2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_error.h"
32 #include "xfs_da_btree.h"
33 #include "xfs_bmap_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_ialloc_btree.h"
36 #include "xfs_dir_sf.h"
37 #include "xfs_dir2_sf.h"
38 #include "xfs_attr_sf.h"
39 #include "xfs_dinode.h"
40 #include "xfs_inode.h"
41 #include "xfs_btree.h"
42 #include "xfs_ialloc.h"
43 #include "xfs_alloc.h"
45 #include "xfs_quota.h"
46 #include "xfs_trans_priv.h"
47 #include "xfs_trans_space.h"
50 STATIC void xfs_trans_apply_sb_deltas(xfs_trans_t *);
51 STATIC uint xfs_trans_count_vecs(xfs_trans_t *);
52 STATIC void xfs_trans_fill_vecs(xfs_trans_t *, xfs_log_iovec_t *);
53 STATIC void xfs_trans_uncommit(xfs_trans_t *, uint);
54 STATIC void xfs_trans_committed(xfs_trans_t *, int);
55 STATIC void xfs_trans_chunk_committed(xfs_log_item_chunk_t *, xfs_lsn_t, int);
56 STATIC void xfs_trans_free(xfs_trans_t *);
58 kmem_zone_t *xfs_trans_zone;
62 * Initialize the precomputed transaction reservation values
63 * in the mount structure.
69 xfs_trans_reservations_t *resp;
71 resp = &(mp->m_reservations);
73 (uint)(XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
75 (uint)(XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
77 (uint)(XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
78 resp->tr_link = (uint)XFS_CALC_LINK_LOG_RES(mp);
80 (uint)(XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
82 (uint)(XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
84 (uint)(XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
86 (uint)(XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
88 (uint)(XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
90 (uint)(XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
91 resp->tr_growdata = (uint)XFS_CALC_GROWDATA_LOG_RES(mp);
92 resp->tr_swrite = (uint)XFS_CALC_SWRITE_LOG_RES(mp);
93 resp->tr_writeid = (uint)XFS_CALC_WRITEID_LOG_RES(mp);
95 (uint)(XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
96 resp->tr_attrinval = (uint)XFS_CALC_ATTRINVAL_LOG_RES(mp);
98 (uint)(XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
100 (uint)(XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp));
101 resp->tr_clearagi = (uint)XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
102 resp->tr_growrtalloc = (uint)XFS_CALC_GROWRTALLOC_LOG_RES(mp);
103 resp->tr_growrtzero = (uint)XFS_CALC_GROWRTZERO_LOG_RES(mp);
104 resp->tr_growrtfree = (uint)XFS_CALC_GROWRTFREE_LOG_RES(mp);
108 * This routine is called to allocate a transaction structure.
109 * The type parameter indicates the type of the transaction. These
110 * are enumerated in xfs_trans.h.
112 * Dynamically allocate the transaction structure from the transaction
113 * zone, initialize it, and return it to the caller.
120 fs_check_frozen(XFS_MTOVFS(mp), SB_FREEZE_TRANS);
121 atomic_inc(&mp->m_active_trans);
123 return (_xfs_trans_alloc(mp, type));
134 ASSERT(xfs_trans_zone != NULL);
135 tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
138 * Initialize the transaction structure.
140 tp->t_magic = XFS_TRANS_MAGIC;
143 tp->t_items_free = XFS_LIC_NUM_SLOTS;
144 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
145 XFS_LIC_INIT(&(tp->t_items));
146 XFS_LBC_INIT(&(tp->t_busy));
152 * This is called to create a new transaction which will share the
153 * permanent log reservation of the given transaction. The remaining
154 * unused block and rt extent reservations are also inherited. This
155 * implies that the original transaction is no longer allowed to allocate
156 * blocks. Locks and log items, however, are no inherited. They must
157 * be added to the new transaction explicitly.
165 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
168 * Initialize the new transaction structure.
170 ntp->t_magic = XFS_TRANS_MAGIC;
171 ntp->t_type = tp->t_type;
172 ntp->t_mountp = tp->t_mountp;
173 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
174 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
175 XFS_LIC_INIT(&(ntp->t_items));
176 XFS_LBC_INIT(&(ntp->t_busy));
178 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
180 #if defined(XLOG_NOLOG) || defined(DEBUG)
181 ASSERT(!xlog_debug || tp->t_ticket != NULL);
183 ASSERT(tp->t_ticket != NULL);
185 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
186 ntp->t_ticket = tp->t_ticket;
187 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
188 tp->t_blk_res = tp->t_blk_res_used;
189 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
190 tp->t_rtx_res = tp->t_rtx_res_used;
191 PFLAGS_DUP(&tp->t_pflags, &ntp->t_pflags);
193 XFS_TRANS_DUP_DQINFO(tp->t_mountp, tp, ntp);
195 atomic_inc(&tp->t_mountp->m_active_trans);
200 * This is called to reserve free disk blocks and log space for the
201 * given transaction. This must be done before allocating any resources
202 * within the transaction.
204 * This will return ENOSPC if there are not enough blocks available.
205 * It will sleep waiting for available log space.
206 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
207 * is used by long running transactions. If any one of the reservations
208 * fails then they will all be backed out.
210 * This does not do quota reservations. That typically is done by the
227 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
229 /* Mark this thread as being in a transaction */
230 PFLAGS_SET_FSTRANS(&tp->t_pflags);
233 * Attempt to reserve the needed disk blocks by decrementing
234 * the number needed from the number available. This will
235 * fail if the count would go below zero.
238 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
241 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
242 return (XFS_ERROR(ENOSPC));
244 tp->t_blk_res += blocks;
248 * Reserve the log space needed for this transaction.
251 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
252 ASSERT((tp->t_log_count == 0) ||
253 (tp->t_log_count == logcount));
254 if (flags & XFS_TRANS_PERM_LOG_RES) {
255 log_flags = XFS_LOG_PERM_RESERV;
256 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
258 ASSERT(tp->t_ticket == NULL);
259 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
263 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
265 XFS_TRANSACTION, log_flags, tp->t_type);
269 tp->t_log_res = logspace;
270 tp->t_log_count = logcount;
274 * Attempt to reserve the needed realtime extents by decrementing
275 * the number needed from the number available. This will
276 * fail if the count would go below zero.
279 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
282 error = XFS_ERROR(ENOSPC);
285 tp->t_rtx_res += rtextents;
291 * Error cases jump to one of these labels to undo any
292 * reservations which have already been performed.
296 if (flags & XFS_TRANS_PERM_LOG_RES) {
297 log_flags = XFS_LOG_REL_PERM_RESERV;
301 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
304 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
309 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
314 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
321 * Record the indicated change to the given field for application
322 * to the file system's superblock when the transaction commits.
323 * For now, just store the change in the transaction structure.
325 * Mark the transaction structure to indicate that the superblock
326 * needs to be updated before committing.
336 case XFS_TRANS_SB_ICOUNT:
337 tp->t_icount_delta += delta;
339 case XFS_TRANS_SB_IFREE:
340 tp->t_ifree_delta += delta;
342 case XFS_TRANS_SB_FDBLOCKS:
344 * Track the number of blocks allocated in the
345 * transaction. Make sure it does not exceed the
349 tp->t_blk_res_used += (uint)-delta;
350 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
352 tp->t_fdblocks_delta += delta;
354 case XFS_TRANS_SB_RES_FDBLOCKS:
356 * The allocation has already been applied to the
357 * in-core superblock's counter. This should only
358 * be applied to the on-disk superblock.
361 tp->t_res_fdblocks_delta += delta;
363 case XFS_TRANS_SB_FREXTENTS:
365 * Track the number of blocks allocated in the
366 * transaction. Make sure it does not exceed the
370 tp->t_rtx_res_used += (uint)-delta;
371 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
373 tp->t_frextents_delta += delta;
375 case XFS_TRANS_SB_RES_FREXTENTS:
377 * The allocation has already been applied to the
378 * in-core superblocks's counter. This should only
379 * be applied to the on-disk superblock.
382 tp->t_res_frextents_delta += delta;
384 case XFS_TRANS_SB_DBLOCKS:
386 tp->t_dblocks_delta += delta;
388 case XFS_TRANS_SB_AGCOUNT:
390 tp->t_agcount_delta += delta;
392 case XFS_TRANS_SB_IMAXPCT:
393 tp->t_imaxpct_delta += delta;
395 case XFS_TRANS_SB_REXTSIZE:
396 tp->t_rextsize_delta += delta;
398 case XFS_TRANS_SB_RBMBLOCKS:
399 tp->t_rbmblocks_delta += delta;
401 case XFS_TRANS_SB_RBLOCKS:
402 tp->t_rblocks_delta += delta;
404 case XFS_TRANS_SB_REXTENTS:
405 tp->t_rextents_delta += delta;
407 case XFS_TRANS_SB_REXTSLOG:
408 tp->t_rextslog_delta += delta;
415 tp->t_flags |= (XFS_TRANS_SB_DIRTY | XFS_TRANS_DIRTY);
419 * xfs_trans_apply_sb_deltas() is called from the commit code
420 * to bring the superblock buffer into the current transaction
421 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
423 * For now we just look at each field allowed to change and change
427 xfs_trans_apply_sb_deltas(
434 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
435 sbp = XFS_BUF_TO_SBP(bp);
438 * Check that superblock mods match the mods made to AGF counters.
440 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
441 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
442 tp->t_ag_btree_delta));
444 if (tp->t_icount_delta != 0) {
445 INT_MOD(sbp->sb_icount, ARCH_CONVERT, tp->t_icount_delta);
447 if (tp->t_ifree_delta != 0) {
448 INT_MOD(sbp->sb_ifree, ARCH_CONVERT, tp->t_ifree_delta);
451 if (tp->t_fdblocks_delta != 0) {
452 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_fdblocks_delta);
454 if (tp->t_res_fdblocks_delta != 0) {
455 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_res_fdblocks_delta);
458 if (tp->t_frextents_delta != 0) {
459 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_frextents_delta);
461 if (tp->t_res_frextents_delta != 0) {
462 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_res_frextents_delta);
464 if (tp->t_dblocks_delta != 0) {
465 INT_MOD(sbp->sb_dblocks, ARCH_CONVERT, tp->t_dblocks_delta);
468 if (tp->t_agcount_delta != 0) {
469 INT_MOD(sbp->sb_agcount, ARCH_CONVERT, tp->t_agcount_delta);
472 if (tp->t_imaxpct_delta != 0) {
473 INT_MOD(sbp->sb_imax_pct, ARCH_CONVERT, tp->t_imaxpct_delta);
476 if (tp->t_rextsize_delta != 0) {
477 INT_MOD(sbp->sb_rextsize, ARCH_CONVERT, tp->t_rextsize_delta);
480 if (tp->t_rbmblocks_delta != 0) {
481 INT_MOD(sbp->sb_rbmblocks, ARCH_CONVERT, tp->t_rbmblocks_delta);
484 if (tp->t_rblocks_delta != 0) {
485 INT_MOD(sbp->sb_rblocks, ARCH_CONVERT, tp->t_rblocks_delta);
488 if (tp->t_rextents_delta != 0) {
489 INT_MOD(sbp->sb_rextents, ARCH_CONVERT, tp->t_rextents_delta);
492 if (tp->t_rextslog_delta != 0) {
493 INT_MOD(sbp->sb_rextslog, ARCH_CONVERT, tp->t_rextslog_delta);
499 * Log the whole thing, the fields are discontiguous.
501 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_sb_t) - 1);
504 * Since all the modifiable fields are contiguous, we
505 * can get away with this.
507 xfs_trans_log_buf(tp, bp, offsetof(xfs_sb_t, sb_icount),
508 offsetof(xfs_sb_t, sb_frextents) +
509 sizeof(sbp->sb_frextents) - 1);
511 XFS_MTOVFS(tp->t_mountp)->vfs_super->s_dirt = 1;
515 * xfs_trans_unreserve_and_mod_sb() is called to release unused
516 * reservations and apply superblock counter changes to the in-core
519 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
522 xfs_trans_unreserve_and_mod_sb(
525 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
532 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
535 * Release any reserved blocks. Any that were allocated
536 * will be taken back again by fdblocks_delta below.
538 if (tp->t_blk_res > 0) {
539 msbp->msb_field = XFS_SBS_FDBLOCKS;
540 msbp->msb_delta = tp->t_blk_res;
545 * Release any reserved real time extents . Any that were
546 * allocated will be taken back again by frextents_delta below.
548 if (tp->t_rtx_res > 0) {
549 msbp->msb_field = XFS_SBS_FREXTENTS;
550 msbp->msb_delta = tp->t_rtx_res;
555 * Apply any superblock modifications to the in-core version.
556 * The t_res_fdblocks_delta and t_res_frextents_delta fields are
557 * explicity NOT applied to the in-core superblock.
558 * The idea is that that has already been done.
560 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
561 if (tp->t_icount_delta != 0) {
562 msbp->msb_field = XFS_SBS_ICOUNT;
563 msbp->msb_delta = (int)tp->t_icount_delta;
566 if (tp->t_ifree_delta != 0) {
567 msbp->msb_field = XFS_SBS_IFREE;
568 msbp->msb_delta = (int)tp->t_ifree_delta;
571 if (tp->t_fdblocks_delta != 0) {
572 msbp->msb_field = XFS_SBS_FDBLOCKS;
573 msbp->msb_delta = (int)tp->t_fdblocks_delta;
576 if (tp->t_frextents_delta != 0) {
577 msbp->msb_field = XFS_SBS_FREXTENTS;
578 msbp->msb_delta = (int)tp->t_frextents_delta;
581 if (tp->t_dblocks_delta != 0) {
582 msbp->msb_field = XFS_SBS_DBLOCKS;
583 msbp->msb_delta = (int)tp->t_dblocks_delta;
586 if (tp->t_agcount_delta != 0) {
587 msbp->msb_field = XFS_SBS_AGCOUNT;
588 msbp->msb_delta = (int)tp->t_agcount_delta;
591 if (tp->t_imaxpct_delta != 0) {
592 msbp->msb_field = XFS_SBS_IMAX_PCT;
593 msbp->msb_delta = (int)tp->t_imaxpct_delta;
596 if (tp->t_rextsize_delta != 0) {
597 msbp->msb_field = XFS_SBS_REXTSIZE;
598 msbp->msb_delta = (int)tp->t_rextsize_delta;
601 if (tp->t_rbmblocks_delta != 0) {
602 msbp->msb_field = XFS_SBS_RBMBLOCKS;
603 msbp->msb_delta = (int)tp->t_rbmblocks_delta;
606 if (tp->t_rblocks_delta != 0) {
607 msbp->msb_field = XFS_SBS_RBLOCKS;
608 msbp->msb_delta = (int)tp->t_rblocks_delta;
611 if (tp->t_rextents_delta != 0) {
612 msbp->msb_field = XFS_SBS_REXTENTS;
613 msbp->msb_delta = (int)tp->t_rextents_delta;
616 if (tp->t_rextslog_delta != 0) {
617 msbp->msb_field = XFS_SBS_REXTSLOG;
618 msbp->msb_delta = (int)tp->t_rextslog_delta;
624 * If we need to change anything, do it.
627 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
628 (uint)(msbp - msb), rsvd);
637 * Commit the given transaction to the log a/synchronously.
639 * XFS disk error handling mechanism is not based on a typical
640 * transaction abort mechanism. Logically after the filesystem
641 * gets marked 'SHUTDOWN', we can't let any new transactions
642 * be durable - ie. committed to disk - because some metadata might
643 * be inconsistent. In such cases, this returns an error, and the
644 * caller may assume that all locked objects joined to the transaction
645 * have already been unlocked as if the commit had succeeded.
646 * Do not reference the transaction structure after this call.
653 xfs_lsn_t *commit_lsn_p,
656 xfs_log_iovec_t *log_vector;
659 xfs_lsn_t commit_lsn;
664 #define XFS_TRANS_LOGVEC_COUNT 16
665 xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
666 #if defined(XLOG_NOLOG) || defined(DEBUG)
667 static xfs_lsn_t trans_lsn = 1;
675 * Determine whether this commit is releasing a permanent
676 * log reservation or not.
678 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
679 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
680 log_flags = XFS_LOG_REL_PERM_RESERV;
687 * If there is nothing to be logged by the transaction,
688 * then unlock all of the items associated with the
689 * transaction and free the transaction structure.
690 * Also make sure to return any reserved blocks to
694 shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
695 if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
696 xfs_trans_unreserve_and_mod_sb(tp);
698 * It is indeed possible for the transaction to be
699 * not dirty but the dqinfo portion to be. All that
700 * means is that we have some (non-persistent) quota
701 * reservations that need to be unreserved.
703 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
705 commit_lsn = xfs_log_done(mp, tp->t_ticket,
707 if (commit_lsn == -1 && !shutdown)
708 shutdown = XFS_ERROR(EIO);
710 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
711 xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
712 xfs_trans_free_busy(tp);
714 XFS_STATS_INC(xs_trans_empty);
716 *commit_lsn_p = commit_lsn;
719 #if defined(XLOG_NOLOG) || defined(DEBUG)
720 ASSERT(!xlog_debug || tp->t_ticket != NULL);
722 ASSERT(tp->t_ticket != NULL);
726 * If we need to update the superblock, then do it now.
728 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
729 xfs_trans_apply_sb_deltas(tp);
731 XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);
734 * Ask each log item how many log_vector entries it will
735 * need so we can figure out how many to allocate.
736 * Try to avoid the kmem_alloc() call in the common case
737 * by using a vector from the stack when it fits.
739 nvec = xfs_trans_count_vecs(tp);
742 xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
747 if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
748 log_vector = log_vector_fast;
750 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
751 sizeof(xfs_log_iovec_t),
756 * Fill in the log_vector and pin the logged items, and
757 * then write the transaction to the log.
759 xfs_trans_fill_vecs(tp, log_vector);
762 * Ignore errors here. xfs_log_done would do the right thing.
763 * We need to put the ticket, etc. away.
765 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket,
768 #if defined(XLOG_NOLOG) || defined(DEBUG)
770 commit_lsn = xfs_log_done(mp, tp->t_ticket,
771 &commit_iclog, log_flags);
774 tp->t_lsn = trans_lsn++;
778 * This is the regular case. At this point (after the call finishes),
779 * the transaction is committed incore and could go out to disk at
780 * any time. However, all the items associated with the transaction
781 * are still locked and pinned in memory.
783 commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
786 tp->t_commit_lsn = commit_lsn;
787 if (nvec > XFS_TRANS_LOGVEC_COUNT) {
788 kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
792 *commit_lsn_p = commit_lsn;
795 * If we got a log write error. Unpin the logitems that we
796 * had pinned, clean up, free trans structure, and return error.
798 if (error || commit_lsn == -1) {
799 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
800 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
801 return XFS_ERROR(EIO);
805 * Once the transaction has committed, unused
806 * reservations need to be released and changes to
807 * the superblock need to be reflected in the in-core
808 * version. Do that now.
810 xfs_trans_unreserve_and_mod_sb(tp);
812 sync = tp->t_flags & XFS_TRANS_SYNC;
815 * Tell the LM to call the transaction completion routine
816 * when the log write with LSN commit_lsn completes (e.g.
817 * when the transaction commit really hits the on-disk log).
818 * After this call we cannot reference tp, because the call
819 * can happen at any time and the call will free the transaction
820 * structure pointed to by tp. The only case where we call
821 * the completion routine (xfs_trans_committed) directly is
822 * if the log is turned off on a debug kernel or we're
823 * running in simulation mode (the log is explicitly turned
826 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
827 tp->t_logcb.cb_arg = tp;
830 * We need to pass the iclog buffer which was used for the
831 * transaction commit record into this function, and attach
832 * the callback to it. The callback must be attached before
833 * the items are unlocked to avoid racing with other threads
834 * waiting for an item to unlock.
836 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
839 * Mark this thread as no longer being in a transaction
841 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
844 * Once all the items of the transaction have been copied
845 * to the in core log and the callback is attached, the
846 * items can be unlocked.
848 * This will free descriptors pointing to items which were
849 * not logged since there is nothing more to do with them.
850 * For items which were logged, we will keep pointers to them
851 * so they can be unpinned after the transaction commits to disk.
852 * This will also stamp each modified meta-data item with
853 * the commit lsn of this transaction for dependency tracking
856 xfs_trans_unlock_items(tp, commit_lsn);
859 * If we detected a log error earlier, finish committing
860 * the transaction now (unpin log items, etc).
862 * Order is critical here, to avoid using the transaction
863 * pointer after its been freed (by xfs_trans_committed
864 * either here now, or as a callback). We cannot do this
865 * step inside xfs_log_notify as was done earlier because
869 xfs_trans_committed(tp, XFS_LI_ABORTED);
872 * Now that the xfs_trans_committed callback has been attached,
873 * and the items are released we can finally allow the iclog to
876 error = xfs_log_release_iclog(mp, commit_iclog);
879 * If the transaction needs to be synchronous, then force the
880 * log out now and wait for it.
884 error = _xfs_log_force(mp, commit_lsn,
885 XFS_LOG_FORCE | XFS_LOG_SYNC,
888 XFS_STATS_INC(xs_trans_sync);
890 XFS_STATS_INC(xs_trans_async);
898 * Total up the number of log iovecs needed to commit this
899 * transaction. The transaction itself needs one for the
900 * transaction header. Ask each dirty item in turn how many
901 * it needs to get the total.
904 xfs_trans_count_vecs(
908 xfs_log_item_desc_t *lidp;
911 lidp = xfs_trans_first_item(tp);
912 ASSERT(lidp != NULL);
914 /* In the non-debug case we need to start bailing out if we
915 * didn't find a log_item here, return zero and let trans_commit
921 while (lidp != NULL) {
923 * Skip items which aren't dirty in this transaction.
925 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
926 lidp = xfs_trans_next_item(tp, lidp);
929 lidp->lid_size = IOP_SIZE(lidp->lid_item);
930 nvecs += lidp->lid_size;
931 lidp = xfs_trans_next_item(tp, lidp);
938 * Called from the trans_commit code when we notice that
939 * the filesystem is in the middle of a forced shutdown.
946 xfs_log_item_desc_t *lidp;
948 for (lidp = xfs_trans_first_item(tp);
950 lidp = xfs_trans_next_item(tp, lidp)) {
952 * Unpin all but those that aren't dirty.
954 if (lidp->lid_flags & XFS_LID_DIRTY)
955 IOP_UNPIN_REMOVE(lidp->lid_item, tp);
958 xfs_trans_unreserve_and_mod_sb(tp);
959 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
961 xfs_trans_free_items(tp, flags);
962 xfs_trans_free_busy(tp);
967 * Fill in the vector with pointers to data to be logged
968 * by this transaction. The transaction header takes
969 * the first vector, and then each dirty item takes the
970 * number of vectors it indicated it needed in xfs_trans_count_vecs().
972 * As each item fills in the entries it needs, also pin the item
973 * so that it cannot be flushed out until the log write completes.
978 xfs_log_iovec_t *log_vector)
980 xfs_log_item_desc_t *lidp;
981 xfs_log_iovec_t *vecp;
985 * Skip over the entry for the transaction header, we'll
986 * fill that in at the end.
988 vecp = log_vector + 1; /* pointer arithmetic */
991 lidp = xfs_trans_first_item(tp);
992 ASSERT(lidp != NULL);
993 while (lidp != NULL) {
995 * Skip items which aren't dirty in this transaction.
997 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
998 lidp = xfs_trans_next_item(tp, lidp);
1002 * The item may be marked dirty but not log anything.
1003 * This can be used to get called when a transaction
1006 if (lidp->lid_size) {
1009 IOP_FORMAT(lidp->lid_item, vecp);
1010 vecp += lidp->lid_size; /* pointer arithmetic */
1011 IOP_PIN(lidp->lid_item);
1012 lidp = xfs_trans_next_item(tp, lidp);
1016 * Now that we've counted the number of items in this
1017 * transaction, fill in the transaction header.
1019 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1020 tp->t_header.th_type = tp->t_type;
1021 tp->t_header.th_num_items = nitems;
1022 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1023 log_vector->i_len = sizeof(xfs_trans_header_t);
1024 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR);
1029 * Unlock all of the transaction's items and free the transaction.
1030 * The transaction must not have modified any of its items, because
1031 * there is no way to restore them to their previous state.
1033 * If the transaction has made a log reservation, make sure to release
1043 xfs_log_item_chunk_t *licp;
1044 xfs_log_item_desc_t *lidp;
1045 xfs_log_item_t *lip;
1050 * See if the caller is being too lazy to figure out if
1051 * the transaction really needs an abort.
1053 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1054 flags &= ~XFS_TRANS_ABORT;
1056 * See if the caller is relying on us to shut down the
1057 * filesystem. This happens in paths where we detect
1058 * corruption and decide to give up.
1060 if ((tp->t_flags & XFS_TRANS_DIRTY) &&
1061 !XFS_FORCED_SHUTDOWN(tp->t_mountp))
1062 xfs_force_shutdown(tp->t_mountp, XFS_CORRUPT_INCORE);
1064 if (!(flags & XFS_TRANS_ABORT)) {
1065 licp = &(tp->t_items);
1066 while (licp != NULL) {
1067 lidp = licp->lic_descs;
1068 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1069 if (XFS_LIC_ISFREE(licp, i)) {
1073 lip = lidp->lid_item;
1074 if (!XFS_FORCED_SHUTDOWN(tp->t_mountp))
1075 ASSERT(!(lip->li_type == XFS_LI_EFD));
1077 licp = licp->lic_next;
1081 xfs_trans_unreserve_and_mod_sb(tp);
1082 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
1085 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1086 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1087 log_flags = XFS_LOG_REL_PERM_RESERV;
1091 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
1094 /* mark this thread as no longer being in a transaction */
1095 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
1097 xfs_trans_free_items(tp, flags);
1098 xfs_trans_free_busy(tp);
1104 * Free the transaction structure. If there is more clean up
1105 * to do when the structure is freed, add it here.
1111 atomic_dec(&tp->t_mountp->m_active_trans);
1112 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
1113 kmem_zone_free(xfs_trans_zone, tp);
1118 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1120 * This is typically called by the LM when a transaction has been fully
1121 * committed to disk. It needs to unpin the items which have
1122 * been logged by the transaction and update their positions
1123 * in the AIL if necessary.
1124 * This also gets called when the transactions didn't get written out
1125 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1127 * Call xfs_trans_chunk_committed() to process the items in
1131 xfs_trans_committed(
1135 xfs_log_item_chunk_t *licp;
1136 xfs_log_item_chunk_t *next_licp;
1137 xfs_log_busy_chunk_t *lbcp;
1138 xfs_log_busy_slot_t *lbsp;
1142 * Call the transaction's completion callback if there
1145 if (tp->t_callback != NULL) {
1146 tp->t_callback(tp, tp->t_callarg);
1150 * Special case the chunk embedded in the transaction.
1152 licp = &(tp->t_items);
1153 if (!(XFS_LIC_ARE_ALL_FREE(licp))) {
1154 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1158 * Process the items in each chunk in turn.
1160 licp = licp->lic_next;
1161 while (licp != NULL) {
1162 ASSERT(!XFS_LIC_ARE_ALL_FREE(licp));
1163 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1164 next_licp = licp->lic_next;
1165 kmem_free(licp, sizeof(xfs_log_item_chunk_t));
1170 * Clear all the per-AG busy list items listed in this transaction
1173 while (lbcp != NULL) {
1174 for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1175 if (!XFS_LBC_ISFREE(lbcp, i)) {
1176 xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1180 lbcp = lbcp->lbc_next;
1182 xfs_trans_free_busy(tp);
1185 * That's it for the transaction structure. Free it.
1191 * This is called to perform the commit processing for each
1192 * item described by the given chunk.
1194 * The commit processing consists of unlocking items which were
1195 * held locked with the SYNC_UNLOCK attribute, calling the committed
1196 * routine of each logged item, updating the item's position in the AIL
1197 * if necessary, and unpinning each item. If the committed routine
1198 * returns -1, then do nothing further with the item because it
1199 * may have been freed.
1201 * Since items are unlocked when they are copied to the incore
1202 * log, it is possible for two transactions to be completing
1203 * and manipulating the same item simultaneously. The AIL lock
1204 * will protect the lsn field of each item. The value of this
1205 * field can never go backwards.
1207 * We unpin the items after repositioning them in the AIL, because
1208 * otherwise they could be immediately flushed and we'd have to race
1209 * with the flusher trying to pull the item from the AIL as we add it.
1212 xfs_trans_chunk_committed(
1213 xfs_log_item_chunk_t *licp,
1217 xfs_log_item_desc_t *lidp;
1218 xfs_log_item_t *lip;
1220 struct xfs_mount *mp;
1224 lidp = licp->lic_descs;
1225 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1226 if (XFS_LIC_ISFREE(licp, i)) {
1230 lip = lidp->lid_item;
1232 lip->li_flags |= XFS_LI_ABORTED;
1235 * Send in the ABORTED flag to the COMMITTED routine
1236 * so that it knows whether the transaction was aborted
1239 item_lsn = IOP_COMMITTED(lip, lsn);
1242 * If the committed routine returns -1, make
1243 * no more references to the item.
1245 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1250 * If the returned lsn is greater than what it
1251 * contained before, update the location of the
1252 * item in the AIL. If it is not, then do nothing.
1253 * Items can never move backwards in the AIL.
1255 * While the new lsn should usually be greater, it
1256 * is possible that a later transaction completing
1257 * simultaneously with an earlier one using the
1258 * same item could complete first with a higher lsn.
1259 * This would cause the earlier transaction to fail
1262 mp = lip->li_mountp;
1264 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1266 * This will set the item's lsn to item_lsn
1267 * and update the position of the item in
1270 * xfs_trans_update_ail() drops the AIL lock.
1272 xfs_trans_update_ail(mp, lip, item_lsn, s);
1278 * Now that we've repositioned the item in the AIL,
1279 * unpin it so it can be flushed. Pass information
1280 * about buffer stale state down from the log item
1281 * flags, if anyone else stales the buffer we do not
1282 * want to pay any attention to it.
1284 IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);