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"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_error.h"
32 #include "xfs_log_priv.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_bmap_btree.h"
35 #include "xfs_alloc_btree.h"
36 #include "xfs_ialloc_btree.h"
37 #include "xfs_log_recover.h"
38 #include "xfs_trans_priv.h"
39 #include "xfs_dir_sf.h"
40 #include "xfs_dir2_sf.h"
41 #include "xfs_attr_sf.h"
42 #include "xfs_dinode.h"
43 #include "xfs_inode.h"
47 #define xlog_write_adv_cnt(ptr, len, off, bytes) \
52 /* Local miscellaneous function prototypes */
53 STATIC
int xlog_bdstrat_cb(struct xfs_buf
*);
54 STATIC
int xlog_commit_record(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
,
55 xlog_in_core_t
**, xfs_lsn_t
*);
56 STATIC xlog_t
* xlog_alloc_log(xfs_mount_t
*mp
,
57 xfs_buftarg_t
*log_target
,
58 xfs_daddr_t blk_offset
,
60 STATIC
int xlog_space_left(xlog_t
*log
, int cycle
, int bytes
);
61 STATIC
int xlog_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
62 STATIC
void xlog_unalloc_log(xlog_t
*log
);
63 STATIC
int xlog_write(xfs_mount_t
*mp
, xfs_log_iovec_t region
[],
64 int nentries
, xfs_log_ticket_t tic
,
66 xlog_in_core_t
**commit_iclog
,
69 /* local state machine functions */
70 STATIC
void xlog_state_done_syncing(xlog_in_core_t
*iclog
, int);
71 STATIC
void xlog_state_do_callback(xlog_t
*log
,int aborted
, xlog_in_core_t
*iclog
);
72 STATIC
int xlog_state_get_iclog_space(xlog_t
*log
,
74 xlog_in_core_t
**iclog
,
75 xlog_ticket_t
*ticket
,
78 STATIC
void xlog_state_put_ticket(xlog_t
*log
,
80 STATIC
int xlog_state_release_iclog(xlog_t
*log
,
81 xlog_in_core_t
*iclog
);
82 STATIC
void xlog_state_switch_iclogs(xlog_t
*log
,
83 xlog_in_core_t
*iclog
,
85 STATIC
int xlog_state_sync(xlog_t
*log
,
89 STATIC
int xlog_state_sync_all(xlog_t
*log
, uint flags
, int *log_flushed
);
90 STATIC
void xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
);
92 /* local functions to manipulate grant head */
93 STATIC
int xlog_grant_log_space(xlog_t
*log
,
95 STATIC
void xlog_grant_push_ail(xfs_mount_t
*mp
,
97 STATIC
void xlog_regrant_reserve_log_space(xlog_t
*log
,
98 xlog_ticket_t
*ticket
);
99 STATIC
int xlog_regrant_write_log_space(xlog_t
*log
,
100 xlog_ticket_t
*ticket
);
101 STATIC
void xlog_ungrant_log_space(xlog_t
*log
,
102 xlog_ticket_t
*ticket
);
105 /* local ticket functions */
106 STATIC
void xlog_state_ticket_alloc(xlog_t
*log
);
107 STATIC xlog_ticket_t
*xlog_ticket_get(xlog_t
*log
,
112 STATIC
void xlog_ticket_put(xlog_t
*log
, xlog_ticket_t
*ticket
);
115 STATIC
void xlog_verify_dest_ptr(xlog_t
*log
, __psint_t ptr
);
116 STATIC
void xlog_verify_grant_head(xlog_t
*log
, int equals
);
117 STATIC
void xlog_verify_iclog(xlog_t
*log
, xlog_in_core_t
*iclog
,
118 int count
, boolean_t syncing
);
119 STATIC
void xlog_verify_tail_lsn(xlog_t
*log
, xlog_in_core_t
*iclog
,
122 #define xlog_verify_dest_ptr(a,b)
123 #define xlog_verify_grant_head(a,b)
124 #define xlog_verify_iclog(a,b,c,d)
125 #define xlog_verify_tail_lsn(a,b,c)
128 STATIC
int xlog_iclogs_empty(xlog_t
*log
);
130 #if defined(XFS_LOG_TRACE)
132 xlog_trace_loggrant(xlog_t
*log
, xlog_ticket_t
*tic
, xfs_caddr_t string
)
136 if (!log
->l_grant_trace
) {
137 log
->l_grant_trace
= ktrace_alloc(2048, KM_NOSLEEP
);
138 if (!log
->l_grant_trace
)
141 /* ticket counts are 1 byte each */
142 cnts
= ((unsigned long)tic
->t_ocnt
) | ((unsigned long)tic
->t_cnt
) << 8;
144 ktrace_enter(log
->l_grant_trace
,
146 (void *)log
->l_reserve_headq
,
147 (void *)log
->l_write_headq
,
148 (void *)((unsigned long)log
->l_grant_reserve_cycle
),
149 (void *)((unsigned long)log
->l_grant_reserve_bytes
),
150 (void *)((unsigned long)log
->l_grant_write_cycle
),
151 (void *)((unsigned long)log
->l_grant_write_bytes
),
152 (void *)((unsigned long)log
->l_curr_cycle
),
153 (void *)((unsigned long)log
->l_curr_block
),
154 (void *)((unsigned long)CYCLE_LSN(log
->l_tail_lsn
)),
155 (void *)((unsigned long)BLOCK_LSN(log
->l_tail_lsn
)),
157 (void *)((unsigned long)tic
->t_trans_type
),
159 (void *)((unsigned long)tic
->t_curr_res
),
160 (void *)((unsigned long)tic
->t_unit_res
));
164 xlog_trace_iclog(xlog_in_core_t
*iclog
, uint state
)
166 if (!iclog
->ic_trace
)
167 iclog
->ic_trace
= ktrace_alloc(256, KM_SLEEP
);
168 ktrace_enter(iclog
->ic_trace
,
169 (void *)((unsigned long)state
),
170 (void *)((unsigned long)current_pid()),
171 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
172 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
173 (void *)NULL
, (void *)NULL
, (void *)NULL
, (void *)NULL
,
174 (void *)NULL
, (void *)NULL
);
177 #define xlog_trace_loggrant(log,tic,string)
178 #define xlog_trace_iclog(iclog,state)
179 #endif /* XFS_LOG_TRACE */
184 * 1. currblock field gets updated at startup and after in-core logs
185 * marked as with WANT_SYNC.
189 * This routine is called when a user of a log manager ticket is done with
190 * the reservation. If the ticket was ever used, then a commit record for
191 * the associated transaction is written out as a log operation header with
192 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
193 * a given ticket. If the ticket was one with a permanent reservation, then
194 * a few operations are done differently. Permanent reservation tickets by
195 * default don't release the reservation. They just commit the current
196 * transaction with the belief that the reservation is still needed. A flag
197 * must be passed in before permanent reservations are actually released.
198 * When these type of tickets are not released, they need to be set into
199 * the inited state again. By doing this, a start record will be written
200 * out when the next write occurs.
203 xfs_log_done(xfs_mount_t
*mp
,
204 xfs_log_ticket_t xtic
,
208 xlog_t
*log
= mp
->m_log
;
209 xlog_ticket_t
*ticket
= (xfs_log_ticket_t
) xtic
;
212 if (XLOG_FORCED_SHUTDOWN(log
) ||
214 * If nothing was ever written, don't write out commit record.
215 * If we get an error, just continue and give back the log ticket.
217 (((ticket
->t_flags
& XLOG_TIC_INITED
) == 0) &&
218 (xlog_commit_record(mp
, ticket
,
219 (xlog_in_core_t
**)iclog
, &lsn
)))) {
220 lsn
= (xfs_lsn_t
) -1;
221 if (ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) {
222 flags
|= XFS_LOG_REL_PERM_RESERV
;
227 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) == 0 ||
228 (flags
& XFS_LOG_REL_PERM_RESERV
)) {
230 * Release ticket if not permanent reservation or a specifc
231 * request has been made to release a permanent reservation.
233 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (non-permanent)");
234 xlog_ungrant_log_space(log
, ticket
);
235 xlog_state_put_ticket(log
, ticket
);
237 xlog_trace_loggrant(log
, ticket
, "xfs_log_done: (permanent)");
238 xlog_regrant_reserve_log_space(log
, ticket
);
241 /* If this ticket was a permanent reservation and we aren't
242 * trying to release it, reset the inited flags; so next time
243 * we write, a start record will be written out.
245 if ((ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) &&
246 (flags
& XFS_LOG_REL_PERM_RESERV
) == 0)
247 ticket
->t_flags
|= XLOG_TIC_INITED
;
254 * Force the in-core log to disk. If flags == XFS_LOG_SYNC,
255 * the force is done synchronously.
257 * Asynchronous forces are implemented by setting the WANT_SYNC
258 * bit in the appropriate in-core log and then returning.
260 * Synchronous forces are implemented with a semaphore. All callers
261 * to force a given lsn to disk will wait on a semaphore attached to the
262 * specific in-core log. When given in-core log finally completes its
263 * write to disk, that thread will wake up all threads waiting on the
273 xlog_t
*log
= mp
->m_log
;
277 log_flushed
= &dummy
;
279 ASSERT(flags
& XFS_LOG_FORCE
);
281 XFS_STATS_INC(xs_log_force
);
283 if (log
->l_flags
& XLOG_IO_ERROR
)
284 return XFS_ERROR(EIO
);
286 return xlog_state_sync_all(log
, flags
, log_flushed
);
288 return xlog_state_sync(log
, lsn
, flags
, log_flushed
);
289 } /* xfs_log_force */
292 * Attaches a new iclog I/O completion callback routine during
293 * transaction commit. If the log is in error state, a non-zero
294 * return code is handed back and the caller is responsible for
295 * executing the callback at an appropriate time.
298 xfs_log_notify(xfs_mount_t
*mp
, /* mount of partition */
299 void *iclog_hndl
, /* iclog to hang callback off */
300 xfs_log_callback_t
*cb
)
302 xlog_t
*log
= mp
->m_log
;
303 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
308 abortflg
= (iclog
->ic_state
& XLOG_STATE_IOERROR
);
310 ASSERT_ALWAYS((iclog
->ic_state
== XLOG_STATE_ACTIVE
) ||
311 (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
));
313 *(iclog
->ic_callback_tail
) = cb
;
314 iclog
->ic_callback_tail
= &(cb
->cb_next
);
316 LOG_UNLOCK(log
, spl
);
318 } /* xfs_log_notify */
321 xfs_log_release_iclog(xfs_mount_t
*mp
,
324 xlog_t
*log
= mp
->m_log
;
325 xlog_in_core_t
*iclog
= (xlog_in_core_t
*)iclog_hndl
;
327 if (xlog_state_release_iclog(log
, iclog
)) {
328 xfs_force_shutdown(mp
, XFS_LOG_IO_ERROR
);
336 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
337 * to the reservation.
338 * 2. Potentially, push buffers at tail of log to disk.
340 * Each reservation is going to reserve extra space for a log record header.
341 * When writes happen to the on-disk log, we don't subtract the length of the
342 * log record header from any reservation. By wasting space in each
343 * reservation, we prevent over allocation problems.
346 xfs_log_reserve(xfs_mount_t
*mp
,
349 xfs_log_ticket_t
*ticket
,
354 xlog_t
*log
= mp
->m_log
;
355 xlog_ticket_t
*internal_ticket
;
358 ASSERT(client
== XFS_TRANSACTION
|| client
== XFS_LOG
);
359 ASSERT((flags
& XFS_LOG_NOSLEEP
) == 0);
361 if (XLOG_FORCED_SHUTDOWN(log
))
362 return XFS_ERROR(EIO
);
364 XFS_STATS_INC(xs_try_logspace
);
366 if (*ticket
!= NULL
) {
367 ASSERT(flags
& XFS_LOG_PERM_RESERV
);
368 internal_ticket
= (xlog_ticket_t
*)*ticket
;
369 xlog_trace_loggrant(log
, internal_ticket
, "xfs_log_reserve: existing ticket (permanent trans)");
370 xlog_grant_push_ail(mp
, internal_ticket
->t_unit_res
);
371 retval
= xlog_regrant_write_log_space(log
, internal_ticket
);
373 /* may sleep if need to allocate more tickets */
374 internal_ticket
= xlog_ticket_get(log
, unit_bytes
, cnt
,
376 internal_ticket
->t_trans_type
= t_type
;
377 *ticket
= internal_ticket
;
378 xlog_trace_loggrant(log
, internal_ticket
,
379 (internal_ticket
->t_flags
& XLOG_TIC_PERM_RESERV
) ?
380 "xfs_log_reserve: create new ticket (permanent trans)" :
381 "xfs_log_reserve: create new ticket");
382 xlog_grant_push_ail(mp
,
383 (internal_ticket
->t_unit_res
*
384 internal_ticket
->t_cnt
));
385 retval
= xlog_grant_log_space(log
, internal_ticket
);
389 } /* xfs_log_reserve */
393 * Mount a log filesystem
395 * mp - ubiquitous xfs mount point structure
396 * log_target - buftarg of on-disk log device
397 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
398 * num_bblocks - Number of BBSIZE blocks in on-disk log
400 * Return error or zero.
403 xfs_log_mount(xfs_mount_t
*mp
,
404 xfs_buftarg_t
*log_target
,
405 xfs_daddr_t blk_offset
,
408 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
409 cmn_err(CE_NOTE
, "XFS mounting filesystem %s", mp
->m_fsname
);
412 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
414 ASSERT(XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
);
417 mp
->m_log
= xlog_alloc_log(mp
, log_target
, blk_offset
, num_bblks
);
420 * skip log recovery on a norecovery mount. pretend it all
423 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
)) {
425 vfs_t
*vfsp
= XFS_MTOVFS(mp
);
426 int readonly
= (vfsp
->vfs_flag
& VFS_RDONLY
);
429 vfsp
->vfs_flag
&= ~VFS_RDONLY
;
431 error
= xlog_recover(mp
->m_log
, readonly
);
434 vfsp
->vfs_flag
|= VFS_RDONLY
;
436 cmn_err(CE_WARN
, "XFS: log mount/recovery failed: error %d", error
);
437 xlog_unalloc_log(mp
->m_log
);
442 /* Normal transactions can now occur */
443 mp
->m_log
->l_flags
&= ~XLOG_ACTIVE_RECOVERY
;
445 /* End mounting message in xfs_log_mount_finish */
447 } /* xfs_log_mount */
450 * Finish the recovery of the file system. This is separate from
451 * the xfs_log_mount() call, because it depends on the code in
452 * xfs_mountfs() to read in the root and real-time bitmap inodes
453 * between calling xfs_log_mount() and here.
455 * mp - ubiquitous xfs mount point structure
458 xfs_log_mount_finish(xfs_mount_t
*mp
, int mfsi_flags
)
462 if (!(mp
->m_flags
& XFS_MOUNT_NORECOVERY
))
463 error
= xlog_recover_finish(mp
->m_log
, mfsi_flags
);
466 ASSERT(XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
);
473 * Unmount processing for the log.
476 xfs_log_unmount(xfs_mount_t
*mp
)
480 error
= xfs_log_unmount_write(mp
);
481 xfs_log_unmount_dealloc(mp
);
486 * Final log writes as part of unmount.
488 * Mark the filesystem clean as unmount happens. Note that during relocation
489 * this routine needs to be executed as part of source-bag while the
490 * deallocation must not be done until source-end.
494 * Unmount record used to have a string "Unmount filesystem--" in the
495 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
496 * We just write the magic number now since that particular field isn't
497 * currently architecture converted and "nUmount" is a bit foo.
498 * As far as I know, there weren't any dependencies on the old behaviour.
502 xfs_log_unmount_write(xfs_mount_t
*mp
)
504 xlog_t
*log
= mp
->m_log
;
505 xlog_in_core_t
*iclog
;
507 xlog_in_core_t
*first_iclog
;
509 xfs_log_iovec_t reg
[1];
510 xfs_log_ticket_t tic
= NULL
;
515 /* the data section must be 32 bit size aligned */
519 __uint32_t pad2
; /* may as well make it 64 bits */
520 } magic
= { XLOG_UNMOUNT_TYPE
, 0, 0 };
523 * Don't write out unmount record on read-only mounts.
524 * Or, if we are doing a forced umount (typically because of IO errors).
526 if (XFS_MTOVFS(mp
)->vfs_flag
& VFS_RDONLY
)
529 xfs_log_force(mp
, 0, XFS_LOG_FORCE
|XFS_LOG_SYNC
);
532 first_iclog
= iclog
= log
->l_iclog
;
534 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
535 ASSERT(iclog
->ic_state
& XLOG_STATE_ACTIVE
);
536 ASSERT(iclog
->ic_offset
== 0);
538 iclog
= iclog
->ic_next
;
539 } while (iclog
!= first_iclog
);
541 if (! (XLOG_FORCED_SHUTDOWN(log
))) {
542 reg
[0].i_addr
= (void*)&magic
;
543 reg
[0].i_len
= sizeof(magic
);
544 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_UNMOUNT
);
546 error
= xfs_log_reserve(mp
, 600, 1, &tic
, XFS_LOG
, 0, 0);
548 /* remove inited flag */
549 ((xlog_ticket_t
*)tic
)->t_flags
= 0;
550 error
= xlog_write(mp
, reg
, 1, tic
, &lsn
,
551 NULL
, XLOG_UNMOUNT_TRANS
);
553 * At this point, we're umounting anyway,
554 * so there's no point in transitioning log state
555 * to IOERROR. Just continue...
560 xfs_fs_cmn_err(CE_ALERT
, mp
,
561 "xfs_log_unmount: unmount record failed");
566 iclog
= log
->l_iclog
;
569 xlog_state_want_sync(log
, iclog
);
570 (void) xlog_state_release_iclog(log
, iclog
);
573 if (!(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
574 iclog
->ic_state
== XLOG_STATE_DIRTY
)) {
575 if (!XLOG_FORCED_SHUTDOWN(log
)) {
576 sv_wait(&iclog
->ic_forcesema
, PMEM
,
577 &log
->l_icloglock
, s
);
585 xlog_state_put_ticket(log
, tic
);
588 * We're already in forced_shutdown mode, couldn't
589 * even attempt to write out the unmount transaction.
591 * Go through the motions of sync'ing and releasing
592 * the iclog, even though no I/O will actually happen,
593 * we need to wait for other log I/O's that may already
594 * be in progress. Do this as a separate section of
595 * code so we'll know if we ever get stuck here that
596 * we're in this odd situation of trying to unmount
597 * a file system that went into forced_shutdown as
598 * the result of an unmount..
601 iclog
= log
->l_iclog
;
605 xlog_state_want_sync(log
, iclog
);
606 (void) xlog_state_release_iclog(log
, iclog
);
610 if ( ! ( iclog
->ic_state
== XLOG_STATE_ACTIVE
611 || iclog
->ic_state
== XLOG_STATE_DIRTY
612 || iclog
->ic_state
== XLOG_STATE_IOERROR
) ) {
614 sv_wait(&iclog
->ic_forcesema
, PMEM
,
615 &log
->l_icloglock
, s
);
622 } /* xfs_log_unmount_write */
625 * Deallocate log structures for unmount/relocation.
628 xfs_log_unmount_dealloc(xfs_mount_t
*mp
)
630 xlog_unalloc_log(mp
->m_log
);
634 * Write region vectors to log. The write happens using the space reservation
635 * of the ticket (tic). It is not a requirement that all writes for a given
636 * transaction occur with one call to xfs_log_write().
639 xfs_log_write(xfs_mount_t
* mp
,
640 xfs_log_iovec_t reg
[],
642 xfs_log_ticket_t tic
,
643 xfs_lsn_t
*start_lsn
)
646 xlog_t
*log
= mp
->m_log
;
648 if (XLOG_FORCED_SHUTDOWN(log
))
649 return XFS_ERROR(EIO
);
651 if ((error
= xlog_write(mp
, reg
, nentries
, tic
, start_lsn
, NULL
, 0))) {
652 xfs_force_shutdown(mp
, XFS_LOG_IO_ERROR
);
655 } /* xfs_log_write */
659 xfs_log_move_tail(xfs_mount_t
*mp
,
663 xlog_t
*log
= mp
->m_log
;
664 int need_bytes
, free_bytes
, cycle
, bytes
;
667 if (XLOG_FORCED_SHUTDOWN(log
))
669 ASSERT(!XFS_FORCED_SHUTDOWN(mp
));
672 /* needed since sync_lsn is 64 bits */
674 tail_lsn
= log
->l_last_sync_lsn
;
680 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
684 log
->l_tail_lsn
= tail_lsn
;
687 if ((tic
= log
->l_write_headq
)) {
689 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
690 panic("Recovery problem");
692 cycle
= log
->l_grant_write_cycle
;
693 bytes
= log
->l_grant_write_bytes
;
694 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
696 ASSERT(tic
->t_flags
& XLOG_TIC_PERM_RESERV
);
698 if (free_bytes
< tic
->t_unit_res
&& tail_lsn
!= 1)
701 free_bytes
-= tic
->t_unit_res
;
702 sv_signal(&tic
->t_sema
);
704 } while (tic
!= log
->l_write_headq
);
706 if ((tic
= log
->l_reserve_headq
)) {
708 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
709 panic("Recovery problem");
711 cycle
= log
->l_grant_reserve_cycle
;
712 bytes
= log
->l_grant_reserve_bytes
;
713 free_bytes
= xlog_space_left(log
, cycle
, bytes
);
715 if (tic
->t_flags
& XLOG_TIC_PERM_RESERV
)
716 need_bytes
= tic
->t_unit_res
*tic
->t_cnt
;
718 need_bytes
= tic
->t_unit_res
;
719 if (free_bytes
< need_bytes
&& tail_lsn
!= 1)
722 free_bytes
-= need_bytes
;
723 sv_signal(&tic
->t_sema
);
725 } while (tic
!= log
->l_reserve_headq
);
727 GRANT_UNLOCK(log
, s
);
728 } /* xfs_log_move_tail */
731 * Determine if we have a transaction that has gone to disk
732 * that needs to be covered. Log activity needs to be idle (no AIL and
733 * nothing in the iclogs). And, we need to be in the right state indicating
734 * something has gone out.
737 xfs_log_need_covered(xfs_mount_t
*mp
)
741 xlog_t
*log
= mp
->m_log
;
742 vfs_t
*vfsp
= XFS_MTOVFS(mp
);
744 if (fs_frozen(vfsp
) || XFS_FORCED_SHUTDOWN(mp
) ||
745 (vfsp
->vfs_flag
& VFS_RDONLY
))
749 if (((log
->l_covered_state
== XLOG_STATE_COVER_NEED
) ||
750 (log
->l_covered_state
== XLOG_STATE_COVER_NEED2
))
751 && !xfs_trans_first_ail(mp
, &gen
)
752 && xlog_iclogs_empty(log
)) {
753 if (log
->l_covered_state
== XLOG_STATE_COVER_NEED
)
754 log
->l_covered_state
= XLOG_STATE_COVER_DONE
;
756 ASSERT(log
->l_covered_state
== XLOG_STATE_COVER_NEED2
);
757 log
->l_covered_state
= XLOG_STATE_COVER_DONE2
;
765 /******************************************************************************
769 ******************************************************************************
772 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
773 * The log manager must keep track of the last LR which was committed
774 * to disk. The lsn of this LR will become the new tail_lsn whenever
775 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
776 * the situation where stuff could be written into the log but nothing
777 * was ever in the AIL when asked. Eventually, we panic since the
778 * tail hits the head.
780 * We may be holding the log iclog lock upon entering this routine.
783 xlog_assign_tail_lsn(xfs_mount_t
*mp
)
787 xlog_t
*log
= mp
->m_log
;
789 tail_lsn
= xfs_trans_tail_ail(mp
);
792 log
->l_tail_lsn
= tail_lsn
;
794 tail_lsn
= log
->l_tail_lsn
= log
->l_last_sync_lsn
;
796 GRANT_UNLOCK(log
, s
);
799 } /* xlog_assign_tail_lsn */
803 * Return the space in the log between the tail and the head. The head
804 * is passed in the cycle/bytes formal parms. In the special case where
805 * the reserve head has wrapped passed the tail, this calculation is no
806 * longer valid. In this case, just return 0 which means there is no space
807 * in the log. This works for all places where this function is called
808 * with the reserve head. Of course, if the write head were to ever
809 * wrap the tail, we should blow up. Rather than catch this case here,
810 * we depend on other ASSERTions in other parts of the code. XXXmiken
812 * This code also handles the case where the reservation head is behind
813 * the tail. The details of this case are described below, but the end
814 * result is that we return the size of the log as the amount of space left.
817 xlog_space_left(xlog_t
*log
, int cycle
, int bytes
)
823 tail_bytes
= BBTOB(BLOCK_LSN(log
->l_tail_lsn
));
824 tail_cycle
= CYCLE_LSN(log
->l_tail_lsn
);
825 if ((tail_cycle
== cycle
) && (bytes
>= tail_bytes
)) {
826 free_bytes
= log
->l_logsize
- (bytes
- tail_bytes
);
827 } else if ((tail_cycle
+ 1) < cycle
) {
829 } else if (tail_cycle
< cycle
) {
830 ASSERT(tail_cycle
== (cycle
- 1));
831 free_bytes
= tail_bytes
- bytes
;
834 * The reservation head is behind the tail.
835 * In this case we just want to return the size of the
836 * log as the amount of space left.
838 xfs_fs_cmn_err(CE_ALERT
, log
->l_mp
,
839 "xlog_space_left: head behind tail\n"
840 " tail_cycle = %d, tail_bytes = %d\n"
841 " GH cycle = %d, GH bytes = %d",
842 tail_cycle
, tail_bytes
, cycle
, bytes
);
844 free_bytes
= log
->l_logsize
;
847 } /* xlog_space_left */
851 * Log function which is called when an io completes.
853 * The log manager needs its own routine, in order to control what
854 * happens with the buffer after the write completes.
857 xlog_iodone(xfs_buf_t
*bp
)
859 xlog_in_core_t
*iclog
;
863 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
864 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long) 2);
865 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
869 * Some versions of cpp barf on the recursive definition of
870 * ic_log -> hic_fields.ic_log and expand ic_log twice when
871 * it is passed through two macros. Workaround broken cpp.
876 * Race to shutdown the filesystem if we see an error.
878 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp
)), l
->l_mp
,
879 XFS_ERRTAG_IODONE_IOERR
, XFS_RANDOM_IODONE_IOERR
)) {
880 xfs_ioerror_alert("xlog_iodone", l
->l_mp
, bp
, XFS_BUF_ADDR(bp
));
882 xfs_force_shutdown(l
->l_mp
, XFS_LOG_IO_ERROR
);
884 * This flag will be propagated to the trans-committed
885 * callback routines to let them know that the log-commit
888 aborted
= XFS_LI_ABORTED
;
889 } else if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
890 aborted
= XFS_LI_ABORTED
;
892 xlog_state_done_syncing(iclog
, aborted
);
893 if (!(XFS_BUF_ISASYNC(bp
))) {
895 * Corresponding psema() will be done in bwrite(). If we don't
896 * vsema() here, panic.
898 XFS_BUF_V_IODONESEMA(bp
);
903 * The bdstrat callback function for log bufs. This gives us a central
904 * place to trap bufs in case we get hit by a log I/O error and need to
905 * shutdown. Actually, in practice, even when we didn't get a log error,
906 * we transition the iclogs to IOERROR state *after* flushing all existing
907 * iclogs to disk. This is because we don't want anymore new transactions to be
908 * started or completed afterwards.
911 xlog_bdstrat_cb(struct xfs_buf
*bp
)
913 xlog_in_core_t
*iclog
;
915 iclog
= XFS_BUF_FSPRIVATE(bp
, xlog_in_core_t
*);
917 if ((iclog
->ic_state
& XLOG_STATE_IOERROR
) == 0) {
918 /* note for irix bstrat will need struct bdevsw passed
919 * Fix the following macro if the code ever is merged
925 xfs_buftrace("XLOG__BDSTRAT IOERROR", bp
);
926 XFS_BUF_ERROR(bp
, EIO
);
929 return (XFS_ERROR(EIO
));
935 * Return size of each in-core log record buffer.
937 * Low memory machines only get 2 16KB buffers. We don't want to waste
938 * memory here. However, all other machines get at least 2 32KB buffers.
939 * The number is hard coded because we don't care about the minimum
940 * memory size, just 32MB systems.
942 * If the filesystem blocksize is too large, we may need to choose a
943 * larger size since the directory code currently logs entire blocks.
947 xlog_get_iclog_buffer_size(xfs_mount_t
*mp
,
953 if (mp
->m_logbufs
<= 0) {
954 if (xfs_physmem
<= btoc(128*1024*1024)) {
955 log
->l_iclog_bufs
= XLOG_MIN_ICLOGS
;
956 } else if (xfs_physmem
<= btoc(400*1024*1024)) {
957 log
->l_iclog_bufs
= XLOG_MED_ICLOGS
;
958 } else { /* 256K with 32K bufs */
959 log
->l_iclog_bufs
= XLOG_MAX_ICLOGS
;
962 log
->l_iclog_bufs
= mp
->m_logbufs
;
966 * Buffer size passed in from mount system call.
968 if (mp
->m_logbsize
> 0) {
969 size
= log
->l_iclog_size
= mp
->m_logbsize
;
970 log
->l_iclog_size_log
= 0;
972 log
->l_iclog_size_log
++;
976 if (XFS_SB_VERSION_HASLOGV2(&mp
->m_sb
)) {
977 /* # headers = size / 32K
978 * one header holds cycles from 32K of data
981 xhdrs
= mp
->m_logbsize
/ XLOG_HEADER_CYCLE_SIZE
;
982 if (mp
->m_logbsize
% XLOG_HEADER_CYCLE_SIZE
)
984 log
->l_iclog_hsize
= xhdrs
<< BBSHIFT
;
985 log
->l_iclog_heads
= xhdrs
;
987 ASSERT(mp
->m_logbsize
<= XLOG_BIG_RECORD_BSIZE
);
988 log
->l_iclog_hsize
= BBSIZE
;
989 log
->l_iclog_heads
= 1;
995 * Special case machines that have less than 32MB of memory.
996 * All machines with more memory use 32KB buffers.
998 if (xfs_physmem
<= btoc(32*1024*1024)) {
999 /* Don't change; min configuration */
1000 log
->l_iclog_size
= XLOG_RECORD_BSIZE
; /* 16k */
1001 log
->l_iclog_size_log
= XLOG_RECORD_BSHIFT
;
1003 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
; /* 32k */
1004 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1007 /* the default log size is 16k or 32k which is one header sector */
1008 log
->l_iclog_hsize
= BBSIZE
;
1009 log
->l_iclog_heads
= 1;
1012 * For 16KB, we use 3 32KB buffers. For 32KB block sizes, we use
1013 * 4 32KB buffers. For 64KB block sizes, we use 8 32KB buffers.
1015 if (mp
->m_sb
.sb_blocksize
>= 16*1024) {
1016 log
->l_iclog_size
= XLOG_BIG_RECORD_BSIZE
;
1017 log
->l_iclog_size_log
= XLOG_BIG_RECORD_BSHIFT
;
1018 if (mp
->m_logbufs
<= 0) {
1019 switch (mp
->m_sb
.sb_blocksize
) {
1020 case 16*1024: /* 16 KB */
1021 log
->l_iclog_bufs
= 3;
1023 case 32*1024: /* 32 KB */
1024 log
->l_iclog_bufs
= 4;
1026 case 64*1024: /* 64 KB */
1027 log
->l_iclog_bufs
= 8;
1030 xlog_panic("XFS: Invalid blocksize");
1036 done
: /* are we being asked to make the sizes selected above visible? */
1037 if (mp
->m_logbufs
== 0)
1038 mp
->m_logbufs
= log
->l_iclog_bufs
;
1039 if (mp
->m_logbsize
== 0)
1040 mp
->m_logbsize
= log
->l_iclog_size
;
1041 } /* xlog_get_iclog_buffer_size */
1045 * This routine initializes some of the log structure for a given mount point.
1046 * Its primary purpose is to fill in enough, so recovery can occur. However,
1047 * some other stuff may be filled in too.
1050 xlog_alloc_log(xfs_mount_t
*mp
,
1051 xfs_buftarg_t
*log_target
,
1052 xfs_daddr_t blk_offset
,
1056 xlog_rec_header_t
*head
;
1057 xlog_in_core_t
**iclogp
;
1058 xlog_in_core_t
*iclog
, *prev_iclog
=NULL
;
1063 log
= (xlog_t
*)kmem_zalloc(sizeof(xlog_t
), KM_SLEEP
);
1066 log
->l_targ
= log_target
;
1067 log
->l_logsize
= BBTOB(num_bblks
);
1068 log
->l_logBBstart
= blk_offset
;
1069 log
->l_logBBsize
= num_bblks
;
1070 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1071 log
->l_flags
|= XLOG_ACTIVE_RECOVERY
;
1073 log
->l_prev_block
= -1;
1074 ASSIGN_ANY_LSN_HOST(log
->l_tail_lsn
, 1, 0);
1075 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1076 log
->l_last_sync_lsn
= log
->l_tail_lsn
;
1077 log
->l_curr_cycle
= 1; /* 0 is bad since this is initial value */
1078 log
->l_grant_reserve_cycle
= 1;
1079 log
->l_grant_write_cycle
= 1;
1081 if (XFS_SB_VERSION_HASSECTOR(&mp
->m_sb
)) {
1082 log
->l_sectbb_log
= mp
->m_sb
.sb_logsectlog
- BBSHIFT
;
1083 ASSERT(log
->l_sectbb_log
<= mp
->m_sectbb_log
);
1084 /* for larger sector sizes, must have v2 or external log */
1085 ASSERT(log
->l_sectbb_log
== 0 ||
1086 log
->l_logBBstart
== 0 ||
1087 XFS_SB_VERSION_HASLOGV2(&mp
->m_sb
));
1088 ASSERT(mp
->m_sb
.sb_logsectlog
>= BBSHIFT
);
1090 log
->l_sectbb_mask
= (1 << log
->l_sectbb_log
) - 1;
1092 xlog_get_iclog_buffer_size(mp
, log
);
1094 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1095 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1096 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1097 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1098 ASSERT(XFS_BUF_ISBUSY(bp
));
1099 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
1102 spinlock_init(&log
->l_icloglock
, "iclog");
1103 spinlock_init(&log
->l_grant_lock
, "grhead_iclog");
1104 initnsema(&log
->l_flushsema
, 0, "ic-flush");
1105 xlog_state_ticket_alloc(log
); /* wait until after icloglock inited */
1107 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1108 ASSERT((XFS_BUF_SIZE(bp
) & BBMASK
) == 0);
1110 iclogp
= &log
->l_iclog
;
1112 * The amount of memory to allocate for the iclog structure is
1113 * rather funky due to the way the structure is defined. It is
1114 * done this way so that we can use different sizes for machines
1115 * with different amounts of memory. See the definition of
1116 * xlog_in_core_t in xfs_log_priv.h for details.
1118 iclogsize
= log
->l_iclog_size
;
1119 ASSERT(log
->l_iclog_size
>= 4096);
1120 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
1121 *iclogp
= (xlog_in_core_t
*)
1122 kmem_zalloc(sizeof(xlog_in_core_t
), KM_SLEEP
);
1124 iclog
->hic_data
= (xlog_in_core_2_t
*)
1125 kmem_zalloc(iclogsize
, KM_SLEEP
);
1127 iclog
->ic_prev
= prev_iclog
;
1129 log
->l_iclog_bak
[i
] = (xfs_caddr_t
)&(iclog
->ic_header
);
1131 head
= &iclog
->ic_header
;
1132 memset(head
, 0, sizeof(xlog_rec_header_t
));
1133 INT_SET(head
->h_magicno
, ARCH_CONVERT
, XLOG_HEADER_MAGIC_NUM
);
1134 INT_SET(head
->h_version
, ARCH_CONVERT
,
1135 XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) ? 2 : 1);
1136 INT_SET(head
->h_size
, ARCH_CONVERT
, log
->l_iclog_size
);
1138 INT_SET(head
->h_fmt
, ARCH_CONVERT
, XLOG_FMT
);
1139 memcpy(&head
->h_fs_uuid
, &mp
->m_sb
.sb_uuid
, sizeof(uuid_t
));
1141 bp
= xfs_buf_get_empty(log
->l_iclog_size
, mp
->m_logdev_targp
);
1142 XFS_BUF_SET_IODONE_FUNC(bp
, xlog_iodone
);
1143 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xlog_bdstrat_cb
);
1144 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)1);
1147 iclog
->ic_size
= XFS_BUF_SIZE(bp
) - log
->l_iclog_hsize
;
1148 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1149 iclog
->ic_log
= log
;
1150 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
1151 iclog
->ic_datap
= (char *)iclog
->hic_data
+ log
->l_iclog_hsize
;
1153 ASSERT(XFS_BUF_ISBUSY(iclog
->ic_bp
));
1154 ASSERT(XFS_BUF_VALUSEMA(iclog
->ic_bp
) <= 0);
1155 sv_init(&iclog
->ic_forcesema
, SV_DEFAULT
, "iclog-force");
1156 sv_init(&iclog
->ic_writesema
, SV_DEFAULT
, "iclog-write");
1158 iclogp
= &iclog
->ic_next
;
1160 *iclogp
= log
->l_iclog
; /* complete ring */
1161 log
->l_iclog
->ic_prev
= prev_iclog
; /* re-write 1st prev ptr */
1164 } /* xlog_alloc_log */
1168 * Write out the commit record of a transaction associated with the given
1169 * ticket. Return the lsn of the commit record.
1172 xlog_commit_record(xfs_mount_t
*mp
,
1173 xlog_ticket_t
*ticket
,
1174 xlog_in_core_t
**iclog
,
1175 xfs_lsn_t
*commitlsnp
)
1178 xfs_log_iovec_t reg
[1];
1180 reg
[0].i_addr
= NULL
;
1182 XLOG_VEC_SET_TYPE(®
[0], XLOG_REG_TYPE_COMMIT
);
1184 ASSERT_ALWAYS(iclog
);
1185 if ((error
= xlog_write(mp
, reg
, 1, ticket
, commitlsnp
,
1186 iclog
, XLOG_COMMIT_TRANS
))) {
1187 xfs_force_shutdown(mp
, XFS_LOG_IO_ERROR
);
1190 } /* xlog_commit_record */
1194 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1195 * log space. This code pushes on the lsn which would supposedly free up
1196 * the 25% which we want to leave free. We may need to adopt a policy which
1197 * pushes on an lsn which is further along in the log once we reach the high
1198 * water mark. In this manner, we would be creating a low water mark.
1201 xlog_grant_push_ail(xfs_mount_t
*mp
,
1204 xlog_t
*log
= mp
->m_log
; /* pointer to the log */
1205 xfs_lsn_t tail_lsn
; /* lsn of the log tail */
1206 xfs_lsn_t threshold_lsn
= 0; /* lsn we'd like to be at */
1207 int free_blocks
; /* free blocks left to write to */
1208 int free_bytes
; /* free bytes left to write to */
1209 int threshold_block
; /* block in lsn we'd like to be at */
1210 int threshold_cycle
; /* lsn cycle we'd like to be at */
1214 ASSERT(BTOBB(need_bytes
) < log
->l_logBBsize
);
1216 s
= GRANT_LOCK(log
);
1217 free_bytes
= xlog_space_left(log
,
1218 log
->l_grant_reserve_cycle
,
1219 log
->l_grant_reserve_bytes
);
1220 tail_lsn
= log
->l_tail_lsn
;
1221 free_blocks
= BTOBBT(free_bytes
);
1224 * Set the threshold for the minimum number of free blocks in the
1225 * log to the maximum of what the caller needs, one quarter of the
1226 * log, and 256 blocks.
1228 free_threshold
= BTOBB(need_bytes
);
1229 free_threshold
= MAX(free_threshold
, (log
->l_logBBsize
>> 2));
1230 free_threshold
= MAX(free_threshold
, 256);
1231 if (free_blocks
< free_threshold
) {
1232 threshold_block
= BLOCK_LSN(tail_lsn
) + free_threshold
;
1233 threshold_cycle
= CYCLE_LSN(tail_lsn
);
1234 if (threshold_block
>= log
->l_logBBsize
) {
1235 threshold_block
-= log
->l_logBBsize
;
1236 threshold_cycle
+= 1;
1238 ASSIGN_ANY_LSN_HOST(threshold_lsn
, threshold_cycle
,
1241 /* Don't pass in an lsn greater than the lsn of the last
1242 * log record known to be on disk.
1244 if (XFS_LSN_CMP(threshold_lsn
, log
->l_last_sync_lsn
) > 0)
1245 threshold_lsn
= log
->l_last_sync_lsn
;
1247 GRANT_UNLOCK(log
, s
);
1250 * Get the transaction layer to kick the dirty buffers out to
1251 * disk asynchronously. No point in trying to do this if
1252 * the filesystem is shutting down.
1254 if (threshold_lsn
&&
1255 !XLOG_FORCED_SHUTDOWN(log
))
1256 xfs_trans_push_ail(mp
, threshold_lsn
);
1257 } /* xlog_grant_push_ail */
1261 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1262 * fashion. Previously, we should have moved the current iclog
1263 * ptr in the log to point to the next available iclog. This allows further
1264 * write to continue while this code syncs out an iclog ready to go.
1265 * Before an in-core log can be written out, the data section must be scanned
1266 * to save away the 1st word of each BBSIZE block into the header. We replace
1267 * it with the current cycle count. Each BBSIZE block is tagged with the
1268 * cycle count because there in an implicit assumption that drives will
1269 * guarantee that entire 512 byte blocks get written at once. In other words,
1270 * we can't have part of a 512 byte block written and part not written. By
1271 * tagging each block, we will know which blocks are valid when recovering
1272 * after an unclean shutdown.
1274 * This routine is single threaded on the iclog. No other thread can be in
1275 * this routine with the same iclog. Changing contents of iclog can there-
1276 * fore be done without grabbing the state machine lock. Updating the global
1277 * log will require grabbing the lock though.
1279 * The entire log manager uses a logical block numbering scheme. Only
1280 * log_sync (and then only bwrite()) know about the fact that the log may
1281 * not start with block zero on a given device. The log block start offset
1282 * is added immediately before calling bwrite().
1286 xlog_sync(xlog_t
*log
,
1287 xlog_in_core_t
*iclog
)
1289 xfs_caddr_t dptr
; /* pointer to byte sized element */
1292 uint count
; /* byte count of bwrite */
1293 uint count_init
; /* initial count before roundup */
1294 int roundoff
; /* roundoff to BB or stripe */
1295 int split
= 0; /* split write into two regions */
1298 int v2
= XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
);
1300 XFS_STATS_INC(xs_log_writes
);
1301 ASSERT(iclog
->ic_refcnt
== 0);
1303 /* Add for LR header */
1304 count_init
= log
->l_iclog_hsize
+ iclog
->ic_offset
;
1306 /* Round out the log write size */
1307 if (v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1) {
1308 /* we have a v2 stripe unit to use */
1309 count
= XLOG_LSUNITTOB(log
, XLOG_BTOLSUNIT(log
, count_init
));
1311 count
= BBTOB(BTOBB(count_init
));
1313 roundoff
= count
- count_init
;
1314 ASSERT(roundoff
>= 0);
1315 ASSERT((v2
&& log
->l_mp
->m_sb
.sb_logsunit
> 1 &&
1316 roundoff
< log
->l_mp
->m_sb
.sb_logsunit
)
1318 (log
->l_mp
->m_sb
.sb_logsunit
<= 1 &&
1319 roundoff
< BBTOB(1)));
1321 /* move grant heads by roundoff in sync */
1322 s
= GRANT_LOCK(log
);
1323 XLOG_GRANT_ADD_SPACE(log
, roundoff
, 'w');
1324 XLOG_GRANT_ADD_SPACE(log
, roundoff
, 'r');
1325 GRANT_UNLOCK(log
, s
);
1327 /* put cycle number in every block */
1328 xlog_pack_data(log
, iclog
, roundoff
);
1330 /* real byte length */
1332 INT_SET(iclog
->ic_header
.h_len
,
1334 iclog
->ic_offset
+ roundoff
);
1336 INT_SET(iclog
->ic_header
.h_len
, ARCH_CONVERT
, iclog
->ic_offset
);
1339 /* put ops count in correct order */
1340 ops
= iclog
->ic_header
.h_num_logops
;
1341 INT_SET(iclog
->ic_header
.h_num_logops
, ARCH_CONVERT
, ops
);
1344 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) == (unsigned long)1);
1345 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1346 XFS_BUF_SET_ADDR(bp
, BLOCK_LSN(INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
)));
1348 XFS_STATS_ADD(xs_log_blocks
, BTOBB(count
));
1350 /* Do we need to split this write into 2 parts? */
1351 if (XFS_BUF_ADDR(bp
) + BTOBB(count
) > log
->l_logBBsize
) {
1352 split
= count
- (BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
)));
1353 count
= BBTOB(log
->l_logBBsize
- XFS_BUF_ADDR(bp
));
1354 iclog
->ic_bwritecnt
= 2; /* split into 2 writes */
1356 iclog
->ic_bwritecnt
= 1;
1358 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
) &(iclog
->ic_header
), count
);
1359 XFS_BUF_SET_FSPRIVATE(bp
, iclog
); /* save for later */
1363 * Do an ordered write for the log block.
1365 * It may not be needed to flush the first split block in the log wrap
1366 * case, but do it anyways to be safe -AK
1368 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1369 XFS_BUF_ORDERED(bp
);
1371 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1372 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1374 xlog_verify_iclog(log
, iclog
, count
, B_TRUE
);
1376 /* account for log which doesn't start at block #0 */
1377 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1379 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1384 if ((error
= XFS_bwrite(bp
))) {
1385 xfs_ioerror_alert("xlog_sync", log
->l_mp
, bp
,
1390 bp
= iclog
->ic_log
->l_xbuf
;
1391 ASSERT(XFS_BUF_FSPRIVATE2(bp
, unsigned long) ==
1393 XFS_BUF_SET_FSPRIVATE2(bp
, (unsigned long)2);
1394 XFS_BUF_SET_ADDR(bp
, 0); /* logical 0 */
1395 XFS_BUF_SET_PTR(bp
, (xfs_caddr_t
)((__psint_t
)&(iclog
->ic_header
)+
1396 (__psint_t
)count
), split
);
1397 XFS_BUF_SET_FSPRIVATE(bp
, iclog
);
1400 if (log
->l_mp
->m_flags
& XFS_MOUNT_BARRIER
)
1401 XFS_BUF_ORDERED(bp
);
1402 dptr
= XFS_BUF_PTR(bp
);
1404 * Bump the cycle numbers at the start of each block
1405 * since this part of the buffer is at the start of
1406 * a new cycle. Watch out for the header magic number
1409 for (i
=0; i
<split
; i
+= BBSIZE
) {
1410 INT_MOD(*(uint
*)dptr
, ARCH_CONVERT
, +1);
1411 if (INT_GET(*(uint
*)dptr
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
)
1412 INT_MOD(*(uint
*)dptr
, ARCH_CONVERT
, +1);
1416 ASSERT(XFS_BUF_ADDR(bp
) <= log
->l_logBBsize
-1);
1417 ASSERT(XFS_BUF_ADDR(bp
) + BTOBB(count
) <= log
->l_logBBsize
);
1419 /* account for internal log which does't start at block #0 */
1420 XFS_BUF_SET_ADDR(bp
, XFS_BUF_ADDR(bp
) + log
->l_logBBstart
);
1422 if ((error
= XFS_bwrite(bp
))) {
1423 xfs_ioerror_alert("xlog_sync (split)", log
->l_mp
,
1424 bp
, XFS_BUF_ADDR(bp
));
1433 * Unallocate a log structure
1436 xlog_unalloc_log(xlog_t
*log
)
1438 xlog_in_core_t
*iclog
, *next_iclog
;
1439 xlog_ticket_t
*tic
, *next_tic
;
1443 iclog
= log
->l_iclog
;
1444 for (i
=0; i
<log
->l_iclog_bufs
; i
++) {
1445 sv_destroy(&iclog
->ic_forcesema
);
1446 sv_destroy(&iclog
->ic_writesema
);
1447 xfs_buf_free(iclog
->ic_bp
);
1448 #ifdef XFS_LOG_TRACE
1449 if (iclog
->ic_trace
!= NULL
) {
1450 ktrace_free(iclog
->ic_trace
);
1453 next_iclog
= iclog
->ic_next
;
1454 kmem_free(iclog
->hic_data
, log
->l_iclog_size
);
1455 kmem_free(iclog
, sizeof(xlog_in_core_t
));
1458 freesema(&log
->l_flushsema
);
1459 spinlock_destroy(&log
->l_icloglock
);
1460 spinlock_destroy(&log
->l_grant_lock
);
1462 /* XXXsup take a look at this again. */
1463 if ((log
->l_ticket_cnt
!= log
->l_ticket_tcnt
) &&
1464 !XLOG_FORCED_SHUTDOWN(log
)) {
1465 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
1466 "xlog_unalloc_log: (cnt: %d, total: %d)",
1467 log
->l_ticket_cnt
, log
->l_ticket_tcnt
);
1468 /* ASSERT(log->l_ticket_cnt == log->l_ticket_tcnt); */
1471 tic
= log
->l_unmount_free
;
1473 next_tic
= tic
->t_next
;
1474 kmem_free(tic
, NBPP
);
1478 xfs_buf_free(log
->l_xbuf
);
1479 #ifdef XFS_LOG_TRACE
1480 if (log
->l_trace
!= NULL
) {
1481 ktrace_free(log
->l_trace
);
1483 if (log
->l_grant_trace
!= NULL
) {
1484 ktrace_free(log
->l_grant_trace
);
1487 log
->l_mp
->m_log
= NULL
;
1488 kmem_free(log
, sizeof(xlog_t
));
1489 } /* xlog_unalloc_log */
1492 * Update counters atomically now that memcpy is done.
1496 xlog_state_finish_copy(xlog_t
*log
,
1497 xlog_in_core_t
*iclog
,
1505 iclog
->ic_header
.h_num_logops
+= record_cnt
;
1506 iclog
->ic_offset
+= copy_bytes
;
1509 } /* xlog_state_finish_copy */
1515 * print out info relating to regions written which consume
1518 #if defined(XFS_LOG_RES_DEBUG)
1520 xlog_print_tic_res(xfs_mount_t
*mp
, xlog_ticket_t
*ticket
)
1523 uint ophdr_spc
= ticket
->t_res_num_ophdrs
* (uint
)sizeof(xlog_op_header_t
);
1525 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1526 static char *res_type_str
[XLOG_REG_TYPE_MAX
] = {
1547 static char *trans_type_str
[XFS_TRANS_TYPE_MAX
] = {
1590 xfs_fs_cmn_err(CE_WARN
, mp
,
1591 "xfs_log_write: reservation summary:\n"
1592 " trans type = %s (%u)\n"
1593 " unit res = %d bytes\n"
1594 " current res = %d bytes\n"
1595 " total reg = %u bytes (o/flow = %u bytes)\n"
1596 " ophdrs = %u (ophdr space = %u bytes)\n"
1597 " ophdr + reg = %u bytes\n"
1598 " num regions = %u\n",
1599 ((ticket
->t_trans_type
<= 0 ||
1600 ticket
->t_trans_type
> XFS_TRANS_TYPE_MAX
) ?
1601 "bad-trans-type" : trans_type_str
[ticket
->t_trans_type
-1]),
1602 ticket
->t_trans_type
,
1605 ticket
->t_res_arr_sum
, ticket
->t_res_o_flow
,
1606 ticket
->t_res_num_ophdrs
, ophdr_spc
,
1607 ticket
->t_res_arr_sum
+
1608 ticket
->t_res_o_flow
+ ophdr_spc
,
1611 for (i
= 0; i
< ticket
->t_res_num
; i
++) {
1612 uint r_type
= ticket
->t_res_arr
[i
].r_type
;
1614 "region[%u]: %s - %u bytes\n",
1616 ((r_type
<= 0 || r_type
> XLOG_REG_TYPE_MAX
) ?
1617 "bad-rtype" : res_type_str
[r_type
-1]),
1618 ticket
->t_res_arr
[i
].r_len
);
1622 #define xlog_print_tic_res(mp, ticket)
1626 * Write some region out to in-core log
1628 * This will be called when writing externally provided regions or when
1629 * writing out a commit record for a given transaction.
1631 * General algorithm:
1632 * 1. Find total length of this write. This may include adding to the
1633 * lengths passed in.
1634 * 2. Check whether we violate the tickets reservation.
1635 * 3. While writing to this iclog
1636 * A. Reserve as much space in this iclog as can get
1637 * B. If this is first write, save away start lsn
1638 * C. While writing this region:
1639 * 1. If first write of transaction, write start record
1640 * 2. Write log operation header (header per region)
1641 * 3. Find out if we can fit entire region into this iclog
1642 * 4. Potentially, verify destination memcpy ptr
1643 * 5. Memcpy (partial) region
1644 * 6. If partial copy, release iclog; otherwise, continue
1645 * copying more regions into current iclog
1646 * 4. Mark want sync bit (in simulation mode)
1647 * 5. Release iclog for potential flush to on-disk log.
1650 * 1. Panic if reservation is overrun. This should never happen since
1651 * reservation amounts are generated internal to the filesystem.
1653 * 1. Tickets are single threaded data structures.
1654 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1655 * syncing routine. When a single log_write region needs to span
1656 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1657 * on all log operation writes which don't contain the end of the
1658 * region. The XLOG_END_TRANS bit is used for the in-core log
1659 * operation which contains the end of the continued log_write region.
1660 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1661 * we don't really know exactly how much space will be used. As a result,
1662 * we don't update ic_offset until the end when we know exactly how many
1663 * bytes have been written out.
1666 xlog_write(xfs_mount_t
* mp
,
1667 xfs_log_iovec_t reg
[],
1669 xfs_log_ticket_t tic
,
1670 xfs_lsn_t
*start_lsn
,
1671 xlog_in_core_t
**commit_iclog
,
1674 xlog_t
*log
= mp
->m_log
;
1675 xlog_ticket_t
*ticket
= (xlog_ticket_t
*)tic
;
1676 xlog_op_header_t
*logop_head
; /* ptr to log operation header */
1677 xlog_in_core_t
*iclog
; /* ptr to current in-core log */
1678 __psint_t ptr
; /* copy address into data region */
1679 int len
; /* # xlog_write() bytes 2 still copy */
1680 int index
; /* region index currently copying */
1681 int log_offset
; /* offset (from 0) into data region */
1682 int start_rec_copy
; /* # bytes to copy for start record */
1683 int partial_copy
; /* did we split a region? */
1684 int partial_copy_len
;/* # bytes copied if split region */
1685 int need_copy
; /* # bytes need to memcpy this region */
1686 int copy_len
; /* # bytes actually memcpy'ing */
1687 int copy_off
; /* # bytes from entry start */
1688 int contwr
; /* continued write of in-core log? */
1690 int record_cnt
= 0, data_cnt
= 0;
1692 partial_copy_len
= partial_copy
= 0;
1694 /* Calculate potential maximum space. Each region gets its own
1695 * xlog_op_header_t and may need to be double word aligned.
1698 if (ticket
->t_flags
& XLOG_TIC_INITED
) { /* acct for start rec of xact */
1699 len
+= sizeof(xlog_op_header_t
);
1700 XLOG_TIC_ADD_OPHDR(ticket
);
1703 for (index
= 0; index
< nentries
; index
++) {
1704 len
+= sizeof(xlog_op_header_t
); /* each region gets >= 1 */
1705 XLOG_TIC_ADD_OPHDR(ticket
);
1706 len
+= reg
[index
].i_len
;
1707 XLOG_TIC_ADD_REGION(ticket
, reg
[index
].i_len
, reg
[index
].i_type
);
1709 contwr
= *start_lsn
= 0;
1711 if (ticket
->t_curr_res
< len
) {
1712 xlog_print_tic_res(mp
, ticket
);
1715 "xfs_log_write: reservation ran out. Need to up reservation");
1717 /* Customer configurable panic */
1718 xfs_cmn_err(XFS_PTAG_LOGRES
, CE_ALERT
, mp
,
1719 "xfs_log_write: reservation ran out. Need to up reservation");
1720 /* If we did not panic, shutdown the filesystem */
1721 xfs_force_shutdown(mp
, XFS_CORRUPT_INCORE
);
1724 ticket
->t_curr_res
-= len
;
1726 for (index
= 0; index
< nentries
; ) {
1727 if ((error
= xlog_state_get_iclog_space(log
, len
, &iclog
, ticket
,
1728 &contwr
, &log_offset
)))
1731 ASSERT(log_offset
<= iclog
->ic_size
- 1);
1732 ptr
= (__psint_t
) ((char *)iclog
->ic_datap
+log_offset
);
1734 /* start_lsn is the first lsn written to. That's all we need. */
1736 *start_lsn
= INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
);
1738 /* This loop writes out as many regions as can fit in the amount
1739 * of space which was allocated by xlog_state_get_iclog_space().
1741 while (index
< nentries
) {
1742 ASSERT(reg
[index
].i_len
% sizeof(__int32_t
) == 0);
1743 ASSERT((__psint_t
)ptr
% sizeof(__int32_t
) == 0);
1746 /* If first write for transaction, insert start record.
1747 * We can't be trying to commit if we are inited. We can't
1748 * have any "partial_copy" if we are inited.
1750 if (ticket
->t_flags
& XLOG_TIC_INITED
) {
1751 logop_head
= (xlog_op_header_t
*)ptr
;
1752 INT_SET(logop_head
->oh_tid
, ARCH_CONVERT
, ticket
->t_tid
);
1753 logop_head
->oh_clientid
= ticket
->t_clientid
;
1754 logop_head
->oh_len
= 0;
1755 logop_head
->oh_flags
= XLOG_START_TRANS
;
1756 logop_head
->oh_res2
= 0;
1757 ticket
->t_flags
&= ~XLOG_TIC_INITED
; /* clear bit */
1760 start_rec_copy
= sizeof(xlog_op_header_t
);
1761 xlog_write_adv_cnt(ptr
, len
, log_offset
, start_rec_copy
);
1764 /* Copy log operation header directly into data section */
1765 logop_head
= (xlog_op_header_t
*)ptr
;
1766 INT_SET(logop_head
->oh_tid
, ARCH_CONVERT
, ticket
->t_tid
);
1767 logop_head
->oh_clientid
= ticket
->t_clientid
;
1768 logop_head
->oh_res2
= 0;
1770 /* header copied directly */
1771 xlog_write_adv_cnt(ptr
, len
, log_offset
, sizeof(xlog_op_header_t
));
1773 /* are we copying a commit or unmount record? */
1774 logop_head
->oh_flags
= flags
;
1777 * We've seen logs corrupted with bad transaction client
1778 * ids. This makes sure that XFS doesn't generate them on.
1779 * Turn this into an EIO and shut down the filesystem.
1781 switch (logop_head
->oh_clientid
) {
1782 case XFS_TRANSACTION
:
1787 xfs_fs_cmn_err(CE_WARN
, mp
,
1788 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1789 logop_head
->oh_clientid
, tic
);
1790 return XFS_ERROR(EIO
);
1793 /* Partial write last time? => (partial_copy != 0)
1794 * need_copy is the amount we'd like to copy if everything could
1795 * fit in the current memcpy.
1797 need_copy
= reg
[index
].i_len
- partial_copy_len
;
1799 copy_off
= partial_copy_len
;
1800 if (need_copy
<= iclog
->ic_size
- log_offset
) { /*complete write */
1801 INT_SET(logop_head
->oh_len
, ARCH_CONVERT
, copy_len
= need_copy
);
1803 logop_head
->oh_flags
|= (XLOG_END_TRANS
|XLOG_WAS_CONT_TRANS
);
1804 partial_copy_len
= partial_copy
= 0;
1805 } else { /* partial write */
1806 copy_len
= iclog
->ic_size
- log_offset
;
1807 INT_SET(logop_head
->oh_len
, ARCH_CONVERT
, copy_len
);
1808 logop_head
->oh_flags
|= XLOG_CONTINUE_TRANS
;
1810 logop_head
->oh_flags
|= XLOG_WAS_CONT_TRANS
;
1811 partial_copy_len
+= copy_len
;
1813 len
+= sizeof(xlog_op_header_t
); /* from splitting of region */
1814 /* account for new log op header */
1815 ticket
->t_curr_res
-= sizeof(xlog_op_header_t
);
1816 XLOG_TIC_ADD_OPHDR(ticket
);
1818 xlog_verify_dest_ptr(log
, ptr
);
1821 ASSERT(copy_len
>= 0);
1822 memcpy((xfs_caddr_t
)ptr
, reg
[index
].i_addr
+ copy_off
, copy_len
);
1823 xlog_write_adv_cnt(ptr
, len
, log_offset
, copy_len
);
1825 /* make copy_len total bytes copied, including headers */
1826 copy_len
+= start_rec_copy
+ sizeof(xlog_op_header_t
);
1828 data_cnt
+= contwr
? copy_len
: 0;
1829 if (partial_copy
) { /* copied partial region */
1830 /* already marked WANT_SYNC by xlog_state_get_iclog_space */
1831 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1832 record_cnt
= data_cnt
= 0;
1833 if ((error
= xlog_state_release_iclog(log
, iclog
)))
1835 break; /* don't increment index */
1836 } else { /* copied entire region */
1838 partial_copy_len
= partial_copy
= 0;
1840 if (iclog
->ic_size
- log_offset
<= sizeof(xlog_op_header_t
)) {
1841 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1842 record_cnt
= data_cnt
= 0;
1843 xlog_state_want_sync(log
, iclog
);
1845 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1846 *commit_iclog
= iclog
;
1847 } else if ((error
= xlog_state_release_iclog(log
, iclog
)))
1849 if (index
== nentries
)
1850 return 0; /* we are done */
1854 } /* if (partial_copy) */
1855 } /* while (index < nentries) */
1856 } /* for (index = 0; index < nentries; ) */
1859 xlog_state_finish_copy(log
, iclog
, record_cnt
, data_cnt
);
1861 ASSERT(flags
& XLOG_COMMIT_TRANS
);
1862 *commit_iclog
= iclog
;
1865 return (xlog_state_release_iclog(log
, iclog
));
1869 /*****************************************************************************
1871 * State Machine functions
1873 *****************************************************************************
1876 /* Clean iclogs starting from the head. This ordering must be
1877 * maintained, so an iclog doesn't become ACTIVE beyond one that
1878 * is SYNCING. This is also required to maintain the notion that we use
1879 * a counting semaphore to hold off would be writers to the log when every
1880 * iclog is trying to sync to disk.
1882 * State Change: DIRTY -> ACTIVE
1885 xlog_state_clean_log(xlog_t
*log
)
1887 xlog_in_core_t
*iclog
;
1890 iclog
= log
->l_iclog
;
1892 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
1893 iclog
->ic_state
= XLOG_STATE_ACTIVE
;
1894 iclog
->ic_offset
= 0;
1895 iclog
->ic_callback
= NULL
; /* don't need to free */
1897 * If the number of ops in this iclog indicate it just
1898 * contains the dummy transaction, we can
1899 * change state into IDLE (the second time around).
1900 * Otherwise we should change the state into
1902 * We don't need to cover the dummy.
1905 (INT_GET(iclog
->ic_header
.h_num_logops
, ARCH_CONVERT
) == XLOG_COVER_OPS
)) {
1909 * We have two dirty iclogs so start over
1910 * This could also be num of ops indicates
1911 * this is not the dummy going out.
1915 iclog
->ic_header
.h_num_logops
= 0;
1916 memset(iclog
->ic_header
.h_cycle_data
, 0,
1917 sizeof(iclog
->ic_header
.h_cycle_data
));
1918 iclog
->ic_header
.h_lsn
= 0;
1919 } else if (iclog
->ic_state
== XLOG_STATE_ACTIVE
)
1922 break; /* stop cleaning */
1923 iclog
= iclog
->ic_next
;
1924 } while (iclog
!= log
->l_iclog
);
1926 /* log is locked when we are called */
1928 * Change state for the dummy log recording.
1929 * We usually go to NEED. But we go to NEED2 if the changed indicates
1930 * we are done writing the dummy record.
1931 * If we are done with the second dummy recored (DONE2), then
1935 switch (log
->l_covered_state
) {
1936 case XLOG_STATE_COVER_IDLE
:
1937 case XLOG_STATE_COVER_NEED
:
1938 case XLOG_STATE_COVER_NEED2
:
1939 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
1942 case XLOG_STATE_COVER_DONE
:
1944 log
->l_covered_state
= XLOG_STATE_COVER_NEED2
;
1946 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
1949 case XLOG_STATE_COVER_DONE2
:
1951 log
->l_covered_state
= XLOG_STATE_COVER_IDLE
;
1953 log
->l_covered_state
= XLOG_STATE_COVER_NEED
;
1960 } /* xlog_state_clean_log */
1963 xlog_get_lowest_lsn(
1966 xlog_in_core_t
*lsn_log
;
1967 xfs_lsn_t lowest_lsn
, lsn
;
1969 lsn_log
= log
->l_iclog
;
1972 if (!(lsn_log
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
))) {
1973 lsn
= INT_GET(lsn_log
->ic_header
.h_lsn
, ARCH_CONVERT
);
1974 if ((lsn
&& !lowest_lsn
) ||
1975 (XFS_LSN_CMP(lsn
, lowest_lsn
) < 0)) {
1979 lsn_log
= lsn_log
->ic_next
;
1980 } while (lsn_log
!= log
->l_iclog
);
1986 xlog_state_do_callback(
1989 xlog_in_core_t
*ciclog
)
1991 xlog_in_core_t
*iclog
;
1992 xlog_in_core_t
*first_iclog
; /* used to know when we've
1993 * processed all iclogs once */
1994 xfs_log_callback_t
*cb
, *cb_next
;
1996 xfs_lsn_t lowest_lsn
;
1997 int ioerrors
; /* counter: iclogs with errors */
1998 int loopdidcallbacks
; /* flag: inner loop did callbacks*/
1999 int funcdidcallbacks
; /* flag: function did callbacks */
2000 int repeats
; /* for issuing console warnings if
2001 * looping too many times */
2005 first_iclog
= iclog
= log
->l_iclog
;
2007 funcdidcallbacks
= 0;
2012 * Scan all iclogs starting with the one pointed to by the
2013 * log. Reset this starting point each time the log is
2014 * unlocked (during callbacks).
2016 * Keep looping through iclogs until one full pass is made
2017 * without running any callbacks.
2019 first_iclog
= log
->l_iclog
;
2020 iclog
= log
->l_iclog
;
2021 loopdidcallbacks
= 0;
2026 /* skip all iclogs in the ACTIVE & DIRTY states */
2027 if (iclog
->ic_state
&
2028 (XLOG_STATE_ACTIVE
|XLOG_STATE_DIRTY
)) {
2029 iclog
= iclog
->ic_next
;
2034 * Between marking a filesystem SHUTDOWN and stopping
2035 * the log, we do flush all iclogs to disk (if there
2036 * wasn't a log I/O error). So, we do want things to
2037 * go smoothly in case of just a SHUTDOWN w/o a
2040 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
2042 * Can only perform callbacks in order. Since
2043 * this iclog is not in the DONE_SYNC/
2044 * DO_CALLBACK state, we skip the rest and
2045 * just try to clean up. If we set our iclog
2046 * to DO_CALLBACK, we will not process it when
2047 * we retry since a previous iclog is in the
2048 * CALLBACK and the state cannot change since
2049 * we are holding the LOG_LOCK.
2051 if (!(iclog
->ic_state
&
2052 (XLOG_STATE_DONE_SYNC
|
2053 XLOG_STATE_DO_CALLBACK
))) {
2054 if (ciclog
&& (ciclog
->ic_state
==
2055 XLOG_STATE_DONE_SYNC
)) {
2056 ciclog
->ic_state
= XLOG_STATE_DO_CALLBACK
;
2061 * We now have an iclog that is in either the
2062 * DO_CALLBACK or DONE_SYNC states. The other
2063 * states (WANT_SYNC, SYNCING, or CALLBACK were
2064 * caught by the above if and are going to
2065 * clean (i.e. we aren't doing their callbacks)
2070 * We will do one more check here to see if we
2071 * have chased our tail around.
2074 lowest_lsn
= xlog_get_lowest_lsn(log
);
2078 INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
)
2080 iclog
= iclog
->ic_next
;
2081 continue; /* Leave this iclog for
2085 iclog
->ic_state
= XLOG_STATE_CALLBACK
;
2089 /* l_last_sync_lsn field protected by
2090 * GRANT_LOCK. Don't worry about iclog's lsn.
2091 * No one else can be here except us.
2093 s
= GRANT_LOCK(log
);
2095 log
->l_last_sync_lsn
,
2096 INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
)
2098 log
->l_last_sync_lsn
= INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
);
2099 GRANT_UNLOCK(log
, s
);
2102 * Keep processing entries in the callback list
2103 * until we come around and it is empty. We
2104 * need to atomically see that the list is
2105 * empty and change the state to DIRTY so that
2106 * we don't miss any more callbacks being added.
2112 cb
= iclog
->ic_callback
;
2115 iclog
->ic_callback_tail
= &(iclog
->ic_callback
);
2116 iclog
->ic_callback
= NULL
;
2119 /* perform callbacks in the order given */
2120 for (; cb
!= 0; cb
= cb_next
) {
2121 cb_next
= cb
->cb_next
;
2122 cb
->cb_func(cb
->cb_arg
, aborted
);
2125 cb
= iclog
->ic_callback
;
2131 ASSERT(iclog
->ic_callback
== 0);
2132 if (!(iclog
->ic_state
& XLOG_STATE_IOERROR
))
2133 iclog
->ic_state
= XLOG_STATE_DIRTY
;
2136 * Transition from DIRTY to ACTIVE if applicable.
2137 * NOP if STATE_IOERROR.
2139 xlog_state_clean_log(log
);
2141 /* wake up threads waiting in xfs_log_force() */
2142 sv_broadcast(&iclog
->ic_forcesema
);
2144 iclog
= iclog
->ic_next
;
2145 } while (first_iclog
!= iclog
);
2146 if (repeats
&& (repeats
% 10) == 0) {
2147 xfs_fs_cmn_err(CE_WARN
, log
->l_mp
,
2148 "xlog_state_do_callback: looping %d", repeats
);
2150 } while (!ioerrors
&& loopdidcallbacks
);
2153 * make one last gasp attempt to see if iclogs are being left in
2157 if (funcdidcallbacks
) {
2158 first_iclog
= iclog
= log
->l_iclog
;
2160 ASSERT(iclog
->ic_state
!= XLOG_STATE_DO_CALLBACK
);
2162 * Terminate the loop if iclogs are found in states
2163 * which will cause other threads to clean up iclogs.
2165 * SYNCING - i/o completion will go through logs
2166 * DONE_SYNC - interrupt thread should be waiting for
2168 * IOERROR - give up hope all ye who enter here
2170 if (iclog
->ic_state
== XLOG_STATE_WANT_SYNC
||
2171 iclog
->ic_state
== XLOG_STATE_SYNCING
||
2172 iclog
->ic_state
== XLOG_STATE_DONE_SYNC
||
2173 iclog
->ic_state
== XLOG_STATE_IOERROR
)
2175 iclog
= iclog
->ic_next
;
2176 } while (first_iclog
!= iclog
);
2180 if (log
->l_iclog
->ic_state
& (XLOG_STATE_ACTIVE
|XLOG_STATE_IOERROR
)) {
2181 flushcnt
= log
->l_flushcnt
;
2182 log
->l_flushcnt
= 0;
2186 vsema(&log
->l_flushsema
);
2187 } /* xlog_state_do_callback */
2191 * Finish transitioning this iclog to the dirty state.
2193 * Make sure that we completely execute this routine only when this is
2194 * the last call to the iclog. There is a good chance that iclog flushes,
2195 * when we reach the end of the physical log, get turned into 2 separate
2196 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2197 * routine. By using the reference count bwritecnt, we guarantee that only
2198 * the second completion goes through.
2200 * Callbacks could take time, so they are done outside the scope of the
2201 * global state machine log lock. Assume that the calls to cvsema won't
2202 * take a long time. At least we know it won't sleep.
2205 xlog_state_done_syncing(
2206 xlog_in_core_t
*iclog
,
2209 xlog_t
*log
= iclog
->ic_log
;
2214 ASSERT(iclog
->ic_state
== XLOG_STATE_SYNCING
||
2215 iclog
->ic_state
== XLOG_STATE_IOERROR
);
2216 ASSERT(iclog
->ic_refcnt
== 0);
2217 ASSERT(iclog
->ic_bwritecnt
== 1 || iclog
->ic_bwritecnt
== 2);
2221 * If we got an error, either on the first buffer, or in the case of
2222 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2223 * and none should ever be attempted to be written to disk
2226 if (iclog
->ic_state
!= XLOG_STATE_IOERROR
) {
2227 if (--iclog
->ic_bwritecnt
== 1) {
2231 iclog
->ic_state
= XLOG_STATE_DONE_SYNC
;
2235 * Someone could be sleeping prior to writing out the next
2236 * iclog buffer, we wake them all, one will get to do the
2237 * I/O, the others get to wait for the result.
2239 sv_broadcast(&iclog
->ic_writesema
);
2241 xlog_state_do_callback(log
, aborted
, iclog
); /* also cleans log */
2242 } /* xlog_state_done_syncing */
2246 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2247 * sleep. The flush semaphore is set to the number of in-core buffers and
2248 * decremented around disk syncing. Therefore, if all buffers are syncing,
2249 * this semaphore will cause new writes to sleep until a sync completes.
2250 * Otherwise, this code just does p() followed by v(). This approximates
2251 * a sleep/wakeup except we can't race.
2253 * The in-core logs are used in a circular fashion. They are not used
2254 * out-of-order even when an iclog past the head is free.
2257 * * log_offset where xlog_write() can start writing into the in-core
2259 * * in-core log pointer to which xlog_write() should write.
2260 * * boolean indicating this is a continued write to an in-core log.
2261 * If this is the last write, then the in-core log's offset field
2262 * needs to be incremented, depending on the amount of data which
2266 xlog_state_get_iclog_space(xlog_t
*log
,
2268 xlog_in_core_t
**iclogp
,
2269 xlog_ticket_t
*ticket
,
2270 int *continued_write
,
2275 xlog_rec_header_t
*head
;
2276 xlog_in_core_t
*iclog
;
2281 if (XLOG_FORCED_SHUTDOWN(log
)) {
2283 return XFS_ERROR(EIO
);
2286 iclog
= log
->l_iclog
;
2287 if (! (iclog
->ic_state
== XLOG_STATE_ACTIVE
)) {
2290 xlog_trace_iclog(iclog
, XLOG_TRACE_SLEEP_FLUSH
);
2291 XFS_STATS_INC(xs_log_noiclogs
);
2292 /* Ensure that log writes happen */
2293 psema(&log
->l_flushsema
, PINOD
);
2296 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2297 head
= &iclog
->ic_header
;
2299 iclog
->ic_refcnt
++; /* prevents sync */
2300 log_offset
= iclog
->ic_offset
;
2302 /* On the 1st write to an iclog, figure out lsn. This works
2303 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2304 * committing to. If the offset is set, that's how many blocks
2307 if (log_offset
== 0) {
2308 ticket
->t_curr_res
-= log
->l_iclog_hsize
;
2309 XLOG_TIC_ADD_REGION(ticket
,
2311 XLOG_REG_TYPE_LRHEADER
);
2312 INT_SET(head
->h_cycle
, ARCH_CONVERT
, log
->l_curr_cycle
);
2313 ASSIGN_LSN(head
->h_lsn
, log
);
2314 ASSERT(log
->l_curr_block
>= 0);
2317 /* If there is enough room to write everything, then do it. Otherwise,
2318 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2319 * bit is on, so this will get flushed out. Don't update ic_offset
2320 * until you know exactly how many bytes get copied. Therefore, wait
2321 * until later to update ic_offset.
2323 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2324 * can fit into remaining data section.
2326 if (iclog
->ic_size
- iclog
->ic_offset
< 2*sizeof(xlog_op_header_t
)) {
2327 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2329 /* If I'm the only one writing to this iclog, sync it to disk */
2330 if (iclog
->ic_refcnt
== 1) {
2332 if ((error
= xlog_state_release_iclog(log
, iclog
)))
2341 /* Do we have enough room to write the full amount in the remainder
2342 * of this iclog? Or must we continue a write on the next iclog and
2343 * mark this iclog as completely taken? In the case where we switch
2344 * iclogs (to mark it taken), this particular iclog will release/sync
2345 * to disk in xlog_write().
2347 if (len
<= iclog
->ic_size
- iclog
->ic_offset
) {
2348 *continued_write
= 0;
2349 iclog
->ic_offset
+= len
;
2351 *continued_write
= 1;
2352 xlog_state_switch_iclogs(log
, iclog
, iclog
->ic_size
);
2356 ASSERT(iclog
->ic_offset
<= iclog
->ic_size
);
2359 *logoffsetp
= log_offset
;
2361 } /* xlog_state_get_iclog_space */
2364 * Atomically get the log space required for a log ticket.
2366 * Once a ticket gets put onto the reserveq, it will only return after
2367 * the needed reservation is satisfied.
2370 xlog_grant_log_space(xlog_t
*log
,
2382 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2383 panic("grant Recovery problem");
2386 /* Is there space or do we need to sleep? */
2387 s
= GRANT_LOCK(log
);
2388 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: enter");
2390 /* something is already sleeping; insert new transaction at end */
2391 if (log
->l_reserve_headq
) {
2392 XLOG_INS_TICKETQ(log
->l_reserve_headq
, tic
);
2393 xlog_trace_loggrant(log
, tic
,
2394 "xlog_grant_log_space: sleep 1");
2396 * Gotta check this before going to sleep, while we're
2397 * holding the grant lock.
2399 if (XLOG_FORCED_SHUTDOWN(log
))
2402 XFS_STATS_INC(xs_sleep_logspace
);
2403 sv_wait(&tic
->t_sema
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2405 * If we got an error, and the filesystem is shutting down,
2406 * we'll catch it down below. So just continue...
2408 xlog_trace_loggrant(log
, tic
,
2409 "xlog_grant_log_space: wake 1");
2410 s
= GRANT_LOCK(log
);
2412 if (tic
->t_flags
& XFS_LOG_PERM_RESERV
)
2413 need_bytes
= tic
->t_unit_res
*tic
->t_ocnt
;
2415 need_bytes
= tic
->t_unit_res
;
2418 if (XLOG_FORCED_SHUTDOWN(log
))
2421 free_bytes
= xlog_space_left(log
, log
->l_grant_reserve_cycle
,
2422 log
->l_grant_reserve_bytes
);
2423 if (free_bytes
< need_bytes
) {
2424 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2425 XLOG_INS_TICKETQ(log
->l_reserve_headq
, tic
);
2426 xlog_trace_loggrant(log
, tic
,
2427 "xlog_grant_log_space: sleep 2");
2428 XFS_STATS_INC(xs_sleep_logspace
);
2429 sv_wait(&tic
->t_sema
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2431 if (XLOG_FORCED_SHUTDOWN(log
)) {
2432 s
= GRANT_LOCK(log
);
2436 xlog_trace_loggrant(log
, tic
,
2437 "xlog_grant_log_space: wake 2");
2438 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2439 s
= GRANT_LOCK(log
);
2441 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2442 XLOG_DEL_TICKETQ(log
->l_reserve_headq
, tic
);
2444 /* we've got enough space */
2445 XLOG_GRANT_ADD_SPACE(log
, need_bytes
, 'w');
2446 XLOG_GRANT_ADD_SPACE(log
, need_bytes
, 'r');
2448 tail_lsn
= log
->l_tail_lsn
;
2450 * Check to make sure the grant write head didn't just over lap the
2451 * tail. If the cycles are the same, we can't be overlapping.
2452 * Otherwise, make sure that the cycles differ by exactly one and
2453 * check the byte count.
2455 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2456 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2457 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2460 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: exit");
2461 xlog_verify_grant_head(log
, 1);
2462 GRANT_UNLOCK(log
, s
);
2466 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2467 XLOG_DEL_TICKETQ(log
->l_reserve_headq
, tic
);
2468 xlog_trace_loggrant(log
, tic
, "xlog_grant_log_space: err_ret");
2470 * If we are failing, make sure the ticket doesn't have any
2471 * current reservations. We don't want to add this back when
2472 * the ticket/transaction gets cancelled.
2474 tic
->t_curr_res
= 0;
2475 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2476 GRANT_UNLOCK(log
, s
);
2477 return XFS_ERROR(EIO
);
2478 } /* xlog_grant_log_space */
2482 * Replenish the byte reservation required by moving the grant write head.
2487 xlog_regrant_write_log_space(xlog_t
*log
,
2491 int free_bytes
, need_bytes
;
2492 xlog_ticket_t
*ntic
;
2497 tic
->t_curr_res
= tic
->t_unit_res
;
2498 XLOG_TIC_RESET_RES(tic
);
2504 if (log
->l_flags
& XLOG_ACTIVE_RECOVERY
)
2505 panic("regrant Recovery problem");
2508 s
= GRANT_LOCK(log
);
2509 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: enter");
2511 if (XLOG_FORCED_SHUTDOWN(log
))
2514 /* If there are other waiters on the queue then give them a
2515 * chance at logspace before us. Wake up the first waiters,
2516 * if we do not wake up all the waiters then go to sleep waiting
2517 * for more free space, otherwise try to get some space for
2521 if ((ntic
= log
->l_write_headq
)) {
2522 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2523 log
->l_grant_write_bytes
);
2525 ASSERT(ntic
->t_flags
& XLOG_TIC_PERM_RESERV
);
2527 if (free_bytes
< ntic
->t_unit_res
)
2529 free_bytes
-= ntic
->t_unit_res
;
2530 sv_signal(&ntic
->t_sema
);
2531 ntic
= ntic
->t_next
;
2532 } while (ntic
!= log
->l_write_headq
);
2534 if (ntic
!= log
->l_write_headq
) {
2535 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2536 XLOG_INS_TICKETQ(log
->l_write_headq
, tic
);
2538 xlog_trace_loggrant(log
, tic
,
2539 "xlog_regrant_write_log_space: sleep 1");
2540 XFS_STATS_INC(xs_sleep_logspace
);
2541 sv_wait(&tic
->t_sema
, PINOD
|PLTWAIT
,
2542 &log
->l_grant_lock
, s
);
2544 /* If we're shutting down, this tic is already
2546 if (XLOG_FORCED_SHUTDOWN(log
)) {
2547 s
= GRANT_LOCK(log
);
2551 xlog_trace_loggrant(log
, tic
,
2552 "xlog_regrant_write_log_space: wake 1");
2553 xlog_grant_push_ail(log
->l_mp
, tic
->t_unit_res
);
2554 s
= GRANT_LOCK(log
);
2558 need_bytes
= tic
->t_unit_res
;
2561 if (XLOG_FORCED_SHUTDOWN(log
))
2564 free_bytes
= xlog_space_left(log
, log
->l_grant_write_cycle
,
2565 log
->l_grant_write_bytes
);
2566 if (free_bytes
< need_bytes
) {
2567 if ((tic
->t_flags
& XLOG_TIC_IN_Q
) == 0)
2568 XLOG_INS_TICKETQ(log
->l_write_headq
, tic
);
2569 XFS_STATS_INC(xs_sleep_logspace
);
2570 sv_wait(&tic
->t_sema
, PINOD
|PLTWAIT
, &log
->l_grant_lock
, s
);
2572 /* If we're shutting down, this tic is already off the queue */
2573 if (XLOG_FORCED_SHUTDOWN(log
)) {
2574 s
= GRANT_LOCK(log
);
2578 xlog_trace_loggrant(log
, tic
,
2579 "xlog_regrant_write_log_space: wake 2");
2580 xlog_grant_push_ail(log
->l_mp
, need_bytes
);
2581 s
= GRANT_LOCK(log
);
2583 } else if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2584 XLOG_DEL_TICKETQ(log
->l_write_headq
, tic
);
2586 XLOG_GRANT_ADD_SPACE(log
, need_bytes
, 'w'); /* we've got enough space */
2588 tail_lsn
= log
->l_tail_lsn
;
2589 if (CYCLE_LSN(tail_lsn
) != log
->l_grant_write_cycle
) {
2590 ASSERT(log
->l_grant_write_cycle
-1 == CYCLE_LSN(tail_lsn
));
2591 ASSERT(log
->l_grant_write_bytes
<= BBTOB(BLOCK_LSN(tail_lsn
)));
2595 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: exit");
2596 xlog_verify_grant_head(log
, 1);
2597 GRANT_UNLOCK(log
, s
);
2602 if (tic
->t_flags
& XLOG_TIC_IN_Q
)
2603 XLOG_DEL_TICKETQ(log
->l_reserve_headq
, tic
);
2604 xlog_trace_loggrant(log
, tic
, "xlog_regrant_write_log_space: err_ret");
2606 * If we are failing, make sure the ticket doesn't have any
2607 * current reservations. We don't want to add this back when
2608 * the ticket/transaction gets cancelled.
2610 tic
->t_curr_res
= 0;
2611 tic
->t_cnt
= 0; /* ungrant will give back unit_res * t_cnt. */
2612 GRANT_UNLOCK(log
, s
);
2613 return XFS_ERROR(EIO
);
2614 } /* xlog_regrant_write_log_space */
2617 /* The first cnt-1 times through here we don't need to
2618 * move the grant write head because the permanent
2619 * reservation has reserved cnt times the unit amount.
2620 * Release part of current permanent unit reservation and
2621 * reset current reservation to be one units worth. Also
2622 * move grant reservation head forward.
2625 xlog_regrant_reserve_log_space(xlog_t
*log
,
2626 xlog_ticket_t
*ticket
)
2630 xlog_trace_loggrant(log
, ticket
,
2631 "xlog_regrant_reserve_log_space: enter");
2632 if (ticket
->t_cnt
> 0)
2635 s
= GRANT_LOCK(log
);
2636 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_curr_res
, 'w');
2637 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_curr_res
, 'r');
2638 ticket
->t_curr_res
= ticket
->t_unit_res
;
2639 XLOG_TIC_RESET_RES(ticket
);
2640 xlog_trace_loggrant(log
, ticket
,
2641 "xlog_regrant_reserve_log_space: sub current res");
2642 xlog_verify_grant_head(log
, 1);
2644 /* just return if we still have some of the pre-reserved space */
2645 if (ticket
->t_cnt
> 0) {
2646 GRANT_UNLOCK(log
, s
);
2650 XLOG_GRANT_ADD_SPACE(log
, ticket
->t_unit_res
, 'r');
2651 xlog_trace_loggrant(log
, ticket
,
2652 "xlog_regrant_reserve_log_space: exit");
2653 xlog_verify_grant_head(log
, 0);
2654 GRANT_UNLOCK(log
, s
);
2655 ticket
->t_curr_res
= ticket
->t_unit_res
;
2656 XLOG_TIC_RESET_RES(ticket
);
2657 } /* xlog_regrant_reserve_log_space */
2661 * Give back the space left from a reservation.
2663 * All the information we need to make a correct determination of space left
2664 * is present. For non-permanent reservations, things are quite easy. The
2665 * count should have been decremented to zero. We only need to deal with the
2666 * space remaining in the current reservation part of the ticket. If the
2667 * ticket contains a permanent reservation, there may be left over space which
2668 * needs to be released. A count of N means that N-1 refills of the current
2669 * reservation can be done before we need to ask for more space. The first
2670 * one goes to fill up the first current reservation. Once we run out of
2671 * space, the count will stay at zero and the only space remaining will be
2672 * in the current reservation field.
2675 xlog_ungrant_log_space(xlog_t
*log
,
2676 xlog_ticket_t
*ticket
)
2680 if (ticket
->t_cnt
> 0)
2683 s
= GRANT_LOCK(log
);
2684 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: enter");
2686 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_curr_res
, 'w');
2687 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_curr_res
, 'r');
2689 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: sub current");
2691 /* If this is a permanent reservation ticket, we may be able to free
2692 * up more space based on the remaining count.
2694 if (ticket
->t_cnt
> 0) {
2695 ASSERT(ticket
->t_flags
& XLOG_TIC_PERM_RESERV
);
2696 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_unit_res
*ticket
->t_cnt
,'w');
2697 XLOG_GRANT_SUB_SPACE(log
, ticket
->t_unit_res
*ticket
->t_cnt
,'r');
2700 xlog_trace_loggrant(log
, ticket
, "xlog_ungrant_log_space: exit");
2701 xlog_verify_grant_head(log
, 1);
2702 GRANT_UNLOCK(log
, s
);
2703 xfs_log_move_tail(log
->l_mp
, 1);
2704 } /* xlog_ungrant_log_space */
2708 * Atomically put back used ticket.
2711 xlog_state_put_ticket(xlog_t
*log
,
2717 xlog_ticket_put(log
, tic
);
2719 } /* xlog_state_put_ticket */
2722 * Flush iclog to disk if this is the last reference to the given iclog and
2723 * the WANT_SYNC bit is set.
2725 * When this function is entered, the iclog is not necessarily in the
2726 * WANT_SYNC state. It may be sitting around waiting to get filled.
2731 xlog_state_release_iclog(xlog_t
*log
,
2732 xlog_in_core_t
*iclog
)
2735 int sync
= 0; /* do we sync? */
2737 xlog_assign_tail_lsn(log
->l_mp
);
2741 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2743 return XFS_ERROR(EIO
);
2746 ASSERT(iclog
->ic_refcnt
> 0);
2747 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2748 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
);
2750 if (--iclog
->ic_refcnt
== 0 &&
2751 iclog
->ic_state
== XLOG_STATE_WANT_SYNC
) {
2753 iclog
->ic_state
= XLOG_STATE_SYNCING
;
2754 INT_SET(iclog
->ic_header
.h_tail_lsn
, ARCH_CONVERT
, log
->l_tail_lsn
);
2755 xlog_verify_tail_lsn(log
, iclog
, log
->l_tail_lsn
);
2756 /* cycle incremented when incrementing curr_block */
2762 * We let the log lock go, so it's possible that we hit a log I/O
2763 * error or someother SHUTDOWN condition that marks the iclog
2764 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2765 * this iclog has consistent data, so we ignore IOERROR
2766 * flags after this point.
2769 return xlog_sync(log
, iclog
);
2773 } /* xlog_state_release_iclog */
2777 * This routine will mark the current iclog in the ring as WANT_SYNC
2778 * and move the current iclog pointer to the next iclog in the ring.
2779 * When this routine is called from xlog_state_get_iclog_space(), the
2780 * exact size of the iclog has not yet been determined. All we know is
2781 * that every data block. We have run out of space in this log record.
2784 xlog_state_switch_iclogs(xlog_t
*log
,
2785 xlog_in_core_t
*iclog
,
2788 ASSERT(iclog
->ic_state
== XLOG_STATE_ACTIVE
);
2790 eventual_size
= iclog
->ic_offset
;
2791 iclog
->ic_state
= XLOG_STATE_WANT_SYNC
;
2792 INT_SET(iclog
->ic_header
.h_prev_block
, ARCH_CONVERT
, log
->l_prev_block
);
2793 log
->l_prev_block
= log
->l_curr_block
;
2794 log
->l_prev_cycle
= log
->l_curr_cycle
;
2796 /* roll log?: ic_offset changed later */
2797 log
->l_curr_block
+= BTOBB(eventual_size
)+BTOBB(log
->l_iclog_hsize
);
2799 /* Round up to next log-sunit */
2800 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) &&
2801 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
2802 __uint32_t sunit_bb
= BTOBB(log
->l_mp
->m_sb
.sb_logsunit
);
2803 log
->l_curr_block
= roundup(log
->l_curr_block
, sunit_bb
);
2806 if (log
->l_curr_block
>= log
->l_logBBsize
) {
2807 log
->l_curr_cycle
++;
2808 if (log
->l_curr_cycle
== XLOG_HEADER_MAGIC_NUM
)
2809 log
->l_curr_cycle
++;
2810 log
->l_curr_block
-= log
->l_logBBsize
;
2811 ASSERT(log
->l_curr_block
>= 0);
2813 ASSERT(iclog
== log
->l_iclog
);
2814 log
->l_iclog
= iclog
->ic_next
;
2815 } /* xlog_state_switch_iclogs */
2819 * Write out all data in the in-core log as of this exact moment in time.
2821 * Data may be written to the in-core log during this call. However,
2822 * we don't guarantee this data will be written out. A change from past
2823 * implementation means this routine will *not* write out zero length LRs.
2825 * Basically, we try and perform an intelligent scan of the in-core logs.
2826 * If we determine there is no flushable data, we just return. There is no
2827 * flushable data if:
2829 * 1. the current iclog is active and has no data; the previous iclog
2830 * is in the active or dirty state.
2831 * 2. the current iclog is drity, and the previous iclog is in the
2832 * active or dirty state.
2834 * We may sleep (call psema) if:
2836 * 1. the current iclog is not in the active nor dirty state.
2837 * 2. the current iclog dirty, and the previous iclog is not in the
2838 * active nor dirty state.
2839 * 3. the current iclog is active, and there is another thread writing
2840 * to this particular iclog.
2841 * 4. a) the current iclog is active and has no other writers
2842 * b) when we return from flushing out this iclog, it is still
2843 * not in the active nor dirty state.
2846 xlog_state_sync_all(xlog_t
*log
, uint flags
, int *log_flushed
)
2848 xlog_in_core_t
*iclog
;
2854 iclog
= log
->l_iclog
;
2855 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2857 return XFS_ERROR(EIO
);
2860 /* If the head iclog is not active nor dirty, we just attach
2861 * ourselves to the head and go to sleep.
2863 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2864 iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2866 * If the head is dirty or (active and empty), then
2867 * we need to look at the previous iclog. If the previous
2868 * iclog is active or dirty we are done. There is nothing
2869 * to sync out. Otherwise, we attach ourselves to the
2870 * previous iclog and go to sleep.
2872 if (iclog
->ic_state
== XLOG_STATE_DIRTY
||
2873 (iclog
->ic_refcnt
== 0 && iclog
->ic_offset
== 0)) {
2874 iclog
= iclog
->ic_prev
;
2875 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
||
2876 iclog
->ic_state
== XLOG_STATE_DIRTY
)
2881 if (iclog
->ic_refcnt
== 0) {
2882 /* We are the only one with access to this
2883 * iclog. Flush it out now. There should
2884 * be a roundoff of zero to show that someone
2885 * has already taken care of the roundoff from
2886 * the previous sync.
2889 lsn
= INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
);
2890 xlog_state_switch_iclogs(log
, iclog
, 0);
2893 if (xlog_state_release_iclog(log
, iclog
))
2894 return XFS_ERROR(EIO
);
2897 if (INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
) == lsn
&&
2898 iclog
->ic_state
!= XLOG_STATE_DIRTY
)
2903 /* Someone else is writing to this iclog.
2904 * Use its call to flush out the data. However,
2905 * the other thread may not force out this LR,
2906 * so we mark it WANT_SYNC.
2908 xlog_state_switch_iclogs(log
, iclog
, 0);
2914 /* By the time we come around again, the iclog could've been filled
2915 * which would give it another lsn. If we have a new lsn, just
2916 * return because the relevant data has been flushed.
2919 if (flags
& XFS_LOG_SYNC
) {
2921 * We must check if we're shutting down here, before
2922 * we wait, while we're holding the LOG_LOCK.
2923 * Then we check again after waking up, in case our
2924 * sleep was disturbed by a bad news.
2926 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2928 return XFS_ERROR(EIO
);
2930 XFS_STATS_INC(xs_log_force_sleep
);
2931 sv_wait(&iclog
->ic_forcesema
, PINOD
, &log
->l_icloglock
, s
);
2933 * No need to grab the log lock here since we're
2934 * only deciding whether or not to return EIO
2935 * and the memory read should be atomic.
2937 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
2938 return XFS_ERROR(EIO
);
2947 } /* xlog_state_sync_all */
2951 * Used by code which implements synchronous log forces.
2953 * Find in-core log with lsn.
2954 * If it is in the DIRTY state, just return.
2955 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
2956 * state and go to sleep or return.
2957 * If it is in any other state, go to sleep or return.
2959 * If filesystem activity goes to zero, the iclog will get flushed only by
2963 xlog_state_sync(xlog_t
*log
,
2968 xlog_in_core_t
*iclog
;
2969 int already_slept
= 0;
2975 iclog
= log
->l_iclog
;
2977 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
2979 return XFS_ERROR(EIO
);
2983 if (INT_GET(iclog
->ic_header
.h_lsn
, ARCH_CONVERT
) != lsn
) {
2984 iclog
= iclog
->ic_next
;
2988 if (iclog
->ic_state
== XLOG_STATE_DIRTY
) {
2993 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
2995 * We sleep here if we haven't already slept (e.g.
2996 * this is the first time we've looked at the correct
2997 * iclog buf) and the buffer before us is going to
2998 * be sync'ed. The reason for this is that if we
2999 * are doing sync transactions here, by waiting for
3000 * the previous I/O to complete, we can allow a few
3001 * more transactions into this iclog before we close
3004 * Otherwise, we mark the buffer WANT_SYNC, and bump
3005 * up the refcnt so we can release the log (which drops
3006 * the ref count). The state switch keeps new transaction
3007 * commits from using this buffer. When the current commits
3008 * finish writing into the buffer, the refcount will drop to
3009 * zero and the buffer will go out then.
3011 if (!already_slept
&&
3012 (iclog
->ic_prev
->ic_state
& (XLOG_STATE_WANT_SYNC
|
3013 XLOG_STATE_SYNCING
))) {
3014 ASSERT(!(iclog
->ic_state
& XLOG_STATE_IOERROR
));
3015 XFS_STATS_INC(xs_log_force_sleep
);
3016 sv_wait(&iclog
->ic_prev
->ic_writesema
, PSWP
,
3017 &log
->l_icloglock
, s
);
3023 xlog_state_switch_iclogs(log
, iclog
, 0);
3025 if (xlog_state_release_iclog(log
, iclog
))
3026 return XFS_ERROR(EIO
);
3032 if ((flags
& XFS_LOG_SYNC
) && /* sleep */
3033 !(iclog
->ic_state
& (XLOG_STATE_ACTIVE
| XLOG_STATE_DIRTY
))) {
3036 * Don't wait on the forcesema if we know that we've
3037 * gotten a log write error.
3039 if (iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3041 return XFS_ERROR(EIO
);
3043 XFS_STATS_INC(xs_log_force_sleep
);
3044 sv_wait(&iclog
->ic_forcesema
, PSWP
, &log
->l_icloglock
, s
);
3046 * No need to grab the log lock here since we're
3047 * only deciding whether or not to return EIO
3048 * and the memory read should be atomic.
3050 if (iclog
->ic_state
& XLOG_STATE_IOERROR
)
3051 return XFS_ERROR(EIO
);
3053 } else { /* just return */
3058 } while (iclog
!= log
->l_iclog
);
3062 } /* xlog_state_sync */
3066 * Called when we want to mark the current iclog as being ready to sync to
3070 xlog_state_want_sync(xlog_t
*log
, xlog_in_core_t
*iclog
)
3076 if (iclog
->ic_state
== XLOG_STATE_ACTIVE
) {
3077 xlog_state_switch_iclogs(log
, iclog
, 0);
3079 ASSERT(iclog
->ic_state
&
3080 (XLOG_STATE_WANT_SYNC
|XLOG_STATE_IOERROR
));
3084 } /* xlog_state_want_sync */
3088 /*****************************************************************************
3092 *****************************************************************************
3096 * Algorithm doesn't take into account page size. ;-(
3099 xlog_state_ticket_alloc(xlog_t
*log
)
3101 xlog_ticket_t
*t_list
;
3102 xlog_ticket_t
*next
;
3104 uint i
= (NBPP
/ sizeof(xlog_ticket_t
)) - 2;
3108 * The kmem_zalloc may sleep, so we shouldn't be holding the
3109 * global lock. XXXmiken: may want to use zone allocator.
3111 buf
= (xfs_caddr_t
) kmem_zalloc(NBPP
, KM_SLEEP
);
3115 /* Attach 1st ticket to Q, so we can keep track of allocated memory */
3116 t_list
= (xlog_ticket_t
*)buf
;
3117 t_list
->t_next
= log
->l_unmount_free
;
3118 log
->l_unmount_free
= t_list
++;
3119 log
->l_ticket_cnt
++;
3120 log
->l_ticket_tcnt
++;
3122 /* Next ticket becomes first ticket attached to ticket free list */
3123 if (log
->l_freelist
!= NULL
) {
3124 ASSERT(log
->l_tail
!= NULL
);
3125 log
->l_tail
->t_next
= t_list
;
3127 log
->l_freelist
= t_list
;
3129 log
->l_ticket_cnt
++;
3130 log
->l_ticket_tcnt
++;
3132 /* Cycle through rest of alloc'ed memory, building up free Q */
3133 for ( ; i
> 0; i
--) {
3135 t_list
->t_next
= next
;
3137 log
->l_ticket_cnt
++;
3138 log
->l_ticket_tcnt
++;
3140 t_list
->t_next
= NULL
;
3141 log
->l_tail
= t_list
;
3143 } /* xlog_state_ticket_alloc */
3147 * Put ticket into free list
3149 * Assumption: log lock is held around this call.
3152 xlog_ticket_put(xlog_t
*log
,
3153 xlog_ticket_t
*ticket
)
3155 sv_destroy(&ticket
->t_sema
);
3158 * Don't think caching will make that much difference. It's
3159 * more important to make debug easier.
3162 /* real code will want to use LIFO for caching */
3163 ticket
->t_next
= log
->l_freelist
;
3164 log
->l_freelist
= ticket
;
3165 /* no need to clear fields */
3167 /* When we debug, it is easier if tickets are cycled */
3168 ticket
->t_next
= NULL
;
3169 if (log
->l_tail
!= 0) {
3170 log
->l_tail
->t_next
= ticket
;
3172 ASSERT(log
->l_freelist
== 0);
3173 log
->l_freelist
= ticket
;
3175 log
->l_tail
= ticket
;
3177 log
->l_ticket_cnt
++;
3178 } /* xlog_ticket_put */
3182 * Grab ticket off freelist or allocation some more
3185 xlog_ticket_get(xlog_t
*log
,
3196 if (log
->l_freelist
== NULL
)
3197 xlog_state_ticket_alloc(log
); /* potentially sleep */
3200 if (log
->l_freelist
== NULL
) {
3204 tic
= log
->l_freelist
;
3205 log
->l_freelist
= tic
->t_next
;
3206 if (log
->l_freelist
== NULL
)
3208 log
->l_ticket_cnt
--;
3212 * Permanent reservations have up to 'cnt'-1 active log operations
3213 * in the log. A unit in this case is the amount of space for one
3214 * of these log operations. Normal reservations have a cnt of 1
3215 * and their unit amount is the total amount of space required.
3217 * The following lines of code account for non-transaction data
3218 * which occupy space in the on-disk log.
3220 * Normal form of a transaction is:
3221 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3222 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3224 * We need to account for all the leadup data and trailer data
3225 * around the transaction data.
3226 * And then we need to account for the worst case in terms of using
3228 * The worst case will happen if:
3229 * - the placement of the transaction happens to be such that the
3230 * roundoff is at its maximum
3231 * - the transaction data is synced before the commit record is synced
3232 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3233 * Therefore the commit record is in its own Log Record.
3234 * This can happen as the commit record is called with its
3235 * own region to xlog_write().
3236 * This then means that in the worst case, roundoff can happen for
3237 * the commit-rec as well.
3238 * The commit-rec is smaller than padding in this scenario and so it is
3239 * not added separately.
3242 /* for trans header */
3243 unit_bytes
+= sizeof(xlog_op_header_t
);
3244 unit_bytes
+= sizeof(xfs_trans_header_t
);
3247 unit_bytes
+= sizeof(xlog_op_header_t
);
3249 /* for LR headers */
3250 num_headers
= ((unit_bytes
+ log
->l_iclog_size
-1) >> log
->l_iclog_size_log
);
3251 unit_bytes
+= log
->l_iclog_hsize
* num_headers
;
3253 /* for commit-rec LR header - note: padding will subsume the ophdr */
3254 unit_bytes
+= log
->l_iclog_hsize
;
3256 /* for split-recs - ophdrs added when data split over LRs */
3257 unit_bytes
+= sizeof(xlog_op_header_t
) * num_headers
;
3259 /* for roundoff padding for transaction data and one for commit record */
3260 if (XFS_SB_VERSION_HASLOGV2(&log
->l_mp
->m_sb
) &&
3261 log
->l_mp
->m_sb
.sb_logsunit
> 1) {
3262 /* log su roundoff */
3263 unit_bytes
+= 2*log
->l_mp
->m_sb
.sb_logsunit
;
3266 unit_bytes
+= 2*BBSIZE
;
3269 tic
->t_unit_res
= unit_bytes
;
3270 tic
->t_curr_res
= unit_bytes
;
3273 tic
->t_tid
= (xlog_tid_t
)((__psint_t
)tic
& 0xffffffff);
3274 tic
->t_clientid
= client
;
3275 tic
->t_flags
= XLOG_TIC_INITED
;
3276 tic
->t_trans_type
= 0;
3277 if (xflags
& XFS_LOG_PERM_RESERV
)
3278 tic
->t_flags
|= XLOG_TIC_PERM_RESERV
;
3279 sv_init(&(tic
->t_sema
), SV_DEFAULT
, "logtick");
3281 XLOG_TIC_RESET_RES(tic
);
3284 } /* xlog_ticket_get */
3287 /******************************************************************************
3289 * Log debug routines
3291 ******************************************************************************
3295 * Make sure that the destination ptr is within the valid data region of
3296 * one of the iclogs. This uses backup pointers stored in a different
3297 * part of the log in case we trash the log structure.
3300 xlog_verify_dest_ptr(xlog_t
*log
,
3306 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3307 if (ptr
>= (__psint_t
)log
->l_iclog_bak
[i
] &&
3308 ptr
<= (__psint_t
)log
->l_iclog_bak
[i
]+log
->l_iclog_size
)
3312 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3313 } /* xlog_verify_dest_ptr */
3316 xlog_verify_grant_head(xlog_t
*log
, int equals
)
3318 if (log
->l_grant_reserve_cycle
== log
->l_grant_write_cycle
) {
3320 ASSERT(log
->l_grant_reserve_bytes
>= log
->l_grant_write_bytes
);
3322 ASSERT(log
->l_grant_reserve_bytes
> log
->l_grant_write_bytes
);
3324 ASSERT(log
->l_grant_reserve_cycle
-1 == log
->l_grant_write_cycle
);
3325 ASSERT(log
->l_grant_write_bytes
>= log
->l_grant_reserve_bytes
);
3327 } /* xlog_verify_grant_head */
3329 /* check if it will fit */
3331 xlog_verify_tail_lsn(xlog_t
*log
,
3332 xlog_in_core_t
*iclog
,
3337 if (CYCLE_LSN(tail_lsn
) == log
->l_prev_cycle
) {
3339 log
->l_logBBsize
- (log
->l_prev_block
- BLOCK_LSN(tail_lsn
));
3340 if (blocks
< BTOBB(iclog
->ic_offset
)+BTOBB(log
->l_iclog_hsize
))
3341 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3343 ASSERT(CYCLE_LSN(tail_lsn
)+1 == log
->l_prev_cycle
);
3345 if (BLOCK_LSN(tail_lsn
) == log
->l_prev_block
)
3346 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3348 blocks
= BLOCK_LSN(tail_lsn
) - log
->l_prev_block
;
3349 if (blocks
< BTOBB(iclog
->ic_offset
) + 1)
3350 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3352 } /* xlog_verify_tail_lsn */
3355 * Perform a number of checks on the iclog before writing to disk.
3357 * 1. Make sure the iclogs are still circular
3358 * 2. Make sure we have a good magic number
3359 * 3. Make sure we don't have magic numbers in the data
3360 * 4. Check fields of each log operation header for:
3361 * A. Valid client identifier
3362 * B. tid ptr value falls in valid ptr space (user space code)
3363 * C. Length in log record header is correct according to the
3364 * individual operation headers within record.
3365 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3366 * log, check the preceding blocks of the physical log to make sure all
3367 * the cycle numbers agree with the current cycle number.
3370 xlog_verify_iclog(xlog_t
*log
,
3371 xlog_in_core_t
*iclog
,
3375 xlog_op_header_t
*ophead
;
3376 xlog_in_core_t
*icptr
;
3377 xlog_in_core_2_t
*xhdr
;
3379 xfs_caddr_t base_ptr
;
3380 __psint_t field_offset
;
3382 int len
, i
, j
, k
, op_len
;
3386 /* check validity of iclog pointers */
3388 icptr
= log
->l_iclog
;
3389 for (i
=0; i
< log
->l_iclog_bufs
; i
++) {
3391 xlog_panic("xlog_verify_iclog: invalid ptr");
3392 icptr
= icptr
->ic_next
;
3394 if (icptr
!= log
->l_iclog
)
3395 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
3398 /* check log magic numbers */
3399 ptr
= (xfs_caddr_t
) &(iclog
->ic_header
);
3400 if (INT_GET(*(uint
*)ptr
, ARCH_CONVERT
) != XLOG_HEADER_MAGIC_NUM
)
3401 xlog_panic("xlog_verify_iclog: invalid magic num");
3403 for (ptr
+= BBSIZE
; ptr
< ((xfs_caddr_t
)&(iclog
->ic_header
))+count
;
3405 if (INT_GET(*(uint
*)ptr
, ARCH_CONVERT
) == XLOG_HEADER_MAGIC_NUM
)
3406 xlog_panic("xlog_verify_iclog: unexpected magic num");
3410 len
= INT_GET(iclog
->ic_header
.h_num_logops
, ARCH_CONVERT
);
3411 ptr
= iclog
->ic_datap
;
3413 ophead
= (xlog_op_header_t
*)ptr
;
3414 xhdr
= (xlog_in_core_2_t
*)&iclog
->ic_header
;
3415 for (i
= 0; i
< len
; i
++) {
3416 ophead
= (xlog_op_header_t
*)ptr
;
3418 /* clientid is only 1 byte */
3419 field_offset
= (__psint_t
)
3420 ((xfs_caddr_t
)&(ophead
->oh_clientid
) - base_ptr
);
3421 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3422 clientid
= ophead
->oh_clientid
;
3424 idx
= BTOBBT((xfs_caddr_t
)&(ophead
->oh_clientid
) - iclog
->ic_datap
);
3425 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3426 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3427 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3428 clientid
= GET_CLIENT_ID(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
], ARCH_CONVERT
);
3430 clientid
= GET_CLIENT_ID(iclog
->ic_header
.h_cycle_data
[idx
], ARCH_CONVERT
);
3433 if (clientid
!= XFS_TRANSACTION
&& clientid
!= XFS_LOG
)
3434 cmn_err(CE_WARN
, "xlog_verify_iclog: "
3435 "invalid clientid %d op 0x%p offset 0x%lx",
3436 clientid
, ophead
, (unsigned long)field_offset
);
3439 field_offset
= (__psint_t
)
3440 ((xfs_caddr_t
)&(ophead
->oh_len
) - base_ptr
);
3441 if (syncing
== B_FALSE
|| (field_offset
& 0x1ff)) {
3442 op_len
= INT_GET(ophead
->oh_len
, ARCH_CONVERT
);
3444 idx
= BTOBBT((__psint_t
)&ophead
->oh_len
-
3445 (__psint_t
)iclog
->ic_datap
);
3446 if (idx
>= (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
)) {
3447 j
= idx
/ (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3448 k
= idx
% (XLOG_HEADER_CYCLE_SIZE
/ BBSIZE
);
3449 op_len
= INT_GET(xhdr
[j
].hic_xheader
.xh_cycle_data
[k
], ARCH_CONVERT
);
3451 op_len
= INT_GET(iclog
->ic_header
.h_cycle_data
[idx
], ARCH_CONVERT
);
3454 ptr
+= sizeof(xlog_op_header_t
) + op_len
;
3456 } /* xlog_verify_iclog */
3460 * Mark all iclogs IOERROR. LOG_LOCK is held by the caller.
3466 xlog_in_core_t
*iclog
, *ic
;
3468 iclog
= log
->l_iclog
;
3469 if (! (iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3471 * Mark all the incore logs IOERROR.
3472 * From now on, no log flushes will result.
3476 ic
->ic_state
= XLOG_STATE_IOERROR
;
3478 } while (ic
!= iclog
);
3482 * Return non-zero, if state transition has already happened.
3488 * This is called from xfs_force_shutdown, when we're forcibly
3489 * shutting down the filesystem, typically because of an IO error.
3490 * Our main objectives here are to make sure that:
3491 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3492 * parties to find out, 'atomically'.
3493 * b. those who're sleeping on log reservations, pinned objects and
3494 * other resources get woken up, and be told the bad news.
3495 * c. nothing new gets queued up after (a) and (b) are done.
3496 * d. if !logerror, flush the iclogs to disk, then seal them off
3500 xfs_log_force_umount(
3501 struct xfs_mount
*mp
,
3514 * If this happens during log recovery, don't worry about
3515 * locking; the log isn't open for business yet.
3518 log
->l_flags
& XLOG_ACTIVE_RECOVERY
) {
3519 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3520 XFS_BUF_DONE(mp
->m_sb_bp
);
3525 * Somebody could've already done the hard work for us.
3526 * No need to get locks for this.
3528 if (logerror
&& log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
) {
3529 ASSERT(XLOG_FORCED_SHUTDOWN(log
));
3534 * We must hold both the GRANT lock and the LOG lock,
3535 * before we mark the filesystem SHUTDOWN and wake
3536 * everybody up to tell the bad news.
3538 s
= GRANT_LOCK(log
);
3540 mp
->m_flags
|= XFS_MOUNT_FS_SHUTDOWN
;
3541 XFS_BUF_DONE(mp
->m_sb_bp
);
3543 * This flag is sort of redundant because of the mount flag, but
3544 * it's good to maintain the separation between the log and the rest
3547 log
->l_flags
|= XLOG_IO_ERROR
;
3550 * If we hit a log error, we want to mark all the iclogs IOERROR
3551 * while we're still holding the loglock.
3554 retval
= xlog_state_ioerror(log
);
3555 LOG_UNLOCK(log
, s2
);
3558 * We don't want anybody waiting for log reservations
3559 * after this. That means we have to wake up everybody
3560 * queued up on reserve_headq as well as write_headq.
3561 * In addition, we make sure in xlog_{re}grant_log_space
3562 * that we don't enqueue anything once the SHUTDOWN flag
3563 * is set, and this action is protected by the GRANTLOCK.
3565 if ((tic
= log
->l_reserve_headq
)) {
3567 sv_signal(&tic
->t_sema
);
3569 } while (tic
!= log
->l_reserve_headq
);
3572 if ((tic
= log
->l_write_headq
)) {
3574 sv_signal(&tic
->t_sema
);
3576 } while (tic
!= log
->l_write_headq
);
3578 GRANT_UNLOCK(log
, s
);
3580 if (! (log
->l_iclog
->ic_state
& XLOG_STATE_IOERROR
)) {
3583 * Force the incore logs to disk before shutting the
3584 * log down completely.
3586 xlog_state_sync_all(log
, XFS_LOG_FORCE
|XFS_LOG_SYNC
, &dummy
);
3588 retval
= xlog_state_ioerror(log
);
3589 LOG_UNLOCK(log
, s2
);
3592 * Wake up everybody waiting on xfs_log_force.
3593 * Callback all log item committed functions as if the
3594 * log writes were completed.
3596 xlog_state_do_callback(log
, XFS_LI_ABORTED
, NULL
);
3598 #ifdef XFSERRORDEBUG
3600 xlog_in_core_t
*iclog
;
3603 iclog
= log
->l_iclog
;
3605 ASSERT(iclog
->ic_callback
== 0);
3606 iclog
= iclog
->ic_next
;
3607 } while (iclog
!= log
->l_iclog
);
3611 /* return non-zero if log IOERROR transition had already happened */
3616 xlog_iclogs_empty(xlog_t
*log
)
3618 xlog_in_core_t
*iclog
;
3620 iclog
= log
->l_iclog
;
3622 /* endianness does not matter here, zero is zero in
3625 if (iclog
->ic_header
.h_num_logops
)
3627 iclog
= iclog
->ic_next
;
3628 } while (iclog
!= log
->l_iclog
);