2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_error.h"
31 #include "xfs_log_priv.h"
32 #include "xfs_buf_item.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_log_recover.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_dir2_sf.h"
39 #include "xfs_attr_sf.h"
40 #include "xfs_dinode.h"
41 #include "xfs_inode.h"
45 #define xlog_write_adv_cnt(ptr, len, off, bytes) \
50 /* Local miscellaneous function prototypes */
51 STATIC int xlog_bdstrat_cb(struct xfs_buf *);
52 STATIC int xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket,
53 xlog_in_core_t **, xfs_lsn_t *);
54 STATIC xlog_t * xlog_alloc_log(xfs_mount_t *mp,
55 xfs_buftarg_t *log_target,
56 xfs_daddr_t blk_offset,
58 STATIC int xlog_space_left(xlog_t *log, int cycle, int bytes);
59 STATIC int xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
60 STATIC void xlog_dealloc_log(xlog_t *log);
61 STATIC int xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[],
62 int nentries, xfs_log_ticket_t tic,
64 xlog_in_core_t **commit_iclog,
67 /* local state machine functions */
68 STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
69 STATIC void xlog_state_do_callback(xlog_t *log,int aborted, xlog_in_core_t *iclog);
70 STATIC int xlog_state_get_iclog_space(xlog_t *log,
72 xlog_in_core_t **iclog,
73 xlog_ticket_t *ticket,
76 STATIC void xlog_state_put_ticket(xlog_t *log,
78 STATIC int xlog_state_release_iclog(xlog_t *log,
79 xlog_in_core_t *iclog);
80 STATIC void xlog_state_switch_iclogs(xlog_t *log,
81 xlog_in_core_t *iclog,
83 STATIC int xlog_state_sync(xlog_t *log,
87 STATIC int xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed);
88 STATIC void xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog);
90 /* local functions to manipulate grant head */
91 STATIC int xlog_grant_log_space(xlog_t *log,
93 STATIC void xlog_grant_push_ail(xfs_mount_t *mp,
95 STATIC void xlog_regrant_reserve_log_space(xlog_t *log,
96 xlog_ticket_t *ticket);
97 STATIC int xlog_regrant_write_log_space(xlog_t *log,
98 xlog_ticket_t *ticket);
99 STATIC void xlog_ungrant_log_space(xlog_t *log,
100 xlog_ticket_t *ticket);
103 /* local ticket functions */
104 STATIC void xlog_state_ticket_alloc(xlog_t *log);
105 STATIC xlog_ticket_t *xlog_ticket_get(xlog_t *log,
110 STATIC void xlog_ticket_put(xlog_t *log, xlog_ticket_t *ticket);
113 STATIC void xlog_verify_dest_ptr(xlog_t *log, __psint_t ptr);
114 STATIC void xlog_verify_grant_head(xlog_t *log, int equals);
115 STATIC void xlog_verify_iclog(xlog_t *log, xlog_in_core_t *iclog,
116 int count, boolean_t syncing);
117 STATIC void xlog_verify_tail_lsn(xlog_t *log, xlog_in_core_t *iclog,
120 #define xlog_verify_dest_ptr(a,b)
121 #define xlog_verify_grant_head(a,b)
122 #define xlog_verify_iclog(a,b,c,d)
123 #define xlog_verify_tail_lsn(a,b,c)
126 STATIC int xlog_iclogs_empty(xlog_t *log);
128 #if defined(XFS_LOG_TRACE)
130 xlog_trace_loggrant(xlog_t *log, xlog_ticket_t *tic, xfs_caddr_t string)
134 if (!log->l_grant_trace) {
135 log->l_grant_trace = ktrace_alloc(2048, KM_NOSLEEP);
136 if (!log->l_grant_trace)
139 /* ticket counts are 1 byte each */
140 cnts = ((unsigned long)tic->t_ocnt) | ((unsigned long)tic->t_cnt) << 8;
142 ktrace_enter(log->l_grant_trace,
144 (void *)log->l_reserve_headq,
145 (void *)log->l_write_headq,
146 (void *)((unsigned long)log->l_grant_reserve_cycle),
147 (void *)((unsigned long)log->l_grant_reserve_bytes),
148 (void *)((unsigned long)log->l_grant_write_cycle),
149 (void *)((unsigned long)log->l_grant_write_bytes),
150 (void *)((unsigned long)log->l_curr_cycle),
151 (void *)((unsigned long)log->l_curr_block),
152 (void *)((unsigned long)CYCLE_LSN(log->l_tail_lsn)),
153 (void *)((unsigned long)BLOCK_LSN(log->l_tail_lsn)),
155 (void *)((unsigned long)tic->t_trans_type),
157 (void *)((unsigned long)tic->t_curr_res),
158 (void *)((unsigned long)tic->t_unit_res));
162 xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
164 if (!iclog->ic_trace)
165 iclog->ic_trace = ktrace_alloc(256, KM_SLEEP);
166 ktrace_enter(iclog->ic_trace,
167 (void *)((unsigned long)state),
168 (void *)((unsigned long)current_pid()),
169 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
170 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
171 (void *)NULL, (void *)NULL, (void *)NULL, (void *)NULL,
172 (void *)NULL, (void *)NULL);
175 #define xlog_trace_loggrant(log,tic,string)
176 #define xlog_trace_iclog(iclog,state)
177 #endif /* XFS_LOG_TRACE */
181 xlog_ins_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
185 tic->t_prev = (*qp)->t_prev;
186 (*qp)->t_prev->t_next = tic;
189 tic->t_prev = tic->t_next = tic;
193 tic->t_flags |= XLOG_TIC_IN_Q;
197 xlog_del_ticketq(struct xlog_ticket **qp, struct xlog_ticket *tic)
199 if (tic == tic->t_next) {
203 tic->t_next->t_prev = tic->t_prev;
204 tic->t_prev->t_next = tic->t_next;
207 tic->t_next = tic->t_prev = NULL;
208 tic->t_flags &= ~XLOG_TIC_IN_Q;
212 xlog_grant_sub_space(struct log *log, int bytes)
214 log->l_grant_write_bytes -= bytes;
215 if (log->l_grant_write_bytes < 0) {
216 log->l_grant_write_bytes += log->l_logsize;
217 log->l_grant_write_cycle--;
220 log->l_grant_reserve_bytes -= bytes;
221 if ((log)->l_grant_reserve_bytes < 0) {
222 log->l_grant_reserve_bytes += log->l_logsize;
223 log->l_grant_reserve_cycle--;
229 xlog_grant_add_space_write(struct log *log, int bytes)
231 log->l_grant_write_bytes += bytes;
232 if (log->l_grant_write_bytes > log->l_logsize) {
233 log->l_grant_write_bytes -= log->l_logsize;
234 log->l_grant_write_cycle++;
239 xlog_grant_add_space_reserve(struct log *log, int bytes)
241 log->l_grant_reserve_bytes += bytes;
242 if (log->l_grant_reserve_bytes > log->l_logsize) {
243 log->l_grant_reserve_bytes -= log->l_logsize;
244 log->l_grant_reserve_cycle++;
249 xlog_grant_add_space(struct log *log, int bytes)
251 xlog_grant_add_space_write(log, bytes);
252 xlog_grant_add_space_reserve(log, bytes);
256 xlog_tic_reset_res(xlog_ticket_t *tic)
259 tic->t_res_arr_sum = 0;
260 tic->t_res_num_ophdrs = 0;
264 xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
266 if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
267 /* add to overflow and start again */
268 tic->t_res_o_flow += tic->t_res_arr_sum;
270 tic->t_res_arr_sum = 0;
273 tic->t_res_arr[tic->t_res_num].r_len = len;
274 tic->t_res_arr[tic->t_res_num].r_type = type;
275 tic->t_res_arr_sum += len;
282 * 1. currblock field gets updated at startup and after in-core logs
283 * marked as with WANT_SYNC.
287 * This routine is called when a user of a log manager ticket is done with
288 * the reservation. If the ticket was ever used, then a commit record for
289 * the associated transaction is written out as a log operation header with
290 * no data. The flag XLOG_TIC_INITED is set when the first write occurs with
291 * a given ticket. If the ticket was one with a permanent reservation, then
292 * a few operations are done differently. Permanent reservation tickets by
293 * default don't release the reservation. They just commit the current
294 * transaction with the belief that the reservation is still needed. A flag
295 * must be passed in before permanent reservations are actually released.
296 * When these type of tickets are not released, they need to be set into
297 * the inited state again. By doing this, a start record will be written
298 * out when the next write occurs.
301 xfs_log_done(xfs_mount_t *mp,
302 xfs_log_ticket_t xtic,
306 xlog_t *log = mp->m_log;
307 xlog_ticket_t *ticket = (xfs_log_ticket_t) xtic;
310 if (XLOG_FORCED_SHUTDOWN(log) ||
312 * If nothing was ever written, don't write out commit record.
313 * If we get an error, just continue and give back the log ticket.
315 (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
316 (xlog_commit_record(mp, ticket,
317 (xlog_in_core_t **)iclog, &lsn)))) {
318 lsn = (xfs_lsn_t) -1;
319 if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
320 flags |= XFS_LOG_REL_PERM_RESERV;
325 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) == 0 ||
326 (flags & XFS_LOG_REL_PERM_RESERV)) {
328 * Release ticket if not permanent reservation or a specific
329 * request has been made to release a permanent reservation.
331 xlog_trace_loggrant(log, ticket, "xfs_log_done: (non-permanent)");
332 xlog_ungrant_log_space(log, ticket);
333 xlog_state_put_ticket(log, ticket);
335 xlog_trace_loggrant(log, ticket, "xfs_log_done: (permanent)");
336 xlog_regrant_reserve_log_space(log, ticket);
339 /* If this ticket was a permanent reservation and we aren't
340 * trying to release it, reset the inited flags; so next time
341 * we write, a start record will be written out.
343 if ((ticket->t_flags & XLOG_TIC_PERM_RESERV) &&
344 (flags & XFS_LOG_REL_PERM_RESERV) == 0)
345 ticket->t_flags |= XLOG_TIC_INITED;
352 * Force the in-core log to disk. If flags == XFS_LOG_SYNC,
353 * the force is done synchronously.
355 * Asynchronous forces are implemented by setting the WANT_SYNC
356 * bit in the appropriate in-core log and then returning.
358 * Synchronous forces are implemented with a semaphore. All callers
359 * to force a given lsn to disk will wait on a semaphore attached to the
360 * specific in-core log. When given in-core log finally completes its
361 * write to disk, that thread will wake up all threads waiting on the
371 xlog_t *log = mp->m_log;
375 log_flushed = &dummy;
377 ASSERT(flags & XFS_LOG_FORCE);
379 XFS_STATS_INC(xs_log_force);
381 if (log->l_flags & XLOG_IO_ERROR)
382 return XFS_ERROR(EIO);
384 return xlog_state_sync_all(log, flags, log_flushed);
386 return xlog_state_sync(log, lsn, flags, log_flushed);
387 } /* xfs_log_force */
390 * Attaches a new iclog I/O completion callback routine during
391 * transaction commit. If the log is in error state, a non-zero
392 * return code is handed back and the caller is responsible for
393 * executing the callback at an appropriate time.
396 xfs_log_notify(xfs_mount_t *mp, /* mount of partition */
397 void *iclog_hndl, /* iclog to hang callback off */
398 xfs_log_callback_t *cb)
400 xlog_t *log = mp->m_log;
401 xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl;
405 spin_lock(&log->l_icloglock);
406 abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
408 ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
409 (iclog->ic_state == XLOG_STATE_WANT_SYNC));
411 *(iclog->ic_callback_tail) = cb;
412 iclog->ic_callback_tail = &(cb->cb_next);
414 spin_unlock(&log->l_icloglock);
416 } /* xfs_log_notify */
419 xfs_log_release_iclog(xfs_mount_t *mp,
422 xlog_t *log = mp->m_log;
423 xlog_in_core_t *iclog = (xlog_in_core_t *)iclog_hndl;
425 if (xlog_state_release_iclog(log, iclog)) {
426 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
434 * 1. Reserve an amount of on-disk log space and return a ticket corresponding
435 * to the reservation.
436 * 2. Potentially, push buffers at tail of log to disk.
438 * Each reservation is going to reserve extra space for a log record header.
439 * When writes happen to the on-disk log, we don't subtract the length of the
440 * log record header from any reservation. By wasting space in each
441 * reservation, we prevent over allocation problems.
444 xfs_log_reserve(xfs_mount_t *mp,
447 xfs_log_ticket_t *ticket,
452 xlog_t *log = mp->m_log;
453 xlog_ticket_t *internal_ticket;
456 ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);
457 ASSERT((flags & XFS_LOG_NOSLEEP) == 0);
459 if (XLOG_FORCED_SHUTDOWN(log))
460 return XFS_ERROR(EIO);
462 XFS_STATS_INC(xs_try_logspace);
464 if (*ticket != NULL) {
465 ASSERT(flags & XFS_LOG_PERM_RESERV);
466 internal_ticket = (xlog_ticket_t *)*ticket;
467 xlog_trace_loggrant(log, internal_ticket, "xfs_log_reserve: existing ticket (permanent trans)");
468 xlog_grant_push_ail(mp, internal_ticket->t_unit_res);
469 retval = xlog_regrant_write_log_space(log, internal_ticket);
471 /* may sleep if need to allocate more tickets */
472 internal_ticket = xlog_ticket_get(log, unit_bytes, cnt,
474 internal_ticket->t_trans_type = t_type;
475 *ticket = internal_ticket;
476 xlog_trace_loggrant(log, internal_ticket,
477 (internal_ticket->t_flags & XLOG_TIC_PERM_RESERV) ?
478 "xfs_log_reserve: create new ticket (permanent trans)" :
479 "xfs_log_reserve: create new ticket");
480 xlog_grant_push_ail(mp,
481 (internal_ticket->t_unit_res *
482 internal_ticket->t_cnt));
483 retval = xlog_grant_log_space(log, internal_ticket);
487 } /* xfs_log_reserve */
491 * Mount a log filesystem
493 * mp - ubiquitous xfs mount point structure
494 * log_target - buftarg of on-disk log device
495 * blk_offset - Start block # where block size is 512 bytes (BBSIZE)
496 * num_bblocks - Number of BBSIZE blocks in on-disk log
498 * Return error or zero.
501 xfs_log_mount(xfs_mount_t *mp,
502 xfs_buftarg_t *log_target,
503 xfs_daddr_t blk_offset,
506 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
507 cmn_err(CE_NOTE, "XFS mounting filesystem %s", mp->m_fsname);
510 "!Mounting filesystem \"%s\" in no-recovery mode. Filesystem will be inconsistent.",
512 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
515 mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
518 * skip log recovery on a norecovery mount. pretend it all
521 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
522 int error, readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
525 mp->m_flags &= ~XFS_MOUNT_RDONLY;
527 error = xlog_recover(mp->m_log);
530 mp->m_flags |= XFS_MOUNT_RDONLY;
532 cmn_err(CE_WARN, "XFS: log mount/recovery failed: error %d", error);
533 xlog_dealloc_log(mp->m_log);
538 /* Normal transactions can now occur */
539 mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;
541 /* End mounting message in xfs_log_mount_finish */
543 } /* xfs_log_mount */
546 * Finish the recovery of the file system. This is separate from
547 * the xfs_log_mount() call, because it depends on the code in
548 * xfs_mountfs() to read in the root and real-time bitmap inodes
549 * between calling xfs_log_mount() and here.
551 * mp - ubiquitous xfs mount point structure
554 xfs_log_mount_finish(xfs_mount_t *mp, int mfsi_flags)
558 if (!(mp->m_flags & XFS_MOUNT_NORECOVERY))
559 error = xlog_recover_finish(mp->m_log, mfsi_flags);
562 ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
569 * Unmount processing for the log.
572 xfs_log_unmount(xfs_mount_t *mp)
576 error = xfs_log_unmount_write(mp);
577 xfs_log_unmount_dealloc(mp);
582 * Final log writes as part of unmount.
584 * Mark the filesystem clean as unmount happens. Note that during relocation
585 * this routine needs to be executed as part of source-bag while the
586 * deallocation must not be done until source-end.
590 * Unmount record used to have a string "Unmount filesystem--" in the
591 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
592 * We just write the magic number now since that particular field isn't
593 * currently architecture converted and "nUmount" is a bit foo.
594 * As far as I know, there weren't any dependencies on the old behaviour.
598 xfs_log_unmount_write(xfs_mount_t *mp)
600 xlog_t *log = mp->m_log;
601 xlog_in_core_t *iclog;
603 xlog_in_core_t *first_iclog;
605 xfs_log_iovec_t reg[1];
606 xfs_log_ticket_t tic = NULL;
610 /* the data section must be 32 bit size aligned */
614 __uint32_t pad2; /* may as well make it 64 bits */
615 } magic = { XLOG_UNMOUNT_TYPE, 0, 0 };
618 * Don't write out unmount record on read-only mounts.
619 * Or, if we are doing a forced umount (typically because of IO errors).
621 if (mp->m_flags & XFS_MOUNT_RDONLY)
624 xfs_log_force(mp, 0, XFS_LOG_FORCE|XFS_LOG_SYNC);
627 first_iclog = iclog = log->l_iclog;
629 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
630 ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
631 ASSERT(iclog->ic_offset == 0);
633 iclog = iclog->ic_next;
634 } while (iclog != first_iclog);
636 if (! (XLOG_FORCED_SHUTDOWN(log))) {
637 reg[0].i_addr = (void*)&magic;
638 reg[0].i_len = sizeof(magic);
639 XLOG_VEC_SET_TYPE(®[0], XLOG_REG_TYPE_UNMOUNT);
641 error = xfs_log_reserve(mp, 600, 1, &tic,
642 XFS_LOG, 0, XLOG_UNMOUNT_REC_TYPE);
644 /* remove inited flag */
645 ((xlog_ticket_t *)tic)->t_flags = 0;
646 error = xlog_write(mp, reg, 1, tic, &lsn,
647 NULL, XLOG_UNMOUNT_TRANS);
649 * At this point, we're umounting anyway,
650 * so there's no point in transitioning log state
651 * to IOERROR. Just continue...
656 xfs_fs_cmn_err(CE_ALERT, mp,
657 "xfs_log_unmount: unmount record failed");
661 spin_lock(&log->l_icloglock);
662 iclog = log->l_iclog;
664 spin_unlock(&log->l_icloglock);
665 xlog_state_want_sync(log, iclog);
666 (void) xlog_state_release_iclog(log, iclog);
668 spin_lock(&log->l_icloglock);
669 if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
670 iclog->ic_state == XLOG_STATE_DIRTY)) {
671 if (!XLOG_FORCED_SHUTDOWN(log)) {
672 sv_wait(&iclog->ic_forcesema, PMEM,
673 &log->l_icloglock, s);
675 spin_unlock(&log->l_icloglock);
678 spin_unlock(&log->l_icloglock);
681 xlog_trace_loggrant(log, tic, "unmount rec");
682 xlog_ungrant_log_space(log, tic);
683 xlog_state_put_ticket(log, tic);
687 * We're already in forced_shutdown mode, couldn't
688 * even attempt to write out the unmount transaction.
690 * Go through the motions of sync'ing and releasing
691 * the iclog, even though no I/O will actually happen,
692 * we need to wait for other log I/Os that may already
693 * be in progress. Do this as a separate section of
694 * code so we'll know if we ever get stuck here that
695 * we're in this odd situation of trying to unmount
696 * a file system that went into forced_shutdown as
697 * the result of an unmount..
699 spin_lock(&log->l_icloglock);
700 iclog = log->l_iclog;
702 spin_unlock(&log->l_icloglock);
704 xlog_state_want_sync(log, iclog);
705 (void) xlog_state_release_iclog(log, iclog);
707 spin_lock(&log->l_icloglock);
709 if ( ! ( iclog->ic_state == XLOG_STATE_ACTIVE
710 || iclog->ic_state == XLOG_STATE_DIRTY
711 || iclog->ic_state == XLOG_STATE_IOERROR) ) {
713 sv_wait(&iclog->ic_forcesema, PMEM,
714 &log->l_icloglock, s);
716 spin_unlock(&log->l_icloglock);
721 } /* xfs_log_unmount_write */
724 * Deallocate log structures for unmount/relocation.
727 xfs_log_unmount_dealloc(xfs_mount_t *mp)
729 xlog_dealloc_log(mp->m_log);
733 * Write region vectors to log. The write happens using the space reservation
734 * of the ticket (tic). It is not a requirement that all writes for a given
735 * transaction occur with one call to xfs_log_write().
738 xfs_log_write(xfs_mount_t * mp,
739 xfs_log_iovec_t reg[],
741 xfs_log_ticket_t tic,
742 xfs_lsn_t *start_lsn)
745 xlog_t *log = mp->m_log;
747 if (XLOG_FORCED_SHUTDOWN(log))
748 return XFS_ERROR(EIO);
750 if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
751 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
754 } /* xfs_log_write */
758 xfs_log_move_tail(xfs_mount_t *mp,
762 xlog_t *log = mp->m_log;
763 int need_bytes, free_bytes, cycle, bytes;
765 if (XLOG_FORCED_SHUTDOWN(log))
769 /* needed since sync_lsn is 64 bits */
770 spin_lock(&log->l_icloglock);
771 tail_lsn = log->l_last_sync_lsn;
772 spin_unlock(&log->l_icloglock);
775 spin_lock(&log->l_grant_lock);
777 /* Also an invalid lsn. 1 implies that we aren't passing in a valid
781 log->l_tail_lsn = tail_lsn;
784 if ((tic = log->l_write_headq)) {
786 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
787 panic("Recovery problem");
789 cycle = log->l_grant_write_cycle;
790 bytes = log->l_grant_write_bytes;
791 free_bytes = xlog_space_left(log, cycle, bytes);
793 ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
795 if (free_bytes < tic->t_unit_res && tail_lsn != 1)
798 free_bytes -= tic->t_unit_res;
799 sv_signal(&tic->t_sema);
801 } while (tic != log->l_write_headq);
803 if ((tic = log->l_reserve_headq)) {
805 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
806 panic("Recovery problem");
808 cycle = log->l_grant_reserve_cycle;
809 bytes = log->l_grant_reserve_bytes;
810 free_bytes = xlog_space_left(log, cycle, bytes);
812 if (tic->t_flags & XLOG_TIC_PERM_RESERV)
813 need_bytes = tic->t_unit_res*tic->t_cnt;
815 need_bytes = tic->t_unit_res;
816 if (free_bytes < need_bytes && tail_lsn != 1)
819 free_bytes -= need_bytes;
820 sv_signal(&tic->t_sema);
822 } while (tic != log->l_reserve_headq);
824 spin_unlock(&log->l_grant_lock);
825 } /* xfs_log_move_tail */
828 * Determine if we have a transaction that has gone to disk
829 * that needs to be covered. Log activity needs to be idle (no AIL and
830 * nothing in the iclogs). And, we need to be in the right state indicating
831 * something has gone out.
834 xfs_log_need_covered(xfs_mount_t *mp)
837 xlog_t *log = mp->m_log;
839 if (!xfs_fs_writable(mp))
842 spin_lock(&log->l_icloglock);
843 if (((log->l_covered_state == XLOG_STATE_COVER_NEED) ||
844 (log->l_covered_state == XLOG_STATE_COVER_NEED2))
845 && !xfs_trans_first_ail(mp, &gen)
846 && xlog_iclogs_empty(log)) {
847 if (log->l_covered_state == XLOG_STATE_COVER_NEED)
848 log->l_covered_state = XLOG_STATE_COVER_DONE;
850 ASSERT(log->l_covered_state == XLOG_STATE_COVER_NEED2);
851 log->l_covered_state = XLOG_STATE_COVER_DONE2;
855 spin_unlock(&log->l_icloglock);
859 /******************************************************************************
863 ******************************************************************************
866 /* xfs_trans_tail_ail returns 0 when there is nothing in the list.
867 * The log manager must keep track of the last LR which was committed
868 * to disk. The lsn of this LR will become the new tail_lsn whenever
869 * xfs_trans_tail_ail returns 0. If we don't do this, we run into
870 * the situation where stuff could be written into the log but nothing
871 * was ever in the AIL when asked. Eventually, we panic since the
872 * tail hits the head.
874 * We may be holding the log iclog lock upon entering this routine.
877 xlog_assign_tail_lsn(xfs_mount_t *mp)
880 xlog_t *log = mp->m_log;
882 tail_lsn = xfs_trans_tail_ail(mp);
883 spin_lock(&log->l_grant_lock);
885 log->l_tail_lsn = tail_lsn;
887 tail_lsn = log->l_tail_lsn = log->l_last_sync_lsn;
889 spin_unlock(&log->l_grant_lock);
892 } /* xlog_assign_tail_lsn */
896 * Return the space in the log between the tail and the head. The head
897 * is passed in the cycle/bytes formal parms. In the special case where
898 * the reserve head has wrapped passed the tail, this calculation is no
899 * longer valid. In this case, just return 0 which means there is no space
900 * in the log. This works for all places where this function is called
901 * with the reserve head. Of course, if the write head were to ever
902 * wrap the tail, we should blow up. Rather than catch this case here,
903 * we depend on other ASSERTions in other parts of the code. XXXmiken
905 * This code also handles the case where the reservation head is behind
906 * the tail. The details of this case are described below, but the end
907 * result is that we return the size of the log as the amount of space left.
910 xlog_space_left(xlog_t *log, int cycle, int bytes)
916 tail_bytes = BBTOB(BLOCK_LSN(log->l_tail_lsn));
917 tail_cycle = CYCLE_LSN(log->l_tail_lsn);
918 if ((tail_cycle == cycle) && (bytes >= tail_bytes)) {
919 free_bytes = log->l_logsize - (bytes - tail_bytes);
920 } else if ((tail_cycle + 1) < cycle) {
922 } else if (tail_cycle < cycle) {
923 ASSERT(tail_cycle == (cycle - 1));
924 free_bytes = tail_bytes - bytes;
927 * The reservation head is behind the tail.
928 * In this case we just want to return the size of the
929 * log as the amount of space left.
931 xfs_fs_cmn_err(CE_ALERT, log->l_mp,
932 "xlog_space_left: head behind tail\n"
933 " tail_cycle = %d, tail_bytes = %d\n"
934 " GH cycle = %d, GH bytes = %d",
935 tail_cycle, tail_bytes, cycle, bytes);
937 free_bytes = log->l_logsize;
940 } /* xlog_space_left */
944 * Log function which is called when an io completes.
946 * The log manager needs its own routine, in order to control what
947 * happens with the buffer after the write completes.
950 xlog_iodone(xfs_buf_t *bp)
952 xlog_in_core_t *iclog;
956 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
957 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long) 2);
958 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
962 * Some versions of cpp barf on the recursive definition of
963 * ic_log -> hic_fields.ic_log and expand ic_log twice when
964 * it is passed through two macros. Workaround broken cpp.
969 * If the ordered flag has been removed by a lower
970 * layer, it means the underlyin device no longer supports
971 * barrier I/O. Warn loudly and turn off barriers.
973 if ((l->l_mp->m_flags & XFS_MOUNT_BARRIER) && !XFS_BUF_ORDERED(bp)) {
974 l->l_mp->m_flags &= ~XFS_MOUNT_BARRIER;
975 xfs_fs_cmn_err(CE_WARN, l->l_mp,
976 "xlog_iodone: Barriers are no longer supported"
977 " by device. Disabling barriers\n");
978 xfs_buftrace("XLOG_IODONE BARRIERS OFF", bp);
982 * Race to shutdown the filesystem if we see an error.
984 if (XFS_TEST_ERROR((XFS_BUF_GETERROR(bp)), l->l_mp,
985 XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
986 xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
988 xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
990 * This flag will be propagated to the trans-committed
991 * callback routines to let them know that the log-commit
994 aborted = XFS_LI_ABORTED;
995 } else if (iclog->ic_state & XLOG_STATE_IOERROR) {
996 aborted = XFS_LI_ABORTED;
999 /* log I/O is always issued ASYNC */
1000 ASSERT(XFS_BUF_ISASYNC(bp));
1001 xlog_state_done_syncing(iclog, aborted);
1003 * do not reference the buffer (bp) here as we could race
1004 * with it being freed after writing the unmount record to the
1011 * The bdstrat callback function for log bufs. This gives us a central
1012 * place to trap bufs in case we get hit by a log I/O error and need to
1013 * shutdown. Actually, in practice, even when we didn't get a log error,
1014 * we transition the iclogs to IOERROR state *after* flushing all existing
1015 * iclogs to disk. This is because we don't want anymore new transactions to be
1016 * started or completed afterwards.
1019 xlog_bdstrat_cb(struct xfs_buf *bp)
1021 xlog_in_core_t *iclog;
1023 iclog = XFS_BUF_FSPRIVATE(bp, xlog_in_core_t *);
1025 if ((iclog->ic_state & XLOG_STATE_IOERROR) == 0) {
1026 /* note for irix bstrat will need struct bdevsw passed
1027 * Fix the following macro if the code ever is merged
1033 xfs_buftrace("XLOG__BDSTRAT IOERROR", bp);
1034 XFS_BUF_ERROR(bp, EIO);
1037 return XFS_ERROR(EIO);
1043 * Return size of each in-core log record buffer.
1045 * All machines get 8 x 32KB buffers by default, unless tuned otherwise.
1047 * If the filesystem blocksize is too large, we may need to choose a
1048 * larger size since the directory code currently logs entire blocks.
1052 xlog_get_iclog_buffer_size(xfs_mount_t *mp,
1058 if (mp->m_logbufs <= 0)
1059 log->l_iclog_bufs = XLOG_MAX_ICLOGS;
1061 log->l_iclog_bufs = mp->m_logbufs;
1064 * Buffer size passed in from mount system call.
1066 if (mp->m_logbsize > 0) {
1067 size = log->l_iclog_size = mp->m_logbsize;
1068 log->l_iclog_size_log = 0;
1070 log->l_iclog_size_log++;
1074 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
1075 /* # headers = size / 32K
1076 * one header holds cycles from 32K of data
1079 xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
1080 if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
1082 log->l_iclog_hsize = xhdrs << BBSHIFT;
1083 log->l_iclog_heads = xhdrs;
1085 ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
1086 log->l_iclog_hsize = BBSIZE;
1087 log->l_iclog_heads = 1;
1092 /* All machines use 32KB buffers by default. */
1093 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1094 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
1096 /* the default log size is 16k or 32k which is one header sector */
1097 log->l_iclog_hsize = BBSIZE;
1098 log->l_iclog_heads = 1;
1101 * For 16KB, we use 3 32KB buffers. For 32KB block sizes, we use
1102 * 4 32KB buffers. For 64KB block sizes, we use 8 32KB buffers.
1104 if (mp->m_sb.sb_blocksize >= 16*1024) {
1105 log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
1106 log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
1107 if (mp->m_logbufs <= 0) {
1108 switch (mp->m_sb.sb_blocksize) {
1109 case 16*1024: /* 16 KB */
1110 log->l_iclog_bufs = 3;
1112 case 32*1024: /* 32 KB */
1113 log->l_iclog_bufs = 4;
1115 case 64*1024: /* 64 KB */
1116 log->l_iclog_bufs = 8;
1119 xlog_panic("XFS: Invalid blocksize");
1125 done: /* are we being asked to make the sizes selected above visible? */
1126 if (mp->m_logbufs == 0)
1127 mp->m_logbufs = log->l_iclog_bufs;
1128 if (mp->m_logbsize == 0)
1129 mp->m_logbsize = log->l_iclog_size;
1130 } /* xlog_get_iclog_buffer_size */
1134 * This routine initializes some of the log structure for a given mount point.
1135 * Its primary purpose is to fill in enough, so recovery can occur. However,
1136 * some other stuff may be filled in too.
1139 xlog_alloc_log(xfs_mount_t *mp,
1140 xfs_buftarg_t *log_target,
1141 xfs_daddr_t blk_offset,
1145 xlog_rec_header_t *head;
1146 xlog_in_core_t **iclogp;
1147 xlog_in_core_t *iclog, *prev_iclog=NULL;
1152 log = (xlog_t *)kmem_zalloc(sizeof(xlog_t), KM_SLEEP);
1155 log->l_targ = log_target;
1156 log->l_logsize = BBTOB(num_bblks);
1157 log->l_logBBstart = blk_offset;
1158 log->l_logBBsize = num_bblks;
1159 log->l_covered_state = XLOG_STATE_COVER_IDLE;
1160 log->l_flags |= XLOG_ACTIVE_RECOVERY;
1162 log->l_prev_block = -1;
1163 log->l_tail_lsn = xlog_assign_lsn(1, 0);
1164 /* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1165 log->l_last_sync_lsn = log->l_tail_lsn;
1166 log->l_curr_cycle = 1; /* 0 is bad since this is initial value */
1167 log->l_grant_reserve_cycle = 1;
1168 log->l_grant_write_cycle = 1;
1170 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb)) {
1171 log->l_sectbb_log = mp->m_sb.sb_logsectlog - BBSHIFT;
1172 ASSERT(log->l_sectbb_log <= mp->m_sectbb_log);
1173 /* for larger sector sizes, must have v2 or external log */
1174 ASSERT(log->l_sectbb_log == 0 ||
1175 log->l_logBBstart == 0 ||
1176 XFS_SB_VERSION_HASLOGV2(&mp->m_sb));
1177 ASSERT(mp->m_sb.sb_logsectlog >= BBSHIFT);
1179 log->l_sectbb_mask = (1 << log->l_sectbb_log) - 1;
1181 xlog_get_iclog_buffer_size(mp, log);
1183 bp = xfs_buf_get_empty(log->l_iclog_size, mp->m_logdev_targp);
1184 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
1185 XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
1186 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1187 ASSERT(XFS_BUF_ISBUSY(bp));
1188 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
1191 spin_lock_init(&log->l_icloglock);
1192 spin_lock_init(&log->l_grant_lock);
1193 initnsema(&log->l_flushsema, 0, "ic-flush");
1194 xlog_state_ticket_alloc(log); /* wait until after icloglock inited */
1196 /* log record size must be multiple of BBSIZE; see xlog_rec_header_t */
1197 ASSERT((XFS_BUF_SIZE(bp) & BBMASK) == 0);
1199 iclogp = &log->l_iclog;
1201 * The amount of memory to allocate for the iclog structure is
1202 * rather funky due to the way the structure is defined. It is
1203 * done this way so that we can use different sizes for machines
1204 * with different amounts of memory. See the definition of
1205 * xlog_in_core_t in xfs_log_priv.h for details.
1207 iclogsize = log->l_iclog_size;
1208 ASSERT(log->l_iclog_size >= 4096);
1209 for (i=0; i < log->l_iclog_bufs; i++) {
1210 *iclogp = (xlog_in_core_t *)
1211 kmem_zalloc(sizeof(xlog_in_core_t), KM_SLEEP);
1213 iclog->ic_prev = prev_iclog;
1216 bp = xfs_buf_get_noaddr(log->l_iclog_size, mp->m_logdev_targp);
1217 if (!XFS_BUF_CPSEMA(bp))
1219 XFS_BUF_SET_IODONE_FUNC(bp, xlog_iodone);
1220 XFS_BUF_SET_BDSTRAT_FUNC(bp, xlog_bdstrat_cb);
1221 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)1);
1223 iclog->hic_data = bp->b_addr;
1225 log->l_iclog_bak[i] = (xfs_caddr_t)&(iclog->ic_header);
1227 head = &iclog->ic_header;
1228 memset(head, 0, sizeof(xlog_rec_header_t));
1229 head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
1230 head->h_version = cpu_to_be32(
1231 XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) ? 2 : 1);
1232 head->h_size = cpu_to_be32(log->l_iclog_size);
1234 head->h_fmt = cpu_to_be32(XLOG_FMT);
1235 memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));
1238 iclog->ic_size = XFS_BUF_SIZE(bp) - log->l_iclog_hsize;
1239 iclog->ic_state = XLOG_STATE_ACTIVE;
1240 iclog->ic_log = log;
1241 iclog->ic_callback_tail = &(iclog->ic_callback);
1242 iclog->ic_datap = (char *)iclog->hic_data + log->l_iclog_hsize;
1244 ASSERT(XFS_BUF_ISBUSY(iclog->ic_bp));
1245 ASSERT(XFS_BUF_VALUSEMA(iclog->ic_bp) <= 0);
1246 sv_init(&iclog->ic_forcesema, SV_DEFAULT, "iclog-force");
1247 sv_init(&iclog->ic_writesema, SV_DEFAULT, "iclog-write");
1249 iclogp = &iclog->ic_next;
1251 *iclogp = log->l_iclog; /* complete ring */
1252 log->l_iclog->ic_prev = prev_iclog; /* re-write 1st prev ptr */
1255 } /* xlog_alloc_log */
1259 * Write out the commit record of a transaction associated with the given
1260 * ticket. Return the lsn of the commit record.
1263 xlog_commit_record(xfs_mount_t *mp,
1264 xlog_ticket_t *ticket,
1265 xlog_in_core_t **iclog,
1266 xfs_lsn_t *commitlsnp)
1269 xfs_log_iovec_t reg[1];
1271 reg[0].i_addr = NULL;
1273 XLOG_VEC_SET_TYPE(®[0], XLOG_REG_TYPE_COMMIT);
1275 ASSERT_ALWAYS(iclog);
1276 if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
1277 iclog, XLOG_COMMIT_TRANS))) {
1278 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1281 } /* xlog_commit_record */
1285 * Push on the buffer cache code if we ever use more than 75% of the on-disk
1286 * log space. This code pushes on the lsn which would supposedly free up
1287 * the 25% which we want to leave free. We may need to adopt a policy which
1288 * pushes on an lsn which is further along in the log once we reach the high
1289 * water mark. In this manner, we would be creating a low water mark.
1292 xlog_grant_push_ail(xfs_mount_t *mp,
1295 xlog_t *log = mp->m_log; /* pointer to the log */
1296 xfs_lsn_t tail_lsn; /* lsn of the log tail */
1297 xfs_lsn_t threshold_lsn = 0; /* lsn we'd like to be at */
1298 int free_blocks; /* free blocks left to write to */
1299 int free_bytes; /* free bytes left to write to */
1300 int threshold_block; /* block in lsn we'd like to be at */
1301 int threshold_cycle; /* lsn cycle we'd like to be at */
1304 ASSERT(BTOBB(need_bytes) < log->l_logBBsize);
1306 spin_lock(&log->l_grant_lock);
1307 free_bytes = xlog_space_left(log,
1308 log->l_grant_reserve_cycle,
1309 log->l_grant_reserve_bytes);
1310 tail_lsn = log->l_tail_lsn;
1311 free_blocks = BTOBBT(free_bytes);
1314 * Set the threshold for the minimum number of free blocks in the
1315 * log to the maximum of what the caller needs, one quarter of the
1316 * log, and 256 blocks.
1318 free_threshold = BTOBB(need_bytes);
1319 free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
1320 free_threshold = MAX(free_threshold, 256);
1321 if (free_blocks < free_threshold) {
1322 threshold_block = BLOCK_LSN(tail_lsn) + free_threshold;
1323 threshold_cycle = CYCLE_LSN(tail_lsn);
1324 if (threshold_block >= log->l_logBBsize) {
1325 threshold_block -= log->l_logBBsize;
1326 threshold_cycle += 1;
1328 threshold_lsn = xlog_assign_lsn(threshold_cycle, threshold_block);
1330 /* Don't pass in an lsn greater than the lsn of the last
1331 * log record known to be on disk.
1333 if (XFS_LSN_CMP(threshold_lsn, log->l_last_sync_lsn) > 0)
1334 threshold_lsn = log->l_last_sync_lsn;
1336 spin_unlock(&log->l_grant_lock);
1339 * Get the transaction layer to kick the dirty buffers out to
1340 * disk asynchronously. No point in trying to do this if
1341 * the filesystem is shutting down.
1343 if (threshold_lsn &&
1344 !XLOG_FORCED_SHUTDOWN(log))
1345 xfs_trans_push_ail(mp, threshold_lsn);
1346 } /* xlog_grant_push_ail */
1350 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous
1351 * fashion. Previously, we should have moved the current iclog
1352 * ptr in the log to point to the next available iclog. This allows further
1353 * write to continue while this code syncs out an iclog ready to go.
1354 * Before an in-core log can be written out, the data section must be scanned
1355 * to save away the 1st word of each BBSIZE block into the header. We replace
1356 * it with the current cycle count. Each BBSIZE block is tagged with the
1357 * cycle count because there in an implicit assumption that drives will
1358 * guarantee that entire 512 byte blocks get written at once. In other words,
1359 * we can't have part of a 512 byte block written and part not written. By
1360 * tagging each block, we will know which blocks are valid when recovering
1361 * after an unclean shutdown.
1363 * This routine is single threaded on the iclog. No other thread can be in
1364 * this routine with the same iclog. Changing contents of iclog can there-
1365 * fore be done without grabbing the state machine lock. Updating the global
1366 * log will require grabbing the lock though.
1368 * The entire log manager uses a logical block numbering scheme. Only
1369 * log_sync (and then only bwrite()) know about the fact that the log may
1370 * not start with block zero on a given device. The log block start offset
1371 * is added immediately before calling bwrite().
1375 xlog_sync(xlog_t *log,
1376 xlog_in_core_t *iclog)
1378 xfs_caddr_t dptr; /* pointer to byte sized element */
1381 uint count; /* byte count of bwrite */
1382 uint count_init; /* initial count before roundup */
1383 int roundoff; /* roundoff to BB or stripe */
1384 int split = 0; /* split write into two regions */
1386 int v2 = XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb);
1388 XFS_STATS_INC(xs_log_writes);
1389 ASSERT(iclog->ic_refcnt == 0);
1391 /* Add for LR header */
1392 count_init = log->l_iclog_hsize + iclog->ic_offset;
1394 /* Round out the log write size */
1395 if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
1396 /* we have a v2 stripe unit to use */
1397 count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
1399 count = BBTOB(BTOBB(count_init));
1401 roundoff = count - count_init;
1402 ASSERT(roundoff >= 0);
1403 ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 &&
1404 roundoff < log->l_mp->m_sb.sb_logsunit)
1406 (log->l_mp->m_sb.sb_logsunit <= 1 &&
1407 roundoff < BBTOB(1)));
1409 /* move grant heads by roundoff in sync */
1410 spin_lock(&log->l_grant_lock);
1411 xlog_grant_add_space(log, roundoff);
1412 spin_unlock(&log->l_grant_lock);
1414 /* put cycle number in every block */
1415 xlog_pack_data(log, iclog, roundoff);
1417 /* real byte length */
1419 iclog->ic_header.h_len =
1420 cpu_to_be32(iclog->ic_offset + roundoff);
1422 iclog->ic_header.h_len =
1423 cpu_to_be32(iclog->ic_offset);
1427 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) == (unsigned long)1);
1428 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1429 XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
1431 XFS_STATS_ADD(xs_log_blocks, BTOBB(count));
1433 /* Do we need to split this write into 2 parts? */
1434 if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1435 split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
1436 count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1437 iclog->ic_bwritecnt = 2; /* split into 2 writes */
1439 iclog->ic_bwritecnt = 1;
1441 XFS_BUF_SET_COUNT(bp, count);
1442 XFS_BUF_SET_FSPRIVATE(bp, iclog); /* save for later */
1443 XFS_BUF_ZEROFLAGS(bp);
1447 * Do an ordered write for the log block.
1448 * Its unnecessary to flush the first split block in the log wrap case.
1450 if (!split && (log->l_mp->m_flags & XFS_MOUNT_BARRIER))
1451 XFS_BUF_ORDERED(bp);
1453 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1454 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1456 xlog_verify_iclog(log, iclog, count, B_TRUE);
1458 /* account for log which doesn't start at block #0 */
1459 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1461 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
1466 if ((error = XFS_bwrite(bp))) {
1467 xfs_ioerror_alert("xlog_sync", log->l_mp, bp,
1472 bp = iclog->ic_log->l_xbuf;
1473 ASSERT(XFS_BUF_FSPRIVATE2(bp, unsigned long) ==
1475 XFS_BUF_SET_FSPRIVATE2(bp, (unsigned long)2);
1476 XFS_BUF_SET_ADDR(bp, 0); /* logical 0 */
1477 XFS_BUF_SET_PTR(bp, (xfs_caddr_t)((__psint_t)&(iclog->ic_header)+
1478 (__psint_t)count), split);
1479 XFS_BUF_SET_FSPRIVATE(bp, iclog);
1480 XFS_BUF_ZEROFLAGS(bp);
1483 if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1484 XFS_BUF_ORDERED(bp);
1485 dptr = XFS_BUF_PTR(bp);
1487 * Bump the cycle numbers at the start of each block
1488 * since this part of the buffer is at the start of
1489 * a new cycle. Watch out for the header magic number
1492 for (i = 0; i < split; i += BBSIZE) {
1493 be32_add((__be32 *)dptr, 1);
1494 if (be32_to_cpu(*(__be32 *)dptr) == XLOG_HEADER_MAGIC_NUM)
1495 be32_add((__be32 *)dptr, 1);
1499 ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
1500 ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);
1502 /* account for internal log which doesn't start at block #0 */
1503 XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1505 if ((error = XFS_bwrite(bp))) {
1506 xfs_ioerror_alert("xlog_sync (split)", log->l_mp,
1507 bp, XFS_BUF_ADDR(bp));
1516 * Deallocate a log structure
1519 xlog_dealloc_log(xlog_t *log)
1521 xlog_in_core_t *iclog, *next_iclog;
1522 xlog_ticket_t *tic, *next_tic;
1525 iclog = log->l_iclog;
1526 for (i=0; i<log->l_iclog_bufs; i++) {
1527 sv_destroy(&iclog->ic_forcesema);
1528 sv_destroy(&iclog->ic_writesema);
1529 xfs_buf_free(iclog->ic_bp);
1530 #ifdef XFS_LOG_TRACE
1531 if (iclog->ic_trace != NULL) {
1532 ktrace_free(iclog->ic_trace);
1535 next_iclog = iclog->ic_next;
1536 kmem_free(iclog, sizeof(xlog_in_core_t));
1539 freesema(&log->l_flushsema);
1540 spinlock_destroy(&log->l_icloglock);
1541 spinlock_destroy(&log->l_grant_lock);
1543 /* XXXsup take a look at this again. */
1544 if ((log->l_ticket_cnt != log->l_ticket_tcnt) &&
1545 !XLOG_FORCED_SHUTDOWN(log)) {
1546 xfs_fs_cmn_err(CE_WARN, log->l_mp,
1547 "xlog_dealloc_log: (cnt: %d, total: %d)",
1548 log->l_ticket_cnt, log->l_ticket_tcnt);
1549 /* ASSERT(log->l_ticket_cnt == log->l_ticket_tcnt); */
1552 tic = log->l_unmount_free;
1554 next_tic = tic->t_next;
1555 kmem_free(tic, PAGE_SIZE);
1559 xfs_buf_free(log->l_xbuf);
1560 #ifdef XFS_LOG_TRACE
1561 if (log->l_trace != NULL) {
1562 ktrace_free(log->l_trace);
1564 if (log->l_grant_trace != NULL) {
1565 ktrace_free(log->l_grant_trace);
1568 log->l_mp->m_log = NULL;
1569 kmem_free(log, sizeof(xlog_t));
1570 } /* xlog_dealloc_log */
1573 * Update counters atomically now that memcpy is done.
1577 xlog_state_finish_copy(xlog_t *log,
1578 xlog_in_core_t *iclog,
1582 spin_lock(&log->l_icloglock);
1584 be32_add(&iclog->ic_header.h_num_logops, record_cnt);
1585 iclog->ic_offset += copy_bytes;
1587 spin_unlock(&log->l_icloglock);
1588 } /* xlog_state_finish_copy */
1594 * print out info relating to regions written which consume
1598 xlog_print_tic_res(xfs_mount_t *mp, xlog_ticket_t *ticket)
1601 uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);
1603 /* match with XLOG_REG_TYPE_* in xfs_log.h */
1604 static char *res_type_str[XLOG_REG_TYPE_MAX] = {
1625 static char *trans_type_str[XFS_TRANS_TYPE_MAX] = {
1668 xfs_fs_cmn_err(CE_WARN, mp,
1669 "xfs_log_write: reservation summary:\n"
1670 " trans type = %s (%u)\n"
1671 " unit res = %d bytes\n"
1672 " current res = %d bytes\n"
1673 " total reg = %u bytes (o/flow = %u bytes)\n"
1674 " ophdrs = %u (ophdr space = %u bytes)\n"
1675 " ophdr + reg = %u bytes\n"
1676 " num regions = %u\n",
1677 ((ticket->t_trans_type <= 0 ||
1678 ticket->t_trans_type > XFS_TRANS_TYPE_MAX) ?
1679 "bad-trans-type" : trans_type_str[ticket->t_trans_type-1]),
1680 ticket->t_trans_type,
1683 ticket->t_res_arr_sum, ticket->t_res_o_flow,
1684 ticket->t_res_num_ophdrs, ophdr_spc,
1685 ticket->t_res_arr_sum +
1686 ticket->t_res_o_flow + ophdr_spc,
1689 for (i = 0; i < ticket->t_res_num; i++) {
1690 uint r_type = ticket->t_res_arr[i].r_type;
1692 "region[%u]: %s - %u bytes\n",
1694 ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
1695 "bad-rtype" : res_type_str[r_type-1]),
1696 ticket->t_res_arr[i].r_len);
1701 * Write some region out to in-core log
1703 * This will be called when writing externally provided regions or when
1704 * writing out a commit record for a given transaction.
1706 * General algorithm:
1707 * 1. Find total length of this write. This may include adding to the
1708 * lengths passed in.
1709 * 2. Check whether we violate the tickets reservation.
1710 * 3. While writing to this iclog
1711 * A. Reserve as much space in this iclog as can get
1712 * B. If this is first write, save away start lsn
1713 * C. While writing this region:
1714 * 1. If first write of transaction, write start record
1715 * 2. Write log operation header (header per region)
1716 * 3. Find out if we can fit entire region into this iclog
1717 * 4. Potentially, verify destination memcpy ptr
1718 * 5. Memcpy (partial) region
1719 * 6. If partial copy, release iclog; otherwise, continue
1720 * copying more regions into current iclog
1721 * 4. Mark want sync bit (in simulation mode)
1722 * 5. Release iclog for potential flush to on-disk log.
1725 * 1. Panic if reservation is overrun. This should never happen since
1726 * reservation amounts are generated internal to the filesystem.
1728 * 1. Tickets are single threaded data structures.
1729 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
1730 * syncing routine. When a single log_write region needs to span
1731 * multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
1732 * on all log operation writes which don't contain the end of the
1733 * region. The XLOG_END_TRANS bit is used for the in-core log
1734 * operation which contains the end of the continued log_write region.
1735 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
1736 * we don't really know exactly how much space will be used. As a result,
1737 * we don't update ic_offset until the end when we know exactly how many
1738 * bytes have been written out.
1741 xlog_write(xfs_mount_t * mp,
1742 xfs_log_iovec_t reg[],
1744 xfs_log_ticket_t tic,
1745 xfs_lsn_t *start_lsn,
1746 xlog_in_core_t **commit_iclog,
1749 xlog_t *log = mp->m_log;
1750 xlog_ticket_t *ticket = (xlog_ticket_t *)tic;
1751 xlog_in_core_t *iclog = NULL; /* ptr to current in-core log */
1752 xlog_op_header_t *logop_head; /* ptr to log operation header */
1753 __psint_t ptr; /* copy address into data region */
1754 int len; /* # xlog_write() bytes 2 still copy */
1755 int index; /* region index currently copying */
1756 int log_offset; /* offset (from 0) into data region */
1757 int start_rec_copy; /* # bytes to copy for start record */
1758 int partial_copy; /* did we split a region? */
1759 int partial_copy_len;/* # bytes copied if split region */
1760 int need_copy; /* # bytes need to memcpy this region */
1761 int copy_len; /* # bytes actually memcpy'ing */
1762 int copy_off; /* # bytes from entry start */
1763 int contwr; /* continued write of in-core log? */
1765 int record_cnt = 0, data_cnt = 0;
1767 partial_copy_len = partial_copy = 0;
1769 /* Calculate potential maximum space. Each region gets its own
1770 * xlog_op_header_t and may need to be double word aligned.
1773 if (ticket->t_flags & XLOG_TIC_INITED) { /* acct for start rec of xact */
1774 len += sizeof(xlog_op_header_t);
1775 ticket->t_res_num_ophdrs++;
1778 for (index = 0; index < nentries; index++) {
1779 len += sizeof(xlog_op_header_t); /* each region gets >= 1 */
1780 ticket->t_res_num_ophdrs++;
1781 len += reg[index].i_len;
1782 xlog_tic_add_region(ticket, reg[index].i_len, reg[index].i_type);
1784 contwr = *start_lsn = 0;
1786 if (ticket->t_curr_res < len) {
1787 xlog_print_tic_res(mp, ticket);
1790 "xfs_log_write: reservation ran out. Need to up reservation");
1792 /* Customer configurable panic */
1793 xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
1794 "xfs_log_write: reservation ran out. Need to up reservation");
1795 /* If we did not panic, shutdown the filesystem */
1796 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1799 ticket->t_curr_res -= len;
1801 for (index = 0; index < nentries; ) {
1802 if ((error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
1803 &contwr, &log_offset)))
1806 ASSERT(log_offset <= iclog->ic_size - 1);
1807 ptr = (__psint_t) ((char *)iclog->ic_datap+log_offset);
1809 /* start_lsn is the first lsn written to. That's all we need. */
1811 *start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
1813 /* This loop writes out as many regions as can fit in the amount
1814 * of space which was allocated by xlog_state_get_iclog_space().
1816 while (index < nentries) {
1817 ASSERT(reg[index].i_len % sizeof(__int32_t) == 0);
1818 ASSERT((__psint_t)ptr % sizeof(__int32_t) == 0);
1821 /* If first write for transaction, insert start record.
1822 * We can't be trying to commit if we are inited. We can't
1823 * have any "partial_copy" if we are inited.
1825 if (ticket->t_flags & XLOG_TIC_INITED) {
1826 logop_head = (xlog_op_header_t *)ptr;
1827 logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
1828 logop_head->oh_clientid = ticket->t_clientid;
1829 logop_head->oh_len = 0;
1830 logop_head->oh_flags = XLOG_START_TRANS;
1831 logop_head->oh_res2 = 0;
1832 ticket->t_flags &= ~XLOG_TIC_INITED; /* clear bit */
1835 start_rec_copy = sizeof(xlog_op_header_t);
1836 xlog_write_adv_cnt(ptr, len, log_offset, start_rec_copy);
1839 /* Copy log operation header directly into data section */
1840 logop_head = (xlog_op_header_t *)ptr;
1841 logop_head->oh_tid = cpu_to_be32(ticket->t_tid);
1842 logop_head->oh_clientid = ticket->t_clientid;
1843 logop_head->oh_res2 = 0;
1845 /* header copied directly */
1846 xlog_write_adv_cnt(ptr, len, log_offset, sizeof(xlog_op_header_t));
1848 /* are we copying a commit or unmount record? */
1849 logop_head->oh_flags = flags;
1852 * We've seen logs corrupted with bad transaction client
1853 * ids. This makes sure that XFS doesn't generate them on.
1854 * Turn this into an EIO and shut down the filesystem.
1856 switch (logop_head->oh_clientid) {
1857 case XFS_TRANSACTION:
1862 xfs_fs_cmn_err(CE_WARN, mp,
1863 "Bad XFS transaction clientid 0x%x in ticket 0x%p",
1864 logop_head->oh_clientid, tic);
1865 return XFS_ERROR(EIO);
1868 /* Partial write last time? => (partial_copy != 0)
1869 * need_copy is the amount we'd like to copy if everything could
1870 * fit in the current memcpy.
1872 need_copy = reg[index].i_len - partial_copy_len;
1874 copy_off = partial_copy_len;
1875 if (need_copy <= iclog->ic_size - log_offset) { /*complete write */
1876 copy_len = need_copy;
1877 logop_head->oh_len = cpu_to_be32(copy_len);
1879 logop_head->oh_flags|= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
1880 partial_copy_len = partial_copy = 0;
1881 } else { /* partial write */
1882 copy_len = iclog->ic_size - log_offset;
1883 logop_head->oh_len = cpu_to_be32(copy_len);
1884 logop_head->oh_flags |= XLOG_CONTINUE_TRANS;
1886 logop_head->oh_flags |= XLOG_WAS_CONT_TRANS;
1887 partial_copy_len += copy_len;
1889 len += sizeof(xlog_op_header_t); /* from splitting of region */
1890 /* account for new log op header */
1891 ticket->t_curr_res -= sizeof(xlog_op_header_t);
1892 ticket->t_res_num_ophdrs++;
1894 xlog_verify_dest_ptr(log, ptr);
1897 ASSERT(copy_len >= 0);
1898 memcpy((xfs_caddr_t)ptr, reg[index].i_addr + copy_off, copy_len);
1899 xlog_write_adv_cnt(ptr, len, log_offset, copy_len);
1901 /* make copy_len total bytes copied, including headers */
1902 copy_len += start_rec_copy + sizeof(xlog_op_header_t);
1904 data_cnt += contwr ? copy_len : 0;
1905 if (partial_copy) { /* copied partial region */
1906 /* already marked WANT_SYNC by xlog_state_get_iclog_space */
1907 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1908 record_cnt = data_cnt = 0;
1909 if ((error = xlog_state_release_iclog(log, iclog)))
1911 break; /* don't increment index */
1912 } else { /* copied entire region */
1914 partial_copy_len = partial_copy = 0;
1916 if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
1917 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1918 record_cnt = data_cnt = 0;
1919 xlog_state_want_sync(log, iclog);
1921 ASSERT(flags & XLOG_COMMIT_TRANS);
1922 *commit_iclog = iclog;
1923 } else if ((error = xlog_state_release_iclog(log, iclog)))
1925 if (index == nentries)
1926 return 0; /* we are done */
1930 } /* if (partial_copy) */
1931 } /* while (index < nentries) */
1932 } /* for (index = 0; index < nentries; ) */
1935 xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
1937 ASSERT(flags & XLOG_COMMIT_TRANS);
1938 *commit_iclog = iclog;
1941 return xlog_state_release_iclog(log, iclog);
1945 /*****************************************************************************
1947 * State Machine functions
1949 *****************************************************************************
1952 /* Clean iclogs starting from the head. This ordering must be
1953 * maintained, so an iclog doesn't become ACTIVE beyond one that
1954 * is SYNCING. This is also required to maintain the notion that we use
1955 * a counting semaphore to hold off would be writers to the log when every
1956 * iclog is trying to sync to disk.
1958 * State Change: DIRTY -> ACTIVE
1961 xlog_state_clean_log(xlog_t *log)
1963 xlog_in_core_t *iclog;
1966 iclog = log->l_iclog;
1968 if (iclog->ic_state == XLOG_STATE_DIRTY) {
1969 iclog->ic_state = XLOG_STATE_ACTIVE;
1970 iclog->ic_offset = 0;
1971 iclog->ic_callback = NULL; /* don't need to free */
1973 * If the number of ops in this iclog indicate it just
1974 * contains the dummy transaction, we can
1975 * change state into IDLE (the second time around).
1976 * Otherwise we should change the state into
1978 * We don't need to cover the dummy.
1981 (be32_to_cpu(iclog->ic_header.h_num_logops) ==
1986 * We have two dirty iclogs so start over
1987 * This could also be num of ops indicates
1988 * this is not the dummy going out.
1992 iclog->ic_header.h_num_logops = 0;
1993 memset(iclog->ic_header.h_cycle_data, 0,
1994 sizeof(iclog->ic_header.h_cycle_data));
1995 iclog->ic_header.h_lsn = 0;
1996 } else if (iclog->ic_state == XLOG_STATE_ACTIVE)
1999 break; /* stop cleaning */
2000 iclog = iclog->ic_next;
2001 } while (iclog != log->l_iclog);
2003 /* log is locked when we are called */
2005 * Change state for the dummy log recording.
2006 * We usually go to NEED. But we go to NEED2 if the changed indicates
2007 * we are done writing the dummy record.
2008 * If we are done with the second dummy recored (DONE2), then
2012 switch (log->l_covered_state) {
2013 case XLOG_STATE_COVER_IDLE:
2014 case XLOG_STATE_COVER_NEED:
2015 case XLOG_STATE_COVER_NEED2:
2016 log->l_covered_state = XLOG_STATE_COVER_NEED;
2019 case XLOG_STATE_COVER_DONE:
2021 log->l_covered_state = XLOG_STATE_COVER_NEED2;
2023 log->l_covered_state = XLOG_STATE_COVER_NEED;
2026 case XLOG_STATE_COVER_DONE2:
2028 log->l_covered_state = XLOG_STATE_COVER_IDLE;
2030 log->l_covered_state = XLOG_STATE_COVER_NEED;
2037 } /* xlog_state_clean_log */
2040 xlog_get_lowest_lsn(
2043 xlog_in_core_t *lsn_log;
2044 xfs_lsn_t lowest_lsn, lsn;
2046 lsn_log = log->l_iclog;
2049 if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
2050 lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
2051 if ((lsn && !lowest_lsn) ||
2052 (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
2056 lsn_log = lsn_log->ic_next;
2057 } while (lsn_log != log->l_iclog);
2063 xlog_state_do_callback(
2066 xlog_in_core_t *ciclog)
2068 xlog_in_core_t *iclog;
2069 xlog_in_core_t *first_iclog; /* used to know when we've
2070 * processed all iclogs once */
2071 xfs_log_callback_t *cb, *cb_next;
2073 xfs_lsn_t lowest_lsn;
2074 int ioerrors; /* counter: iclogs with errors */
2075 int loopdidcallbacks; /* flag: inner loop did callbacks*/
2076 int funcdidcallbacks; /* flag: function did callbacks */
2077 int repeats; /* for issuing console warnings if
2078 * looping too many times */
2080 spin_lock(&log->l_icloglock);
2081 first_iclog = iclog = log->l_iclog;
2083 funcdidcallbacks = 0;
2088 * Scan all iclogs starting with the one pointed to by the
2089 * log. Reset this starting point each time the log is
2090 * unlocked (during callbacks).
2092 * Keep looping through iclogs until one full pass is made
2093 * without running any callbacks.
2095 first_iclog = log->l_iclog;
2096 iclog = log->l_iclog;
2097 loopdidcallbacks = 0;
2102 /* skip all iclogs in the ACTIVE & DIRTY states */
2103 if (iclog->ic_state &
2104 (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
2105 iclog = iclog->ic_next;
2110 * Between marking a filesystem SHUTDOWN and stopping
2111 * the log, we do flush all iclogs to disk (if there
2112 * wasn't a log I/O error). So, we do want things to
2113 * go smoothly in case of just a SHUTDOWN w/o a
2116 if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
2118 * Can only perform callbacks in order. Since
2119 * this iclog is not in the DONE_SYNC/
2120 * DO_CALLBACK state, we skip the rest and
2121 * just try to clean up. If we set our iclog
2122 * to DO_CALLBACK, we will not process it when
2123 * we retry since a previous iclog is in the
2124 * CALLBACK and the state cannot change since
2125 * we are holding the l_icloglock.
2127 if (!(iclog->ic_state &
2128 (XLOG_STATE_DONE_SYNC |
2129 XLOG_STATE_DO_CALLBACK))) {
2130 if (ciclog && (ciclog->ic_state ==
2131 XLOG_STATE_DONE_SYNC)) {
2132 ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
2137 * We now have an iclog that is in either the
2138 * DO_CALLBACK or DONE_SYNC states. The other
2139 * states (WANT_SYNC, SYNCING, or CALLBACK were
2140 * caught by the above if and are going to
2141 * clean (i.e. we aren't doing their callbacks)
2146 * We will do one more check here to see if we
2147 * have chased our tail around.
2150 lowest_lsn = xlog_get_lowest_lsn(log);
2152 XFS_LSN_CMP(lowest_lsn,
2153 be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
2154 iclog = iclog->ic_next;
2155 continue; /* Leave this iclog for
2159 iclog->ic_state = XLOG_STATE_CALLBACK;
2161 spin_unlock(&log->l_icloglock);
2163 /* l_last_sync_lsn field protected by
2164 * l_grant_lock. Don't worry about iclog's lsn.
2165 * No one else can be here except us.
2167 spin_lock(&log->l_grant_lock);
2168 ASSERT(XFS_LSN_CMP(log->l_last_sync_lsn,
2169 be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2170 log->l_last_sync_lsn =
2171 be64_to_cpu(iclog->ic_header.h_lsn);
2172 spin_unlock(&log->l_grant_lock);
2175 * Keep processing entries in the callback list
2176 * until we come around and it is empty. We
2177 * need to atomically see that the list is
2178 * empty and change the state to DIRTY so that
2179 * we don't miss any more callbacks being added.
2181 spin_lock(&log->l_icloglock);
2185 cb = iclog->ic_callback;
2188 iclog->ic_callback_tail = &(iclog->ic_callback);
2189 iclog->ic_callback = NULL;
2190 spin_unlock(&log->l_icloglock);
2192 /* perform callbacks in the order given */
2193 for (; cb; cb = cb_next) {
2194 cb_next = cb->cb_next;
2195 cb->cb_func(cb->cb_arg, aborted);
2197 spin_lock(&log->l_icloglock);
2198 cb = iclog->ic_callback;
2204 ASSERT(iclog->ic_callback == NULL);
2205 if (!(iclog->ic_state & XLOG_STATE_IOERROR))
2206 iclog->ic_state = XLOG_STATE_DIRTY;
2209 * Transition from DIRTY to ACTIVE if applicable.
2210 * NOP if STATE_IOERROR.
2212 xlog_state_clean_log(log);
2214 /* wake up threads waiting in xfs_log_force() */
2215 sv_broadcast(&iclog->ic_forcesema);
2217 iclog = iclog->ic_next;
2218 } while (first_iclog != iclog);
2220 if (repeats > 5000) {
2221 flushcnt += repeats;
2223 xfs_fs_cmn_err(CE_WARN, log->l_mp,
2224 "%s: possible infinite loop (%d iterations)",
2225 __FUNCTION__, flushcnt);
2227 } while (!ioerrors && loopdidcallbacks);
2230 * make one last gasp attempt to see if iclogs are being left in
2234 if (funcdidcallbacks) {
2235 first_iclog = iclog = log->l_iclog;
2237 ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
2239 * Terminate the loop if iclogs are found in states
2240 * which will cause other threads to clean up iclogs.
2242 * SYNCING - i/o completion will go through logs
2243 * DONE_SYNC - interrupt thread should be waiting for
2245 * IOERROR - give up hope all ye who enter here
2247 if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2248 iclog->ic_state == XLOG_STATE_SYNCING ||
2249 iclog->ic_state == XLOG_STATE_DONE_SYNC ||
2250 iclog->ic_state == XLOG_STATE_IOERROR )
2252 iclog = iclog->ic_next;
2253 } while (first_iclog != iclog);
2258 if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) {
2259 flushcnt = log->l_flushcnt;
2260 log->l_flushcnt = 0;
2262 spin_unlock(&log->l_icloglock);
2264 vsema(&log->l_flushsema);
2265 } /* xlog_state_do_callback */
2269 * Finish transitioning this iclog to the dirty state.
2271 * Make sure that we completely execute this routine only when this is
2272 * the last call to the iclog. There is a good chance that iclog flushes,
2273 * when we reach the end of the physical log, get turned into 2 separate
2274 * calls to bwrite. Hence, one iclog flush could generate two calls to this
2275 * routine. By using the reference count bwritecnt, we guarantee that only
2276 * the second completion goes through.
2278 * Callbacks could take time, so they are done outside the scope of the
2279 * global state machine log lock. Assume that the calls to cvsema won't
2280 * take a long time. At least we know it won't sleep.
2283 xlog_state_done_syncing(
2284 xlog_in_core_t *iclog,
2287 xlog_t *log = iclog->ic_log;
2289 spin_lock(&log->l_icloglock);
2291 ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
2292 iclog->ic_state == XLOG_STATE_IOERROR);
2293 ASSERT(iclog->ic_refcnt == 0);
2294 ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);
2298 * If we got an error, either on the first buffer, or in the case of
2299 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
2300 * and none should ever be attempted to be written to disk
2303 if (iclog->ic_state != XLOG_STATE_IOERROR) {
2304 if (--iclog->ic_bwritecnt == 1) {
2305 spin_unlock(&log->l_icloglock);
2308 iclog->ic_state = XLOG_STATE_DONE_SYNC;
2312 * Someone could be sleeping prior to writing out the next
2313 * iclog buffer, we wake them all, one will get to do the
2314 * I/O, the others get to wait for the result.
2316 sv_broadcast(&iclog->ic_writesema);
2317 spin_unlock(&log->l_icloglock);
2318 xlog_state_do_callback(log, aborted, iclog); /* also cleans log */
2319 } /* xlog_state_done_syncing */
2323 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2324 * sleep. The flush semaphore is set to the number of in-core buffers and
2325 * decremented around disk syncing. Therefore, if all buffers are syncing,
2326 * this semaphore will cause new writes to sleep until a sync completes.
2327 * Otherwise, this code just does p() followed by v(). This approximates
2328 * a sleep/wakeup except we can't race.
2330 * The in-core logs are used in a circular fashion. They are not used
2331 * out-of-order even when an iclog past the head is free.
2334 * * log_offset where xlog_write() can start writing into the in-core
2336 * * in-core log pointer to which xlog_write() should write.
2337 * * boolean indicating this is a continued write to an in-core log.
2338 * If this is the last write, then the in-core log's offset field
2339 * needs to be incremented, depending on the amount of data which
2343 xlog_state_get_iclog_space(xlog_t *log,
2345 xlog_in_core_t **iclogp,
2346 xlog_ticket_t *ticket,
2347 int *continued_write,
2351 xlog_rec_header_t *head;
2352 xlog_in_core_t *iclog;
2356 spin_lock(&log->l_icloglock);
2357 if (XLOG_FORCED_SHUTDOWN(log)) {
2358 spin_unlock(&log->l_icloglock);
2359 return XFS_ERROR(EIO);
2362 iclog = log->l_iclog;
2363 if (! (iclog->ic_state == XLOG_STATE_ACTIVE)) {
2365 spin_unlock(&log->l_icloglock);
2366 xlog_trace_iclog(iclog, XLOG_TRACE_SLEEP_FLUSH);
2367 XFS_STATS_INC(xs_log_noiclogs);
2368 /* Ensure that log writes happen */
2369 psema(&log->l_flushsema, PINOD);
2372 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2373 head = &iclog->ic_header;
2375 iclog->ic_refcnt++; /* prevents sync */
2376 log_offset = iclog->ic_offset;
2378 /* On the 1st write to an iclog, figure out lsn. This works
2379 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
2380 * committing to. If the offset is set, that's how many blocks
2383 if (log_offset == 0) {
2384 ticket->t_curr_res -= log->l_iclog_hsize;
2385 xlog_tic_add_region(ticket,
2387 XLOG_REG_TYPE_LRHEADER);
2388 head->h_cycle = cpu_to_be32(log->l_curr_cycle);
2389 head->h_lsn = cpu_to_be64(
2390 xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
2391 ASSERT(log->l_curr_block >= 0);
2394 /* If there is enough room to write everything, then do it. Otherwise,
2395 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
2396 * bit is on, so this will get flushed out. Don't update ic_offset
2397 * until you know exactly how many bytes get copied. Therefore, wait
2398 * until later to update ic_offset.
2400 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
2401 * can fit into remaining data section.
2403 if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
2404 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2406 /* If I'm the only one writing to this iclog, sync it to disk */
2407 if (iclog->ic_refcnt == 1) {
2408 spin_unlock(&log->l_icloglock);
2409 if ((error = xlog_state_release_iclog(log, iclog)))
2413 spin_unlock(&log->l_icloglock);
2418 /* Do we have enough room to write the full amount in the remainder
2419 * of this iclog? Or must we continue a write on the next iclog and
2420 * mark this iclog as completely taken? In the case where we switch
2421 * iclogs (to mark it taken), this particular iclog will release/sync
2422 * to disk in xlog_write().
2424 if (len <= iclog->ic_size - iclog->ic_offset) {
2425 *continued_write = 0;
2426 iclog->ic_offset += len;
2428 *continued_write = 1;
2429 xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
2433 ASSERT(iclog->ic_offset <= iclog->ic_size);
2434 spin_unlock(&log->l_icloglock);
2436 *logoffsetp = log_offset;
2438 } /* xlog_state_get_iclog_space */
2441 * Atomically get the log space required for a log ticket.
2443 * Once a ticket gets put onto the reserveq, it will only return after
2444 * the needed reservation is satisfied.
2447 xlog_grant_log_space(xlog_t *log,
2458 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2459 panic("grant Recovery problem");
2462 /* Is there space or do we need to sleep? */
2463 spin_lock(&log->l_grant_lock);
2464 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: enter");
2466 /* something is already sleeping; insert new transaction at end */
2467 if (log->l_reserve_headq) {
2468 xlog_ins_ticketq(&log->l_reserve_headq, tic);
2469 xlog_trace_loggrant(log, tic,
2470 "xlog_grant_log_space: sleep 1");
2472 * Gotta check this before going to sleep, while we're
2473 * holding the grant lock.
2475 if (XLOG_FORCED_SHUTDOWN(log))
2478 XFS_STATS_INC(xs_sleep_logspace);
2479 sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
2481 * If we got an error, and the filesystem is shutting down,
2482 * we'll catch it down below. So just continue...
2484 xlog_trace_loggrant(log, tic,
2485 "xlog_grant_log_space: wake 1");
2486 spin_lock(&log->l_grant_lock);
2488 if (tic->t_flags & XFS_LOG_PERM_RESERV)
2489 need_bytes = tic->t_unit_res*tic->t_ocnt;
2491 need_bytes = tic->t_unit_res;
2494 if (XLOG_FORCED_SHUTDOWN(log))
2497 free_bytes = xlog_space_left(log, log->l_grant_reserve_cycle,
2498 log->l_grant_reserve_bytes);
2499 if (free_bytes < need_bytes) {
2500 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2501 xlog_ins_ticketq(&log->l_reserve_headq, tic);
2502 xlog_trace_loggrant(log, tic,
2503 "xlog_grant_log_space: sleep 2");
2504 XFS_STATS_INC(xs_sleep_logspace);
2505 sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
2507 if (XLOG_FORCED_SHUTDOWN(log)) {
2508 spin_lock(&log->l_grant_lock);
2512 xlog_trace_loggrant(log, tic,
2513 "xlog_grant_log_space: wake 2");
2514 xlog_grant_push_ail(log->l_mp, need_bytes);
2515 spin_lock(&log->l_grant_lock);
2517 } else if (tic->t_flags & XLOG_TIC_IN_Q)
2518 xlog_del_ticketq(&log->l_reserve_headq, tic);
2520 /* we've got enough space */
2521 xlog_grant_add_space(log, need_bytes);
2523 tail_lsn = log->l_tail_lsn;
2525 * Check to make sure the grant write head didn't just over lap the
2526 * tail. If the cycles are the same, we can't be overlapping.
2527 * Otherwise, make sure that the cycles differ by exactly one and
2528 * check the byte count.
2530 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2531 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2532 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2535 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: exit");
2536 xlog_verify_grant_head(log, 1);
2537 spin_unlock(&log->l_grant_lock);
2541 if (tic->t_flags & XLOG_TIC_IN_Q)
2542 xlog_del_ticketq(&log->l_reserve_headq, tic);
2543 xlog_trace_loggrant(log, tic, "xlog_grant_log_space: err_ret");
2545 * If we are failing, make sure the ticket doesn't have any
2546 * current reservations. We don't want to add this back when
2547 * the ticket/transaction gets cancelled.
2549 tic->t_curr_res = 0;
2550 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
2551 spin_unlock(&log->l_grant_lock);
2552 return XFS_ERROR(EIO);
2553 } /* xlog_grant_log_space */
2557 * Replenish the byte reservation required by moving the grant write head.
2562 xlog_regrant_write_log_space(xlog_t *log,
2565 int free_bytes, need_bytes;
2566 xlog_ticket_t *ntic;
2571 tic->t_curr_res = tic->t_unit_res;
2572 xlog_tic_reset_res(tic);
2578 if (log->l_flags & XLOG_ACTIVE_RECOVERY)
2579 panic("regrant Recovery problem");
2582 spin_lock(&log->l_grant_lock);
2583 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: enter");
2585 if (XLOG_FORCED_SHUTDOWN(log))
2588 /* If there are other waiters on the queue then give them a
2589 * chance at logspace before us. Wake up the first waiters,
2590 * if we do not wake up all the waiters then go to sleep waiting
2591 * for more free space, otherwise try to get some space for
2595 if ((ntic = log->l_write_headq)) {
2596 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2597 log->l_grant_write_bytes);
2599 ASSERT(ntic->t_flags & XLOG_TIC_PERM_RESERV);
2601 if (free_bytes < ntic->t_unit_res)
2603 free_bytes -= ntic->t_unit_res;
2604 sv_signal(&ntic->t_sema);
2605 ntic = ntic->t_next;
2606 } while (ntic != log->l_write_headq);
2608 if (ntic != log->l_write_headq) {
2609 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2610 xlog_ins_ticketq(&log->l_write_headq, tic);
2612 xlog_trace_loggrant(log, tic,
2613 "xlog_regrant_write_log_space: sleep 1");
2614 XFS_STATS_INC(xs_sleep_logspace);
2615 sv_wait(&tic->t_sema, PINOD|PLTWAIT,
2616 &log->l_grant_lock, s);
2618 /* If we're shutting down, this tic is already
2620 if (XLOG_FORCED_SHUTDOWN(log)) {
2621 spin_lock(&log->l_grant_lock);
2625 xlog_trace_loggrant(log, tic,
2626 "xlog_regrant_write_log_space: wake 1");
2627 xlog_grant_push_ail(log->l_mp, tic->t_unit_res);
2628 spin_lock(&log->l_grant_lock);
2632 need_bytes = tic->t_unit_res;
2635 if (XLOG_FORCED_SHUTDOWN(log))
2638 free_bytes = xlog_space_left(log, log->l_grant_write_cycle,
2639 log->l_grant_write_bytes);
2640 if (free_bytes < need_bytes) {
2641 if ((tic->t_flags & XLOG_TIC_IN_Q) == 0)
2642 xlog_ins_ticketq(&log->l_write_headq, tic);
2643 XFS_STATS_INC(xs_sleep_logspace);
2644 sv_wait(&tic->t_sema, PINOD|PLTWAIT, &log->l_grant_lock, s);
2646 /* If we're shutting down, this tic is already off the queue */
2647 if (XLOG_FORCED_SHUTDOWN(log)) {
2648 spin_lock(&log->l_grant_lock);
2652 xlog_trace_loggrant(log, tic,
2653 "xlog_regrant_write_log_space: wake 2");
2654 xlog_grant_push_ail(log->l_mp, need_bytes);
2655 spin_lock(&log->l_grant_lock);
2657 } else if (tic->t_flags & XLOG_TIC_IN_Q)
2658 xlog_del_ticketq(&log->l_write_headq, tic);
2660 /* we've got enough space */
2661 xlog_grant_add_space_write(log, need_bytes);
2663 tail_lsn = log->l_tail_lsn;
2664 if (CYCLE_LSN(tail_lsn) != log->l_grant_write_cycle) {
2665 ASSERT(log->l_grant_write_cycle-1 == CYCLE_LSN(tail_lsn));
2666 ASSERT(log->l_grant_write_bytes <= BBTOB(BLOCK_LSN(tail_lsn)));
2670 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: exit");
2671 xlog_verify_grant_head(log, 1);
2672 spin_unlock(&log->l_grant_lock);
2677 if (tic->t_flags & XLOG_TIC_IN_Q)
2678 xlog_del_ticketq(&log->l_reserve_headq, tic);
2679 xlog_trace_loggrant(log, tic, "xlog_regrant_write_log_space: err_ret");
2681 * If we are failing, make sure the ticket doesn't have any
2682 * current reservations. We don't want to add this back when
2683 * the ticket/transaction gets cancelled.
2685 tic->t_curr_res = 0;
2686 tic->t_cnt = 0; /* ungrant will give back unit_res * t_cnt. */
2687 spin_unlock(&log->l_grant_lock);
2688 return XFS_ERROR(EIO);
2689 } /* xlog_regrant_write_log_space */
2692 /* The first cnt-1 times through here we don't need to
2693 * move the grant write head because the permanent
2694 * reservation has reserved cnt times the unit amount.
2695 * Release part of current permanent unit reservation and
2696 * reset current reservation to be one units worth. Also
2697 * move grant reservation head forward.
2700 xlog_regrant_reserve_log_space(xlog_t *log,
2701 xlog_ticket_t *ticket)
2703 xlog_trace_loggrant(log, ticket,
2704 "xlog_regrant_reserve_log_space: enter");
2705 if (ticket->t_cnt > 0)
2708 spin_lock(&log->l_grant_lock);
2709 xlog_grant_sub_space(log, ticket->t_curr_res);
2710 ticket->t_curr_res = ticket->t_unit_res;
2711 xlog_tic_reset_res(ticket);
2712 xlog_trace_loggrant(log, ticket,
2713 "xlog_regrant_reserve_log_space: sub current res");
2714 xlog_verify_grant_head(log, 1);
2716 /* just return if we still have some of the pre-reserved space */
2717 if (ticket->t_cnt > 0) {
2718 spin_unlock(&log->l_grant_lock);
2722 xlog_grant_add_space_reserve(log, ticket->t_unit_res);
2723 xlog_trace_loggrant(log, ticket,
2724 "xlog_regrant_reserve_log_space: exit");
2725 xlog_verify_grant_head(log, 0);
2726 spin_unlock(&log->l_grant_lock);
2727 ticket->t_curr_res = ticket->t_unit_res;
2728 xlog_tic_reset_res(ticket);
2729 } /* xlog_regrant_reserve_log_space */
2733 * Give back the space left from a reservation.
2735 * All the information we need to make a correct determination of space left
2736 * is present. For non-permanent reservations, things are quite easy. The
2737 * count should have been decremented to zero. We only need to deal with the
2738 * space remaining in the current reservation part of the ticket. If the
2739 * ticket contains a permanent reservation, there may be left over space which
2740 * needs to be released. A count of N means that N-1 refills of the current
2741 * reservation can be done before we need to ask for more space. The first
2742 * one goes to fill up the first current reservation. Once we run out of
2743 * space, the count will stay at zero and the only space remaining will be
2744 * in the current reservation field.
2747 xlog_ungrant_log_space(xlog_t *log,
2748 xlog_ticket_t *ticket)
2750 if (ticket->t_cnt > 0)
2753 spin_lock(&log->l_grant_lock);
2754 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: enter");
2756 xlog_grant_sub_space(log, ticket->t_curr_res);
2758 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: sub current");
2760 /* If this is a permanent reservation ticket, we may be able to free
2761 * up more space based on the remaining count.
2763 if (ticket->t_cnt > 0) {
2764 ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
2765 xlog_grant_sub_space(log, ticket->t_unit_res*ticket->t_cnt);
2768 xlog_trace_loggrant(log, ticket, "xlog_ungrant_log_space: exit");
2769 xlog_verify_grant_head(log, 1);
2770 spin_unlock(&log->l_grant_lock);
2771 xfs_log_move_tail(log->l_mp, 1);
2772 } /* xlog_ungrant_log_space */
2776 * Atomically put back used ticket.
2779 xlog_state_put_ticket(xlog_t *log,
2782 spin_lock(&log->l_icloglock);
2783 xlog_ticket_put(log, tic);
2784 spin_unlock(&log->l_icloglock);
2785 } /* xlog_state_put_ticket */
2788 * Flush iclog to disk if this is the last reference to the given iclog and
2789 * the WANT_SYNC bit is set.
2791 * When this function is entered, the iclog is not necessarily in the
2792 * WANT_SYNC state. It may be sitting around waiting to get filled.
2797 xlog_state_release_iclog(xlog_t *log,
2798 xlog_in_core_t *iclog)
2800 int sync = 0; /* do we sync? */
2802 xlog_assign_tail_lsn(log->l_mp);
2804 spin_lock(&log->l_icloglock);
2806 if (iclog->ic_state & XLOG_STATE_IOERROR) {
2807 spin_unlock(&log->l_icloglock);
2808 return XFS_ERROR(EIO);
2811 ASSERT(iclog->ic_refcnt > 0);
2812 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
2813 iclog->ic_state == XLOG_STATE_WANT_SYNC);
2815 if (--iclog->ic_refcnt == 0 &&
2816 iclog->ic_state == XLOG_STATE_WANT_SYNC) {
2818 iclog->ic_state = XLOG_STATE_SYNCING;
2819 iclog->ic_header.h_tail_lsn = cpu_to_be64(log->l_tail_lsn);
2820 xlog_verify_tail_lsn(log, iclog, log->l_tail_lsn);
2821 /* cycle incremented when incrementing curr_block */
2824 spin_unlock(&log->l_icloglock);
2827 * We let the log lock go, so it's possible that we hit a log I/O
2828 * error or some other SHUTDOWN condition that marks the iclog
2829 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
2830 * this iclog has consistent data, so we ignore IOERROR
2831 * flags after this point.
2834 return xlog_sync(log, iclog);
2838 } /* xlog_state_release_iclog */
2842 * This routine will mark the current iclog in the ring as WANT_SYNC
2843 * and move the current iclog pointer to the next iclog in the ring.
2844 * When this routine is called from xlog_state_get_iclog_space(), the
2845 * exact size of the iclog has not yet been determined. All we know is
2846 * that every data block. We have run out of space in this log record.
2849 xlog_state_switch_iclogs(xlog_t *log,
2850 xlog_in_core_t *iclog,
2853 ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
2855 eventual_size = iclog->ic_offset;
2856 iclog->ic_state = XLOG_STATE_WANT_SYNC;
2857 iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
2858 log->l_prev_block = log->l_curr_block;
2859 log->l_prev_cycle = log->l_curr_cycle;
2861 /* roll log?: ic_offset changed later */
2862 log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);
2864 /* Round up to next log-sunit */
2865 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) &&
2866 log->l_mp->m_sb.sb_logsunit > 1) {
2867 __uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
2868 log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
2871 if (log->l_curr_block >= log->l_logBBsize) {
2872 log->l_curr_cycle++;
2873 if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
2874 log->l_curr_cycle++;
2875 log->l_curr_block -= log->l_logBBsize;
2876 ASSERT(log->l_curr_block >= 0);
2878 ASSERT(iclog == log->l_iclog);
2879 log->l_iclog = iclog->ic_next;
2880 } /* xlog_state_switch_iclogs */
2884 * Write out all data in the in-core log as of this exact moment in time.
2886 * Data may be written to the in-core log during this call. However,
2887 * we don't guarantee this data will be written out. A change from past
2888 * implementation means this routine will *not* write out zero length LRs.
2890 * Basically, we try and perform an intelligent scan of the in-core logs.
2891 * If we determine there is no flushable data, we just return. There is no
2892 * flushable data if:
2894 * 1. the current iclog is active and has no data; the previous iclog
2895 * is in the active or dirty state.
2896 * 2. the current iclog is drity, and the previous iclog is in the
2897 * active or dirty state.
2899 * We may sleep (call psema) if:
2901 * 1. the current iclog is not in the active nor dirty state.
2902 * 2. the current iclog dirty, and the previous iclog is not in the
2903 * active nor dirty state.
2904 * 3. the current iclog is active, and there is another thread writing
2905 * to this particular iclog.
2906 * 4. a) the current iclog is active and has no other writers
2907 * b) when we return from flushing out this iclog, it is still
2908 * not in the active nor dirty state.
2911 xlog_state_sync_all(xlog_t *log, uint flags, int *log_flushed)
2913 xlog_in_core_t *iclog;
2916 spin_lock(&log->l_icloglock);
2918 iclog = log->l_iclog;
2919 if (iclog->ic_state & XLOG_STATE_IOERROR) {
2920 spin_unlock(&log->l_icloglock);
2921 return XFS_ERROR(EIO);
2924 /* If the head iclog is not active nor dirty, we just attach
2925 * ourselves to the head and go to sleep.
2927 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2928 iclog->ic_state == XLOG_STATE_DIRTY) {
2930 * If the head is dirty or (active and empty), then
2931 * we need to look at the previous iclog. If the previous
2932 * iclog is active or dirty we are done. There is nothing
2933 * to sync out. Otherwise, we attach ourselves to the
2934 * previous iclog and go to sleep.
2936 if (iclog->ic_state == XLOG_STATE_DIRTY ||
2937 (iclog->ic_refcnt == 0 && iclog->ic_offset == 0)) {
2938 iclog = iclog->ic_prev;
2939 if (iclog->ic_state == XLOG_STATE_ACTIVE ||
2940 iclog->ic_state == XLOG_STATE_DIRTY)
2945 if (iclog->ic_refcnt == 0) {
2946 /* We are the only one with access to this
2947 * iclog. Flush it out now. There should
2948 * be a roundoff of zero to show that someone
2949 * has already taken care of the roundoff from
2950 * the previous sync.
2953 lsn = be64_to_cpu(iclog->ic_header.h_lsn);
2954 xlog_state_switch_iclogs(log, iclog, 0);
2955 spin_unlock(&log->l_icloglock);
2957 if (xlog_state_release_iclog(log, iclog))
2958 return XFS_ERROR(EIO);
2960 spin_lock(&log->l_icloglock);
2961 if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
2962 iclog->ic_state != XLOG_STATE_DIRTY)
2967 /* Someone else is writing to this iclog.
2968 * Use its call to flush out the data. However,
2969 * the other thread may not force out this LR,
2970 * so we mark it WANT_SYNC.
2972 xlog_state_switch_iclogs(log, iclog, 0);
2978 /* By the time we come around again, the iclog could've been filled
2979 * which would give it another lsn. If we have a new lsn, just
2980 * return because the relevant data has been flushed.
2983 if (flags & XFS_LOG_SYNC) {
2985 * We must check if we're shutting down here, before
2986 * we wait, while we're holding the l_icloglock.
2987 * Then we check again after waking up, in case our
2988 * sleep was disturbed by a bad news.
2990 if (iclog->ic_state & XLOG_STATE_IOERROR) {
2991 spin_unlock(&log->l_icloglock);
2992 return XFS_ERROR(EIO);
2994 XFS_STATS_INC(xs_log_force_sleep);
2995 sv_wait(&iclog->ic_forcesema, PINOD, &log->l_icloglock, s);
2997 * No need to grab the log lock here since we're
2998 * only deciding whether or not to return EIO
2999 * and the memory read should be atomic.
3001 if (iclog->ic_state & XLOG_STATE_IOERROR)
3002 return XFS_ERROR(EIO);
3008 spin_unlock(&log->l_icloglock);
3011 } /* xlog_state_sync_all */
3015 * Used by code which implements synchronous log forces.
3017 * Find in-core log with lsn.
3018 * If it is in the DIRTY state, just return.
3019 * If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
3020 * state and go to sleep or return.
3021 * If it is in any other state, go to sleep or return.
3023 * If filesystem activity goes to zero, the iclog will get flushed only by
3027 xlog_state_sync(xlog_t *log,
3032 xlog_in_core_t *iclog;
3033 int already_slept = 0;
3036 spin_lock(&log->l_icloglock);
3037 iclog = log->l_iclog;
3039 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3040 spin_unlock(&log->l_icloglock);
3041 return XFS_ERROR(EIO);
3045 if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
3046 iclog = iclog->ic_next;
3050 if (iclog->ic_state == XLOG_STATE_DIRTY) {
3051 spin_unlock(&log->l_icloglock);
3055 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3057 * We sleep here if we haven't already slept (e.g.
3058 * this is the first time we've looked at the correct
3059 * iclog buf) and the buffer before us is going to
3060 * be sync'ed. The reason for this is that if we
3061 * are doing sync transactions here, by waiting for
3062 * the previous I/O to complete, we can allow a few
3063 * more transactions into this iclog before we close
3066 * Otherwise, we mark the buffer WANT_SYNC, and bump
3067 * up the refcnt so we can release the log (which drops
3068 * the ref count). The state switch keeps new transaction
3069 * commits from using this buffer. When the current commits
3070 * finish writing into the buffer, the refcount will drop to
3071 * zero and the buffer will go out then.
3073 if (!already_slept &&
3074 (iclog->ic_prev->ic_state & (XLOG_STATE_WANT_SYNC |
3075 XLOG_STATE_SYNCING))) {
3076 ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
3077 XFS_STATS_INC(xs_log_force_sleep);
3078 sv_wait(&iclog->ic_prev->ic_writesema, PSWP,
3079 &log->l_icloglock, s);
3085 xlog_state_switch_iclogs(log, iclog, 0);
3086 spin_unlock(&log->l_icloglock);
3087 if (xlog_state_release_iclog(log, iclog))
3088 return XFS_ERROR(EIO);
3090 spin_lock(&log->l_icloglock);
3094 if ((flags & XFS_LOG_SYNC) && /* sleep */
3095 !(iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
3098 * Don't wait on the forcesema if we know that we've
3099 * gotten a log write error.
3101 if (iclog->ic_state & XLOG_STATE_IOERROR) {
3102 spin_unlock(&log->l_icloglock);
3103 return XFS_ERROR(EIO);
3105 XFS_STATS_INC(xs_log_force_sleep);
3106 sv_wait(&iclog->ic_forcesema, PSWP, &log->l_icloglock, s);
3108 * No need to grab the log lock here since we're
3109 * only deciding whether or not to return EIO
3110 * and the memory read should be atomic.
3112 if (iclog->ic_state & XLOG_STATE_IOERROR)
3113 return XFS_ERROR(EIO);
3115 } else { /* just return */
3116 spin_unlock(&log->l_icloglock);
3120 } while (iclog != log->l_iclog);
3122 spin_unlock(&log->l_icloglock);
3124 } /* xlog_state_sync */
3128 * Called when we want to mark the current iclog as being ready to sync to
3132 xlog_state_want_sync(xlog_t *log, xlog_in_core_t *iclog)
3134 spin_lock(&log->l_icloglock);
3136 if (iclog->ic_state == XLOG_STATE_ACTIVE) {
3137 xlog_state_switch_iclogs(log, iclog, 0);
3139 ASSERT(iclog->ic_state &
3140 (XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
3143 spin_unlock(&log->l_icloglock);
3144 } /* xlog_state_want_sync */
3148 /*****************************************************************************
3152 *****************************************************************************
3156 * Algorithm doesn't take into account page size. ;-(
3159 xlog_state_ticket_alloc(xlog_t *log)
3161 xlog_ticket_t *t_list;
3162 xlog_ticket_t *next;
3164 uint i = (PAGE_SIZE / sizeof(xlog_ticket_t)) - 2;
3167 * The kmem_zalloc may sleep, so we shouldn't be holding the
3168 * global lock. XXXmiken: may want to use zone allocator.
3170 buf = (xfs_caddr_t) kmem_zalloc(PAGE_SIZE, KM_SLEEP);
3172 spin_lock(&log->l_icloglock);
3174 /* Attach 1st ticket to Q, so we can keep track of allocated memory */
3175 t_list = (xlog_ticket_t *)buf;
3176 t_list->t_next = log->l_unmount_free;
3177 log->l_unmount_free = t_list++;
3178 log->l_ticket_cnt++;
3179 log->l_ticket_tcnt++;
3181 /* Next ticket becomes first ticket attached to ticket free list */
3182 if (log->l_freelist != NULL) {
3183 ASSERT(log->l_tail != NULL);
3184 log->l_tail->t_next = t_list;
3186 log->l_freelist = t_list;
3188 log->l_ticket_cnt++;
3189 log->l_ticket_tcnt++;
3191 /* Cycle through rest of alloc'ed memory, building up free Q */
3192 for ( ; i > 0; i--) {
3194 t_list->t_next = next;
3196 log->l_ticket_cnt++;
3197 log->l_ticket_tcnt++;
3199 t_list->t_next = NULL;
3200 log->l_tail = t_list;
3201 spin_unlock(&log->l_icloglock);
3202 } /* xlog_state_ticket_alloc */
3206 * Put ticket into free list
3208 * Assumption: log lock is held around this call.
3211 xlog_ticket_put(xlog_t *log,
3212 xlog_ticket_t *ticket)
3214 sv_destroy(&ticket->t_sema);
3217 * Don't think caching will make that much difference. It's
3218 * more important to make debug easier.
3221 /* real code will want to use LIFO for caching */
3222 ticket->t_next = log->l_freelist;
3223 log->l_freelist = ticket;
3224 /* no need to clear fields */
3226 /* When we debug, it is easier if tickets are cycled */
3227 ticket->t_next = NULL;
3229 log->l_tail->t_next = ticket;
3231 ASSERT(log->l_freelist == NULL);
3232 log->l_freelist = ticket;
3234 log->l_tail = ticket;
3236 log->l_ticket_cnt++;
3237 } /* xlog_ticket_put */
3241 * Grab ticket off freelist or allocation some more
3243 STATIC xlog_ticket_t *
3244 xlog_ticket_get(xlog_t *log,
3254 if (log->l_freelist == NULL)
3255 xlog_state_ticket_alloc(log); /* potentially sleep */
3257 spin_lock(&log->l_icloglock);
3258 if (log->l_freelist == NULL) {
3259 spin_unlock(&log->l_icloglock);
3262 tic = log->l_freelist;
3263 log->l_freelist = tic->t_next;
3264 if (log->l_freelist == NULL)
3266 log->l_ticket_cnt--;
3267 spin_unlock(&log->l_icloglock);
3270 * Permanent reservations have up to 'cnt'-1 active log operations
3271 * in the log. A unit in this case is the amount of space for one
3272 * of these log operations. Normal reservations have a cnt of 1
3273 * and their unit amount is the total amount of space required.
3275 * The following lines of code account for non-transaction data
3276 * which occupy space in the on-disk log.
3278 * Normal form of a transaction is:
3279 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
3280 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
3282 * We need to account for all the leadup data and trailer data
3283 * around the transaction data.
3284 * And then we need to account for the worst case in terms of using
3286 * The worst case will happen if:
3287 * - the placement of the transaction happens to be such that the
3288 * roundoff is at its maximum
3289 * - the transaction data is synced before the commit record is synced
3290 * i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
3291 * Therefore the commit record is in its own Log Record.
3292 * This can happen as the commit record is called with its
3293 * own region to xlog_write().
3294 * This then means that in the worst case, roundoff can happen for
3295 * the commit-rec as well.
3296 * The commit-rec is smaller than padding in this scenario and so it is
3297 * not added separately.
3300 /* for trans header */
3301 unit_bytes += sizeof(xlog_op_header_t);
3302 unit_bytes += sizeof(xfs_trans_header_t);
3305 unit_bytes += sizeof(xlog_op_header_t);
3307 /* for LR headers */
3308 num_headers = ((unit_bytes + log->l_iclog_size-1) >> log->l_iclog_size_log);
3309 unit_bytes += log->l_iclog_hsize * num_headers;
3311 /* for commit-rec LR header - note: padding will subsume the ophdr */
3312 unit_bytes += log->l_iclog_hsize;
3314 /* for split-recs - ophdrs added when data split over LRs */
3315 unit_bytes += sizeof(xlog_op_header_t) * num_headers;
3317 /* for roundoff padding for transaction data and one for commit record */
3318 if (XFS_SB_VERSION_HASLOGV2(&log->l_mp->m_sb) &&
3319 log->l_mp->m_sb.sb_logsunit > 1) {
3320 /* log su roundoff */
3321 unit_bytes += 2*log->l_mp->m_sb.sb_logsunit;
3324 unit_bytes += 2*BBSIZE;
3327 tic->t_unit_res = unit_bytes;
3328 tic->t_curr_res = unit_bytes;
3331 tic->t_tid = (xlog_tid_t)((__psint_t)tic & 0xffffffff);
3332 tic->t_clientid = client;
3333 tic->t_flags = XLOG_TIC_INITED;
3334 tic->t_trans_type = 0;
3335 if (xflags & XFS_LOG_PERM_RESERV)
3336 tic->t_flags |= XLOG_TIC_PERM_RESERV;
3337 sv_init(&(tic->t_sema), SV_DEFAULT, "logtick");
3339 xlog_tic_reset_res(tic);
3342 } /* xlog_ticket_get */
3345 /******************************************************************************
3347 * Log debug routines
3349 ******************************************************************************
3353 * Make sure that the destination ptr is within the valid data region of
3354 * one of the iclogs. This uses backup pointers stored in a different
3355 * part of the log in case we trash the log structure.
3358 xlog_verify_dest_ptr(xlog_t *log,
3364 for (i=0; i < log->l_iclog_bufs; i++) {
3365 if (ptr >= (__psint_t)log->l_iclog_bak[i] &&
3366 ptr <= (__psint_t)log->l_iclog_bak[i]+log->l_iclog_size)
3370 xlog_panic("xlog_verify_dest_ptr: invalid ptr");
3371 } /* xlog_verify_dest_ptr */
3374 xlog_verify_grant_head(xlog_t *log, int equals)
3376 if (log->l_grant_reserve_cycle == log->l_grant_write_cycle) {
3378 ASSERT(log->l_grant_reserve_bytes >= log->l_grant_write_bytes);
3380 ASSERT(log->l_grant_reserve_bytes > log->l_grant_write_bytes);
3382 ASSERT(log->l_grant_reserve_cycle-1 == log->l_grant_write_cycle);
3383 ASSERT(log->l_grant_write_bytes >= log->l_grant_reserve_bytes);
3385 } /* xlog_verify_grant_head */
3387 /* check if it will fit */
3389 xlog_verify_tail_lsn(xlog_t *log,
3390 xlog_in_core_t *iclog,
3395 if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
3397 log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
3398 if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3399 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3401 ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);
3403 if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3404 xlog_panic("xlog_verify_tail_lsn: tail wrapped");
3406 blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
3407 if (blocks < BTOBB(iclog->ic_offset) + 1)
3408 xlog_panic("xlog_verify_tail_lsn: ran out of log space");
3410 } /* xlog_verify_tail_lsn */
3413 * Perform a number of checks on the iclog before writing to disk.
3415 * 1. Make sure the iclogs are still circular
3416 * 2. Make sure we have a good magic number
3417 * 3. Make sure we don't have magic numbers in the data
3418 * 4. Check fields of each log operation header for:
3419 * A. Valid client identifier
3420 * B. tid ptr value falls in valid ptr space (user space code)
3421 * C. Length in log record header is correct according to the
3422 * individual operation headers within record.
3423 * 5. When a bwrite will occur within 5 blocks of the front of the physical
3424 * log, check the preceding blocks of the physical log to make sure all
3425 * the cycle numbers agree with the current cycle number.
3428 xlog_verify_iclog(xlog_t *log,
3429 xlog_in_core_t *iclog,
3433 xlog_op_header_t *ophead;
3434 xlog_in_core_t *icptr;
3435 xlog_in_core_2_t *xhdr;
3437 xfs_caddr_t base_ptr;
3438 __psint_t field_offset;
3440 int len, i, j, k, op_len;
3443 /* check validity of iclog pointers */
3444 spin_lock(&log->l_icloglock);
3445 icptr = log->l_iclog;
3446 for (i=0; i < log->l_iclog_bufs; i++) {
3448 xlog_panic("xlog_verify_iclog: invalid ptr");
3449 icptr = icptr->ic_next;
3451 if (icptr != log->l_iclog)
3452 xlog_panic("xlog_verify_iclog: corrupt iclog ring");
3453 spin_unlock(&log->l_icloglock);
3455 /* check log magic numbers */
3456 if (be32_to_cpu(iclog->ic_header.h_magicno) != XLOG_HEADER_MAGIC_NUM)
3457 xlog_panic("xlog_verify_iclog: invalid magic num");
3459 ptr = (xfs_caddr_t) &iclog->ic_header;
3460 for (ptr += BBSIZE; ptr < ((xfs_caddr_t)&iclog->ic_header) + count;
3462 if (be32_to_cpu(*(__be32 *)ptr) == XLOG_HEADER_MAGIC_NUM)
3463 xlog_panic("xlog_verify_iclog: unexpected magic num");
3467 len = be32_to_cpu(iclog->ic_header.h_num_logops);
3468 ptr = iclog->ic_datap;
3470 ophead = (xlog_op_header_t *)ptr;
3471 xhdr = (xlog_in_core_2_t *)&iclog->ic_header;
3472 for (i = 0; i < len; i++) {
3473 ophead = (xlog_op_header_t *)ptr;
3475 /* clientid is only 1 byte */
3476 field_offset = (__psint_t)
3477 ((xfs_caddr_t)&(ophead->oh_clientid) - base_ptr);
3478 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3479 clientid = ophead->oh_clientid;
3481 idx = BTOBBT((xfs_caddr_t)&(ophead->oh_clientid) - iclog->ic_datap);
3482 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3483 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3484 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3485 clientid = xlog_get_client_id(
3486 xhdr[j].hic_xheader.xh_cycle_data[k]);
3488 clientid = xlog_get_client_id(
3489 iclog->ic_header.h_cycle_data[idx]);
3492 if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
3493 cmn_err(CE_WARN, "xlog_verify_iclog: "
3494 "invalid clientid %d op 0x%p offset 0x%lx",
3495 clientid, ophead, (unsigned long)field_offset);
3498 field_offset = (__psint_t)
3499 ((xfs_caddr_t)&(ophead->oh_len) - base_ptr);
3500 if (syncing == B_FALSE || (field_offset & 0x1ff)) {
3501 op_len = be32_to_cpu(ophead->oh_len);
3503 idx = BTOBBT((__psint_t)&ophead->oh_len -
3504 (__psint_t)iclog->ic_datap);
3505 if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
3506 j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3507 k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3508 op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
3510 op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
3513 ptr += sizeof(xlog_op_header_t) + op_len;
3515 } /* xlog_verify_iclog */
3519 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
3525 xlog_in_core_t *iclog, *ic;
3527 iclog = log->l_iclog;
3528 if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
3530 * Mark all the incore logs IOERROR.
3531 * From now on, no log flushes will result.
3535 ic->ic_state = XLOG_STATE_IOERROR;
3537 } while (ic != iclog);
3541 * Return non-zero, if state transition has already happened.
3547 * This is called from xfs_force_shutdown, when we're forcibly
3548 * shutting down the filesystem, typically because of an IO error.
3549 * Our main objectives here are to make sure that:
3550 * a. the filesystem gets marked 'SHUTDOWN' for all interested
3551 * parties to find out, 'atomically'.
3552 * b. those who're sleeping on log reservations, pinned objects and
3553 * other resources get woken up, and be told the bad news.
3554 * c. nothing new gets queued up after (a) and (b) are done.
3555 * d. if !logerror, flush the iclogs to disk, then seal them off
3559 xfs_log_force_umount(
3560 struct xfs_mount *mp,
3571 * If this happens during log recovery, don't worry about
3572 * locking; the log isn't open for business yet.
3575 log->l_flags & XLOG_ACTIVE_RECOVERY) {
3576 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3577 XFS_BUF_DONE(mp->m_sb_bp);
3582 * Somebody could've already done the hard work for us.
3583 * No need to get locks for this.
3585 if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
3586 ASSERT(XLOG_FORCED_SHUTDOWN(log));
3591 * We must hold both the GRANT lock and the LOG lock,
3592 * before we mark the filesystem SHUTDOWN and wake
3593 * everybody up to tell the bad news.
3595 spin_lock(&log->l_grant_lock);
3596 spin_lock(&log->l_icloglock);
3597 mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3598 XFS_BUF_DONE(mp->m_sb_bp);
3600 * This flag is sort of redundant because of the mount flag, but
3601 * it's good to maintain the separation between the log and the rest
3604 log->l_flags |= XLOG_IO_ERROR;
3607 * If we hit a log error, we want to mark all the iclogs IOERROR
3608 * while we're still holding the loglock.
3611 retval = xlog_state_ioerror(log);
3612 spin_unlock(&log->l_icloglock);
3615 * We don't want anybody waiting for log reservations
3616 * after this. That means we have to wake up everybody
3617 * queued up on reserve_headq as well as write_headq.
3618 * In addition, we make sure in xlog_{re}grant_log_space
3619 * that we don't enqueue anything once the SHUTDOWN flag
3620 * is set, and this action is protected by the GRANTLOCK.
3622 if ((tic = log->l_reserve_headq)) {
3624 sv_signal(&tic->t_sema);
3626 } while (tic != log->l_reserve_headq);
3629 if ((tic = log->l_write_headq)) {
3631 sv_signal(&tic->t_sema);
3633 } while (tic != log->l_write_headq);
3635 spin_unlock(&log->l_grant_lock);
3637 if (! (log->l_iclog->ic_state & XLOG_STATE_IOERROR)) {
3640 * Force the incore logs to disk before shutting the
3641 * log down completely.
3643 xlog_state_sync_all(log, XFS_LOG_FORCE|XFS_LOG_SYNC, &dummy);
3644 spin_lock(&log->l_icloglock);
3645 retval = xlog_state_ioerror(log);
3646 spin_unlock(&log->l_icloglock);
3649 * Wake up everybody waiting on xfs_log_force.
3650 * Callback all log item committed functions as if the
3651 * log writes were completed.
3653 xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);
3655 #ifdef XFSERRORDEBUG
3657 xlog_in_core_t *iclog;
3659 spin_lock(&log->l_icloglock);
3660 iclog = log->l_iclog;
3662 ASSERT(iclog->ic_callback == 0);
3663 iclog = iclog->ic_next;
3664 } while (iclog != log->l_iclog);
3665 spin_unlock(&log->l_icloglock);
3668 /* return non-zero if log IOERROR transition had already happened */
3673 xlog_iclogs_empty(xlog_t *log)
3675 xlog_in_core_t *iclog;
3677 iclog = log->l_iclog;
3679 /* endianness does not matter here, zero is zero in
3682 if (iclog->ic_header.h_num_logops)
3684 iclog = iclog->ic_next;
3685 } while (iclog != log->l_iclog);