Find total length of this write. This may include adding to the
1797 * lengths passed in.
1798 * 2. Check whether we violate the tickets reservation.
1799 * 3. While writing to this iclog
1800 * A. Reserve as much space in this iclog as can get
1801 * B. If this is first write, save away start lsn
1802 * C. While writing this region:
1803 * 1. If first write of transaction, write start record
1804 * 2. Write log operation header (header per region)
1805 * 3. Find out if we can fit entire region into this iclog
1806 * 4. Potentially, verify destination memcpy ptr
1807 * 5. Memcpy (partial) region
1808 * 6. If partial copy, release iclog; otherwise, continue
1809 * copying more regions into current iclog
1810 * 4. Mark want sync bit (in simulation mode)
1811 * 5. Release iclog for potential flush to on-disk log.
1814 * 1. Panic if reservation is overrun. This should never happen since
1815 * reservation amounts are generated internal to the filesystem.
1817 * 1. Tickets are single threaded data structures.
1818 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1819 * syncing routine. When a single log_write region needs to span
1820 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1821 * on all log operation writes which don't contain the end of the
1822 * region. The XLOG_END_TRANS bit is used for the in-core log
1823 * operation which contains the end of the continued log_write region.
1824 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1825 * we don't really know exactly how much space will be used. As a result,
1826 * we don't update ic_offset until the end when we know exactly how many
1827 * bytes have been written out.
1832 struct xfs_log_vec
*log_vector
,
1833 struct xlog_ticket
*ticket
,
1834 xfs_lsn_t
*start_lsn
,
1835 struct xlog_in_core
**commit_iclog
,
1838 struct xlog_in_core
*iclog
= NULL
;
1839 struct xfs_log_iovec
*vecp
;
1840 struct xfs_log_vec
*lv
;
1843 int partial_copy
= 0;
1844 int partial_copy_len
= 0;
1852 len
= xlog_write_calc_vec_length(ticket
, log_vector
);
1855 * Region headers and bytes are already accounted for.
1856 * We only need to take into account start records and
1857 * split regions in this function.
1859 if (ticket
->t_flags
& XLOG_TIC_INITED
)
1860 ticket
->t_curr_res
-= sizeof(xlog_op_header_t
);
1863 * Commit record headers need to be accounted for. These
1864 * come in as separate writes so are easy to detect.
1866 if (flags
& (XLOG_COMMIT_TRANS
| XLOG_UNMOUNT_TRANS
))
1867 ticket
->t_curr_res
-= sizeof(xlog_op_header_t
);
1869 ticket
->t_curr_res
-= len
;
1871 if (ticket
->t_curr_res
< 0)
1872 xlog_print_tic_res(log
->l_mp
, ticket
);
1876 vecp
= lv
->lv_iovecp
;
1877 while (lv
&& index
< lv
->lv_niovecs
) {
1881 error
= xlog_state_get_iclog_space(log
, len
, &iclog
, ticket
,
1882 &contwr
, &log_offset
);
1886 ASSERT(log_offset
<= iclog
->ic_size
- 1);
1887 ptr
= iclog
->ic_datap
+ log_offset
;
1889 /* start_lsn is the first lsn written to. That's all we need. */
1891 *start_lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
1894 * This loop writes out as many regions as can fit in the amount
1895 * of space which was allocated by xlog_state_get_iclog_space().
1897 while (lv
&& index
< lv
->lv_niovecs
) {
1898 struct xfs_log_iovec
*reg
= &vecp
[index
];
1899 struct xlog_op_header
*ophdr
;
1904 ASSERT(reg
->i_len
% sizeof(__int32_t
) == 0);
1905 ASSERT((unsigned long)ptr
% sizeof(__int32_t
) == 0);
1907 start_rec_copy
= xlog_write_start_rec(ptr
, ticket
);
1908 if (start_rec_copy
) {
1910 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
,
1914 ophdr
= xlog_write_setup_ophdr(log
, ptr
, ticket
, flags
);
1916 return XFS_ERROR(EIO
);
1918 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
,
1919 sizeof(struct xlog_op_header
));
1921 len
+= xlog_write_setup_copy(ticket
, ophdr
,
1922 iclog
->ic_size
-log_offset
,
1924 ©_off
, ©_len
,
1927 xlog_verify_dest_ptr(log
, ptr
);
1930 ASSERT(copy_len
>= 0);
1931 memcpy(ptr
, reg
->i_addr
+ copy_off
, copy_len
);
1932 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
, copy_len
);
1934 copy_len
+= start_rec_copy
+ sizeof(xlog_op_header_t
);
1936 data_cnt
+= contwr
? copy_len
: 0;
1938 error
= xlog_write_copy_finish(log
, iclog
, flags
,
1939 &record_cnt
, &data_cnt
,
1948 * if we had a partial copy, we need to get more iclog
1949 * space but we don't want to increment the region
1950 * index because there is still more is this region to
1953 * If we completed writing this region, and we flushed
1954 * the iclog (indicated by resetting of the record
1955 * count), then we also need to get more log space. If
1956 * this was the last record, though, we are done and
1962 if (++index
== lv
->lv_niovecs
) {
1966 vecp
= lv
->lv_iovecp
;
1968 if (record_cnt
== 0) {
1978 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1980 return xlog_state_release_iclog(log
, iclog
);
1982 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1983 *commit_iclog
= iclog
;
1988 /*****************************************************************************
1990 * State Machine functions
1992 *****************************************************************************
1995 /* Clean iclogs starting from the head. This ordering must be
1996 * maintained, so an iclog doesn't become ACTIVE beyond one that
1997 * is SYNCING. This is also required to maintain the notion that we use
1998 * a ordered wait queue to hold off would be writers to the log when every
1999 * iclog is trying to sync to disk.
2001 * State Change: DIRTY -> ACTIVE
2004 xlog_state_clean_log(xlog_t
*log
)
2006 xlog_in_core_t
*iclog
;
2009 iclog
= log
->l_iclog
;
2011 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2012 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
2013 iclog
->ic_offset
= 0;
2014 ASSERT(iclog
->ic_callback
== NULL
);
2016 * If the number of ops in this iclog indicate it just
2017 * contains the dummy transaction, we can
2018 * change state into IDLE (the second time around).
2019 * Otherwise we should change the state into
2021 * We don't need to cover the dummy.
2024 (be32_to_cpu(iclog
->ic_header
.h_num_logops
) ==
2029 * We have two dirty iclogs so start over
2030 * This could also be num of ops indicates
2031 * this is not the dummy going out.
2035 iclog
->ic_header
.h_num_logops
= 0;
2036 memset(iclog
->ic_header
.h_cycle_data
, 0,
2037 sizeof(iclog
->ic_header
.h_cycle_data
));
2038 iclog
->ic_header
.h_lsn
= 0;
2039 } else if (iclog
->ic_state
== XLOG_STATE_ACTIVE
)
2042 break; /* stop cleaning */
2043 iclog
= iclog
->ic_next
;
2044 } while (iclog
!= log
->l_iclog
);
2046 /* log is locked when we are called */
2048 * Change state for the dummy log recording.
2049 * We usually go to NEED. But we go to NEED2 if the changed indicates
2050 * we are done writing the dummy record.
2051 * If we are done with the second dummy recored (DONE2), then
2055 switch (log
->l_covered_state
) {
2056 case XLOG_STATE_COVER_IDLE
:
2057 case XLOG_STATE_COVER_NEED
:
2058 case XLOG_STATE_COVER_NEED2
:
2059 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2062 case XLOG_STATE_COVER_DONE
:
2064 log
->l_covered_state
= XLOG_STATE_COVER_NEED2
;
2066 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2069 case XLOG_STATE_COVER_DONE2
:
2071 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
2073 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2080 } /* xlog_state_clean_log */
2083 xlog_get_lowest_lsn(
2086 xlog_in_core_t
*lsn_log
;
2087 xfs_lsn_t lowest_lsn
, lsn
;
2089 lsn_log
= log
->l_iclog
;
2092 if (!(lsn_log
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
))) {
2093 lsn
= be64_to_cpu(lsn_log
->ic_header
.h_lsn
);
2094 if ((lsn
&& !lowest_lsn
) ||
2095 (XFS_LSN_CMP(lsn
, lowest_lsn
) < 0)) {
2099 lsn_log
= lsn_log
->ic_next
;
2100 } while (lsn_log
!= log
->l_iclog
);
2106 xlog_state_do_callback(
2109 xlog_in_core_t
*ciclog
)
2111 xlog_in_core_t
*iclog
;
2112 xlog_in_core_t
*first_iclog
; /* used to know when we've
2113 * processed all iclogs once */
2114 xfs_log_callback_t
*cb
, *cb_next
;
2116 xfs_lsn_t lowest_lsn
;
2117 int ioerrors
; /* counter: iclogs with errors */
2118 int loopdidcallbacks
; /* flag: inner loop did callbacks*/
2119 int funcdidcallbacks
; /* flag: function did callbacks */
2120 int repeats
; /* for issuing console warnings if
2121 * looping too many times */
2124 spin_lock(&log
->l_icloglock
);
2125 first_iclog
= iclog
= log
->l_iclog
;
2127 funcdidcallbacks
= 0;
2132 * Scan all iclogs starting with the one pointed to by the
2133 * log. Reset this starting point each time the log is
2134 * unlocked (during callbacks).
2136 * Keep looping through iclogs until one full pass is made
2137 * without running any callbacks.
2139 first_iclog
= log
->l_iclog
;
2140 iclog
= log
->l_iclog
;
2141 loopdidcallbacks
= 0;
2146 /* skip all iclogs in the ACTIVE & DIRTY states */
2147 if (iclog
->ic_state
&
2148 (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
)) {
2149 iclog
= iclog
->ic_next
;
2154 * Between marking a filesystem SHUTDOWN and stopping
2155 * the log, we do flush all iclogs to disk (if there
2156 * wasn't a log I/O error). So, we do want things to
2157 * go smoothly in case of just a SHUTDOWN w/o a
2160 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
2162 * Can only perform callbacks in order. Since
2163 * this iclog is not in the DONE_SYNC/
2164 * DO_CALLBACK state, we skip the rest and
2165 * just try to clean up. If we set our iclog
2166 * to DO_CALLBACK, we will not process it when
2167 * we retry since a previous iclog is in the
2168 * CALLBACK and the state cannot change since
2169 * we are holding the l_icloglock.
2171 if (!(iclog
->ic_state
&
2172 (XLOG_STATE_DONE_SYNC
|
2173 XLOG_STATE_DO_CALLBACK
))) {
2174 if (ciclog
&& (ciclog
->ic_state
==
2175 XLOG_STATE_DONE_SYNC
)) {
2176 ciclog
->ic_state
= XLOG_STATE_DO_CALLBACK
;
2181 * We now have an iclog that is in either the
2182 * DO_CALLBACK or DONE_SYNC states. The other
2183 * states (WANT_SYNC, SYNCING, or CALLBACK were
2184 * caught by the above if and are going to
2185 * clean (i.e. we aren't doing their callbacks)
2190 * We will do one more check here to see if we
2191 * have chased our tail around.
2194 lowest_lsn
= xlog_get_lowest_lsn(log
);
2196 XFS_LSN_CMP(lowest_lsn
,
2197 be64_to_cpu(iclog
->ic_header
.h_lsn
)) < 0) {
2198 iclog
= iclog
->ic_next
;
2199 continue; /* Leave this iclog for
2203 iclog
->ic_state
= XLOG_STATE_CALLBACK
;
2207 * update the last_sync_lsn before we drop the
2208 * icloglock to ensure we are the only one that
2211 ASSERT(XFS_LSN_CMP(atomic64_read(&log
->l_last_sync_lsn
),
2212 be64_to_cpu(iclog
->ic_header
.h_lsn
)) <= 0);
2213 atomic64_set(&log
->l_last_sync_lsn
,
2214 be64_to_cpu(iclog
->ic_header
.h_lsn
));
2219 spin_unlock(&log
->l_icloglock
);
2222 * Keep processing entries in the callback list until
2223 * we come around and it is empty. We need to
2224 * atomically see that the list is empty and change the
2225 * state to DIRTY so that we don't miss any more
2226 * callbacks being added.
2228 spin_lock(&iclog
->ic_callback_lock
);
2229 cb
= iclog
->ic_callback
;
2231 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
2232 iclog
->ic_callback
= NULL
;
2233 spin_unlock(&iclog
->ic_callback_lock
);
2235 /* perform callbacks in the order given */
2236 for (; cb
; cb
= cb_next
) {
2237 cb_next
= cb
->cb_next
;
2238 cb
->cb_func(cb
->cb_arg
, aborted
);
2240 spin_lock(&iclog
->ic_callback_lock
);
2241 cb
= iclog
->ic_callback
;
2247 spin_lock(&log
->l_icloglock
);
2248 ASSERT(iclog
->ic_callback
== NULL
);
2249 spin_unlock(&iclog
->ic_callback_lock
);
2250 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
))
2251 iclog
->ic_state
= XLOG_STATE_DIRTY
;
2254 * Transition from DIRTY to ACTIVE if applicable.
2255 * NOP if STATE_IOERROR.
2257 xlog_state_clean_log(log
);
2259 /* wake up threads waiting in xfs_log_force() */
2260 wake_up_all(&iclog
->ic_force_wait
);
2262 iclog
= iclog
->ic_next
;
2263 } while (first_iclog
!= iclog
);
2265 if (repeats
> 5000) {
2266 flushcnt
+= repeats
;
2269 "%s: possible infinite loop (%d iterations)",
2270 __func__
, flushcnt
);
2272 } while (!ioerrors
&& loopdidcallbacks
);
2275 * make one last gasp attempt to see if iclogs are being left in
2279 if (funcdidcallbacks
) {
2280 first_iclog
= iclog
= log
->l_iclog
;
2282 ASSERT(iclog
->ic_state
!= XLOG_STATE_DO_CALLBACK
);
2284 * Terminate the loop if iclogs are found in states
2285 * which will cause other threads to clean up iclogs.
2287 * SYNCING - i/o completion will go through logs
2288 * DONE_SYNC - interrupt thread should be waiting for
2290 * IOERROR - give up hope all ye who enter here
2292 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
||
2293 iclog
->ic_state
== XLOG_STATE_SYNCING
||
2294 iclog
->ic_state
== XLOG_STATE_DONE_SYNC
||
2295 iclog
->ic_state
== XLOG_STATE_IOERROR
)
2297 iclog
= iclog
->ic_next
;
2298 } while (first_iclog
!= iclog
);
2302 if (log
->l_iclog
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_IOERROR
))
2304 spin_unlock(&log
->l_icloglock
);
2307 wake_up_all(&log
->l_flush_wait
);
2312 * Finish transitioning this iclog to the dirty state.
2314 * Make sure that we completely execute this routine only when this is
2315 * the last call to the iclog. There is a good chance that iclog flushes,
2316 * when we reach the end of the physical log, get turned into 2 separate
2317 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2318 * routine. By using the reference count bwritecnt, we guarantee that only
2319 * the second completion goes through.
2321 * Callbacks could take time, so they are done outside the scope of the
2322 * global state machine log lock.
2325 xlog_state_done_syncing(
2326 xlog_in_core_t
*iclog
,
2329 xlog_t
*log
= iclog
->ic_log
;
2331 spin_lock(&log
->l_icloglock
);
2333 ASSERT(iclog
->ic_state
== XLOG_STATE_SYNCING
||
2334 iclog
->ic_state
== XLOG_STATE_IOERROR
);
2335 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
2336 ASSERT(iclog
->ic_bwritecnt
== 1 || iclog
->ic_bwritecnt
== 2);
2340 * If we got an error, either on the first buffer, or in the case of
2341 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2342 * and none should ever be attempted to be written to disk
2345 if (iclog
->ic_state
!= XLOG_STATE_IOERROR
) {
2346 if (--iclog
->ic_bwritecnt
== 1) {
2347 spin_unlock(&log
->l_icloglock
);
2350 iclog
->ic_state
= XLOG_STATE_DONE_SYNC
;
2354 * Someone could be sleeping prior to writing out the next
2355 * iclog buffer, we wake them all, one will get to do the
2356 * I/O, the others get to wait for the result.
2358 wake_up_all(&iclog
->ic_write_wait
);
2359 spin_unlock(&log
->l_icloglock
);
2360 xlog_state_do_callback(log
, aborted
, iclog
); /* also cleans log */
2361 } /* xlog_state_done_syncing */
2365 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2366 * sleep. We wait on the flush queue on the head iclog as that should be
2367 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2368 * we will wait here and all new writes will sleep until a sync completes.
2370 * The in-core logs are used in a circular fashion. They are not used
2371 * out-of-order even when an iclog past the head is free.
2374 * * log_offset where xlog_write() can start writing into the in-core
2376 * * in-core log pointer to which xlog_write() should write.
2377 * * boolean indicating this is a continued write to an in-core log.
2378 * If this is the last write, then the in-core log's offset field
2379 * needs to be incremented, depending on the amount of data which
2383 xlog_state_get_iclog_space(xlog_t
*log
,
2385 xlog_in_core_t
**iclogp
,
2386 xlog_ticket_t
*ticket
,
2387 int *continued_write
,
2391 xlog_rec_header_t
*head
;
2392 xlog_in_core_t
*iclog
;
2396 spin_lock(&log
->l_icloglock
);
2397 if (XLOG_FORCED_SHUTDOWN(log
)) {
2398 spin_unlock(&log
->l_icloglock
);
2399 return XFS_ERROR(EIO
);
2402 iclog
= log
->l_iclog
;
2403 if (iclog
->ic_state
!= XLOG_STATE_ACTIVE
) {
2404 XFS_STATS_INC(xs_log_noiclogs
);
2406 /* Wait for log writes to have flushed */
2407 xlog_wait(&log
->l_flush_wait
, &log
->l_icloglock
);
2411 head
= &iclog
->ic_header
;
2413 atomic_inc(&iclog
->ic_refcnt
); /* prevents sync */
2414 log_offset
= iclog
->ic_offset
;
2416 /* On the 1st write to an iclog, figure out lsn. This works
2417 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2418 * committing to. If the offset is set, that's how many blocks
2421 if (log_offset
== 0) {
2422 ticket
->t_curr_res
-= log
->l_iclog_hsize
;
2423 xlog_tic_add_region(ticket
,
2425 XLOG_REG_TYPE_LRHEADER
);
2426 head
->h_cycle
= cpu_to_be32(log
->l_curr_cycle
);
2427 head
->h_lsn
= cpu_to_be64(
2428 xlog_assign_lsn(log
->l_curr_cycle
, log
->l_curr_block
));
2429 ASSERT(log
->l_curr_block
>= 0);
2432 /* If there is enough room to write everything, then do it. Otherwise,
2433 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2434 * bit is on, so this will get flushed out. Don't update ic_offset
2435 * until you know exactly how many bytes get copied. Therefore, wait
2436 * until later to update ic_offset.
2438 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2439 * can fit into remaining data section.
2441 if (iclog
->ic_size
- iclog
->ic_offset
< 2*sizeof(xlog_op_header_t
)) {
2442 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2445 * If I'm the only one writing to this iclog, sync it to disk.
2446 * We need to do an atomic compare and decrement here to avoid
2447 * racing with concurrent atomic_dec_and_lock() calls in
2448 * xlog_state_release_iclog() when there is more than one
2449 * reference to the iclog.
2451 if (!atomic_add_unless(&iclog
->ic_refcnt
, -1, 1)) {
2452 /* we are the only one */
2453 spin_unlock(&log
->l_icloglock
);
2454 error
= xlog_state_release_iclog(log
, iclog
);
2458 spin_unlock(&log
->l_icloglock
);
2463 /* Do we have enough room to write the full amount in the remainder
2464 * of this iclog? Or must we continue a write on the next iclog and
2465 * mark this iclog as completely taken? In the case where we switch
2466 * iclogs (to mark it taken), this particular iclog will release/sync
2467 * to disk in xlog_write().
2469 if (len
<= iclog
->ic_size
- iclog
->ic_offset
) {
2470 *continued_write
= 0;
2471 iclog
->ic_offset
+= len
;
2473 *continued_write
= 1;
2474 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2478 ASSERT(iclog
->ic_offset
<= iclog
->ic_size
);
2479 spin_unlock(&log
->l_icloglock
);
2481 *logoffsetp
= log_offset
;
2483 } /* xlog_state_get_iclog_space */
2486 * Atomically get the log space required for a log ticket.
2488 * Once a ticket gets put onto the reserveq, it will only return after
2489 * the needed reservation is satisfied.
2491 * This function is structured so that it has a lock free fast path. This is
2492 * necessary because every new transaction reservation will come through this
2493 * path. Hence any lock will be globally hot if we take it unconditionally on
2496 * As tickets are only ever moved on and off the reserveq under the
2497 * l_grant_reserve_lock, we only need to take that lock if we are going
2498 * to add the ticket to the queue and sleep. We can avoid taking the lock if the
2499 * ticket was never added to the reserveq because the t_queue list head will be
2500 * empty and we hold the only reference to it so it can safely be checked
2504 xlog_grant_log_space(xlog_t
*log
,
2511 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2512 panic("grant Recovery problem");
2515 trace_xfs_log_grant_enter(log
, tic
);
2517 need_bytes
= tic
->t_unit_res
;
2518 if (tic
->t_flags
& XFS_LOG_PERM_RESERV
)
2519 need_bytes
*= tic
->t_ocnt
;
2521 /* something is already sleeping; insert new transaction at end */
2522 if (!list_empty_careful(&log
->l_reserveq
)) {
2523 spin_lock(&log
->l_grant_reserve_lock
);
2524 /* recheck the queue now we are locked */
2525 if (list_empty(&log
->l_reserveq
)) {
2526 spin_unlock(&log
->l_grant_reserve_lock
);
2529 list_add_tail(&tic
->t_queue
, &log
->l_reserveq
);
2531 trace_xfs_log_grant_sleep1(log
, tic
);
2534 * Gotta check this before going to sleep, while we're
2535 * holding the grant lock.
2537 if (XLOG_FORCED_SHUTDOWN(log
))
2540 XFS_STATS_INC(xs_sleep_logspace
);
2541 xlog_wait(&tic
->t_wait
, &log
->l_grant_reserve_lock
);
2544 * If we got an error, and the filesystem is shutting down,
2545 * we'll catch it down below. So just continue...
2547 trace_xfs_log_grant_wake1(log
, tic
);
2551 if (XLOG_FORCED_SHUTDOWN(log
))
2552 goto error_return_unlocked
;
2554 free_bytes
= xlog_space_left(log
, &log
->l_grant_reserve_head
);
2555 if (free_bytes
< need_bytes
) {
2556 spin_lock(&log
->l_grant_reserve_lock
);
2557 if (list_empty(&tic
->t_queue
))
2558 list_add_tail(&tic
->t_queue
, &log
->l_reserveq
);
2560 trace_xfs_log_grant_sleep2(log
, tic
);
2562 if (XLOG_FORCED_SHUTDOWN(log
))
2565 xlog_grant_push_ail(log
, need_bytes
);
2567 XFS_STATS_INC(xs_sleep_logspace
);
2568 xlog_wait(&tic
->t_wait
, &log
->l_grant_reserve_lock
);
2570 trace_xfs_log_grant_wake2(log
, tic
);
2574 if (!list_empty(&tic
->t_queue
)) {
2575 spin_lock(&log
->l_grant_reserve_lock
);
2576 list_del_init(&tic
->t_queue
);
2577 spin_unlock(&log
->l_grant_reserve_lock
);
2580 /* we've got enough space */
2581 xlog_grant_add_space(log
, &log
->l_grant_reserve_head
, need_bytes
);
2582 xlog_grant_add_space(log
, &log
->l_grant_write_head
, need_bytes
);
2583 trace_xfs_log_grant_exit(log
, tic
);
2584 xlog_verify_grant_tail(log
);
2587 error_return_unlocked
:
2588 spin_lock(&log
->l_grant_reserve_lock
);
2590 list_del_init(&tic
->t_queue
);
2591 spin_unlock(&log
->l_grant_reserve_lock
);
2592 trace_xfs_log_grant_error(log
, tic
);
2595 * If we are failing, make sure the ticket doesn't have any
2596 * current reservations. We don't want to add this back when
2597 * the ticket/transaction gets cancelled.
2599 tic
->t_curr_res
= 0;
2600 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2601 return XFS_ERROR(EIO
);
2602 } /* xlog_grant_log_space */
2606 * Replenish the byte reservation required by moving the grant write head.
2608 * Similar to xlog_grant_log_space, the function is structured to have a lock
2612 xlog_regrant_write_log_space(xlog_t
*log
,
2615 int free_bytes
, need_bytes
;
2617 tic
->t_curr_res
= tic
->t_unit_res
;
2618 xlog_tic_reset_res(tic
);
2624 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2625 panic("regrant Recovery problem");
2628 trace_xfs_log_regrant_write_enter(log
, tic
);
2629 if (XLOG_FORCED_SHUTDOWN(log
))
2630 goto error_return_unlocked
;
2632 /* If there are other waiters on the queue then give them a
2633 * chance at logspace before us. Wake up the first waiters,
2634 * if we do not wake up all the waiters then go to sleep waiting
2635 * for more free space, otherwise try to get some space for
2638 need_bytes
= tic
->t_unit_res
;
2639 if (!list_empty_careful(&log
->l_writeq
)) {
2640 struct xlog_ticket
*ntic
;
2642 spin_lock(&log
->l_grant_write_lock
);
2643 free_bytes
= xlog_space_left(log
, &log
->l_grant_write_head
);
2644 list_for_each_entry(ntic
, &log
->l_writeq
, t_queue
) {
2645 ASSERT(ntic
->t_flags
& XLOG_TIC_PERM_RESERV
);
2647 if (free_bytes
< ntic
->t_unit_res
)
2649 free_bytes
-= ntic
->t_unit_res
;
2650 wake_up(&ntic
->t_wait
);
2653 if (ntic
!= list_first_entry(&log
->l_writeq
,
2654 struct xlog_ticket
, t_queue
)) {
2655 if (list_empty(&tic
->t_queue
))
2656 list_add_tail(&tic
->t_queue
, &log
->l_writeq
);
2657 trace_xfs_log_regrant_write_sleep1(log
, tic
);
2659 xlog_grant_push_ail(log
, need_bytes
);
2661 XFS_STATS_INC(xs_sleep_logspace
);
2662 xlog_wait(&tic
->t_wait
, &log
->l_grant_write_lock
);
2663 trace_xfs_log_regrant_write_wake1(log
, tic
);
2665 spin_unlock(&log
->l_grant_write_lock
);
2669 if (XLOG_FORCED_SHUTDOWN(log
))
2670 goto error_return_unlocked
;
2672 free_bytes
= xlog_space_left(log
, &log
->l_grant_write_head
);
2673 if (free_bytes
< need_bytes
) {
2674 spin_lock(&log
->l_grant_write_lock
);
2675 if (list_empty(&tic
->t_queue
))
2676 list_add_tail(&tic
->t_queue
, &log
->l_writeq
);
2678 if (XLOG_FORCED_SHUTDOWN(log
))
2681 xlog_grant_push_ail(log
, need_bytes
);
2683 XFS_STATS_INC(xs_sleep_logspace
);
2684 trace_xfs_log_regrant_write_sleep2(log
, tic
);
2685 xlog_wait(&tic
->t_wait
, &log
->l_grant_write_lock
);
2687 trace_xfs_log_regrant_write_wake2(log
, tic
);
2691 if (!list_empty(&tic
->t_queue
)) {
2692 spin_lock(&log
->l_grant_write_lock
);
2693 list_del_init(&tic
->t_queue
);
2694 spin_unlock(&log
->l_grant_write_lock
);
2697 /* we've got enough space */
2698 xlog_grant_add_space(log
, &log
->l_grant_write_head
, need_bytes
);
2699 trace_xfs_log_regrant_write_exit(log
, tic
);
2700 xlog_verify_grant_tail(log
);
2704 error_return_unlocked
:
2705 spin_lock(&log
->l_grant_write_lock
);
2707 list_del_init(&tic
->t_queue
);
2708 spin_unlock(&log
->l_grant_write_lock
);
2709 trace_xfs_log_regrant_write_error(log
, tic
);
2712 * If we are failing, make sure the ticket doesn't have any
2713 * current reservations. We don't want to add this back when
2714 * the ticket/transaction gets cancelled.
2716 tic
->t_curr_res
= 0;
2717 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2718 return XFS_ERROR(EIO
);
2719 } /* xlog_regrant_write_log_space */
2722 /* The first cnt-1 times through here we don't need to
2723 * move the grant write head because the permanent
2724 * reservation has reserved cnt times the unit amount.
2725 * Release part of current permanent unit reservation and
2726 * reset current reservation to be one units worth. Also
2727 * move grant reservation head forward.
2730 xlog_regrant_reserve_log_space(xlog_t
*log
,
2731 xlog_ticket_t
*ticket
)
2733 trace_xfs_log_regrant_reserve_enter(log
, ticket
);
2735 if (ticket
->t_cnt
> 0)
2738 xlog_grant_sub_space(log
, &log
->l_grant_reserve_head
,
2739 ticket
->t_curr_res
);
2740 xlog_grant_sub_space(log
, &log
->l_grant_write_head
,
2741 ticket
->t_curr_res
);
2742 ticket
->t_curr_res
= ticket
->t_unit_res
;
2743 xlog_tic_reset_res(ticket
);
2745 trace_xfs_log_regrant_reserve_sub(log
, ticket
);
2747 /* just return if we still have some of the pre-reserved space */
2748 if (ticket
->t_cnt
> 0)
2751 xlog_grant_add_space(log
, &log
->l_grant_reserve_head
,
2752 ticket
->t_unit_res
);
2754 trace_xfs_log_regrant_reserve_exit(log
, ticket
);
2756 ticket
->t_curr_res
= ticket
->t_unit_res
;
2757 xlog_tic_reset_res(ticket
);
2758 } /* xlog_regrant_reserve_log_space */
2762 * Give back the space left from a reservation.
2764 * All the information we need to make a correct determination of space left
2765 * is present. For non-permanent reservations, things are quite easy. The
2766 * count should have been decremented to zero. We only need to deal with the
2767 * space remaining in the current reservation part of the ticket. If the
2768 * ticket contains a permanent reservation, there may be left over space which
2769 * needs to be released. A count of N means that N-1 refills of the current
2770 * reservation can be done before we need to ask for more space. The first
2771 * one goes to fill up the first current reservation. Once we run out of
2772 * space, the count will stay at zero and the only space remaining will be
2773 * in the current reservation field.
2776 xlog_ungrant_log_space(xlog_t
*log
,
2777 xlog_ticket_t
*ticket
)
2781 if (ticket
->t_cnt
> 0)
2784 trace_xfs_log_ungrant_enter(log
, ticket
);
2785 trace_xfs_log_ungrant_sub(log
, ticket
);
2788 * If this is a permanent reservation ticket, we may be able to free
2789 * up more space based on the remaining count.
2791 bytes
= ticket
->t_curr_res
;
2792 if (ticket
->t_cnt
> 0) {
2793 ASSERT(ticket
->t_flags
& XLOG_TIC_PERM_RESERV
);
2794 bytes
+= ticket
->t_unit_res
*ticket
->t_cnt
;
2797 xlog_grant_sub_space(log
, &log
->l_grant_reserve_head
, bytes
);
2798 xlog_grant_sub_space(log
, &log
->l_grant_write_head
, bytes
);
2800 trace_xfs_log_ungrant_exit(log
, ticket
);
2802 xfs_log_move_tail(log
->l_mp
, 1);
2803 } /* xlog_ungrant_log_space */
2807 * Flush iclog to disk if this is the last reference to the given iclog and
2808 * the WANT_SYNC bit is set.
2810 * When this function is entered, the iclog is not necessarily in the
2811 * WANT_SYNC state. It may be sitting around waiting to get filled.
2816 xlog_state_release_iclog(
2818 xlog_in_core_t
*iclog
)
2820 int sync
= 0; /* do we sync? */
2822 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
2823 return XFS_ERROR(EIO
);
2825 ASSERT(atomic_read(&iclog
->ic_refcnt
) > 0);
2826 if (!atomic_dec_and_lock(&iclog
->ic_refcnt
, &log
->l_icloglock
))
2829 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2830 spin_unlock(&log
->l_icloglock
);
2831 return XFS_ERROR(EIO
);
2833 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2834 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
);
2836 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
) {
2837 /* update tail before writing to iclog */
2838 xfs_lsn_t tail_lsn
= xlog_assign_tail_lsn(log
->l_mp
);
2840 iclog
->ic_state
= XLOG_STATE_SYNCING
;
2841 iclog
->ic_header
.h_tail_lsn
= cpu_to_be64(tail_lsn
);
2842 xlog_verify_tail_lsn(log
, iclog
, tail_lsn
);
2843 /* cycle incremented when incrementing curr_block */
2845 spin_unlock(&log
->l_icloglock
);
2848 * We let the log lock go, so it's possible that we hit a log I/O
2849 * error or some other SHUTDOWN condition that marks the iclog
2850 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2851 * this iclog has consistent data, so we ignore IOERROR
2852 * flags after this point.
2855 return xlog_sync(log
, iclog
);
2857 } /* xlog_state_release_iclog */
2861 * This routine will mark the current iclog in the ring as WANT_SYNC
2862 * and move the current iclog pointer to the next iclog in the ring.
2863 * When this routine is called from xlog_state_get_iclog_space(), the
2864 * exact size of the iclog has not yet been determined. All we know is
2865 * that every data block. We have run out of space in this log record.
2868 xlog_state_switch_iclogs(xlog_t
*log
,
2869 xlog_in_core_t
*iclog
,
2872 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2874 eventual_size
= iclog
->ic_offset
;
2875 iclog
->ic_state
= XLOG_STATE_WANT_SYNC
;
2876 iclog
->ic_header
.h_prev_block
= cpu_to_be32(log
->l_prev_block
);
2877 log
->l_prev_block
= log
->l_curr_block
;
2878 log
->l_prev_cycle
= log
->l_curr_cycle
;
2880 /* roll log?: ic_offset changed later */
2881 log
->l_curr_block
+= BTOBB(eventual_size
)+BTOBB(log
->l_iclog_hsize
);
2883 /* Round up to next log-sunit */
2884 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
2885 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
2886 __uint32_t sunit_bb
= BTOBB(log
->l_mp
->m_sb
.sb_logsunit
);
2887 log
->l_curr_block
= roundup(log
->l_curr_block
, sunit_bb
);
2890 if (log
->l_curr_block
>= log
->l_logBBsize
) {
2891 log
->l_curr_cycle
++;
2892 if (log
->l_curr_cycle
== XLOG_HEADER_MAGIC_NUM
)
2893 log
->l_curr_cycle
++;
2894 log
->l_curr_block
-= log
->l_logBBsize
;
2895 ASSERT(log
->l_curr_block
>= 0);
2897 ASSERT(iclog
== log
->l_iclog
);
2898 log
->l_iclog
= iclog
->ic_next
;
2899 } /* xlog_state_switch_iclogs */
2902 * Write out all data in the in-core log as of this exact moment in time.
2904 * Data may be written to the in-core log during this call. However,
2905 * we don't guarantee this data will be written out. A change from past
2906 * implementation means this routine will *not* write out zero length LRs.
2908 * Basically, we try and perform an intelligent scan of the in-core logs.
2909 * If we determine there is no flushable data, we just return. There is no
2910 * flushable data if:
2912 * 1. the current iclog is active and has no data; the previous iclog
2913 * is in the active or dirty state.
2914 * 2. the current iclog is drity, and the previous iclog is in the
2915 * active or dirty state.
2919 * 1. the current iclog is not in the active nor dirty state.
2920 * 2. the current iclog dirty, and the previous iclog is not in the
2921 * active nor dirty state.
2922 * 3. the current iclog is active, and there is another thread writing
2923 * to this particular iclog.
2924 * 4. a) the current iclog is active and has no other writers
2925 * b) when we return from flushing out this iclog, it is still
2926 * not in the active nor dirty state.
2930 struct xfs_mount
*mp
,
2934 struct log
*log
= mp
->m_log
;
2935 struct xlog_in_core
*iclog
;
2938 XFS_STATS_INC(xs_log_force
);
2941 xlog_cil_force(log
);
2943 spin_lock(&log
->l_icloglock
);
2945 iclog
= log
->l_iclog
;
2946 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2947 spin_unlock(&log
->l_icloglock
);
2948 return XFS_ERROR(EIO
);
2951 /* If the head iclog is not active nor dirty, we just attach
2952 * ourselves to the head and go to sleep.
2954 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2955 iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2957 * If the head is dirty or (active and empty), then
2958 * we need to look at the previous iclog. If the previous
2959 * iclog is active or dirty we are done. There is nothing
2960 * to sync out. Otherwise, we attach ourselves to the
2961 * previous iclog and go to sleep.
2963 if (iclog
->ic_state
== XLOG_STATE_DIRTY
||
2964 (atomic_read(&iclog
->ic_refcnt
) == 0
2965 && iclog
->ic_offset
== 0)) {
2966 iclog
= iclog
->ic_prev
;
2967 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2968 iclog
->ic_state
== XLOG_STATE_DIRTY
)
2973 if (atomic_read(&iclog
->ic_refcnt
) == 0) {
2974 /* We are the only one with access to this
2975 * iclog. Flush it out now. There should
2976 * be a roundoff of zero to show that someone
2977 * has already taken care of the roundoff from
2978 * the previous sync.
2980 atomic_inc(&iclog
->ic_refcnt
);
2981 lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
2982 xlog_state_switch_iclogs(log
, iclog
, 0);
2983 spin_unlock(&log
->l_icloglock
);
2985 if (xlog_state_release_iclog(log
, iclog
))
2986 return XFS_ERROR(EIO
);
2990 spin_lock(&log
->l_icloglock
);
2991 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) == lsn
&&
2992 iclog
->ic_state
!= XLOG_STATE_DIRTY
)
2997 /* Someone else is writing to this iclog.
2998 * Use its call to flush out the data. However,
2999 * the other thread may not force out this LR,
3000 * so we mark it WANT_SYNC.
3002 xlog_state_switch_iclogs(log
, iclog
, 0);
3008 /* By the time we come around again, the iclog could've been filled
3009 * which would give it another lsn. If we have a new lsn, just
3010 * return because the relevant data has been flushed.
3013 if (flags
& XFS_LOG_SYNC
) {
3015 * We must check if we're shutting down here, before
3016 * we wait, while we're holding the l_icloglock.
3017 * Then we check again after waking up, in case our
3018 * sleep was disturbed by a bad news.
3020 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3021 spin_unlock(&log
->l_icloglock
);
3022 return XFS_ERROR(EIO
);
3024 XFS_STATS_INC(xs_log_force_sleep
);
3025 xlog_wait(&iclog
->ic_force_wait
, &log
->l_icloglock
);
3027 * No need to grab the log lock here since we're
3028 * only deciding whether or not to return EIO
3029 * and the memory read should be atomic.
3031 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3032 return XFS_ERROR(EIO
);
3038 spin_unlock(&log
->l_icloglock
);
3044 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3045 * about errors or whether the log was flushed or not. This is the normal
3046 * interface to use when trying to unpin items or move the log forward.
3055 error
= _xfs_log_force(mp
, flags
, NULL
);
3057 xfs_warn(mp
, "%s: error %d returned.", __func__
, error
);
3061 * Force the in-core log to disk for a specific LSN.
3063 * Find in-core log with lsn.
3064 * If it is in the DIRTY state, just return.
3065 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3066 * state and go to sleep or return.
3067 * If it is in any other state, go to sleep or return.
3069 * Synchronous forces are implemented with a signal variable. All callers
3070 * to force a given lsn to disk will wait on a the sv attached to the
3071 * specific in-core log. When given in-core log finally completes its
3072 * write to disk, that thread will wake up all threads waiting on the
3077 struct xfs_mount
*mp
,
3082 struct log
*log
= mp
->m_log
;
3083 struct xlog_in_core
*iclog
;
3084 int already_slept
= 0;
3088 XFS_STATS_INC(xs_log_force
);
3091 lsn
= xlog_cil_force_lsn(log
, lsn
);
3092 if (lsn
== NULLCOMMITLSN
)
3097 spin_lock(&log
->l_icloglock
);
3098 iclog
= log
->l_iclog
;
3099 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3100 spin_unlock(&log
->l_icloglock
);
3101 return XFS_ERROR(EIO
);
3105 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) != lsn
) {
3106 iclog
= iclog
->ic_next
;
3110 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
3111 spin_unlock(&log
->l_icloglock
);
3115 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3117 * We sleep here if we haven't already slept (e.g.
3118 * this is the first time we've looked at the correct
3119 * iclog buf) and the buffer before us is going to
3120 * be sync'ed. The reason for this is that if we
3121 * are doing sync transactions here, by waiting for
3122 * the previous I/O to complete, we can allow a few
3123 * more transactions into this iclog before we close
3126 * Otherwise, we mark the buffer WANT_SYNC, and bump
3127 * up the refcnt so we can release the log (which
3128 * drops the ref count). The state switch keeps new
3129 * transaction commits from using this buffer. When
3130 * the current commits finish writing into the buffer,
3131 * the refcount will drop to zero and the buffer will
3134 if (!already_slept
&&
3135 (iclog
->ic_prev
->ic_state
&
3136 (XLOG_STATE_WANT_SYNC
| XLOG_STATE_SYNCING
))) {
3137 ASSERT(!(iclog
->ic_state
& XLOG_STATE_IOERROR
));
3139 XFS_STATS_INC(xs_log_force_sleep
);
3141 xlog_wait(&iclog
->ic_prev
->ic_write_wait
,
3148 atomic_inc(&iclog
->ic_refcnt
);
3149 xlog_state_switch_iclogs(log
, iclog
, 0);
3150 spin_unlock(&log
->l_icloglock
);
3151 if (xlog_state_release_iclog(log
, iclog
))
3152 return XFS_ERROR(EIO
);
3155 spin_lock(&log
->l_icloglock
);
3158 if ((flags
& XFS_LOG_SYNC
) && /* sleep */
3160 (XLOG_STATE_ACTIVE
| XLOG_STATE_DIRTY
))) {
3162 * Don't wait on completion if we know that we've
3163 * gotten a log write error.
3165 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3166 spin_unlock(&log
->l_icloglock
);
3167 return XFS_ERROR(EIO
);
3169 XFS_STATS_INC(xs_log_force_sleep
);
3170 xlog_wait(&iclog
->ic_force_wait
, &log
->l_icloglock
);
3172 * No need to grab the log lock here since we're
3173 * only deciding whether or not to return EIO
3174 * and the memory read should be atomic.
3176 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3177 return XFS_ERROR(EIO
);
3181 } else { /* just return */
3182 spin_unlock(&log
->l_icloglock
);
3186 } while (iclog
!= log
->l_iclog
);
3188 spin_unlock(&log
->l_icloglock
);
3193 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3194 * about errors or whether the log was flushed or not. This is the normal
3195 * interface to use when trying to unpin items or move the log forward.
3205 error
= _xfs_log_force_lsn(mp
, lsn
, flags
, NULL
);
3207 xfs_warn(mp
, "%s: error %d returned.", __func__
, error
);
3211 * Called when we want to mark the current iclog as being ready to sync to
3215 xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
)
3217 assert_spin_locked(&log
->l_icloglock
);
3219 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3220 xlog_state_switch_iclogs(log
, iclog
, 0);
3222 ASSERT(iclog
->ic_state
&
3223 (XLOG_STATE_WANT_SYNC
|XLOG_STATE_IOERROR
));
3228 /*****************************************************************************
3232 *****************************************************************************
3236 * Free a used ticket when its refcount falls to zero.
3240 xlog_ticket_t
*ticket
)
3242 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3243 if (atomic_dec_and_test(&ticket
->t_ref
))
3244 kmem_zone_free(xfs_log_ticket_zone
, ticket
);
3249 xlog_ticket_t
*ticket
)
3251 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3252 atomic_inc(&ticket
->t_ref
);
3257 * Allocate and initialise a new log ticket.
3268 struct xlog_ticket
*tic
;
3272 tic
= kmem_zone_zalloc(xfs_log_ticket_zone
, alloc_flags
);
3277 * Permanent reservations have up to 'cnt'-1 active log operations
3278 * in the log. A unit in this case is the amount of space for one
3279 * of these log operations. Normal reservations have a cnt of 1
3280 * and their unit amount is the total amount of space required.
3282 * The following lines of code account for non-transaction data
3283 * which occupy space in the on-disk log.
3285 * Normal form of a transaction is:
3286 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3287 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3289 * We need to account for all the leadup data and trailer data
3290 * around the transaction data.
3291 * And then we need to account for the worst case in terms of using
3293 * The worst case will happen if:
3294 * - the placement of the transaction happens to be such that the
3295 * roundoff is at its maximum
3296 * - the transaction data is synced before the commit record is synced
3297 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3298 * Therefore the commit record is in its own Log Record.
3299 * This can happen as the commit record is called with its
3300 * own region to xlog_write().
3301 * This then means that in the worst case, roundoff can happen for
3302 * the commit-rec as well.
3303 * The commit-rec is smaller than padding in this scenario and so it is
3304 * not added separately.
3307 /* for trans header */
3308 unit_bytes
+= sizeof(xlog_op_header_t
);
3309 unit_bytes
+= sizeof(xfs_trans_header_t
);
3312 unit_bytes
+= sizeof(xlog_op_header_t
);
3315 * for LR headers - the space for data in an iclog is the size minus
3316 * the space used for the headers. If we use the iclog size, then we
3317 * undercalculate the number of headers required.
3319 * Furthermore - the addition of op headers for split-recs might
3320 * increase the space required enough to require more log and op
3321 * headers, so take that into account too.
3323 * IMPORTANT: This reservation makes the assumption that if this
3324 * transaction is the first in an iclog and hence has the LR headers
3325 * accounted to it, then the remaining space in the iclog is
3326 * exclusively for this transaction. i.e. if the transaction is larger
3327 * than the iclog, it will be the only thing in that iclog.
3328 * Fundamentally, this means we must pass the entire log vector to
3329 * xlog_write to guarantee this.
3331 iclog_space
= log
->l_iclog_size
- log
->l_iclog_hsize
;
3332 num_headers
= howmany(unit_bytes
, iclog_space
);
3334 /* for split-recs - ophdrs added when data split over LRs */
3335 unit_bytes
+= sizeof(xlog_op_header_t
) * num_headers
;
3337 /* add extra header reservations if we overrun */
3338 while (!num_headers
||
3339 howmany(unit_bytes
, iclog_space
) > num_headers
) {
3340 unit_bytes
+= sizeof(xlog_op_header_t
);
3343 unit_bytes
+= log
->l_iclog_hsize
* num_headers
;
3345 /* for commit-rec LR header - note: padding will subsume the ophdr */
3346 unit_bytes
+= log
->l_iclog_hsize
;
3348 /* for roundoff padding for transaction data and one for commit record */
3349 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
3350 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
3351 /* log su roundoff */
3352 unit_bytes
+= 2*log
->l_mp
->m_sb
.sb_logsunit
;
3355 unit_bytes
+= 2*BBSIZE
;
3358 atomic_set(&tic
->t_ref
, 1);
3359 INIT_LIST_HEAD(&tic
->t_queue
);
3360 tic
->t_unit_res
= unit_bytes
;
3361 tic
->t_curr_res
= unit_bytes
;
3364 tic
->t_tid
= random32();
3365 tic
->t_clientid
= client
;
3366 tic
->t_flags
= XLOG_TIC_INITED
;
3367 tic
->t_trans_type
= 0;
3368 if (xflags
& XFS_LOG_PERM_RESERV
)
3369 tic
->t_flags
|= XLOG_TIC_PERM_RESERV
;
3370 init_waitqueue_head(&tic
->t_wait
);
3372 xlog_tic_reset_res(tic
);
3378 /******************************************************************************
3380 * Log debug routines
3382 ******************************************************************************
3386 * Make sure that the destination ptr is within the valid data region of
3387 * one of the iclogs. This uses backup pointers stored in a different
3388 * part of the log in case we trash the log structure.
3391 xlog_verify_dest_ptr(
3398 for (i
= 0; i
< log
->l_iclog_bufs
; i
++) {
3399 if (ptr
>= log
->l_iclog_bak
[i
] &&
3400 ptr
<= log
->l_iclog_bak
[i
] + log
->l_iclog_size
)
3405 xfs_emerg(log
->l_mp
, "%s: invalid ptr", __func__
);
3409 * Check to make sure the grant write head didn't just over lap the tail. If
3410 * the cycles are the same, we can't be overlapping. Otherwise, make sure that
3411 * the cycles differ by exactly one and check the byte count.
3413 * This check is run unlocked, so can give false positives. Rather than assert
3414 * on failures, use a warn-once flag and a panic tag to allow the admin to
3415 * determine if they want to panic the machine when such an error occurs. For
3416 * debug kernels this will have the same effect as using an assert but, unlinke
3417 * an assert, it can be turned off at runtime.
3420 xlog_verify_grant_tail(
3423 int tail_cycle
, tail_blocks
;
3426 xlog_crack_grant_head(&log
->l_grant_write_head
, &cycle
, &space
);
3427 xlog_crack_atomic_lsn(&log
->l_tail_lsn
, &tail_cycle
, &tail_blocks
);
3428 if (tail_cycle
!= cycle
) {
3429 if (cycle
- 1 != tail_cycle
&&
3430 !(log
->l_flags
& XLOG_TAIL_WARN
)) {
3431 xfs_alert_tag(log
->l_mp
, XFS_PTAG_LOGRES
,
3432 "%s: cycle - 1 != tail_cycle", __func__
);
3433 log
->l_flags
|= XLOG_TAIL_WARN
;
3436 if (space
> BBTOB(tail_blocks
) &&
3437 !(log
->l_flags
& XLOG_TAIL_WARN
)) {
3438 xfs_alert_tag(log
->l_mp
, XFS_PTAG_LOGRES
,
3439 "%s: space > BBTOB(tail_blocks)", __func__
);
3440 log
->l_flags
|= XLOG_TAIL_WARN
;
3445 /* check if it will fit */
3447 xlog_verify_tail_lsn(xlog_t
*log
,
3448 xlog_in_core_t
*iclog
,
3453 if (CYCLE_LSN(tail_lsn
) == log
->l_prev_cycle
) {
3455 log
->l_logBBsize
- (log
->l_prev_block
- BLOCK_LSN(tail_lsn
));
3456 if (blocks
< BTOBB(iclog
->ic_offset
)+BTOBB(log
->l_iclog_hsize
))
3457 xfs_emerg(log
->l_mp
, "%s: ran out of log space", __func__
);
3459 ASSERT(CYCLE_LSN(tail_lsn
)+1 == log
->l_prev_cycle
);
3461 if (BLOCK_LSN(tail_lsn
) == log
->l_prev_block
)
3462 xfs_emerg(log
->l_mp
, "%s: tail wrapped", __func__
);
3464 blocks
= BLOCK_LSN(tail_lsn
) - log
->l_prev_block
;
3465 if (blocks
< BTOBB(iclog
->ic_offset
) + 1)
3466 xfs_emerg(log
->l_mp
, "%s: ran out of log space", __func__
);
3468 } /* xlog_verify_tail_lsn */
3471 * Perform a number of checks on the iclog before writing to disk.
3473 * 1. Make sure the iclogs are still circular
3474 * 2. Make sure we have a good magic number
3475 * 3. Make sure we don't have magic numbers in the data
3476 * 4. Check fields of each log operation header for:
3477 * A. Valid client identifier
3478 * B. tid ptr value falls in valid ptr space (user space code)
3479 * C. Length in log record header is correct according to the
3480 * individual operation headers within record.
3481 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3482 * log, check the preceding blocks of the physical log to make sure all
3483 * the cycle numbers agree with the current cycle number.
3486 xlog_verify_iclog(xlog_t
*log
,
3487 xlog_in_core_t
*iclog
,
3491 xlog_op_header_t
*ophead
;
3492 xlog_in_core_t
*icptr
;
3493 xlog_in_core_2_t
*xhdr
;
3495 xfs_caddr_t base_ptr
;
3496 __psint_t field_offset
;
3498 int len
, i
, j
, k
, op_len
;
3501 /* check validity of iclog pointers */
3502 spin_lock(&log
->l_icloglock
);
3503 icptr
= log
->l_iclog
;
3504 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3506 xfs_emerg(log
->l_mp
, "%s: invalid ptr", __func__
);
3507 icptr
= icptr
->ic_next
;
3509 if (icptr
!= log
->l_iclog
)
3510 xfs_emerg(log
->l_mp
, "%s: corrupt iclog ring", __func__
);
3511 spin_unlock(&log
->l_icloglock
);
3513 /* check log magic numbers */
3514 if (iclog
->ic_header
.h_magicno
!= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
3515 xfs_emerg(log
->l_mp
, "%s: invalid magic num", __func__
);
3517 ptr
= (xfs_caddr_t
) &iclog
->ic_header
;
3518 for (ptr
+= BBSIZE
; ptr
< ((xfs_caddr_t
)&iclog
->ic_header
) + count
;
3520 if (*(__be32
*)ptr
== cpu_to_be32(XLOG_HEADER_MAGIC_NUM
))
3521 xfs_emerg(log
->l_mp
, "%s: unexpected magic num",
3526 len
= be32_to_cpu(iclog
->ic_header
.h_num_logops
);
3527 ptr
= iclog
->ic_datap
;
3529 ophead
= (xlog_op_header_t
*)ptr
;
3530 xhdr
= iclog
->ic_data
;
3531 for (i
= 0; i
< len
; i
++) {
3532 ophead
= (xlog_op_header_t
*)ptr
;
3534 /* clientid is only 1 byte */
3535 field_offset
= (__psint_t
)
3536 ((xfs_caddr_t
)&(ophead
->oh_clientid
) - base_ptr
);
3537 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3538 clientid
= ophead
->oh_clientid
;
3540 idx
= BTOBBT((xfs_caddr_t
)&(ophead
->oh_clientid
) - iclog
->ic_datap
);
3541 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3542 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3543 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3544 clientid
= xlog_get_client_id(
3545 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3547 clientid
= xlog_get_client_id(
3548 iclog
->ic_header
.h_cycle_data
[idx
]);
3551 if (clientid
!= XFS_TRANSACTION
&& clientid
!= XFS_LOG
)
3553 "%s: invalid clientid %d op 0x%p offset 0x%lx",
3554 __func__
, clientid
, ophead
,
3555 (unsigned long)field_offset
);
3558 field_offset
= (__psint_t
)
3559 ((xfs_caddr_t
)&(ophead
->oh_len
) - base_ptr
);
3560 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3561 op_len
= be32_to_cpu(ophead
->oh_len
);
3563 idx
= BTOBBT((__psint_t
)&ophead
->oh_len
-
3564 (__psint_t
)iclog
->ic_datap
);
3565 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3566 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3567 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3568 op_len
= be32_to_cpu(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3570 op_len
= be32_to_cpu(iclog
->ic_header
.h_cycle_data
[idx
]);
3573 ptr
+= sizeof(xlog_op_header_t
) + op_len
;
3575 } /* xlog_verify_iclog */
3579 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3585 xlog_in_core_t
*iclog
, *ic
;
3587 iclog
= log
->l_iclog
;
3588 if (! (iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3590 * Mark all the incore logs IOERROR.
3591 * From now on, no log flushes will result.
3595 ic
->ic_state
= XLOG_STATE_IOERROR
;
3597 } while (ic
!= iclog
);
3601 * Return non-zero, if state transition has already happened.
3607 * This is called from xfs_force_shutdown, when we're forcibly
3608 * shutting down the filesystem, typically because of an IO error.
3609 * Our main objectives here are to make sure that:
3610 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3611 * parties to find out, 'atomically'.
3612 * b. those who're sleeping on log reservations, pinned objects and
3613 * other resources get woken up, and be told the bad news.
3614 * c. nothing new gets queued up after (a) and (b) are done.
3615 * d. if !logerror, flush the iclogs to disk, then seal them off
3618 * Note: for delayed logging the !logerror case needs to flush the regions
3619 * held in memory out to the iclogs before flushing them to disk. This needs
3620 * to be done before the log is marked as shutdown, otherwise the flush to the
3624 xfs_log_force_umount(
3625 struct xfs_mount
*mp
,
3635 * If this happens during log recovery, don't worry about
3636 * locking; the log isn't open for business yet.
3639 log
->l_flags
& XLOG_ACTIVE_RECOVERY
) {
3640 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3642 XFS_BUF_DONE(mp
->m_sb_bp
);
3647 * Somebody could've already done the hard work for us.
3648 * No need to get locks for this.
3650 if (logerror
&& log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3651 ASSERT(XLOG_FORCED_SHUTDOWN(log
));
3657 * Flush the in memory commit item list before marking the log as
3658 * being shut down. We need to do it in this order to ensure all the
3659 * completed transactions are flushed to disk with the xfs_log_force()
3662 if (!logerror
&& (mp
->m_flags
& XFS_MOUNT_DELAYLOG
))
3663 xlog_cil_force(log
);
3666 * mark the filesystem and the as in a shutdown state and wake
3667 * everybody up to tell them the bad news.
3669 spin_lock(&log
->l_icloglock
);
3670 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3672 XFS_BUF_DONE(mp
->m_sb_bp
);
3675 * This flag is sort of redundant because of the mount flag, but
3676 * it's good to maintain the separation between the log and the rest
3679 log
->l_flags
|= XLOG_IO_ERROR
;
3682 * If we hit a log error, we want to mark all the iclogs IOERROR
3683 * while we're still holding the loglock.
3686 retval
= xlog_state_ioerror(log
);
3687 spin_unlock(&log
->l_icloglock
);
3690 * We don't want anybody waiting for log reservations after this. That
3691 * means we have to wake up everybody queued up on reserveq as well as
3692 * writeq. In addition, we make sure in xlog_{re}grant_log_space that
3693 * we don't enqueue anything once the SHUTDOWN flag is set, and this
3694 * action is protected by the grant locks.
3696 spin_lock(&log
->l_grant_reserve_lock
);
3697 list_for_each_entry(tic
, &log
->l_reserveq
, t_queue
)
3698 wake_up(&tic
->t_wait
);
3699 spin_unlock(&log
->l_grant_reserve_lock
);
3701 spin_lock(&log
->l_grant_write_lock
);
3702 list_for_each_entry(tic
, &log
->l_writeq
, t_queue
)
3703 wake_up(&tic
->t_wait
);
3704 spin_unlock(&log
->l_grant_write_lock
);
3706 if (!(log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3709 * Force the incore logs to disk before shutting the
3710 * log down completely.
3712 _xfs_log_force(mp
, XFS_LOG_SYNC
, NULL
);
3714 spin_lock(&log
->l_icloglock
);
3715 retval
= xlog_state_ioerror(log
);
3716 spin_unlock(&log
->l_icloglock
);
3719 * Wake up everybody waiting on xfs_log_force.
3720 * Callback all log item committed functions as if the
3721 * log writes were completed.
3723 xlog_state_do_callback(log
, XFS_LI_ABORTED
, NULL
);
3725 #ifdef XFSERRORDEBUG
3727 xlog_in_core_t
*iclog
;
3729 spin_lock(&log
->l_icloglock
);
3730 iclog
= log
->l_iclog
;
3732 ASSERT(iclog
->ic_callback
== 0);
3733 iclog
= iclog
->ic_next
;
3734 } while (iclog
!= log
->l_iclog
);
3735 spin_unlock(&log
->l_icloglock
);
3738 /* return non-zero if log IOERROR transition had already happened */
3743 xlog_iclogs_empty(xlog_t
*log
)
3745 xlog_in_core_t
*iclog
;
3747 iclog
= log
->l_iclog
;
3749 /* endianness does not matter here, zero is zero in
3752 if (iclog
->ic_header
.h_num_logops
)
3754 iclog
= iclog
->ic_next
;
3755 } while (iclog
!= log
->l_iclog
);