]>
git.proxmox.com Git - mirror_ubuntu-artful-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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_log_priv.h"
32 #include "xfs_buf_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_log_recover.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_dir2_sf.h"
39 #include "xfs_attr_sf.h"
40 #include "xfs_dinode.h"
41 #include "xfs_inode.h"
43 #include "xfs_trace.h"
45 kmem_zone_t
*xfs_log_ticket_zone
;
47 /* Local miscellaneous function prototypes */
48 STATIC
int xlog_commit_record(struct log
*log
, struct xlog_ticket
*ticket
,
49 xlog_in_core_t
**, xfs_lsn_t
*);
50 STATIC xlog_t
* xlog_alloc_log(xfs_mount_t
*mp
,
51 xfs_buftarg_t
*log_target
,
52 xfs_daddr_t blk_offset
,
54 STATIC
int xlog_space_left(xlog_t
*log
, int cycle
, int bytes
);
55 STATIC
int xlog_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
56 STATIC
void xlog_dealloc_log(xlog_t
*log
);
57 STATIC
int xlog_write(struct log
*log
, struct xfs_log_vec
*log_vector
,
58 struct xlog_ticket
*tic
, xfs_lsn_t
*start_lsn
,
59 xlog_in_core_t
**commit_iclog
, uint flags
);
61 /* local state machine functions */
62 STATIC
void xlog_state_done_syncing(xlog_in_core_t
*iclog
, int);
63 STATIC
void xlog_state_do_callback(xlog_t
*log
,int aborted
, xlog_in_core_t
*iclog
);
64 STATIC
int xlog_state_get_iclog_space(xlog_t
*log
,
66 xlog_in_core_t
**iclog
,
67 xlog_ticket_t
*ticket
,
70 STATIC
int xlog_state_release_iclog(xlog_t
*log
,
71 xlog_in_core_t
*iclog
);
72 STATIC
void xlog_state_switch_iclogs(xlog_t
*log
,
73 xlog_in_core_t
*iclog
,
75 STATIC
void xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
77 /* local functions to manipulate grant head */
78 STATIC
int xlog_grant_log_space(xlog_t
*log
,
80 STATIC
void xlog_grant_push_ail(xfs_mount_t
*mp
,
82 STATIC
void xlog_regrant_reserve_log_space(xlog_t
*log
,
83 xlog_ticket_t
*ticket
);
84 STATIC
int xlog_regrant_write_log_space(xlog_t
*log
,
85 xlog_ticket_t
*ticket
);
86 STATIC
void xlog_ungrant_log_space(xlog_t
*log
,
87 xlog_ticket_t
*ticket
);
90 /* local ticket functions */
91 STATIC xlog_ticket_t
*xlog_ticket_alloc(xlog_t
*log
,
98 STATIC
void xlog_verify_dest_ptr(xlog_t
*log
, char *ptr
);
99 STATIC
void xlog_verify_grant_head(xlog_t
*log
, int equals
);
100 STATIC
void xlog_verify_iclog(xlog_t
*log
, xlog_in_core_t
*iclog
,
101 int count
, boolean_t syncing
);
102 STATIC
void xlog_verify_tail_lsn(xlog_t
*log
, xlog_in_core_t
*iclog
,
105 #define xlog_verify_dest_ptr(a,b)
106 #define xlog_verify_grant_head(a,b)
107 #define xlog_verify_iclog(a,b,c,d)
108 #define xlog_verify_tail_lsn(a,b,c)
111 STATIC
int xlog_iclogs_empty(xlog_t
*log
);
115 xlog_ins_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
119 tic
->t_prev
= (*qp
)->t_prev
;
120 (*qp
)->t_prev
->t_next
= tic
;
123 tic
->t_prev
= tic
->t_next
= tic
;
127 tic
->t_flags
|= XLOG_TIC_IN_Q
;
131 xlog_del_ticketq(struct xlog_ticket
**qp
, struct xlog_ticket
*tic
)
133 if (tic
== tic
->t_next
) {
137 tic
->t_next
->t_prev
= tic
->t_prev
;
138 tic
->t_prev
->t_next
= tic
->t_next
;
141 tic
->t_next
= tic
->t_prev
= NULL
;
142 tic
->t_flags
&= ~XLOG_TIC_IN_Q
;
146 xlog_grant_sub_space(struct log
*log
, int bytes
)
148 log
->l_grant_write_bytes
-= bytes
;
149 if (log
->l_grant_write_bytes
< 0) {
150 log
->l_grant_write_bytes
+= log
->l_logsize
;
151 log
->l_grant_write_cycle
--;
154 log
->l_grant_reserve_bytes
-= bytes
;
155 if ((log
)->l_grant_reserve_bytes
< 0) {
156 log
->l_grant_reserve_bytes
+= log
->l_logsize
;
157 log
->l_grant_reserve_cycle
--;
163 xlog_grant_add_space_write(struct log
*log
, int bytes
)
165 int tmp
= log
->l_logsize
- log
->l_grant_write_bytes
;
167 log
->l_grant_write_bytes
+= bytes
;
169 log
->l_grant_write_cycle
++;
170 log
->l_grant_write_bytes
= bytes
- tmp
;
175 xlog_grant_add_space_reserve(struct log
*log
, int bytes
)
177 int tmp
= log
->l_logsize
- log
->l_grant_reserve_bytes
;
179 log
->l_grant_reserve_bytes
+= bytes
;
181 log
->l_grant_reserve_cycle
++;
182 log
->l_grant_reserve_bytes
= bytes
- tmp
;
187 xlog_grant_add_space(struct log
*log
, int bytes
)
189 xlog_grant_add_space_write(log
, bytes
);
190 xlog_grant_add_space_reserve(log
, bytes
);
194 xlog_tic_reset_res(xlog_ticket_t
*tic
)
197 tic
->t_res_arr_sum
= 0;
198 tic
->t_res_num_ophdrs
= 0;
202 xlog_tic_add_region(xlog_ticket_t
*tic
, uint len
, uint type
)
204 if (tic
->t_res_num
== XLOG_TIC_LEN_MAX
) {
205 /* add to overflow and start again */
206 tic
->t_res_o_flow
+= tic
->t_res_arr_sum
;
208 tic
->t_res_arr_sum
= 0;
211 tic
->t_res_arr
[tic
->t_res_num
].r_len
= len
;
212 tic
->t_res_arr
[tic
->t_res_num
].r_type
= type
;
213 tic
->t_res_arr_sum
+= len
;
220 * 1. currblock field gets updated at startup and after in-core logs
221 * marked as with WANT_SYNC.
225 * This routine is called when a user of a log manager ticket is done with
226 * the reservation. If the ticket was ever used, then a commit record for
227 * the associated transaction is written out as a log operation header with
228 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
229 * a given ticket. If the ticket was one with a permanent reservation, then
230 * a few operations are done differently. Permanent reservation tickets by
231 * default don't release the reservation. They just commit the current
232 * transaction with the belief that the reservation is still needed. A flag
233 * must be passed in before permanent reservations are actually released.
234 * When these type of tickets are not released, they need to be set into
235 * the inited state again. By doing this, a start record will be written
236 * out when the next write occurs.
240 struct xfs_mount
*mp
,
241 struct xlog_ticket
*ticket
,
242 struct xlog_in_core
**iclog
,
245 struct log
*log
= mp
->m_log
;
248 if (XLOG_FORCED_SHUTDOWN(log
) ||
250 * If nothing was ever written, don't write out commit record.
251 * If we get an error, just continue and give back the log ticket.
253 (((ticket
->t_flags
& XLOG_TIC_INITED
) == 0) &&
254 (xlog_commit_record(log
, ticket
, iclog
, &lsn
)))) {
255 lsn
= (xfs_lsn_t
) -1;
256 if (ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) {
257 flags
|= XFS_LOG_REL_PERM_RESERV
;
262 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) == 0 ||
263 (flags
& XFS_LOG_REL_PERM_RESERV
)) {
264 trace_xfs_log_done_nonperm(log
, ticket
);
267 * Release ticket if not permanent reservation or a specific
268 * request has been made to release a permanent reservation.
270 xlog_ungrant_log_space(log
, ticket
);
271 xfs_log_ticket_put(ticket
);
273 trace_xfs_log_done_perm(log
, ticket
);
275 xlog_regrant_reserve_log_space(log
, ticket
);
276 /* If this ticket was a permanent reservation and we aren't
277 * trying to release it, reset the inited flags; so next time
278 * we write, a start record will be written out.
280 ticket
->t_flags
|= XLOG_TIC_INITED
;
287 * Attaches a new iclog I/O completion callback routine during
288 * transaction commit. If the log is in error state, a non-zero
289 * return code is handed back and the caller is responsible for
290 * executing the callback at an appropriate time.
294 struct xfs_mount
*mp
,
295 struct xlog_in_core
*iclog
,
296 xfs_log_callback_t
*cb
)
300 spin_lock(&iclog
->ic_callback_lock
);
301 abortflg
= (iclog
->ic_state
& XLOG_STATE_IOERROR
);
303 ASSERT_ALWAYS((iclog
->ic_state
== XLOG_STATE_ACTIVE
) ||
304 (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
));
306 *(iclog
->ic_callback_tail
) = cb
;
307 iclog
->ic_callback_tail
= &(cb
->cb_next
);
309 spin_unlock(&iclog
->ic_callback_lock
);
314 xfs_log_release_iclog(
315 struct xfs_mount
*mp
,
316 struct xlog_in_core
*iclog
)
318 if (xlog_state_release_iclog(mp
->m_log
, iclog
)) {
319 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
327 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
328 * to the reservation.
329 * 2. Potentially, push buffers at tail of log to disk.
331 * Each reservation is going to reserve extra space for a log record header.
332 * When writes happen to the on-disk log, we don't subtract the length of the
333 * log record header from any reservation. By wasting space in each
334 * reservation, we prevent over allocation problems.
338 struct xfs_mount
*mp
,
341 struct xlog_ticket
**ticket
,
346 struct log
*log
= mp
->m_log
;
347 struct xlog_ticket
*internal_ticket
;
350 ASSERT(client
== XFS_TRANSACTION
|| client
== XFS_LOG
);
351 ASSERT((flags
& XFS_LOG_NOSLEEP
) == 0);
353 if (XLOG_FORCED_SHUTDOWN(log
))
354 return XFS_ERROR(EIO
);
356 XFS_STATS_INC(xs_try_logspace
);
359 if (*ticket
!= NULL
) {
360 ASSERT(flags
& XFS_LOG_PERM_RESERV
);
361 internal_ticket
= *ticket
;
364 * this is a new transaction on the ticket, so we need to
365 * change the transaction ID so that the next transaction has a
366 * different TID in the log. Just add one to the existing tid
367 * so that we can see chains of rolling transactions in the log
370 internal_ticket
->t_tid
++;
372 trace_xfs_log_reserve(log
, internal_ticket
);
374 xlog_grant_push_ail(mp
, internal_ticket
->t_unit_res
);
375 retval
= xlog_regrant_write_log_space(log
, internal_ticket
);
377 /* may sleep if need to allocate more tickets */
378 internal_ticket
= xlog_ticket_alloc(log
, unit_bytes
, cnt
,
380 if (!internal_ticket
)
381 return XFS_ERROR(ENOMEM
);
382 internal_ticket
->t_trans_type
= t_type
;
383 *ticket
= internal_ticket
;
385 trace_xfs_log_reserve(log
, internal_ticket
);
387 xlog_grant_push_ail(mp
,
388 (internal_ticket
->t_unit_res
*
389 internal_ticket
->t_cnt
));
390 retval
= xlog_grant_log_space(log
, internal_ticket
);
394 } /* xfs_log_reserve */
398 * Mount a log filesystem
400 * mp - ubiquitous xfs mount point structure
401 * log_target - buftarg of on-disk log device
402 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
403 * num_bblocks - Number of BBSIZE blocks in on-disk log
405 * Return error or zero.
410 xfs_buftarg_t
*log_target
,
411 xfs_daddr_t blk_offset
,
416 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
417 cmn_err(CE_NOTE
, "XFS mounting filesystem %s", mp
->m_fsname
);
420 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
422 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
425 mp
->m_log
= xlog_alloc_log(mp
, log_target
, blk_offset
, num_bblks
);
426 if (IS_ERR(mp
->m_log
)) {
427 error
= -PTR_ERR(mp
->m_log
);
432 * Initialize the AIL now we have a log.
434 error
= xfs_trans_ail_init(mp
);
436 cmn_err(CE_WARN
, "XFS: AIL initialisation failed: error %d", error
);
439 mp
->m_log
->l_ailp
= mp
->m_ail
;
442 * skip log recovery on a norecovery mount. pretend it all
445 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
)) {
446 int readonly
= (mp
->m_flags
& XFS_MOUNT_RDONLY
);
449 mp
->m_flags
&= ~XFS_MOUNT_RDONLY
;
451 error
= xlog_recover(mp
->m_log
);
454 mp
->m_flags
|= XFS_MOUNT_RDONLY
;
456 cmn_err(CE_WARN
, "XFS: log mount/recovery failed: error %d", error
);
457 goto out_destroy_ail
;
461 /* Normal transactions can now occur */
462 mp
->m_log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
467 xfs_trans_ail_destroy(mp
);
469 xlog_dealloc_log(mp
->m_log
);
475 * Finish the recovery of the file system. This is separate from
476 * the xfs_log_mount() call, because it depends on the code in
477 * xfs_mountfs() to read in the root and real-time bitmap inodes
478 * between calling xfs_log_mount() and here.
480 * mp - ubiquitous xfs mount point structure
483 xfs_log_mount_finish(xfs_mount_t
*mp
)
487 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
488 error
= xlog_recover_finish(mp
->m_log
);
491 ASSERT(mp
->m_flags
& XFS_MOUNT_RDONLY
);
498 * Final log writes as part of unmount.
500 * Mark the filesystem clean as unmount happens. Note that during relocation
501 * this routine needs to be executed as part of source-bag while the
502 * deallocation must not be done until source-end.
506 * Unmount record used to have a string "Unmount filesystem--" in the
507 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
508 * We just write the magic number now since that particular field isn't
509 * currently architecture converted and "nUmount" is a bit foo.
510 * As far as I know, there weren't any dependencies on the old behaviour.
514 xfs_log_unmount_write(xfs_mount_t
*mp
)
516 xlog_t
*log
= mp
->m_log
;
517 xlog_in_core_t
*iclog
;
519 xlog_in_core_t
*first_iclog
;
521 xlog_ticket_t
*tic
= NULL
;
526 * Don't write out unmount record on read-only mounts.
527 * Or, if we are doing a forced umount (typically because of IO errors).
529 if (mp
->m_flags
& XFS_MOUNT_RDONLY
)
532 error
= _xfs_log_force(mp
, XFS_LOG_SYNC
, NULL
);
533 ASSERT(error
|| !(XLOG_FORCED_SHUTDOWN(log
)));
536 first_iclog
= iclog
= log
->l_iclog
;
538 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
539 ASSERT(iclog
->ic_state
& XLOG_STATE_ACTIVE
);
540 ASSERT(iclog
->ic_offset
== 0);
542 iclog
= iclog
->ic_next
;
543 } while (iclog
!= first_iclog
);
545 if (! (XLOG_FORCED_SHUTDOWN(log
))) {
546 error
= xfs_log_reserve(mp
, 600, 1, &tic
,
547 XFS_LOG
, 0, XLOG_UNMOUNT_REC_TYPE
);
549 /* the data section must be 32 bit size aligned */
553 __uint32_t pad2
; /* may as well make it 64 bits */
555 .magic
= XLOG_UNMOUNT_TYPE
,
557 struct xfs_log_iovec reg
= {
558 .i_addr
= (void *)&magic
,
559 .i_len
= sizeof(magic
),
560 .i_type
= XLOG_REG_TYPE_UNMOUNT
,
562 struct xfs_log_vec vec
= {
567 /* remove inited flag */
569 error
= xlog_write(log
, &vec
, tic
, &lsn
,
570 NULL
, XLOG_UNMOUNT_TRANS
);
572 * At this point, we're umounting anyway,
573 * so there's no point in transitioning log state
574 * to IOERROR. Just continue...
579 xfs_fs_cmn_err(CE_ALERT
, mp
,
580 "xfs_log_unmount: unmount record failed");
584 spin_lock(&log
->l_icloglock
);
585 iclog
= log
->l_iclog
;
586 atomic_inc(&iclog
->ic_refcnt
);
587 xlog_state_want_sync(log
, iclog
);
588 spin_unlock(&log
->l_icloglock
);
589 error
= xlog_state_release_iclog(log
, iclog
);
591 spin_lock(&log
->l_icloglock
);
592 if (!(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
593 iclog
->ic_state
== XLOG_STATE_DIRTY
)) {
594 if (!XLOG_FORCED_SHUTDOWN(log
)) {
595 sv_wait(&iclog
->ic_force_wait
, PMEM
,
596 &log
->l_icloglock
, s
);
598 spin_unlock(&log
->l_icloglock
);
601 spin_unlock(&log
->l_icloglock
);
604 trace_xfs_log_umount_write(log
, tic
);
605 xlog_ungrant_log_space(log
, tic
);
606 xfs_log_ticket_put(tic
);
610 * We're already in forced_shutdown mode, couldn't
611 * even attempt to write out the unmount transaction.
613 * Go through the motions of sync'ing and releasing
614 * the iclog, even though no I/O will actually happen,
615 * we need to wait for other log I/Os that may already
616 * be in progress. Do this as a separate section of
617 * code so we'll know if we ever get stuck here that
618 * we're in this odd situation of trying to unmount
619 * a file system that went into forced_shutdown as
620 * the result of an unmount..
622 spin_lock(&log
->l_icloglock
);
623 iclog
= log
->l_iclog
;
624 atomic_inc(&iclog
->ic_refcnt
);
626 xlog_state_want_sync(log
, iclog
);
627 spin_unlock(&log
->l_icloglock
);
628 error
= xlog_state_release_iclog(log
, iclog
);
630 spin_lock(&log
->l_icloglock
);
632 if ( ! ( iclog
->ic_state
== XLOG_STATE_ACTIVE
633 || iclog
->ic_state
== XLOG_STATE_DIRTY
634 || iclog
->ic_state
== XLOG_STATE_IOERROR
) ) {
636 sv_wait(&iclog
->ic_force_wait
, PMEM
,
637 &log
->l_icloglock
, s
);
639 spin_unlock(&log
->l_icloglock
);
644 } /* xfs_log_unmount_write */
647 * Deallocate log structures for unmount/relocation.
649 * We need to stop the aild from running before we destroy
650 * and deallocate the log as the aild references the log.
653 xfs_log_unmount(xfs_mount_t
*mp
)
655 xfs_trans_ail_destroy(mp
);
656 xlog_dealloc_log(mp
->m_log
);
661 struct xfs_mount
*mp
,
662 struct xfs_log_item
*item
,
664 struct xfs_item_ops
*ops
)
666 item
->li_mountp
= mp
;
667 item
->li_ailp
= mp
->m_ail
;
668 item
->li_type
= type
;
673 * Write region vectors to log. The write happens using the space reservation
674 * of the ticket (tic). It is not a requirement that all writes for a given
675 * transaction occur with one call to xfs_log_write(). However, it is important
676 * to note that the transaction reservation code makes an assumption about the
677 * number of log headers a transaction requires that may be violated if you
678 * don't pass all the transaction vectors in one call....
682 struct xfs_mount
*mp
,
683 struct xfs_log_iovec reg
[],
685 struct xlog_ticket
*tic
,
686 xfs_lsn_t
*start_lsn
)
688 struct log
*log
= mp
->m_log
;
690 struct xfs_log_vec vec
= {
691 .lv_niovecs
= nentries
,
695 if (XLOG_FORCED_SHUTDOWN(log
))
696 return XFS_ERROR(EIO
);
698 error
= xlog_write(log
, &vec
, tic
, start_lsn
, NULL
, 0);
700 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
705 xfs_log_move_tail(xfs_mount_t
*mp
,
709 xlog_t
*log
= mp
->m_log
;
710 int need_bytes
, free_bytes
, cycle
, bytes
;
712 if (XLOG_FORCED_SHUTDOWN(log
))
716 /* needed since sync_lsn is 64 bits */
717 spin_lock(&log
->l_icloglock
);
718 tail_lsn
= log
->l_last_sync_lsn
;
719 spin_unlock(&log
->l_icloglock
);
722 spin_lock(&log
->l_grant_lock
);
724 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
728 log
->l_tail_lsn
= tail_lsn
;
731 if ((tic
= log
->l_write_headq
)) {
733 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
734 panic("Recovery problem");
736 cycle
= log
->l_grant_write_cycle
;
737 bytes
= log
->l_grant_write_bytes
;
738 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
740 ASSERT(tic
->t_flags
& XLOG_TIC_PERM_RESERV
);
742 if (free_bytes
< tic
->t_unit_res
&& tail_lsn
!= 1)
745 free_bytes
-= tic
->t_unit_res
;
746 sv_signal(&tic
->t_wait
);
748 } while (tic
!= log
->l_write_headq
);
750 if ((tic
= log
->l_reserve_headq
)) {
752 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
753 panic("Recovery problem");
755 cycle
= log
->l_grant_reserve_cycle
;
756 bytes
= log
->l_grant_reserve_bytes
;
757 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
759 if (tic
->t_flags
& XLOG_TIC_PERM_RESERV
)
760 need_bytes
= tic
->t_unit_res
*tic
->t_cnt
;
762 need_bytes
= tic
->t_unit_res
;
763 if (free_bytes
< need_bytes
&& tail_lsn
!= 1)
766 free_bytes
-= need_bytes
;
767 sv_signal(&tic
->t_wait
);
769 } while (tic
!= log
->l_reserve_headq
);
771 spin_unlock(&log
->l_grant_lock
);
772 } /* xfs_log_move_tail */
775 * Determine if we have a transaction that has gone to disk
776 * that needs to be covered. To begin the transition to the idle state
777 * firstly the log needs to be idle (no AIL and nothing in the iclogs).
778 * If we are then in a state where covering is needed, the caller is informed
779 * that dummy transactions are required to move the log into the idle state.
781 * Because this is called as part of the sync process, we should also indicate
782 * that dummy transactions should be issued in anything but the covered or
783 * idle states. This ensures that the log tail is accurately reflected in
784 * the log at the end of the sync, hence if a crash occurrs avoids replay
785 * of transactions where the metadata is already on disk.
788 xfs_log_need_covered(xfs_mount_t
*mp
)
791 xlog_t
*log
= mp
->m_log
;
793 if (!xfs_fs_writable(mp
))
796 spin_lock(&log
->l_icloglock
);
797 switch (log
->l_covered_state
) {
798 case XLOG_STATE_COVER_DONE
:
799 case XLOG_STATE_COVER_DONE2
:
800 case XLOG_STATE_COVER_IDLE
:
802 case XLOG_STATE_COVER_NEED
:
803 case XLOG_STATE_COVER_NEED2
:
804 if (!xfs_trans_ail_tail(log
->l_ailp
) &&
805 xlog_iclogs_empty(log
)) {
806 if (log
->l_covered_state
== XLOG_STATE_COVER_NEED
)
807 log
->l_covered_state
= XLOG_STATE_COVER_DONE
;
809 log
->l_covered_state
= XLOG_STATE_COVER_DONE2
;
816 spin_unlock(&log
->l_icloglock
);
820 /******************************************************************************
824 ******************************************************************************
827 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
828 * The log manager must keep track of the last LR which was committed
829 * to disk. The lsn of this LR will become the new tail_lsn whenever
830 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
831 * the situation where stuff could be written into the log but nothing
832 * was ever in the AIL when asked. Eventually, we panic since the
833 * tail hits the head.
835 * We may be holding the log iclog lock upon entering this routine.
838 xlog_assign_tail_lsn(xfs_mount_t
*mp
)
841 xlog_t
*log
= mp
->m_log
;
843 tail_lsn
= xfs_trans_ail_tail(mp
->m_ail
);
844 spin_lock(&log
->l_grant_lock
);
846 log
->l_tail_lsn
= tail_lsn
;
848 tail_lsn
= log
->l_tail_lsn
= log
->l_last_sync_lsn
;
850 spin_unlock(&log
->l_grant_lock
);
853 } /* xlog_assign_tail_lsn */
857 * Return the space in the log between the tail and the head. The head
858 * is passed in the cycle/bytes formal parms. In the special case where
859 * the reserve head has wrapped passed the tail, this calculation is no
860 * longer valid. In this case, just return 0 which means there is no space
861 * in the log. This works for all places where this function is called
862 * with the reserve head. Of course, if the write head were to ever
863 * wrap the tail, we should blow up. Rather than catch this case here,
864 * we depend on other ASSERTions in other parts of the code. XXXmiken
866 * This code also handles the case where the reservation head is behind
867 * the tail. The details of this case are described below, but the end
868 * result is that we return the size of the log as the amount of space left.
871 xlog_space_left(xlog_t
*log
, int cycle
, int bytes
)
877 tail_bytes
= BBTOB(BLOCK_LSN(log
->l_tail_lsn
));
878 tail_cycle
= CYCLE_LSN(log
->l_tail_lsn
);
879 if ((tail_cycle
== cycle
) && (bytes
>= tail_bytes
)) {
880 free_bytes
= log
->l_logsize
- (bytes
- tail_bytes
);
881 } else if ((tail_cycle
+ 1) < cycle
) {
883 } else if (tail_cycle
< cycle
) {
884 ASSERT(tail_cycle
== (cycle
- 1));
885 free_bytes
= tail_bytes
- bytes
;
888 * The reservation head is behind the tail.
889 * In this case we just want to return the size of the
890 * log as the amount of space left.
892 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
893 "xlog_space_left: head behind tail\n"
894 " tail_cycle = %d, tail_bytes = %d\n"
895 " GH cycle = %d, GH bytes = %d",
896 tail_cycle
, tail_bytes
, cycle
, bytes
);
898 free_bytes
= log
->l_logsize
;
901 } /* xlog_space_left */
905 * Log function which is called when an io completes.
907 * The log manager needs its own routine, in order to control what
908 * happens with the buffer after the write completes.
911 xlog_iodone(xfs_buf_t
*bp
)
913 xlog_in_core_t
*iclog
;
917 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
918 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long) 2);
919 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
924 * If the _XFS_BARRIER_FAILED flag was set by a lower
925 * layer, it means the underlying device no longer supports
926 * barrier I/O. Warn loudly and turn off barriers.
928 if (bp
->b_flags
& _XFS_BARRIER_FAILED
) {
929 bp
->b_flags
&= ~_XFS_BARRIER_FAILED
;
930 l
->l_mp
->m_flags
&= ~XFS_MOUNT_BARRIER
;
931 xfs_fs_cmn_err(CE_WARN
, l
->l_mp
,
932 "xlog_iodone: Barriers are no longer supported"
933 " by device. Disabling barriers\n");
937 * Race to shutdown the filesystem if we see an error.
939 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp
)), l
->l_mp
,
940 XFS_ERRTAG_IODONE_IOERR
, XFS_RANDOM_IODONE_IOERR
)) {
941 xfs_ioerror_alert("xlog_iodone", l
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
943 xfs_force_shutdown(l
->l_mp
, SHUTDOWN_LOG_IO_ERROR
);
945 * This flag will be propagated to the trans-committed
946 * callback routines to let them know that the log-commit
949 aborted
= XFS_LI_ABORTED
;
950 } else if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
951 aborted
= XFS_LI_ABORTED
;
954 /* log I/O is always issued ASYNC */
955 ASSERT(XFS_BUF_ISASYNC(bp
));
956 xlog_state_done_syncing(iclog
, aborted
);
958 * do not reference the buffer (bp) here as we could race
959 * with it being freed after writing the unmount record to the
966 * Return size of each in-core log record buffer.
968 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
970 * If the filesystem blocksize is too large, we may need to choose a
971 * larger size since the directory code currently logs entire blocks.
975 xlog_get_iclog_buffer_size(xfs_mount_t
*mp
,
981 if (mp
->m_logbufs
<= 0)
982 log
->l_iclog_bufs
= XLOG_MAX_ICLOGS
;
984 log
->l_iclog_bufs
= mp
->m_logbufs
;
987 * Buffer size passed in from mount system call.
989 if (mp
->m_logbsize
> 0) {
990 size
= log
->l_iclog_size
= mp
->m_logbsize
;
991 log
->l_iclog_size_log
= 0;
993 log
->l_iclog_size_log
++;
997 if (xfs_sb_version_haslogv2(&mp
->m_sb
)) {
998 /* # headers = size / 32k
999 * one header holds cycles from 32k of data
1002 xhdrs
= mp
->m_logbsize
/ XLOG_HEADER_CYCLE_SIZE
;
1003 if (mp
->m_logbsize
% XLOG_HEADER_CYCLE_SIZE
)
1005 log
->l_iclog_hsize
= xhdrs
<< BBSHIFT
;
1006 log
->l_iclog_heads
= xhdrs
;
1008 ASSERT(mp
->m_logbsize
<= XLOG_BIG_RECORD_BSIZE
);
1009 log
->l_iclog_hsize
= BBSIZE
;
1010 log
->l_iclog_heads
= 1;
1015 /* All machines use 32kB buffers by default. */
1016 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
1017 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1019 /* the default log size is 16k or 32k which is one header sector */
1020 log
->l_iclog_hsize
= BBSIZE
;
1021 log
->l_iclog_heads
= 1;
1024 /* are we being asked to make the sizes selected above visible? */
1025 if (mp
->m_logbufs
== 0)
1026 mp
->m_logbufs
= log
->l_iclog_bufs
;
1027 if (mp
->m_logbsize
== 0)
1028 mp
->m_logbsize
= log
->l_iclog_size
;
1029 } /* xlog_get_iclog_buffer_size */
1033 * This routine initializes some of the log structure for a given mount point.
1034 * Its primary purpose is to fill in enough, so recovery can occur. However,
1035 * some other stuff may be filled in too.
1038 xlog_alloc_log(xfs_mount_t
*mp
,
1039 xfs_buftarg_t
*log_target
,
1040 xfs_daddr_t blk_offset
,
1044 xlog_rec_header_t
*head
;
1045 xlog_in_core_t
**iclogp
;
1046 xlog_in_core_t
*iclog
, *prev_iclog
=NULL
;
1053 log
= kmem_zalloc(sizeof(xlog_t
), KM_MAYFAIL
);
1055 xlog_warn("XFS: Log allocation failed: No memory!");
1060 log
->l_targ
= log_target
;
1061 log
->l_logsize
= BBTOB(num_bblks
);
1062 log
->l_logBBstart
= blk_offset
;
1063 log
->l_logBBsize
= num_bblks
;
1064 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1065 log
->l_flags
|= XLOG_ACTIVE_RECOVERY
;
1067 log
->l_prev_block
= -1;
1068 log
->l_tail_lsn
= xlog_assign_lsn(1, 0);
1069 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1070 log
->l_last_sync_lsn
= log
->l_tail_lsn
;
1071 log
->l_curr_cycle
= 1; /* 0 is bad since this is initial value */
1072 log
->l_grant_reserve_cycle
= 1;
1073 log
->l_grant_write_cycle
= 1;
1075 error
= EFSCORRUPTED
;
1076 if (xfs_sb_version_hassector(&mp
->m_sb
)) {
1077 log2_size
= mp
->m_sb
.sb_logsectlog
;
1078 if (log2_size
< BBSHIFT
) {
1079 xlog_warn("XFS: Log sector size too small "
1080 "(0x%x < 0x%x)", log2_size
, BBSHIFT
);
1084 log2_size
-= BBSHIFT
;
1085 if (log2_size
> mp
->m_sectbb_log
) {
1086 xlog_warn("XFS: Log sector size too large "
1087 "(0x%x > 0x%x)", log2_size
, mp
->m_sectbb_log
);
1091 /* for larger sector sizes, must have v2 or external log */
1092 if (log2_size
&& log
->l_logBBstart
> 0 &&
1093 !xfs_sb_version_haslogv2(&mp
->m_sb
)) {
1095 xlog_warn("XFS: log sector size (0x%x) invalid "
1096 "for configuration.", log2_size
);
1100 log
->l_sectBBsize
= 1 << log2_size
;
1102 xlog_get_iclog_buffer_size(mp
, log
);
1105 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1108 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1109 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1110 ASSERT(XFS_BUF_ISBUSY(bp
));
1111 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
1114 spin_lock_init(&log
->l_icloglock
);
1115 spin_lock_init(&log
->l_grant_lock
);
1116 sv_init(&log
->l_flush_wait
, 0, "flush_wait");
1118 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1119 ASSERT((XFS_BUF_SIZE(bp
) & BBMASK
) == 0);
1121 iclogp
= &log
->l_iclog
;
1123 * The amount of memory to allocate for the iclog structure is
1124 * rather funky due to the way the structure is defined. It is
1125 * done this way so that we can use different sizes for machines
1126 * with different amounts of memory. See the definition of
1127 * xlog_in_core_t in xfs_log_priv.h for details.
1129 iclogsize
= log
->l_iclog_size
;
1130 ASSERT(log
->l_iclog_size
>= 4096);
1131 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
1132 *iclogp
= kmem_zalloc(sizeof(xlog_in_core_t
), KM_MAYFAIL
);
1134 goto out_free_iclog
;
1137 iclog
->ic_prev
= prev_iclog
;
1140 bp
= xfs_buf_get_noaddr(log
->l_iclog_size
, mp
->m_logdev_targp
);
1142 goto out_free_iclog
;
1143 if (!XFS_BUF_CPSEMA(bp
))
1145 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1146 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1148 iclog
->ic_data
= bp
->b_addr
;
1150 log
->l_iclog_bak
[i
] = (xfs_caddr_t
)&(iclog
->ic_header
);
1152 head
= &iclog
->ic_header
;
1153 memset(head
, 0, sizeof(xlog_rec_header_t
));
1154 head
->h_magicno
= cpu_to_be32(XLOG_HEADER_MAGIC_NUM
);
1155 head
->h_version
= cpu_to_be32(
1156 xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) ? 2 : 1);
1157 head
->h_size
= cpu_to_be32(log
->l_iclog_size
);
1159 head
->h_fmt
= cpu_to_be32(XLOG_FMT
);
1160 memcpy(&head
->h_fs_uuid
, &mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1162 iclog
->ic_size
= XFS_BUF_SIZE(bp
) - log
->l_iclog_hsize
;
1163 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1164 iclog
->ic_log
= log
;
1165 atomic_set(&iclog
->ic_refcnt
, 0);
1166 spin_lock_init(&iclog
->ic_callback_lock
);
1167 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
1168 iclog
->ic_datap
= (char *)iclog
->ic_data
+ log
->l_iclog_hsize
;
1170 ASSERT(XFS_BUF_ISBUSY(iclog
->ic_bp
));
1171 ASSERT(XFS_BUF_VALUSEMA(iclog
->ic_bp
) <= 0);
1172 sv_init(&iclog
->ic_force_wait
, SV_DEFAULT
, "iclog-force");
1173 sv_init(&iclog
->ic_write_wait
, SV_DEFAULT
, "iclog-write");
1175 iclogp
= &iclog
->ic_next
;
1177 *iclogp
= log
->l_iclog
; /* complete ring */
1178 log
->l_iclog
->ic_prev
= prev_iclog
; /* re-write 1st prev ptr */
1183 for (iclog
= log
->l_iclog
; iclog
; iclog
= prev_iclog
) {
1184 prev_iclog
= iclog
->ic_next
;
1186 sv_destroy(&iclog
->ic_force_wait
);
1187 sv_destroy(&iclog
->ic_write_wait
);
1188 xfs_buf_free(iclog
->ic_bp
);
1192 spinlock_destroy(&log
->l_icloglock
);
1193 spinlock_destroy(&log
->l_grant_lock
);
1194 xfs_buf_free(log
->l_xbuf
);
1198 return ERR_PTR(-error
);
1199 } /* xlog_alloc_log */
1203 * Write out the commit record of a transaction associated with the given
1204 * ticket. Return the lsn of the commit record.
1209 struct xlog_ticket
*ticket
,
1210 struct xlog_in_core
**iclog
,
1211 xfs_lsn_t
*commitlsnp
)
1213 struct xfs_mount
*mp
= log
->l_mp
;
1215 struct xfs_log_iovec reg
= {
1218 .i_type
= XLOG_REG_TYPE_COMMIT
,
1220 struct xfs_log_vec vec
= {
1225 ASSERT_ALWAYS(iclog
);
1226 error
= xlog_write(log
, &vec
, ticket
, commitlsnp
, iclog
,
1229 xfs_force_shutdown(mp
, SHUTDOWN_LOG_IO_ERROR
);
1234 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1235 * log space. This code pushes on the lsn which would supposedly free up
1236 * the 25% which we want to leave free. We may need to adopt a policy which
1237 * pushes on an lsn which is further along in the log once we reach the high
1238 * water mark. In this manner, we would be creating a low water mark.
1241 xlog_grant_push_ail(xfs_mount_t
*mp
,
1244 xlog_t
*log
= mp
->m_log
; /* pointer to the log */
1245 xfs_lsn_t tail_lsn
; /* lsn of the log tail */
1246 xfs_lsn_t threshold_lsn
= 0; /* lsn we'd like to be at */
1247 int free_blocks
; /* free blocks left to write to */
1248 int free_bytes
; /* free bytes left to write to */
1249 int threshold_block
; /* block in lsn we'd like to be at */
1250 int threshold_cycle
; /* lsn cycle we'd like to be at */
1253 ASSERT(BTOBB(need_bytes
) < log
->l_logBBsize
);
1255 spin_lock(&log
->l_grant_lock
);
1256 free_bytes
= xlog_space_left(log
,
1257 log
->l_grant_reserve_cycle
,
1258 log
->l_grant_reserve_bytes
);
1259 tail_lsn
= log
->l_tail_lsn
;
1260 free_blocks
= BTOBBT(free_bytes
);
1263 * Set the threshold for the minimum number of free blocks in the
1264 * log to the maximum of what the caller needs, one quarter of the
1265 * log, and 256 blocks.
1267 free_threshold
= BTOBB(need_bytes
);
1268 free_threshold
= MAX(free_threshold
, (log
->l_logBBsize
>> 2));
1269 free_threshold
= MAX(free_threshold
, 256);
1270 if (free_blocks
< free_threshold
) {
1271 threshold_block
= BLOCK_LSN(tail_lsn
) + free_threshold
;
1272 threshold_cycle
= CYCLE_LSN(tail_lsn
);
1273 if (threshold_block
>= log
->l_logBBsize
) {
1274 threshold_block
-= log
->l_logBBsize
;
1275 threshold_cycle
+= 1;
1277 threshold_lsn
= xlog_assign_lsn(threshold_cycle
, threshold_block
);
1279 /* Don't pass in an lsn greater than the lsn of the last
1280 * log record known to be on disk.
1282 if (XFS_LSN_CMP(threshold_lsn
, log
->l_last_sync_lsn
) > 0)
1283 threshold_lsn
= log
->l_last_sync_lsn
;
1285 spin_unlock(&log
->l_grant_lock
);
1288 * Get the transaction layer to kick the dirty buffers out to
1289 * disk asynchronously. No point in trying to do this if
1290 * the filesystem is shutting down.
1292 if (threshold_lsn
&&
1293 !XLOG_FORCED_SHUTDOWN(log
))
1294 xfs_trans_ail_push(log
->l_ailp
, threshold_lsn
);
1295 } /* xlog_grant_push_ail */
1298 * The bdstrat callback function for log bufs. This gives us a central
1299 * place to trap bufs in case we get hit by a log I/O error and need to
1300 * shutdown. Actually, in practice, even when we didn't get a log error,
1301 * we transition the iclogs to IOERROR state *after* flushing all existing
1302 * iclogs to disk. This is because we don't want anymore new transactions to be
1303 * started or completed afterwards.
1309 struct xlog_in_core
*iclog
;
1311 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
1312 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
1313 XFS_BUF_ERROR(bp
, EIO
);
1317 * It would seem logical to return EIO here, but we rely on
1318 * the log state machine to propagate I/O errors instead of
1324 bp
->b_flags
|= _XBF_RUN_QUEUES
;
1325 xfs_buf_iorequest(bp
);
1330 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1331 * fashion. Previously, we should have moved the current iclog
1332 * ptr in the log to point to the next available iclog. This allows further
1333 * write to continue while this code syncs out an iclog ready to go.
1334 * Before an in-core log can be written out, the data section must be scanned
1335 * to save away the 1st word of each BBSIZE block into the header. We replace
1336 * it with the current cycle count. Each BBSIZE block is tagged with the
1337 * cycle count because there in an implicit assumption that drives will
1338 * guarantee that entire 512 byte blocks get written at once. In other words,
1339 * we can't have part of a 512 byte block written and part not written. By
1340 * tagging each block, we will know which blocks are valid when recovering
1341 * after an unclean shutdown.
1343 * This routine is single threaded on the iclog. No other thread can be in
1344 * this routine with the same iclog. Changing contents of iclog can there-
1345 * fore be done without grabbing the state machine lock. Updating the global
1346 * log will require grabbing the lock though.
1348 * The entire log manager uses a logical block numbering scheme. Only
1349 * log_sync (and then only bwrite()) know about the fact that the log may
1350 * not start with block zero on a given device. The log block start offset
1351 * is added immediately before calling bwrite().
1355 xlog_sync(xlog_t
*log
,
1356 xlog_in_core_t
*iclog
)
1358 xfs_caddr_t dptr
; /* pointer to byte sized element */
1361 uint count
; /* byte count of bwrite */
1362 uint count_init
; /* initial count before roundup */
1363 int roundoff
; /* roundoff to BB or stripe */
1364 int split
= 0; /* split write into two regions */
1366 int v2
= xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
);
1368 XFS_STATS_INC(xs_log_writes
);
1369 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
1371 /* Add for LR header */
1372 count_init
= log
->l_iclog_hsize
+ iclog
->ic_offset
;
1374 /* Round out the log write size */
1375 if (v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1) {
1376 /* we have a v2 stripe unit to use */
1377 count
= XLOG_LSUNITTOB(log
, XLOG_BTOLSUNIT(log
, count_init
));
1379 count
= BBTOB(BTOBB(count_init
));
1381 roundoff
= count
- count_init
;
1382 ASSERT(roundoff
>= 0);
1383 ASSERT((v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1 &&
1384 roundoff
< log
->l_mp
->m_sb
.sb_logsunit
)
1386 (log
->l_mp
->m_sb
.sb_logsunit
<= 1 &&
1387 roundoff
< BBTOB(1)));
1389 /* move grant heads by roundoff in sync */
1390 spin_lock(&log
->l_grant_lock
);
1391 xlog_grant_add_space(log
, roundoff
);
1392 spin_unlock(&log
->l_grant_lock
);
1394 /* put cycle number in every block */
1395 xlog_pack_data(log
, iclog
, roundoff
);
1397 /* real byte length */
1399 iclog
->ic_header
.h_len
=
1400 cpu_to_be32(iclog
->ic_offset
+ roundoff
);
1402 iclog
->ic_header
.h_len
=
1403 cpu_to_be32(iclog
->ic_offset
);
1407 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long)1);
1408 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1409 XFS_BUF_SET_ADDR(bp
, BLOCK_LSN(be64_to_cpu(iclog
->ic_header
.h_lsn
)));
1411 XFS_STATS_ADD(xs_log_blocks
, BTOBB(count
));
1413 /* Do we need to split this write into 2 parts? */
1414 if (XFS_BUF_ADDR(bp
) + BTOBB(count
) > log
->l_logBBsize
) {
1415 split
= count
- (BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
)));
1416 count
= BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
));
1417 iclog
->ic_bwritecnt
= 2; /* split into 2 writes */
1419 iclog
->ic_bwritecnt
= 1;
1421 XFS_BUF_SET_COUNT(bp
, count
);
1422 XFS_BUF_SET_FSPRIVATE(bp
, iclog
); /* save for later */
1423 XFS_BUF_ZEROFLAGS(bp
);
1426 bp
->b_flags
|= XBF_LOG_BUFFER
;
1428 * Do an ordered write for the log block.
1429 * Its unnecessary to flush the first split block in the log wrap case.
1431 if (!split
&& (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
))
1432 XFS_BUF_ORDERED(bp
);
1434 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1435 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1437 xlog_verify_iclog(log
, iclog
, count
, B_TRUE
);
1439 /* account for log which doesn't start at block #0 */
1440 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1442 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1447 if ((error
= xlog_bdstrat(bp
))) {
1448 xfs_ioerror_alert("xlog_sync", log
->l_mp
, bp
,
1453 bp
= iclog
->ic_log
->l_xbuf
;
1454 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) ==
1456 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1457 XFS_BUF_SET_ADDR(bp
, 0); /* logical 0 */
1458 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
)((__psint_t
)&(iclog
->ic_header
)+
1459 (__psint_t
)count
), split
);
1460 XFS_BUF_SET_FSPRIVATE(bp
, iclog
);
1461 XFS_BUF_ZEROFLAGS(bp
);
1464 bp
->b_flags
|= XBF_LOG_BUFFER
;
1465 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1466 XFS_BUF_ORDERED(bp
);
1467 dptr
= XFS_BUF_PTR(bp
);
1469 * Bump the cycle numbers at the start of each block
1470 * since this part of the buffer is at the start of
1471 * a new cycle. Watch out for the header magic number
1474 for (i
= 0; i
< split
; i
+= BBSIZE
) {
1475 be32_add_cpu((__be32
*)dptr
, 1);
1476 if (be32_to_cpu(*(__be32
*)dptr
) == XLOG_HEADER_MAGIC_NUM
)
1477 be32_add_cpu((__be32
*)dptr
, 1);
1481 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1482 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1484 /* account for internal log which doesn't start at block #0 */
1485 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1487 if ((error
= xlog_bdstrat(bp
))) {
1488 xfs_ioerror_alert("xlog_sync (split)", log
->l_mp
,
1489 bp
, XFS_BUF_ADDR(bp
));
1498 * Deallocate a log structure
1501 xlog_dealloc_log(xlog_t
*log
)
1503 xlog_in_core_t
*iclog
, *next_iclog
;
1506 iclog
= log
->l_iclog
;
1507 for (i
=0; i
<log
->l_iclog_bufs
; i
++) {
1508 sv_destroy(&iclog
->ic_force_wait
);
1509 sv_destroy(&iclog
->ic_write_wait
);
1510 xfs_buf_free(iclog
->ic_bp
);
1511 next_iclog
= iclog
->ic_next
;
1515 spinlock_destroy(&log
->l_icloglock
);
1516 spinlock_destroy(&log
->l_grant_lock
);
1518 xfs_buf_free(log
->l_xbuf
);
1519 log
->l_mp
->m_log
= NULL
;
1521 } /* xlog_dealloc_log */
1524 * Update counters atomically now that memcpy is done.
1528 xlog_state_finish_copy(xlog_t
*log
,
1529 xlog_in_core_t
*iclog
,
1533 spin_lock(&log
->l_icloglock
);
1535 be32_add_cpu(&iclog
->ic_header
.h_num_logops
, record_cnt
);
1536 iclog
->ic_offset
+= copy_bytes
;
1538 spin_unlock(&log
->l_icloglock
);
1539 } /* xlog_state_finish_copy */
1545 * print out info relating to regions written which consume
1549 xlog_print_tic_res(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
)
1552 uint ophdr_spc
= ticket
->t_res_num_ophdrs
* (uint
)sizeof(xlog_op_header_t
);
1554 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1555 static char *res_type_str
[XLOG_REG_TYPE_MAX
] = {
1576 static char *trans_type_str
[XFS_TRANS_TYPE_MAX
] = {
1619 xfs_fs_cmn_err(CE_WARN
, mp
,
1620 "xfs_log_write: reservation summary:\n"
1621 " trans type = %s (%u)\n"
1622 " unit res = %d bytes\n"
1623 " current res = %d bytes\n"
1624 " total reg = %u bytes (o/flow = %u bytes)\n"
1625 " ophdrs = %u (ophdr space = %u bytes)\n"
1626 " ophdr + reg = %u bytes\n"
1627 " num regions = %u\n",
1628 ((ticket
->t_trans_type
<= 0 ||
1629 ticket
->t_trans_type
> XFS_TRANS_TYPE_MAX
) ?
1630 "bad-trans-type" : trans_type_str
[ticket
->t_trans_type
-1]),
1631 ticket
->t_trans_type
,
1634 ticket
->t_res_arr_sum
, ticket
->t_res_o_flow
,
1635 ticket
->t_res_num_ophdrs
, ophdr_spc
,
1636 ticket
->t_res_arr_sum
+
1637 ticket
->t_res_o_flow
+ ophdr_spc
,
1640 for (i
= 0; i
< ticket
->t_res_num
; i
++) {
1641 uint r_type
= ticket
->t_res_arr
[i
].r_type
;
1643 "region[%u]: %s - %u bytes\n",
1645 ((r_type
<= 0 || r_type
> XLOG_REG_TYPE_MAX
) ?
1646 "bad-rtype" : res_type_str
[r_type
-1]),
1647 ticket
->t_res_arr
[i
].r_len
);
1652 * Calculate the potential space needed by the log vector. Each region gets
1653 * its own xlog_op_header_t and may need to be double word aligned.
1656 xlog_write_calc_vec_length(
1657 struct xlog_ticket
*ticket
,
1658 struct xfs_log_vec
*log_vector
)
1660 struct xfs_log_vec
*lv
;
1665 /* acct for start rec of xact */
1666 if (ticket
->t_flags
& XLOG_TIC_INITED
)
1669 for (lv
= log_vector
; lv
; lv
= lv
->lv_next
) {
1670 headers
+= lv
->lv_niovecs
;
1672 for (i
= 0; i
< lv
->lv_niovecs
; i
++) {
1673 struct xfs_log_iovec
*vecp
= &lv
->lv_iovecp
[i
];
1676 xlog_tic_add_region(ticket
, vecp
->i_len
, vecp
->i_type
);
1680 ticket
->t_res_num_ophdrs
+= headers
;
1681 len
+= headers
* sizeof(struct xlog_op_header
);
1687 * If first write for transaction, insert start record We can't be trying to
1688 * commit if we are inited. We can't have any "partial_copy" if we are inited.
1691 xlog_write_start_rec(
1692 struct xlog_op_header
*ophdr
,
1693 struct xlog_ticket
*ticket
)
1695 if (!(ticket
->t_flags
& XLOG_TIC_INITED
))
1698 ophdr
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1699 ophdr
->oh_clientid
= ticket
->t_clientid
;
1701 ophdr
->oh_flags
= XLOG_START_TRANS
;
1704 ticket
->t_flags
&= ~XLOG_TIC_INITED
;
1706 return sizeof(struct xlog_op_header
);
1709 static xlog_op_header_t
*
1710 xlog_write_setup_ophdr(
1712 struct xlog_op_header
*ophdr
,
1713 struct xlog_ticket
*ticket
,
1716 ophdr
->oh_tid
= cpu_to_be32(ticket
->t_tid
);
1717 ophdr
->oh_clientid
= ticket
->t_clientid
;
1720 /* are we copying a commit or unmount record? */
1721 ophdr
->oh_flags
= flags
;
1724 * We've seen logs corrupted with bad transaction client ids. This
1725 * makes sure that XFS doesn't generate them on. Turn this into an EIO
1726 * and shut down the filesystem.
1728 switch (ophdr
->oh_clientid
) {
1729 case XFS_TRANSACTION
:
1734 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
1735 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1736 ophdr
->oh_clientid
, ticket
);
1744 * Set up the parameters of the region copy into the log. This has
1745 * to handle region write split across multiple log buffers - this
1746 * state is kept external to this function so that this code can
1747 * can be written in an obvious, self documenting manner.
1750 xlog_write_setup_copy(
1751 struct xlog_ticket
*ticket
,
1752 struct xlog_op_header
*ophdr
,
1753 int space_available
,
1757 int *last_was_partial_copy
,
1758 int *bytes_consumed
)
1762 still_to_copy
= space_required
- *bytes_consumed
;
1763 *copy_off
= *bytes_consumed
;
1765 if (still_to_copy
<= space_available
) {
1766 /* write of region completes here */
1767 *copy_len
= still_to_copy
;
1768 ophdr
->oh_len
= cpu_to_be32(*copy_len
);
1769 if (*last_was_partial_copy
)
1770 ophdr
->oh_flags
|= (XLOG_END_TRANS
|XLOG_WAS_CONT_TRANS
);
1771 *last_was_partial_copy
= 0;
1772 *bytes_consumed
= 0;
1776 /* partial write of region, needs extra log op header reservation */
1777 *copy_len
= space_available
;
1778 ophdr
->oh_len
= cpu_to_be32(*copy_len
);
1779 ophdr
->oh_flags
|= XLOG_CONTINUE_TRANS
;
1780 if (*last_was_partial_copy
)
1781 ophdr
->oh_flags
|= XLOG_WAS_CONT_TRANS
;
1782 *bytes_consumed
+= *copy_len
;
1783 (*last_was_partial_copy
)++;
1785 /* account for new log op header */
1786 ticket
->t_curr_res
-= sizeof(struct xlog_op_header
);
1787 ticket
->t_res_num_ophdrs
++;
1789 return sizeof(struct xlog_op_header
);
1793 xlog_write_copy_finish(
1795 struct xlog_in_core
*iclog
,
1800 int *partial_copy_len
,
1802 struct xlog_in_core
**commit_iclog
)
1804 if (*partial_copy
) {
1806 * This iclog has already been marked WANT_SYNC by
1807 * xlog_state_get_iclog_space.
1809 xlog_state_finish_copy(log
, iclog
, *record_cnt
, *data_cnt
);
1812 return xlog_state_release_iclog(log
, iclog
);
1816 *partial_copy_len
= 0;
1818 if (iclog
->ic_size
- log_offset
<= sizeof(xlog_op_header_t
)) {
1819 /* no more space in this iclog - push it. */
1820 xlog_state_finish_copy(log
, iclog
, *record_cnt
, *data_cnt
);
1824 spin_lock(&log
->l_icloglock
);
1825 xlog_state_want_sync(log
, iclog
);
1826 spin_unlock(&log
->l_icloglock
);
1829 return xlog_state_release_iclog(log
, iclog
);
1830 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1831 *commit_iclog
= iclog
;
1838 * Write some region out to in-core log
1840 * This will be called when writing externally provided regions or when
1841 * writing out a commit record for a given transaction.
1843 * General algorithm:
1844 * 1. Find total length of this write. This may include adding to the
1845 * lengths passed in.
1846 * 2. Check whether we violate the tickets reservation.
1847 * 3. While writing to this iclog
1848 * A. Reserve as much space in this iclog as can get
1849 * B. If this is first write, save away start lsn
1850 * C. While writing this region:
1851 * 1. If first write of transaction, write start record
1852 * 2. Write log operation header (header per region)
1853 * 3. Find out if we can fit entire region into this iclog
1854 * 4. Potentially, verify destination memcpy ptr
1855 * 5. Memcpy (partial) region
1856 * 6. If partial copy, release iclog; otherwise, continue
1857 * copying more regions into current iclog
1858 * 4. Mark want sync bit (in simulation mode)
1859 * 5. Release iclog for potential flush to on-disk log.
1862 * 1. Panic if reservation is overrun. This should never happen since
1863 * reservation amounts are generated internal to the filesystem.
1865 * 1. Tickets are single threaded data structures.
1866 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1867 * syncing routine. When a single log_write region needs to span
1868 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1869 * on all log operation writes which don't contain the end of the
1870 * region. The XLOG_END_TRANS bit is used for the in-core log
1871 * operation which contains the end of the continued log_write region.
1872 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1873 * we don't really know exactly how much space will be used. As a result,
1874 * we don't update ic_offset until the end when we know exactly how many
1875 * bytes have been written out.
1880 struct xfs_log_vec
*log_vector
,
1881 struct xlog_ticket
*ticket
,
1882 xfs_lsn_t
*start_lsn
,
1883 struct xlog_in_core
**commit_iclog
,
1886 struct xlog_in_core
*iclog
= NULL
;
1887 struct xfs_log_iovec
*vecp
;
1888 struct xfs_log_vec
*lv
;
1891 int partial_copy
= 0;
1892 int partial_copy_len
= 0;
1900 len
= xlog_write_calc_vec_length(ticket
, log_vector
);
1901 if (ticket
->t_curr_res
< len
) {
1902 xlog_print_tic_res(log
->l_mp
, ticket
);
1905 "xfs_log_write: reservation ran out. Need to up reservation");
1907 /* Customer configurable panic */
1908 xfs_cmn_err(XFS_PTAG_LOGRES
, CE_ALERT
, log
->l_mp
,
1909 "xfs_log_write: reservation ran out. Need to up reservation");
1911 /* If we did not panic, shutdown the filesystem */
1912 xfs_force_shutdown(log
->l_mp
, SHUTDOWN_CORRUPT_INCORE
);
1916 ticket
->t_curr_res
-= len
;
1920 vecp
= lv
->lv_iovecp
;
1921 while (lv
&& index
< lv
->lv_niovecs
) {
1925 error
= xlog_state_get_iclog_space(log
, len
, &iclog
, ticket
,
1926 &contwr
, &log_offset
);
1930 ASSERT(log_offset
<= iclog
->ic_size
- 1);
1931 ptr
= iclog
->ic_datap
+ log_offset
;
1933 /* start_lsn is the first lsn written to. That's all we need. */
1935 *start_lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
1938 * This loop writes out as many regions as can fit in the amount
1939 * of space which was allocated by xlog_state_get_iclog_space().
1941 while (lv
&& index
< lv
->lv_niovecs
) {
1942 struct xfs_log_iovec
*reg
= &vecp
[index
];
1943 struct xlog_op_header
*ophdr
;
1948 ASSERT(reg
->i_len
% sizeof(__int32_t
) == 0);
1949 ASSERT((unsigned long)ptr
% sizeof(__int32_t
) == 0);
1951 start_rec_copy
= xlog_write_start_rec(ptr
, ticket
);
1952 if (start_rec_copy
) {
1954 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
,
1958 ophdr
= xlog_write_setup_ophdr(log
, ptr
, ticket
, flags
);
1960 return XFS_ERROR(EIO
);
1962 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
,
1963 sizeof(struct xlog_op_header
));
1965 len
+= xlog_write_setup_copy(ticket
, ophdr
,
1966 iclog
->ic_size
-log_offset
,
1968 ©_off
, ©_len
,
1971 xlog_verify_dest_ptr(log
, ptr
);
1974 ASSERT(copy_len
>= 0);
1975 memcpy(ptr
, reg
->i_addr
+ copy_off
, copy_len
);
1976 xlog_write_adv_cnt(&ptr
, &len
, &log_offset
, copy_len
);
1978 copy_len
+= start_rec_copy
+ sizeof(xlog_op_header_t
);
1980 data_cnt
+= contwr
? copy_len
: 0;
1982 error
= xlog_write_copy_finish(log
, iclog
, flags
,
1983 &record_cnt
, &data_cnt
,
1992 * if we had a partial copy, we need to get more iclog
1993 * space but we don't want to increment the region
1994 * index because there is still more is this region to
1997 * If we completed writing this region, and we flushed
1998 * the iclog (indicated by resetting of the record
1999 * count), then we also need to get more log space. If
2000 * this was the last record, though, we are done and
2006 if (++index
== lv
->lv_niovecs
) {
2010 vecp
= lv
->lv_iovecp
;
2012 if (record_cnt
== 0) {
2022 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
2024 return xlog_state_release_iclog(log
, iclog
);
2026 ASSERT(flags
& XLOG_COMMIT_TRANS
);
2027 *commit_iclog
= iclog
;
2032 /*****************************************************************************
2034 * State Machine functions
2036 *****************************************************************************
2039 /* Clean iclogs starting from the head. This ordering must be
2040 * maintained, so an iclog doesn't become ACTIVE beyond one that
2041 * is SYNCING. This is also required to maintain the notion that we use
2042 * a ordered wait queue to hold off would be writers to the log when every
2043 * iclog is trying to sync to disk.
2045 * State Change: DIRTY -> ACTIVE
2048 xlog_state_clean_log(xlog_t
*log
)
2050 xlog_in_core_t
*iclog
;
2053 iclog
= log
->l_iclog
;
2055 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2056 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
2057 iclog
->ic_offset
= 0;
2058 ASSERT(iclog
->ic_callback
== NULL
);
2060 * If the number of ops in this iclog indicate it just
2061 * contains the dummy transaction, we can
2062 * change state into IDLE (the second time around).
2063 * Otherwise we should change the state into
2065 * We don't need to cover the dummy.
2068 (be32_to_cpu(iclog
->ic_header
.h_num_logops
) ==
2073 * We have two dirty iclogs so start over
2074 * This could also be num of ops indicates
2075 * this is not the dummy going out.
2079 iclog
->ic_header
.h_num_logops
= 0;
2080 memset(iclog
->ic_header
.h_cycle_data
, 0,
2081 sizeof(iclog
->ic_header
.h_cycle_data
));
2082 iclog
->ic_header
.h_lsn
= 0;
2083 } else if (iclog
->ic_state
== XLOG_STATE_ACTIVE
)
2086 break; /* stop cleaning */
2087 iclog
= iclog
->ic_next
;
2088 } while (iclog
!= log
->l_iclog
);
2090 /* log is locked when we are called */
2092 * Change state for the dummy log recording.
2093 * We usually go to NEED. But we go to NEED2 if the changed indicates
2094 * we are done writing the dummy record.
2095 * If we are done with the second dummy recored (DONE2), then
2099 switch (log
->l_covered_state
) {
2100 case XLOG_STATE_COVER_IDLE
:
2101 case XLOG_STATE_COVER_NEED
:
2102 case XLOG_STATE_COVER_NEED2
:
2103 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2106 case XLOG_STATE_COVER_DONE
:
2108 log
->l_covered_state
= XLOG_STATE_COVER_NEED2
;
2110 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2113 case XLOG_STATE_COVER_DONE2
:
2115 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
2117 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
2124 } /* xlog_state_clean_log */
2127 xlog_get_lowest_lsn(
2130 xlog_in_core_t
*lsn_log
;
2131 xfs_lsn_t lowest_lsn
, lsn
;
2133 lsn_log
= log
->l_iclog
;
2136 if (!(lsn_log
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
))) {
2137 lsn
= be64_to_cpu(lsn_log
->ic_header
.h_lsn
);
2138 if ((lsn
&& !lowest_lsn
) ||
2139 (XFS_LSN_CMP(lsn
, lowest_lsn
) < 0)) {
2143 lsn_log
= lsn_log
->ic_next
;
2144 } while (lsn_log
!= log
->l_iclog
);
2150 xlog_state_do_callback(
2153 xlog_in_core_t
*ciclog
)
2155 xlog_in_core_t
*iclog
;
2156 xlog_in_core_t
*first_iclog
; /* used to know when we've
2157 * processed all iclogs once */
2158 xfs_log_callback_t
*cb
, *cb_next
;
2160 xfs_lsn_t lowest_lsn
;
2161 int ioerrors
; /* counter: iclogs with errors */
2162 int loopdidcallbacks
; /* flag: inner loop did callbacks*/
2163 int funcdidcallbacks
; /* flag: function did callbacks */
2164 int repeats
; /* for issuing console warnings if
2165 * looping too many times */
2168 spin_lock(&log
->l_icloglock
);
2169 first_iclog
= iclog
= log
->l_iclog
;
2171 funcdidcallbacks
= 0;
2176 * Scan all iclogs starting with the one pointed to by the
2177 * log. Reset this starting point each time the log is
2178 * unlocked (during callbacks).
2180 * Keep looping through iclogs until one full pass is made
2181 * without running any callbacks.
2183 first_iclog
= log
->l_iclog
;
2184 iclog
= log
->l_iclog
;
2185 loopdidcallbacks
= 0;
2190 /* skip all iclogs in the ACTIVE & DIRTY states */
2191 if (iclog
->ic_state
&
2192 (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
)) {
2193 iclog
= iclog
->ic_next
;
2198 * Between marking a filesystem SHUTDOWN and stopping
2199 * the log, we do flush all iclogs to disk (if there
2200 * wasn't a log I/O error). So, we do want things to
2201 * go smoothly in case of just a SHUTDOWN w/o a
2204 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
2206 * Can only perform callbacks in order. Since
2207 * this iclog is not in the DONE_SYNC/
2208 * DO_CALLBACK state, we skip the rest and
2209 * just try to clean up. If we set our iclog
2210 * to DO_CALLBACK, we will not process it when
2211 * we retry since a previous iclog is in the
2212 * CALLBACK and the state cannot change since
2213 * we are holding the l_icloglock.
2215 if (!(iclog
->ic_state
&
2216 (XLOG_STATE_DONE_SYNC
|
2217 XLOG_STATE_DO_CALLBACK
))) {
2218 if (ciclog
&& (ciclog
->ic_state
==
2219 XLOG_STATE_DONE_SYNC
)) {
2220 ciclog
->ic_state
= XLOG_STATE_DO_CALLBACK
;
2225 * We now have an iclog that is in either the
2226 * DO_CALLBACK or DONE_SYNC states. The other
2227 * states (WANT_SYNC, SYNCING, or CALLBACK were
2228 * caught by the above if and are going to
2229 * clean (i.e. we aren't doing their callbacks)
2234 * We will do one more check here to see if we
2235 * have chased our tail around.
2238 lowest_lsn
= xlog_get_lowest_lsn(log
);
2240 XFS_LSN_CMP(lowest_lsn
,
2241 be64_to_cpu(iclog
->ic_header
.h_lsn
)) < 0) {
2242 iclog
= iclog
->ic_next
;
2243 continue; /* Leave this iclog for
2247 iclog
->ic_state
= XLOG_STATE_CALLBACK
;
2249 spin_unlock(&log
->l_icloglock
);
2251 /* l_last_sync_lsn field protected by
2252 * l_grant_lock. Don't worry about iclog's lsn.
2253 * No one else can be here except us.
2255 spin_lock(&log
->l_grant_lock
);
2256 ASSERT(XFS_LSN_CMP(log
->l_last_sync_lsn
,
2257 be64_to_cpu(iclog
->ic_header
.h_lsn
)) <= 0);
2258 log
->l_last_sync_lsn
=
2259 be64_to_cpu(iclog
->ic_header
.h_lsn
);
2260 spin_unlock(&log
->l_grant_lock
);
2263 spin_unlock(&log
->l_icloglock
);
2268 * Keep processing entries in the callback list until
2269 * we come around and it is empty. We need to
2270 * atomically see that the list is empty and change the
2271 * state to DIRTY so that we don't miss any more
2272 * callbacks being added.
2274 spin_lock(&iclog
->ic_callback_lock
);
2275 cb
= iclog
->ic_callback
;
2277 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
2278 iclog
->ic_callback
= NULL
;
2279 spin_unlock(&iclog
->ic_callback_lock
);
2281 /* perform callbacks in the order given */
2282 for (; cb
; cb
= cb_next
) {
2283 cb_next
= cb
->cb_next
;
2284 cb
->cb_func(cb
->cb_arg
, aborted
);
2286 spin_lock(&iclog
->ic_callback_lock
);
2287 cb
= iclog
->ic_callback
;
2293 spin_lock(&log
->l_icloglock
);
2294 ASSERT(iclog
->ic_callback
== NULL
);
2295 spin_unlock(&iclog
->ic_callback_lock
);
2296 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
))
2297 iclog
->ic_state
= XLOG_STATE_DIRTY
;
2300 * Transition from DIRTY to ACTIVE if applicable.
2301 * NOP if STATE_IOERROR.
2303 xlog_state_clean_log(log
);
2305 /* wake up threads waiting in xfs_log_force() */
2306 sv_broadcast(&iclog
->ic_force_wait
);
2308 iclog
= iclog
->ic_next
;
2309 } while (first_iclog
!= iclog
);
2311 if (repeats
> 5000) {
2312 flushcnt
+= repeats
;
2314 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
2315 "%s: possible infinite loop (%d iterations)",
2316 __func__
, flushcnt
);
2318 } while (!ioerrors
&& loopdidcallbacks
);
2321 * make one last gasp attempt to see if iclogs are being left in
2325 if (funcdidcallbacks
) {
2326 first_iclog
= iclog
= log
->l_iclog
;
2328 ASSERT(iclog
->ic_state
!= XLOG_STATE_DO_CALLBACK
);
2330 * Terminate the loop if iclogs are found in states
2331 * which will cause other threads to clean up iclogs.
2333 * SYNCING - i/o completion will go through logs
2334 * DONE_SYNC - interrupt thread should be waiting for
2336 * IOERROR - give up hope all ye who enter here
2338 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
||
2339 iclog
->ic_state
== XLOG_STATE_SYNCING
||
2340 iclog
->ic_state
== XLOG_STATE_DONE_SYNC
||
2341 iclog
->ic_state
== XLOG_STATE_IOERROR
)
2343 iclog
= iclog
->ic_next
;
2344 } while (first_iclog
!= iclog
);
2348 if (log
->l_iclog
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_IOERROR
))
2350 spin_unlock(&log
->l_icloglock
);
2353 sv_broadcast(&log
->l_flush_wait
);
2358 * Finish transitioning this iclog to the dirty state.
2360 * Make sure that we completely execute this routine only when this is
2361 * the last call to the iclog. There is a good chance that iclog flushes,
2362 * when we reach the end of the physical log, get turned into 2 separate
2363 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2364 * routine. By using the reference count bwritecnt, we guarantee that only
2365 * the second completion goes through.
2367 * Callbacks could take time, so they are done outside the scope of the
2368 * global state machine log lock.
2371 xlog_state_done_syncing(
2372 xlog_in_core_t
*iclog
,
2375 xlog_t
*log
= iclog
->ic_log
;
2377 spin_lock(&log
->l_icloglock
);
2379 ASSERT(iclog
->ic_state
== XLOG_STATE_SYNCING
||
2380 iclog
->ic_state
== XLOG_STATE_IOERROR
);
2381 ASSERT(atomic_read(&iclog
->ic_refcnt
) == 0);
2382 ASSERT(iclog
->ic_bwritecnt
== 1 || iclog
->ic_bwritecnt
== 2);
2386 * If we got an error, either on the first buffer, or in the case of
2387 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2388 * and none should ever be attempted to be written to disk
2391 if (iclog
->ic_state
!= XLOG_STATE_IOERROR
) {
2392 if (--iclog
->ic_bwritecnt
== 1) {
2393 spin_unlock(&log
->l_icloglock
);
2396 iclog
->ic_state
= XLOG_STATE_DONE_SYNC
;
2400 * Someone could be sleeping prior to writing out the next
2401 * iclog buffer, we wake them all, one will get to do the
2402 * I/O, the others get to wait for the result.
2404 sv_broadcast(&iclog
->ic_write_wait
);
2405 spin_unlock(&log
->l_icloglock
);
2406 xlog_state_do_callback(log
, aborted
, iclog
); /* also cleans log */
2407 } /* xlog_state_done_syncing */
2411 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2412 * sleep. We wait on the flush queue on the head iclog as that should be
2413 * the first iclog to complete flushing. Hence if all iclogs are syncing,
2414 * we will wait here and all new writes will sleep until a sync completes.
2416 * The in-core logs are used in a circular fashion. They are not used
2417 * out-of-order even when an iclog past the head is free.
2420 * * log_offset where xlog_write() can start writing into the in-core
2422 * * in-core log pointer to which xlog_write() should write.
2423 * * boolean indicating this is a continued write to an in-core log.
2424 * If this is the last write, then the in-core log's offset field
2425 * needs to be incremented, depending on the amount of data which
2429 xlog_state_get_iclog_space(xlog_t
*log
,
2431 xlog_in_core_t
**iclogp
,
2432 xlog_ticket_t
*ticket
,
2433 int *continued_write
,
2437 xlog_rec_header_t
*head
;
2438 xlog_in_core_t
*iclog
;
2442 spin_lock(&log
->l_icloglock
);
2443 if (XLOG_FORCED_SHUTDOWN(log
)) {
2444 spin_unlock(&log
->l_icloglock
);
2445 return XFS_ERROR(EIO
);
2448 iclog
= log
->l_iclog
;
2449 if (iclog
->ic_state
!= XLOG_STATE_ACTIVE
) {
2450 XFS_STATS_INC(xs_log_noiclogs
);
2452 /* Wait for log writes to have flushed */
2453 sv_wait(&log
->l_flush_wait
, 0, &log
->l_icloglock
, 0);
2457 head
= &iclog
->ic_header
;
2459 atomic_inc(&iclog
->ic_refcnt
); /* prevents sync */
2460 log_offset
= iclog
->ic_offset
;
2462 /* On the 1st write to an iclog, figure out lsn. This works
2463 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2464 * committing to. If the offset is set, that's how many blocks
2467 if (log_offset
== 0) {
2468 ticket
->t_curr_res
-= log
->l_iclog_hsize
;
2469 xlog_tic_add_region(ticket
,
2471 XLOG_REG_TYPE_LRHEADER
);
2472 head
->h_cycle
= cpu_to_be32(log
->l_curr_cycle
);
2473 head
->h_lsn
= cpu_to_be64(
2474 xlog_assign_lsn(log
->l_curr_cycle
, log
->l_curr_block
));
2475 ASSERT(log
->l_curr_block
>= 0);
2478 /* If there is enough room to write everything, then do it. Otherwise,
2479 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2480 * bit is on, so this will get flushed out. Don't update ic_offset
2481 * until you know exactly how many bytes get copied. Therefore, wait
2482 * until later to update ic_offset.
2484 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2485 * can fit into remaining data section.
2487 if (iclog
->ic_size
- iclog
->ic_offset
< 2*sizeof(xlog_op_header_t
)) {
2488 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2491 * If I'm the only one writing to this iclog, sync it to disk.
2492 * We need to do an atomic compare and decrement here to avoid
2493 * racing with concurrent atomic_dec_and_lock() calls in
2494 * xlog_state_release_iclog() when there is more than one
2495 * reference to the iclog.
2497 if (!atomic_add_unless(&iclog
->ic_refcnt
, -1, 1)) {
2498 /* we are the only one */
2499 spin_unlock(&log
->l_icloglock
);
2500 error
= xlog_state_release_iclog(log
, iclog
);
2504 spin_unlock(&log
->l_icloglock
);
2509 /* Do we have enough room to write the full amount in the remainder
2510 * of this iclog? Or must we continue a write on the next iclog and
2511 * mark this iclog as completely taken? In the case where we switch
2512 * iclogs (to mark it taken), this particular iclog will release/sync
2513 * to disk in xlog_write().
2515 if (len
<= iclog
->ic_size
- iclog
->ic_offset
) {
2516 *continued_write
= 0;
2517 iclog
->ic_offset
+= len
;
2519 *continued_write
= 1;
2520 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2524 ASSERT(iclog
->ic_offset
<= iclog
->ic_size
);
2525 spin_unlock(&log
->l_icloglock
);
2527 *logoffsetp
= log_offset
;
2529 } /* xlog_state_get_iclog_space */
2532 * Atomically get the log space required for a log ticket.
2534 * Once a ticket gets put onto the reserveq, it will only return after
2535 * the needed reservation is satisfied.
2538 xlog_grant_log_space(xlog_t
*log
,
2549 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2550 panic("grant Recovery problem");
2553 /* Is there space or do we need to sleep? */
2554 spin_lock(&log
->l_grant_lock
);
2556 trace_xfs_log_grant_enter(log
, tic
);
2558 /* something is already sleeping; insert new transaction at end */
2559 if (log
->l_reserve_headq
) {
2560 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2562 trace_xfs_log_grant_sleep1(log
, tic
);
2565 * Gotta check this before going to sleep, while we're
2566 * holding the grant lock.
2568 if (XLOG_FORCED_SHUTDOWN(log
))
2571 XFS_STATS_INC(xs_sleep_logspace
);
2572 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2574 * If we got an error, and the filesystem is shutting down,
2575 * we'll catch it down below. So just continue...
2577 trace_xfs_log_grant_wake1(log
, tic
);
2578 spin_lock(&log
->l_grant_lock
);
2580 if (tic
->t_flags
& XFS_LOG_PERM_RESERV
)
2581 need_bytes
= tic
->t_unit_res
*tic
->t_ocnt
;
2583 need_bytes
= tic
->t_unit_res
;
2586 if (XLOG_FORCED_SHUTDOWN(log
))
2589 free_bytes
= xlog_space_left(log
, log
->l_grant_reserve_cycle
,
2590 log
->l_grant_reserve_bytes
);
2591 if (free_bytes
< need_bytes
) {
2592 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2593 xlog_ins_ticketq(&log
->l_reserve_headq
, tic
);
2595 trace_xfs_log_grant_sleep2(log
, tic
);
2597 spin_unlock(&log
->l_grant_lock
);
2598 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2599 spin_lock(&log
->l_grant_lock
);
2601 XFS_STATS_INC(xs_sleep_logspace
);
2602 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2604 spin_lock(&log
->l_grant_lock
);
2605 if (XLOG_FORCED_SHUTDOWN(log
))
2608 trace_xfs_log_grant_wake2(log
, tic
);
2611 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2612 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2614 /* we've got enough space */
2615 xlog_grant_add_space(log
, need_bytes
);
2617 tail_lsn
= log
->l_tail_lsn
;
2619 * Check to make sure the grant write head didn't just over lap the
2620 * tail. If the cycles are the same, we can't be overlapping.
2621 * Otherwise, make sure that the cycles differ by exactly one and
2622 * check the byte count.
2624 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2625 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2626 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2629 trace_xfs_log_grant_exit(log
, tic
);
2630 xlog_verify_grant_head(log
, 1);
2631 spin_unlock(&log
->l_grant_lock
);
2635 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2636 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2638 trace_xfs_log_grant_error(log
, tic
);
2641 * If we are failing, make sure the ticket doesn't have any
2642 * current reservations. We don't want to add this back when
2643 * the ticket/transaction gets cancelled.
2645 tic
->t_curr_res
= 0;
2646 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2647 spin_unlock(&log
->l_grant_lock
);
2648 return XFS_ERROR(EIO
);
2649 } /* xlog_grant_log_space */
2653 * Replenish the byte reservation required by moving the grant write head.
2658 xlog_regrant_write_log_space(xlog_t
*log
,
2661 int free_bytes
, need_bytes
;
2662 xlog_ticket_t
*ntic
;
2667 tic
->t_curr_res
= tic
->t_unit_res
;
2668 xlog_tic_reset_res(tic
);
2674 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2675 panic("regrant Recovery problem");
2678 spin_lock(&log
->l_grant_lock
);
2680 trace_xfs_log_regrant_write_enter(log
, tic
);
2682 if (XLOG_FORCED_SHUTDOWN(log
))
2685 /* If there are other waiters on the queue then give them a
2686 * chance at logspace before us. Wake up the first waiters,
2687 * if we do not wake up all the waiters then go to sleep waiting
2688 * for more free space, otherwise try to get some space for
2691 need_bytes
= tic
->t_unit_res
;
2692 if ((ntic
= log
->l_write_headq
)) {
2693 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2694 log
->l_grant_write_bytes
);
2696 ASSERT(ntic
->t_flags
& XLOG_TIC_PERM_RESERV
);
2698 if (free_bytes
< ntic
->t_unit_res
)
2700 free_bytes
-= ntic
->t_unit_res
;
2701 sv_signal(&ntic
->t_wait
);
2702 ntic
= ntic
->t_next
;
2703 } while (ntic
!= log
->l_write_headq
);
2705 if (ntic
!= log
->l_write_headq
) {
2706 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2707 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2709 trace_xfs_log_regrant_write_sleep1(log
, tic
);
2711 spin_unlock(&log
->l_grant_lock
);
2712 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2713 spin_lock(&log
->l_grant_lock
);
2715 XFS_STATS_INC(xs_sleep_logspace
);
2716 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
,
2717 &log
->l_grant_lock
, s
);
2719 /* If we're shutting down, this tic is already
2721 spin_lock(&log
->l_grant_lock
);
2722 if (XLOG_FORCED_SHUTDOWN(log
))
2725 trace_xfs_log_regrant_write_wake1(log
, tic
);
2730 if (XLOG_FORCED_SHUTDOWN(log
))
2733 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2734 log
->l_grant_write_bytes
);
2735 if (free_bytes
< need_bytes
) {
2736 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2737 xlog_ins_ticketq(&log
->l_write_headq
, tic
);
2738 spin_unlock(&log
->l_grant_lock
);
2739 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2740 spin_lock(&log
->l_grant_lock
);
2742 XFS_STATS_INC(xs_sleep_logspace
);
2743 trace_xfs_log_regrant_write_sleep2(log
, tic
);
2745 sv_wait(&tic
->t_wait
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2747 /* If we're shutting down, this tic is already off the queue */
2748 spin_lock(&log
->l_grant_lock
);
2749 if (XLOG_FORCED_SHUTDOWN(log
))
2752 trace_xfs_log_regrant_write_wake2(log
, tic
);
2754 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2755 xlog_del_ticketq(&log
->l_write_headq
, tic
);
2757 /* we've got enough space */
2758 xlog_grant_add_space_write(log
, need_bytes
);
2760 tail_lsn
= log
->l_tail_lsn
;
2761 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2762 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2763 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2767 trace_xfs_log_regrant_write_exit(log
, tic
);
2769 xlog_verify_grant_head(log
, 1);
2770 spin_unlock(&log
->l_grant_lock
);
2775 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2776 xlog_del_ticketq(&log
->l_reserve_headq
, tic
);
2778 trace_xfs_log_regrant_write_error(log
, tic
);
2781 * If we are failing, make sure the ticket doesn't have any
2782 * current reservations. We don't want to add this back when
2783 * the ticket/transaction gets cancelled.
2785 tic
->t_curr_res
= 0;
2786 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2787 spin_unlock(&log
->l_grant_lock
);
2788 return XFS_ERROR(EIO
);
2789 } /* xlog_regrant_write_log_space */
2792 /* The first cnt-1 times through here we don't need to
2793 * move the grant write head because the permanent
2794 * reservation has reserved cnt times the unit amount.
2795 * Release part of current permanent unit reservation and
2796 * reset current reservation to be one units worth. Also
2797 * move grant reservation head forward.
2800 xlog_regrant_reserve_log_space(xlog_t
*log
,
2801 xlog_ticket_t
*ticket
)
2803 trace_xfs_log_regrant_reserve_enter(log
, ticket
);
2805 if (ticket
->t_cnt
> 0)
2808 spin_lock(&log
->l_grant_lock
);
2809 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2810 ticket
->t_curr_res
= ticket
->t_unit_res
;
2811 xlog_tic_reset_res(ticket
);
2813 trace_xfs_log_regrant_reserve_sub(log
, ticket
);
2815 xlog_verify_grant_head(log
, 1);
2817 /* just return if we still have some of the pre-reserved space */
2818 if (ticket
->t_cnt
> 0) {
2819 spin_unlock(&log
->l_grant_lock
);
2823 xlog_grant_add_space_reserve(log
, ticket
->t_unit_res
);
2825 trace_xfs_log_regrant_reserve_exit(log
, ticket
);
2827 xlog_verify_grant_head(log
, 0);
2828 spin_unlock(&log
->l_grant_lock
);
2829 ticket
->t_curr_res
= ticket
->t_unit_res
;
2830 xlog_tic_reset_res(ticket
);
2831 } /* xlog_regrant_reserve_log_space */
2835 * Give back the space left from a reservation.
2837 * All the information we need to make a correct determination of space left
2838 * is present. For non-permanent reservations, things are quite easy. The
2839 * count should have been decremented to zero. We only need to deal with the
2840 * space remaining in the current reservation part of the ticket. If the
2841 * ticket contains a permanent reservation, there may be left over space which
2842 * needs to be released. A count of N means that N-1 refills of the current
2843 * reservation can be done before we need to ask for more space. The first
2844 * one goes to fill up the first current reservation. Once we run out of
2845 * space, the count will stay at zero and the only space remaining will be
2846 * in the current reservation field.
2849 xlog_ungrant_log_space(xlog_t
*log
,
2850 xlog_ticket_t
*ticket
)
2852 if (ticket
->t_cnt
> 0)
2855 spin_lock(&log
->l_grant_lock
);
2856 trace_xfs_log_ungrant_enter(log
, ticket
);
2858 xlog_grant_sub_space(log
, ticket
->t_curr_res
);
2860 trace_xfs_log_ungrant_sub(log
, ticket
);
2862 /* If this is a permanent reservation ticket, we may be able to free
2863 * up more space based on the remaining count.
2865 if (ticket
->t_cnt
> 0) {
2866 ASSERT(ticket
->t_flags
& XLOG_TIC_PERM_RESERV
);
2867 xlog_grant_sub_space(log
, ticket
->t_unit_res
*ticket
->t_cnt
);
2870 trace_xfs_log_ungrant_exit(log
, ticket
);
2872 xlog_verify_grant_head(log
, 1);
2873 spin_unlock(&log
->l_grant_lock
);
2874 xfs_log_move_tail(log
->l_mp
, 1);
2875 } /* xlog_ungrant_log_space */
2879 * Flush iclog to disk if this is the last reference to the given iclog and
2880 * the WANT_SYNC bit is set.
2882 * When this function is entered, the iclog is not necessarily in the
2883 * WANT_SYNC state. It may be sitting around waiting to get filled.
2888 xlog_state_release_iclog(
2890 xlog_in_core_t
*iclog
)
2892 int sync
= 0; /* do we sync? */
2894 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
2895 return XFS_ERROR(EIO
);
2897 ASSERT(atomic_read(&iclog
->ic_refcnt
) > 0);
2898 if (!atomic_dec_and_lock(&iclog
->ic_refcnt
, &log
->l_icloglock
))
2901 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2902 spin_unlock(&log
->l_icloglock
);
2903 return XFS_ERROR(EIO
);
2905 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2906 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
);
2908 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
) {
2909 /* update tail before writing to iclog */
2910 xlog_assign_tail_lsn(log
->l_mp
);
2912 iclog
->ic_state
= XLOG_STATE_SYNCING
;
2913 iclog
->ic_header
.h_tail_lsn
= cpu_to_be64(log
->l_tail_lsn
);
2914 xlog_verify_tail_lsn(log
, iclog
, log
->l_tail_lsn
);
2915 /* cycle incremented when incrementing curr_block */
2917 spin_unlock(&log
->l_icloglock
);
2920 * We let the log lock go, so it's possible that we hit a log I/O
2921 * error or some other SHUTDOWN condition that marks the iclog
2922 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2923 * this iclog has consistent data, so we ignore IOERROR
2924 * flags after this point.
2927 return xlog_sync(log
, iclog
);
2929 } /* xlog_state_release_iclog */
2933 * This routine will mark the current iclog in the ring as WANT_SYNC
2934 * and move the current iclog pointer to the next iclog in the ring.
2935 * When this routine is called from xlog_state_get_iclog_space(), the
2936 * exact size of the iclog has not yet been determined. All we know is
2937 * that every data block. We have run out of space in this log record.
2940 xlog_state_switch_iclogs(xlog_t
*log
,
2941 xlog_in_core_t
*iclog
,
2944 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2946 eventual_size
= iclog
->ic_offset
;
2947 iclog
->ic_state
= XLOG_STATE_WANT_SYNC
;
2948 iclog
->ic_header
.h_prev_block
= cpu_to_be32(log
->l_prev_block
);
2949 log
->l_prev_block
= log
->l_curr_block
;
2950 log
->l_prev_cycle
= log
->l_curr_cycle
;
2952 /* roll log?: ic_offset changed later */
2953 log
->l_curr_block
+= BTOBB(eventual_size
)+BTOBB(log
->l_iclog_hsize
);
2955 /* Round up to next log-sunit */
2956 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
2957 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
2958 __uint32_t sunit_bb
= BTOBB(log
->l_mp
->m_sb
.sb_logsunit
);
2959 log
->l_curr_block
= roundup(log
->l_curr_block
, sunit_bb
);
2962 if (log
->l_curr_block
>= log
->l_logBBsize
) {
2963 log
->l_curr_cycle
++;
2964 if (log
->l_curr_cycle
== XLOG_HEADER_MAGIC_NUM
)
2965 log
->l_curr_cycle
++;
2966 log
->l_curr_block
-= log
->l_logBBsize
;
2967 ASSERT(log
->l_curr_block
>= 0);
2969 ASSERT(iclog
== log
->l_iclog
);
2970 log
->l_iclog
= iclog
->ic_next
;
2971 } /* xlog_state_switch_iclogs */
2974 * Write out all data in the in-core log as of this exact moment in time.
2976 * Data may be written to the in-core log during this call. However,
2977 * we don't guarantee this data will be written out. A change from past
2978 * implementation means this routine will *not* write out zero length LRs.
2980 * Basically, we try and perform an intelligent scan of the in-core logs.
2981 * If we determine there is no flushable data, we just return. There is no
2982 * flushable data if:
2984 * 1. the current iclog is active and has no data; the previous iclog
2985 * is in the active or dirty state.
2986 * 2. the current iclog is drity, and the previous iclog is in the
2987 * active or dirty state.
2991 * 1. the current iclog is not in the active nor dirty state.
2992 * 2. the current iclog dirty, and the previous iclog is not in the
2993 * active nor dirty state.
2994 * 3. the current iclog is active, and there is another thread writing
2995 * to this particular iclog.
2996 * 4. a) the current iclog is active and has no other writers
2997 * b) when we return from flushing out this iclog, it is still
2998 * not in the active nor dirty state.
3002 struct xfs_mount
*mp
,
3006 struct log
*log
= mp
->m_log
;
3007 struct xlog_in_core
*iclog
;
3010 XFS_STATS_INC(xs_log_force
);
3012 spin_lock(&log
->l_icloglock
);
3014 iclog
= log
->l_iclog
;
3015 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3016 spin_unlock(&log
->l_icloglock
);
3017 return XFS_ERROR(EIO
);
3020 /* If the head iclog is not active nor dirty, we just attach
3021 * ourselves to the head and go to sleep.
3023 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
3024 iclog
->ic_state
== XLOG_STATE_DIRTY
) {
3026 * If the head is dirty or (active and empty), then
3027 * we need to look at the previous iclog. If the previous
3028 * iclog is active or dirty we are done. There is nothing
3029 * to sync out. Otherwise, we attach ourselves to the
3030 * previous iclog and go to sleep.
3032 if (iclog
->ic_state
== XLOG_STATE_DIRTY
||
3033 (atomic_read(&iclog
->ic_refcnt
) == 0
3034 && iclog
->ic_offset
== 0)) {
3035 iclog
= iclog
->ic_prev
;
3036 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
3037 iclog
->ic_state
== XLOG_STATE_DIRTY
)
3042 if (atomic_read(&iclog
->ic_refcnt
) == 0) {
3043 /* We are the only one with access to this
3044 * iclog. Flush it out now. There should
3045 * be a roundoff of zero to show that someone
3046 * has already taken care of the roundoff from
3047 * the previous sync.
3049 atomic_inc(&iclog
->ic_refcnt
);
3050 lsn
= be64_to_cpu(iclog
->ic_header
.h_lsn
);
3051 xlog_state_switch_iclogs(log
, iclog
, 0);
3052 spin_unlock(&log
->l_icloglock
);
3054 if (xlog_state_release_iclog(log
, iclog
))
3055 return XFS_ERROR(EIO
);
3059 spin_lock(&log
->l_icloglock
);
3060 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) == lsn
&&
3061 iclog
->ic_state
!= XLOG_STATE_DIRTY
)
3066 /* Someone else is writing to this iclog.
3067 * Use its call to flush out the data. However,
3068 * the other thread may not force out this LR,
3069 * so we mark it WANT_SYNC.
3071 xlog_state_switch_iclogs(log
, iclog
, 0);
3077 /* By the time we come around again, the iclog could've been filled
3078 * which would give it another lsn. If we have a new lsn, just
3079 * return because the relevant data has been flushed.
3082 if (flags
& XFS_LOG_SYNC
) {
3084 * We must check if we're shutting down here, before
3085 * we wait, while we're holding the l_icloglock.
3086 * Then we check again after waking up, in case our
3087 * sleep was disturbed by a bad news.
3089 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3090 spin_unlock(&log
->l_icloglock
);
3091 return XFS_ERROR(EIO
);
3093 XFS_STATS_INC(xs_log_force_sleep
);
3094 sv_wait(&iclog
->ic_force_wait
, PINOD
, &log
->l_icloglock
, s
);
3096 * No need to grab the log lock here since we're
3097 * only deciding whether or not to return EIO
3098 * and the memory read should be atomic.
3100 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3101 return XFS_ERROR(EIO
);
3107 spin_unlock(&log
->l_icloglock
);
3113 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
3114 * about errors or whether the log was flushed or not. This is the normal
3115 * interface to use when trying to unpin items or move the log forward.
3124 error
= _xfs_log_force(mp
, flags
, NULL
);
3126 xfs_fs_cmn_err(CE_WARN
, mp
, "xfs_log_force: "
3127 "error %d returned.", error
);
3132 * Force the in-core log to disk for a specific LSN.
3134 * Find in-core log with lsn.
3135 * If it is in the DIRTY state, just return.
3136 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3137 * state and go to sleep or return.
3138 * If it is in any other state, go to sleep or return.
3140 * Synchronous forces are implemented with a signal variable. All callers
3141 * to force a given lsn to disk will wait on a the sv attached to the
3142 * specific in-core log. When given in-core log finally completes its
3143 * write to disk, that thread will wake up all threads waiting on the
3148 struct xfs_mount
*mp
,
3153 struct log
*log
= mp
->m_log
;
3154 struct xlog_in_core
*iclog
;
3155 int already_slept
= 0;
3159 XFS_STATS_INC(xs_log_force
);
3162 spin_lock(&log
->l_icloglock
);
3163 iclog
= log
->l_iclog
;
3164 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3165 spin_unlock(&log
->l_icloglock
);
3166 return XFS_ERROR(EIO
);
3170 if (be64_to_cpu(iclog
->ic_header
.h_lsn
) != lsn
) {
3171 iclog
= iclog
->ic_next
;
3175 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
3176 spin_unlock(&log
->l_icloglock
);
3180 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3182 * We sleep here if we haven't already slept (e.g.
3183 * this is the first time we've looked at the correct
3184 * iclog buf) and the buffer before us is going to
3185 * be sync'ed. The reason for this is that if we
3186 * are doing sync transactions here, by waiting for
3187 * the previous I/O to complete, we can allow a few
3188 * more transactions into this iclog before we close
3191 * Otherwise, we mark the buffer WANT_SYNC, and bump
3192 * up the refcnt so we can release the log (which
3193 * drops the ref count). The state switch keeps new
3194 * transaction commits from using this buffer. When
3195 * the current commits finish writing into the buffer,
3196 * the refcount will drop to zero and the buffer will
3199 if (!already_slept
&&
3200 (iclog
->ic_prev
->ic_state
&
3201 (XLOG_STATE_WANT_SYNC
| XLOG_STATE_SYNCING
))) {
3202 ASSERT(!(iclog
->ic_state
& XLOG_STATE_IOERROR
));
3204 XFS_STATS_INC(xs_log_force_sleep
);
3206 sv_wait(&iclog
->ic_prev
->ic_write_wait
,
3207 PSWP
, &log
->l_icloglock
, s
);
3213 atomic_inc(&iclog
->ic_refcnt
);
3214 xlog_state_switch_iclogs(log
, iclog
, 0);
3215 spin_unlock(&log
->l_icloglock
);
3216 if (xlog_state_release_iclog(log
, iclog
))
3217 return XFS_ERROR(EIO
);
3220 spin_lock(&log
->l_icloglock
);
3223 if ((flags
& XFS_LOG_SYNC
) && /* sleep */
3225 (XLOG_STATE_ACTIVE
| XLOG_STATE_DIRTY
))) {
3227 * Don't wait on completion if we know that we've
3228 * gotten a log write error.
3230 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3231 spin_unlock(&log
->l_icloglock
);
3232 return XFS_ERROR(EIO
);
3234 XFS_STATS_INC(xs_log_force_sleep
);
3235 sv_wait(&iclog
->ic_force_wait
, PSWP
, &log
->l_icloglock
, s
);
3237 * No need to grab the log lock here since we're
3238 * only deciding whether or not to return EIO
3239 * and the memory read should be atomic.
3241 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3242 return XFS_ERROR(EIO
);
3246 } else { /* just return */
3247 spin_unlock(&log
->l_icloglock
);
3251 } while (iclog
!= log
->l_iclog
);
3253 spin_unlock(&log
->l_icloglock
);
3258 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
3259 * about errors or whether the log was flushed or not. This is the normal
3260 * interface to use when trying to unpin items or move the log forward.
3270 error
= _xfs_log_force_lsn(mp
, lsn
, flags
, NULL
);
3272 xfs_fs_cmn_err(CE_WARN
, mp
, "xfs_log_force: "
3273 "error %d returned.", error
);
3278 * Called when we want to mark the current iclog as being ready to sync to
3282 xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
)
3284 assert_spin_locked(&log
->l_icloglock
);
3286 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3287 xlog_state_switch_iclogs(log
, iclog
, 0);
3289 ASSERT(iclog
->ic_state
&
3290 (XLOG_STATE_WANT_SYNC
|XLOG_STATE_IOERROR
));
3295 /*****************************************************************************
3299 *****************************************************************************
3303 * Free a used ticket when its refcount falls to zero.
3307 xlog_ticket_t
*ticket
)
3309 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3310 if (atomic_dec_and_test(&ticket
->t_ref
)) {
3311 sv_destroy(&ticket
->t_wait
);
3312 kmem_zone_free(xfs_log_ticket_zone
, ticket
);
3318 xlog_ticket_t
*ticket
)
3320 ASSERT(atomic_read(&ticket
->t_ref
) > 0);
3321 atomic_inc(&ticket
->t_ref
);
3326 * Allocate and initialise a new log ticket.
3328 STATIC xlog_ticket_t
*
3336 struct xlog_ticket
*tic
;
3340 tic
= kmem_zone_zalloc(xfs_log_ticket_zone
, KM_SLEEP
|KM_MAYFAIL
);
3345 * Permanent reservations have up to 'cnt'-1 active log operations
3346 * in the log. A unit in this case is the amount of space for one
3347 * of these log operations. Normal reservations have a cnt of 1
3348 * and their unit amount is the total amount of space required.
3350 * The following lines of code account for non-transaction data
3351 * which occupy space in the on-disk log.
3353 * Normal form of a transaction is:
3354 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3355 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3357 * We need to account for all the leadup data and trailer data
3358 * around the transaction data.
3359 * And then we need to account for the worst case in terms of using
3361 * The worst case will happen if:
3362 * - the placement of the transaction happens to be such that the
3363 * roundoff is at its maximum
3364 * - the transaction data is synced before the commit record is synced
3365 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3366 * Therefore the commit record is in its own Log Record.
3367 * This can happen as the commit record is called with its
3368 * own region to xlog_write().
3369 * This then means that in the worst case, roundoff can happen for
3370 * the commit-rec as well.
3371 * The commit-rec is smaller than padding in this scenario and so it is
3372 * not added separately.
3375 /* for trans header */
3376 unit_bytes
+= sizeof(xlog_op_header_t
);
3377 unit_bytes
+= sizeof(xfs_trans_header_t
);
3380 unit_bytes
+= sizeof(xlog_op_header_t
);
3383 * for LR headers - the space for data in an iclog is the size minus
3384 * the space used for the headers. If we use the iclog size, then we
3385 * undercalculate the number of headers required.
3387 * Furthermore - the addition of op headers for split-recs might
3388 * increase the space required enough to require more log and op
3389 * headers, so take that into account too.
3391 * IMPORTANT: This reservation makes the assumption that if this
3392 * transaction is the first in an iclog and hence has the LR headers
3393 * accounted to it, then the remaining space in the iclog is
3394 * exclusively for this transaction. i.e. if the transaction is larger
3395 * than the iclog, it will be the only thing in that iclog.
3396 * Fundamentally, this means we must pass the entire log vector to
3397 * xlog_write to guarantee this.
3399 iclog_space
= log
->l_iclog_size
- log
->l_iclog_hsize
;
3400 num_headers
= howmany(unit_bytes
, iclog_space
);
3402 /* for split-recs - ophdrs added when data split over LRs */
3403 unit_bytes
+= sizeof(xlog_op_header_t
) * num_headers
;
3405 /* add extra header reservations if we overrun */
3406 while (!num_headers
||
3407 howmany(unit_bytes
, iclog_space
) > num_headers
) {
3408 unit_bytes
+= sizeof(xlog_op_header_t
);
3411 unit_bytes
+= log
->l_iclog_hsize
* num_headers
;
3413 /* for commit-rec LR header - note: padding will subsume the ophdr */
3414 unit_bytes
+= log
->l_iclog_hsize
;
3416 /* for roundoff padding for transaction data and one for commit record */
3417 if (xfs_sb_version_haslogv2(&log
->l_mp
->m_sb
) &&
3418 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
3419 /* log su roundoff */
3420 unit_bytes
+= 2*log
->l_mp
->m_sb
.sb_logsunit
;
3423 unit_bytes
+= 2*BBSIZE
;
3426 atomic_set(&tic
->t_ref
, 1);
3427 tic
->t_unit_res
= unit_bytes
;
3428 tic
->t_curr_res
= unit_bytes
;
3431 tic
->t_tid
= random32();
3432 tic
->t_clientid
= client
;
3433 tic
->t_flags
= XLOG_TIC_INITED
;
3434 tic
->t_trans_type
= 0;
3435 if (xflags
& XFS_LOG_PERM_RESERV
)
3436 tic
->t_flags
|= XLOG_TIC_PERM_RESERV
;
3437 sv_init(&tic
->t_wait
, SV_DEFAULT
, "logtick");
3439 xlog_tic_reset_res(tic
);
3445 /******************************************************************************
3447 * Log debug routines
3449 ******************************************************************************
3453 * Make sure that the destination ptr is within the valid data region of
3454 * one of the iclogs. This uses backup pointers stored in a different
3455 * part of the log in case we trash the log structure.
3458 xlog_verify_dest_ptr(
3465 for (i
= 0; i
< log
->l_iclog_bufs
; i
++) {
3466 if (ptr
>= log
->l_iclog_bak
[i
] &&
3467 ptr
<= log
->l_iclog_bak
[i
] + log
->l_iclog_size
)
3472 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3476 xlog_verify_grant_head(xlog_t
*log
, int equals
)
3478 if (log
->l_grant_reserve_cycle
== log
->l_grant_write_cycle
) {
3480 ASSERT(log
->l_grant_reserve_bytes
>= log
->l_grant_write_bytes
);
3482 ASSERT(log
->l_grant_reserve_bytes
> log
->l_grant_write_bytes
);
3484 ASSERT(log
->l_grant_reserve_cycle
-1 == log
->l_grant_write_cycle
);
3485 ASSERT(log
->l_grant_write_bytes
>= log
->l_grant_reserve_bytes
);
3487 } /* xlog_verify_grant_head */
3489 /* check if it will fit */
3491 xlog_verify_tail_lsn(xlog_t
*log
,
3492 xlog_in_core_t
*iclog
,
3497 if (CYCLE_LSN(tail_lsn
) == log
->l_prev_cycle
) {
3499 log
->l_logBBsize
- (log
->l_prev_block
- BLOCK_LSN(tail_lsn
));
3500 if (blocks
< BTOBB(iclog
->ic_offset
)+BTOBB(log
->l_iclog_hsize
))
3501 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3503 ASSERT(CYCLE_LSN(tail_lsn
)+1 == log
->l_prev_cycle
);
3505 if (BLOCK_LSN(tail_lsn
) == log
->l_prev_block
)
3506 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3508 blocks
= BLOCK_LSN(tail_lsn
) - log
->l_prev_block
;
3509 if (blocks
< BTOBB(iclog
->ic_offset
) + 1)
3510 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3512 } /* xlog_verify_tail_lsn */
3515 * Perform a number of checks on the iclog before writing to disk.
3517 * 1. Make sure the iclogs are still circular
3518 * 2. Make sure we have a good magic number
3519 * 3. Make sure we don't have magic numbers in the data
3520 * 4. Check fields of each log operation header for:
3521 * A. Valid client identifier
3522 * B. tid ptr value falls in valid ptr space (user space code)
3523 * C. Length in log record header is correct according to the
3524 * individual operation headers within record.
3525 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3526 * log, check the preceding blocks of the physical log to make sure all
3527 * the cycle numbers agree with the current cycle number.
3530 xlog_verify_iclog(xlog_t
*log
,
3531 xlog_in_core_t
*iclog
,
3535 xlog_op_header_t
*ophead
;
3536 xlog_in_core_t
*icptr
;
3537 xlog_in_core_2_t
*xhdr
;
3539 xfs_caddr_t base_ptr
;
3540 __psint_t field_offset
;
3542 int len
, i
, j
, k
, op_len
;
3545 /* check validity of iclog pointers */
3546 spin_lock(&log
->l_icloglock
);
3547 icptr
= log
->l_iclog
;
3548 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3550 xlog_panic("xlog_verify_iclog: invalid ptr");
3551 icptr
= icptr
->ic_next
;
3553 if (icptr
!= log
->l_iclog
)
3554 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
3555 spin_unlock(&log
->l_icloglock
);
3557 /* check log magic numbers */
3558 if (be32_to_cpu(iclog
->ic_header
.h_magicno
) != XLOG_HEADER_MAGIC_NUM
)
3559 xlog_panic("xlog_verify_iclog: invalid magic num");
3561 ptr
= (xfs_caddr_t
) &iclog
->ic_header
;
3562 for (ptr
+= BBSIZE
; ptr
< ((xfs_caddr_t
)&iclog
->ic_header
) + count
;
3564 if (be32_to_cpu(*(__be32
*)ptr
) == XLOG_HEADER_MAGIC_NUM
)
3565 xlog_panic("xlog_verify_iclog: unexpected magic num");
3569 len
= be32_to_cpu(iclog
->ic_header
.h_num_logops
);
3570 ptr
= iclog
->ic_datap
;
3572 ophead
= (xlog_op_header_t
*)ptr
;
3573 xhdr
= iclog
->ic_data
;
3574 for (i
= 0; i
< len
; i
++) {
3575 ophead
= (xlog_op_header_t
*)ptr
;
3577 /* clientid is only 1 byte */
3578 field_offset
= (__psint_t
)
3579 ((xfs_caddr_t
)&(ophead
->oh_clientid
) - base_ptr
);
3580 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3581 clientid
= ophead
->oh_clientid
;
3583 idx
= BTOBBT((xfs_caddr_t
)&(ophead
->oh_clientid
) - iclog
->ic_datap
);
3584 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3585 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3586 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3587 clientid
= xlog_get_client_id(
3588 xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3590 clientid
= xlog_get_client_id(
3591 iclog
->ic_header
.h_cycle_data
[idx
]);
3594 if (clientid
!= XFS_TRANSACTION
&& clientid
!= XFS_LOG
)
3595 cmn_err(CE_WARN
, "xlog_verify_iclog: "
3596 "invalid clientid %d op 0x%p offset 0x%lx",
3597 clientid
, ophead
, (unsigned long)field_offset
);
3600 field_offset
= (__psint_t
)
3601 ((xfs_caddr_t
)&(ophead
->oh_len
) - base_ptr
);
3602 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3603 op_len
= be32_to_cpu(ophead
->oh_len
);
3605 idx
= BTOBBT((__psint_t
)&ophead
->oh_len
-
3606 (__psint_t
)iclog
->ic_datap
);
3607 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3608 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3609 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3610 op_len
= be32_to_cpu(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
]);
3612 op_len
= be32_to_cpu(iclog
->ic_header
.h_cycle_data
[idx
]);
3615 ptr
+= sizeof(xlog_op_header_t
) + op_len
;
3617 } /* xlog_verify_iclog */
3621 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3627 xlog_in_core_t
*iclog
, *ic
;
3629 iclog
= log
->l_iclog
;
3630 if (! (iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3632 * Mark all the incore logs IOERROR.
3633 * From now on, no log flushes will result.
3637 ic
->ic_state
= XLOG_STATE_IOERROR
;
3639 } while (ic
!= iclog
);
3643 * Return non-zero, if state transition has already happened.
3649 * This is called from xfs_force_shutdown, when we're forcibly
3650 * shutting down the filesystem, typically because of an IO error.
3651 * Our main objectives here are to make sure that:
3652 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3653 * parties to find out, 'atomically'.
3654 * b. those who're sleeping on log reservations, pinned objects and
3655 * other resources get woken up, and be told the bad news.
3656 * c. nothing new gets queued up after (a) and (b) are done.
3657 * d. if !logerror, flush the iclogs to disk, then seal them off
3661 xfs_log_force_umount(
3662 struct xfs_mount
*mp
,
3672 * If this happens during log recovery, don't worry about
3673 * locking; the log isn't open for business yet.
3676 log
->l_flags
& XLOG_ACTIVE_RECOVERY
) {
3677 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3679 XFS_BUF_DONE(mp
->m_sb_bp
);
3684 * Somebody could've already done the hard work for us.
3685 * No need to get locks for this.
3687 if (logerror
&& log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3688 ASSERT(XLOG_FORCED_SHUTDOWN(log
));
3693 * We must hold both the GRANT lock and the LOG lock,
3694 * before we mark the filesystem SHUTDOWN and wake
3695 * everybody up to tell the bad news.
3697 spin_lock(&log
->l_icloglock
);
3698 spin_lock(&log
->l_grant_lock
);
3699 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3701 XFS_BUF_DONE(mp
->m_sb_bp
);
3704 * This flag is sort of redundant because of the mount flag, but
3705 * it's good to maintain the separation between the log and the rest
3708 log
->l_flags
|= XLOG_IO_ERROR
;
3711 * If we hit a log error, we want to mark all the iclogs IOERROR
3712 * while we're still holding the loglock.
3715 retval
= xlog_state_ioerror(log
);
3716 spin_unlock(&log
->l_icloglock
);
3719 * We don't want anybody waiting for log reservations
3720 * after this. That means we have to wake up everybody
3721 * queued up on reserve_headq as well as write_headq.
3722 * In addition, we make sure in xlog_{re}grant_log_space
3723 * that we don't enqueue anything once the SHUTDOWN flag
3724 * is set, and this action is protected by the GRANTLOCK.
3726 if ((tic
= log
->l_reserve_headq
)) {
3728 sv_signal(&tic
->t_wait
);
3730 } while (tic
!= log
->l_reserve_headq
);
3733 if ((tic
= log
->l_write_headq
)) {
3735 sv_signal(&tic
->t_wait
);
3737 } while (tic
!= log
->l_write_headq
);
3739 spin_unlock(&log
->l_grant_lock
);
3741 if (!(log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3744 * Force the incore logs to disk before shutting the
3745 * log down completely.
3747 _xfs_log_force(mp
, XFS_LOG_SYNC
, NULL
);
3749 spin_lock(&log
->l_icloglock
);
3750 retval
= xlog_state_ioerror(log
);
3751 spin_unlock(&log
->l_icloglock
);
3754 * Wake up everybody waiting on xfs_log_force.
3755 * Callback all log item committed functions as if the
3756 * log writes were completed.
3758 xlog_state_do_callback(log
, XFS_LI_ABORTED
, NULL
);
3760 #ifdef XFSERRORDEBUG
3762 xlog_in_core_t
*iclog
;
3764 spin_lock(&log
->l_icloglock
);
3765 iclog
= log
->l_iclog
;
3767 ASSERT(iclog
->ic_callback
== 0);
3768 iclog
= iclog
->ic_next
;
3769 } while (iclog
!= log
->l_iclog
);
3770 spin_unlock(&log
->l_icloglock
);
3773 /* return non-zero if log IOERROR transition had already happened */
3778 xlog_iclogs_empty(xlog_t
*log
)
3780 xlog_in_core_t
*iclog
;
3782 iclog
= log
->l_iclog
;
3784 /* endianness does not matter here, zero is zero in
3787 if (iclog
->ic_header
.h_num_logops
)
3789 iclog
= iclog
->ic_next
;
3790 } while (iclog
!= log
->l_iclog
);