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 * Reservation functions here avoid a huge stack in xfs_trans_init
63 * due to register overflow from temporaries in the calculations.
67 xfs_calc_write_reservation(xfs_mount_t *mp)
69 return XFS_CALC_WRITE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
73 xfs_calc_itruncate_reservation(xfs_mount_t *mp)
75 return XFS_CALC_ITRUNCATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
79 xfs_calc_rename_reservation(xfs_mount_t *mp)
81 return XFS_CALC_RENAME_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
85 xfs_calc_link_reservation(xfs_mount_t *mp)
87 return XFS_CALC_LINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
91 xfs_calc_remove_reservation(xfs_mount_t *mp)
93 return XFS_CALC_REMOVE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
97 xfs_calc_symlink_reservation(xfs_mount_t *mp)
99 return XFS_CALC_SYMLINK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
103 xfs_calc_create_reservation(xfs_mount_t *mp)
105 return XFS_CALC_CREATE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
109 xfs_calc_mkdir_reservation(xfs_mount_t *mp)
111 return XFS_CALC_MKDIR_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
115 xfs_calc_ifree_reservation(xfs_mount_t *mp)
117 return XFS_CALC_IFREE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
121 xfs_calc_ichange_reservation(xfs_mount_t *mp)
123 return XFS_CALC_ICHANGE_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
127 xfs_calc_growdata_reservation(xfs_mount_t *mp)
129 return XFS_CALC_GROWDATA_LOG_RES(mp);
133 xfs_calc_growrtalloc_reservation(xfs_mount_t *mp)
135 return XFS_CALC_GROWRTALLOC_LOG_RES(mp);
139 xfs_calc_growrtzero_reservation(xfs_mount_t *mp)
141 return XFS_CALC_GROWRTZERO_LOG_RES(mp);
145 xfs_calc_growrtfree_reservation(xfs_mount_t *mp)
147 return XFS_CALC_GROWRTFREE_LOG_RES(mp);
151 xfs_calc_swrite_reservation(xfs_mount_t *mp)
153 return XFS_CALC_SWRITE_LOG_RES(mp);
157 xfs_calc_writeid_reservation(xfs_mount_t *mp)
159 return XFS_CALC_WRITEID_LOG_RES(mp);
163 xfs_calc_addafork_reservation(xfs_mount_t *mp)
165 return XFS_CALC_ADDAFORK_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
169 xfs_calc_attrinval_reservation(xfs_mount_t *mp)
171 return XFS_CALC_ATTRINVAL_LOG_RES(mp);
175 xfs_calc_attrset_reservation(xfs_mount_t *mp)
177 return XFS_CALC_ATTRSET_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
181 xfs_calc_attrrm_reservation(xfs_mount_t *mp)
183 return XFS_CALC_ATTRRM_LOG_RES(mp) + XFS_DQUOT_LOGRES(mp);
187 xfs_calc_clear_agi_bucket_reservation(xfs_mount_t *mp)
189 return XFS_CALC_CLEAR_AGI_BUCKET_LOG_RES(mp);
193 * Initialize the precomputed transaction reservation values
194 * in the mount structure.
200 xfs_trans_reservations_t *resp;
202 resp = &(mp->m_reservations);
203 resp->tr_write = xfs_calc_write_reservation(mp);
204 resp->tr_itruncate = xfs_calc_itruncate_reservation(mp);
205 resp->tr_rename = xfs_calc_rename_reservation(mp);
206 resp->tr_link = xfs_calc_link_reservation(mp);
207 resp->tr_remove = xfs_calc_remove_reservation(mp);
208 resp->tr_symlink = xfs_calc_symlink_reservation(mp);
209 resp->tr_create = xfs_calc_create_reservation(mp);
210 resp->tr_mkdir = xfs_calc_mkdir_reservation(mp);
211 resp->tr_ifree = xfs_calc_ifree_reservation(mp);
212 resp->tr_ichange = xfs_calc_ichange_reservation(mp);
213 resp->tr_growdata = xfs_calc_growdata_reservation(mp);
214 resp->tr_swrite = xfs_calc_swrite_reservation(mp);
215 resp->tr_writeid = xfs_calc_writeid_reservation(mp);
216 resp->tr_addafork = xfs_calc_addafork_reservation(mp);
217 resp->tr_attrinval = xfs_calc_attrinval_reservation(mp);
218 resp->tr_attrset = xfs_calc_attrset_reservation(mp);
219 resp->tr_attrrm = xfs_calc_attrrm_reservation(mp);
220 resp->tr_clearagi = xfs_calc_clear_agi_bucket_reservation(mp);
221 resp->tr_growrtalloc = xfs_calc_growrtalloc_reservation(mp);
222 resp->tr_growrtzero = xfs_calc_growrtzero_reservation(mp);
223 resp->tr_growrtfree = xfs_calc_growrtfree_reservation(mp);
227 * This routine is called to allocate a transaction structure.
228 * The type parameter indicates the type of the transaction. These
229 * are enumerated in xfs_trans.h.
231 * Dynamically allocate the transaction structure from the transaction
232 * zone, initialize it, and return it to the caller.
240 fs_check_frozen(XFS_MTOVFS(mp), SB_FREEZE_TRANS);
242 atomic_inc(&mp->m_active_trans);
244 return (_xfs_trans_alloc(mp, type));
255 ASSERT(xfs_trans_zone != NULL);
256 tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
259 * Initialize the transaction structure.
261 tp->t_magic = XFS_TRANS_MAGIC;
264 tp->t_items_free = XFS_LIC_NUM_SLOTS;
265 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
266 XFS_LIC_INIT(&(tp->t_items));
267 XFS_LBC_INIT(&(tp->t_busy));
273 * This is called to create a new transaction which will share the
274 * permanent log reservation of the given transaction. The remaining
275 * unused block and rt extent reservations are also inherited. This
276 * implies that the original transaction is no longer allowed to allocate
277 * blocks. Locks and log items, however, are no inherited. They must
278 * be added to the new transaction explicitly.
286 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
289 * Initialize the new transaction structure.
291 ntp->t_magic = XFS_TRANS_MAGIC;
292 ntp->t_type = tp->t_type;
293 ntp->t_mountp = tp->t_mountp;
294 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
295 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
296 XFS_LIC_INIT(&(ntp->t_items));
297 XFS_LBC_INIT(&(ntp->t_busy));
299 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
300 ASSERT(tp->t_ticket != NULL);
302 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
303 ntp->t_ticket = tp->t_ticket;
304 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
305 tp->t_blk_res = tp->t_blk_res_used;
306 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
307 tp->t_rtx_res = tp->t_rtx_res_used;
308 PFLAGS_DUP(&tp->t_pflags, &ntp->t_pflags);
310 XFS_TRANS_DUP_DQINFO(tp->t_mountp, tp, ntp);
312 atomic_inc(&tp->t_mountp->m_active_trans);
317 * This is called to reserve free disk blocks and log space for the
318 * given transaction. This must be done before allocating any resources
319 * within the transaction.
321 * This will return ENOSPC if there are not enough blocks available.
322 * It will sleep waiting for available log space.
323 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
324 * is used by long running transactions. If any one of the reservations
325 * fails then they will all be backed out.
327 * This does not do quota reservations. That typically is done by the
344 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
346 /* Mark this thread as being in a transaction */
347 PFLAGS_SET_FSTRANS(&tp->t_pflags);
351 * Attempt to reserve the needed disk blocks by decrementing
352 * the number needed from the number available. This will
353 * fail if the count would go below zero.
356 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
359 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
360 return (XFS_ERROR(ENOSPC));
362 tp->t_blk_res += blocks;
366 * Reserve the log space needed for this transaction.
369 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
370 ASSERT((tp->t_log_count == 0) ||
371 (tp->t_log_count == logcount));
372 if (flags & XFS_TRANS_PERM_LOG_RES) {
373 log_flags = XFS_LOG_PERM_RESERV;
374 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
376 ASSERT(tp->t_ticket == NULL);
377 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
381 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
383 XFS_TRANSACTION, log_flags, tp->t_type);
387 tp->t_log_res = logspace;
388 tp->t_log_count = logcount;
392 * Attempt to reserve the needed realtime extents by decrementing
393 * the number needed from the number available. This will
394 * fail if the count would go below zero.
397 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
400 error = XFS_ERROR(ENOSPC);
403 tp->t_rtx_res += rtextents;
409 * Error cases jump to one of these labels to undo any
410 * reservations which have already been performed.
414 if (flags & XFS_TRANS_PERM_LOG_RES) {
415 log_flags = XFS_LOG_REL_PERM_RESERV;
419 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
422 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
427 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
432 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
439 * Record the indicated change to the given field for application
440 * to the file system's superblock when the transaction commits.
441 * For now, just store the change in the transaction structure.
443 * Mark the transaction structure to indicate that the superblock
444 * needs to be updated before committing.
454 case XFS_TRANS_SB_ICOUNT:
455 tp->t_icount_delta += delta;
457 case XFS_TRANS_SB_IFREE:
458 tp->t_ifree_delta += delta;
460 case XFS_TRANS_SB_FDBLOCKS:
462 * Track the number of blocks allocated in the
463 * transaction. Make sure it does not exceed the
467 tp->t_blk_res_used += (uint)-delta;
468 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
470 tp->t_fdblocks_delta += delta;
472 case XFS_TRANS_SB_RES_FDBLOCKS:
474 * The allocation has already been applied to the
475 * in-core superblock's counter. This should only
476 * be applied to the on-disk superblock.
479 tp->t_res_fdblocks_delta += delta;
481 case XFS_TRANS_SB_FREXTENTS:
483 * Track the number of blocks allocated in the
484 * transaction. Make sure it does not exceed the
488 tp->t_rtx_res_used += (uint)-delta;
489 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
491 tp->t_frextents_delta += delta;
493 case XFS_TRANS_SB_RES_FREXTENTS:
495 * The allocation has already been applied to the
496 * in-core superblock's counter. This should only
497 * be applied to the on-disk superblock.
500 tp->t_res_frextents_delta += delta;
502 case XFS_TRANS_SB_DBLOCKS:
504 tp->t_dblocks_delta += delta;
506 case XFS_TRANS_SB_AGCOUNT:
508 tp->t_agcount_delta += delta;
510 case XFS_TRANS_SB_IMAXPCT:
511 tp->t_imaxpct_delta += delta;
513 case XFS_TRANS_SB_REXTSIZE:
514 tp->t_rextsize_delta += delta;
516 case XFS_TRANS_SB_RBMBLOCKS:
517 tp->t_rbmblocks_delta += delta;
519 case XFS_TRANS_SB_RBLOCKS:
520 tp->t_rblocks_delta += delta;
522 case XFS_TRANS_SB_REXTENTS:
523 tp->t_rextents_delta += delta;
525 case XFS_TRANS_SB_REXTSLOG:
526 tp->t_rextslog_delta += delta;
533 tp->t_flags |= (XFS_TRANS_SB_DIRTY | XFS_TRANS_DIRTY);
537 * xfs_trans_apply_sb_deltas() is called from the commit code
538 * to bring the superblock buffer into the current transaction
539 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
541 * For now we just look at each field allowed to change and change
545 xfs_trans_apply_sb_deltas(
552 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
553 sbp = XFS_BUF_TO_SBP(bp);
556 * Check that superblock mods match the mods made to AGF counters.
558 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
559 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
560 tp->t_ag_btree_delta));
562 if (tp->t_icount_delta != 0) {
563 INT_MOD(sbp->sb_icount, ARCH_CONVERT, tp->t_icount_delta);
565 if (tp->t_ifree_delta != 0) {
566 INT_MOD(sbp->sb_ifree, ARCH_CONVERT, tp->t_ifree_delta);
569 if (tp->t_fdblocks_delta != 0) {
570 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_fdblocks_delta);
572 if (tp->t_res_fdblocks_delta != 0) {
573 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_res_fdblocks_delta);
576 if (tp->t_frextents_delta != 0) {
577 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_frextents_delta);
579 if (tp->t_res_frextents_delta != 0) {
580 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_res_frextents_delta);
582 if (tp->t_dblocks_delta != 0) {
583 INT_MOD(sbp->sb_dblocks, ARCH_CONVERT, tp->t_dblocks_delta);
586 if (tp->t_agcount_delta != 0) {
587 INT_MOD(sbp->sb_agcount, ARCH_CONVERT, tp->t_agcount_delta);
590 if (tp->t_imaxpct_delta != 0) {
591 INT_MOD(sbp->sb_imax_pct, ARCH_CONVERT, tp->t_imaxpct_delta);
594 if (tp->t_rextsize_delta != 0) {
595 INT_MOD(sbp->sb_rextsize, ARCH_CONVERT, tp->t_rextsize_delta);
598 if (tp->t_rbmblocks_delta != 0) {
599 INT_MOD(sbp->sb_rbmblocks, ARCH_CONVERT, tp->t_rbmblocks_delta);
602 if (tp->t_rblocks_delta != 0) {
603 INT_MOD(sbp->sb_rblocks, ARCH_CONVERT, tp->t_rblocks_delta);
606 if (tp->t_rextents_delta != 0) {
607 INT_MOD(sbp->sb_rextents, ARCH_CONVERT, tp->t_rextents_delta);
610 if (tp->t_rextslog_delta != 0) {
611 INT_MOD(sbp->sb_rextslog, ARCH_CONVERT, tp->t_rextslog_delta);
617 * Log the whole thing, the fields are noncontiguous.
619 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_sb_t) - 1);
622 * Since all the modifiable fields are contiguous, we
623 * can get away with this.
625 xfs_trans_log_buf(tp, bp, offsetof(xfs_sb_t, sb_icount),
626 offsetof(xfs_sb_t, sb_frextents) +
627 sizeof(sbp->sb_frextents) - 1);
630 XFS_MTOVFS(tp->t_mountp)->vfs_super->s_dirt = 1;
635 * xfs_trans_unreserve_and_mod_sb() is called to release unused
636 * reservations and apply superblock counter changes to the in-core
639 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
642 xfs_trans_unreserve_and_mod_sb(
645 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
652 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
655 * Release any reserved blocks. Any that were allocated
656 * will be taken back again by fdblocks_delta below.
658 if (tp->t_blk_res > 0) {
659 msbp->msb_field = XFS_SBS_FDBLOCKS;
660 msbp->msb_delta = tp->t_blk_res;
665 * Release any reserved real time extents . Any that were
666 * allocated will be taken back again by frextents_delta below.
668 if (tp->t_rtx_res > 0) {
669 msbp->msb_field = XFS_SBS_FREXTENTS;
670 msbp->msb_delta = tp->t_rtx_res;
675 * Apply any superblock modifications to the in-core version.
676 * The t_res_fdblocks_delta and t_res_frextents_delta fields are
677 * explicitly NOT applied to the in-core superblock.
678 * The idea is that that has already been done.
680 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
681 if (tp->t_icount_delta != 0) {
682 msbp->msb_field = XFS_SBS_ICOUNT;
683 msbp->msb_delta = (int)tp->t_icount_delta;
686 if (tp->t_ifree_delta != 0) {
687 msbp->msb_field = XFS_SBS_IFREE;
688 msbp->msb_delta = (int)tp->t_ifree_delta;
691 if (tp->t_fdblocks_delta != 0) {
692 msbp->msb_field = XFS_SBS_FDBLOCKS;
693 msbp->msb_delta = (int)tp->t_fdblocks_delta;
696 if (tp->t_frextents_delta != 0) {
697 msbp->msb_field = XFS_SBS_FREXTENTS;
698 msbp->msb_delta = (int)tp->t_frextents_delta;
701 if (tp->t_dblocks_delta != 0) {
702 msbp->msb_field = XFS_SBS_DBLOCKS;
703 msbp->msb_delta = (int)tp->t_dblocks_delta;
706 if (tp->t_agcount_delta != 0) {
707 msbp->msb_field = XFS_SBS_AGCOUNT;
708 msbp->msb_delta = (int)tp->t_agcount_delta;
711 if (tp->t_imaxpct_delta != 0) {
712 msbp->msb_field = XFS_SBS_IMAX_PCT;
713 msbp->msb_delta = (int)tp->t_imaxpct_delta;
716 if (tp->t_rextsize_delta != 0) {
717 msbp->msb_field = XFS_SBS_REXTSIZE;
718 msbp->msb_delta = (int)tp->t_rextsize_delta;
721 if (tp->t_rbmblocks_delta != 0) {
722 msbp->msb_field = XFS_SBS_RBMBLOCKS;
723 msbp->msb_delta = (int)tp->t_rbmblocks_delta;
726 if (tp->t_rblocks_delta != 0) {
727 msbp->msb_field = XFS_SBS_RBLOCKS;
728 msbp->msb_delta = (int)tp->t_rblocks_delta;
731 if (tp->t_rextents_delta != 0) {
732 msbp->msb_field = XFS_SBS_REXTENTS;
733 msbp->msb_delta = (int)tp->t_rextents_delta;
736 if (tp->t_rextslog_delta != 0) {
737 msbp->msb_field = XFS_SBS_REXTSLOG;
738 msbp->msb_delta = (int)tp->t_rextslog_delta;
744 * If we need to change anything, do it.
747 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
748 (uint)(msbp - msb), rsvd);
757 * Commit the given transaction to the log a/synchronously.
759 * XFS disk error handling mechanism is not based on a typical
760 * transaction abort mechanism. Logically after the filesystem
761 * gets marked 'SHUTDOWN', we can't let any new transactions
762 * be durable - ie. committed to disk - because some metadata might
763 * be inconsistent. In such cases, this returns an error, and the
764 * caller may assume that all locked objects joined to the transaction
765 * have already been unlocked as if the commit had succeeded.
766 * Do not reference the transaction structure after this call.
773 xfs_lsn_t *commit_lsn_p,
776 xfs_log_iovec_t *log_vector;
779 xfs_lsn_t commit_lsn;
784 #define XFS_TRANS_LOGVEC_COUNT 16
785 xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
792 * Determine whether this commit is releasing a permanent
793 * log reservation or not.
795 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
796 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
797 log_flags = XFS_LOG_REL_PERM_RESERV;
804 * If there is nothing to be logged by the transaction,
805 * then unlock all of the items associated with the
806 * transaction and free the transaction structure.
807 * Also make sure to return any reserved blocks to
811 shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
812 if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
813 xfs_trans_unreserve_and_mod_sb(tp);
815 * It is indeed possible for the transaction to be
816 * not dirty but the dqinfo portion to be. All that
817 * means is that we have some (non-persistent) quota
818 * reservations that need to be unreserved.
820 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
822 commit_lsn = xfs_log_done(mp, tp->t_ticket,
824 if (commit_lsn == -1 && !shutdown)
825 shutdown = XFS_ERROR(EIO);
827 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
828 xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
829 xfs_trans_free_busy(tp);
831 XFS_STATS_INC(xs_trans_empty);
833 *commit_lsn_p = commit_lsn;
836 ASSERT(tp->t_ticket != NULL);
839 * If we need to update the superblock, then do it now.
841 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
842 xfs_trans_apply_sb_deltas(tp);
844 XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);
847 * Ask each log item how many log_vector entries it will
848 * need so we can figure out how many to allocate.
849 * Try to avoid the kmem_alloc() call in the common case
850 * by using a vector from the stack when it fits.
852 nvec = xfs_trans_count_vecs(tp);
854 xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
856 } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
857 log_vector = log_vector_fast;
859 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
860 sizeof(xfs_log_iovec_t),
865 * Fill in the log_vector and pin the logged items, and
866 * then write the transaction to the log.
868 xfs_trans_fill_vecs(tp, log_vector);
870 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
873 * The transaction is committed incore here, and can go out to disk
874 * at any time after this call. However, all the items associated
875 * with the transaction are still locked and pinned in memory.
877 commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
879 tp->t_commit_lsn = commit_lsn;
880 if (nvec > XFS_TRANS_LOGVEC_COUNT) {
881 kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
885 *commit_lsn_p = commit_lsn;
888 * If we got a log write error. Unpin the logitems that we
889 * had pinned, clean up, free trans structure, and return error.
891 if (error || commit_lsn == -1) {
892 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
893 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
894 return XFS_ERROR(EIO);
898 * Once the transaction has committed, unused
899 * reservations need to be released and changes to
900 * the superblock need to be reflected in the in-core
901 * version. Do that now.
903 xfs_trans_unreserve_and_mod_sb(tp);
905 sync = tp->t_flags & XFS_TRANS_SYNC;
908 * Tell the LM to call the transaction completion routine
909 * when the log write with LSN commit_lsn completes (e.g.
910 * when the transaction commit really hits the on-disk log).
911 * After this call we cannot reference tp, because the call
912 * can happen at any time and the call will free the transaction
913 * structure pointed to by tp. The only case where we call
914 * the completion routine (xfs_trans_committed) directly is
915 * if the log is turned off on a debug kernel or we're
916 * running in simulation mode (the log is explicitly turned
919 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
920 tp->t_logcb.cb_arg = tp;
923 * We need to pass the iclog buffer which was used for the
924 * transaction commit record into this function, and attach
925 * the callback to it. The callback must be attached before
926 * the items are unlocked to avoid racing with other threads
927 * waiting for an item to unlock.
929 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
932 * Mark this thread as no longer being in a transaction
934 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
937 * Once all the items of the transaction have been copied
938 * to the in core log and the callback is attached, the
939 * items can be unlocked.
941 * This will free descriptors pointing to items which were
942 * not logged since there is nothing more to do with them.
943 * For items which were logged, we will keep pointers to them
944 * so they can be unpinned after the transaction commits to disk.
945 * This will also stamp each modified meta-data item with
946 * the commit lsn of this transaction for dependency tracking
949 xfs_trans_unlock_items(tp, commit_lsn);
952 * If we detected a log error earlier, finish committing
953 * the transaction now (unpin log items, etc).
955 * Order is critical here, to avoid using the transaction
956 * pointer after its been freed (by xfs_trans_committed
957 * either here now, or as a callback). We cannot do this
958 * step inside xfs_log_notify as was done earlier because
962 xfs_trans_committed(tp, XFS_LI_ABORTED);
965 * Now that the xfs_trans_committed callback has been attached,
966 * and the items are released we can finally allow the iclog to
969 error = xfs_log_release_iclog(mp, commit_iclog);
972 * If the transaction needs to be synchronous, then force the
973 * log out now and wait for it.
977 error = _xfs_log_force(mp, commit_lsn,
978 XFS_LOG_FORCE | XFS_LOG_SYNC,
981 XFS_STATS_INC(xs_trans_sync);
983 XFS_STATS_INC(xs_trans_async);
991 * Total up the number of log iovecs needed to commit this
992 * transaction. The transaction itself needs one for the
993 * transaction header. Ask each dirty item in turn how many
994 * it needs to get the total.
997 xfs_trans_count_vecs(
1001 xfs_log_item_desc_t *lidp;
1004 lidp = xfs_trans_first_item(tp);
1005 ASSERT(lidp != NULL);
1007 /* In the non-debug case we need to start bailing out if we
1008 * didn't find a log_item here, return zero and let trans_commit
1014 while (lidp != NULL) {
1016 * Skip items which aren't dirty in this transaction.
1018 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1019 lidp = xfs_trans_next_item(tp, lidp);
1022 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1023 nvecs += lidp->lid_size;
1024 lidp = xfs_trans_next_item(tp, lidp);
1031 * Called from the trans_commit code when we notice that
1032 * the filesystem is in the middle of a forced shutdown.
1039 xfs_log_item_desc_t *lidp;
1041 for (lidp = xfs_trans_first_item(tp);
1043 lidp = xfs_trans_next_item(tp, lidp)) {
1045 * Unpin all but those that aren't dirty.
1047 if (lidp->lid_flags & XFS_LID_DIRTY)
1048 IOP_UNPIN_REMOVE(lidp->lid_item, tp);
1051 xfs_trans_unreserve_and_mod_sb(tp);
1052 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
1054 xfs_trans_free_items(tp, flags);
1055 xfs_trans_free_busy(tp);
1060 * Fill in the vector with pointers to data to be logged
1061 * by this transaction. The transaction header takes
1062 * the first vector, and then each dirty item takes the
1063 * number of vectors it indicated it needed in xfs_trans_count_vecs().
1065 * As each item fills in the entries it needs, also pin the item
1066 * so that it cannot be flushed out until the log write completes.
1069 xfs_trans_fill_vecs(
1071 xfs_log_iovec_t *log_vector)
1073 xfs_log_item_desc_t *lidp;
1074 xfs_log_iovec_t *vecp;
1078 * Skip over the entry for the transaction header, we'll
1079 * fill that in at the end.
1081 vecp = log_vector + 1; /* pointer arithmetic */
1084 lidp = xfs_trans_first_item(tp);
1085 ASSERT(lidp != NULL);
1086 while (lidp != NULL) {
1088 * Skip items which aren't dirty in this transaction.
1090 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1091 lidp = xfs_trans_next_item(tp, lidp);
1095 * The item may be marked dirty but not log anything.
1096 * This can be used to get called when a transaction
1099 if (lidp->lid_size) {
1102 IOP_FORMAT(lidp->lid_item, vecp);
1103 vecp += lidp->lid_size; /* pointer arithmetic */
1104 IOP_PIN(lidp->lid_item);
1105 lidp = xfs_trans_next_item(tp, lidp);
1109 * Now that we've counted the number of items in this
1110 * transaction, fill in the transaction header.
1112 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1113 tp->t_header.th_type = tp->t_type;
1114 tp->t_header.th_num_items = nitems;
1115 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1116 log_vector->i_len = sizeof(xfs_trans_header_t);
1117 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR);
1122 * Unlock all of the transaction's items and free the transaction.
1123 * The transaction must not have modified any of its items, because
1124 * there is no way to restore them to their previous state.
1126 * If the transaction has made a log reservation, make sure to release
1136 xfs_log_item_chunk_t *licp;
1137 xfs_log_item_desc_t *lidp;
1138 xfs_log_item_t *lip;
1141 xfs_mount_t *mp = tp->t_mountp;
1144 * See if the caller is being too lazy to figure out if
1145 * the transaction really needs an abort.
1147 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1148 flags &= ~XFS_TRANS_ABORT;
1150 * See if the caller is relying on us to shut down the
1151 * filesystem. This happens in paths where we detect
1152 * corruption and decide to give up.
1154 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1155 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1156 xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
1159 if (!(flags & XFS_TRANS_ABORT)) {
1160 licp = &(tp->t_items);
1161 while (licp != NULL) {
1162 lidp = licp->lic_descs;
1163 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1164 if (XFS_LIC_ISFREE(licp, i)) {
1168 lip = lidp->lid_item;
1169 if (!XFS_FORCED_SHUTDOWN(mp))
1170 ASSERT(!(lip->li_type == XFS_LI_EFD));
1172 licp = licp->lic_next;
1176 xfs_trans_unreserve_and_mod_sb(tp);
1177 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
1180 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1181 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1182 log_flags = XFS_LOG_REL_PERM_RESERV;
1186 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1189 /* mark this thread as no longer being in a transaction */
1190 PFLAGS_RESTORE_FSTRANS(&tp->t_pflags);
1192 xfs_trans_free_items(tp, flags);
1193 xfs_trans_free_busy(tp);
1199 * Free the transaction structure. If there is more clean up
1200 * to do when the structure is freed, add it here.
1206 atomic_dec(&tp->t_mountp->m_active_trans);
1207 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
1208 kmem_zone_free(xfs_trans_zone, tp);
1213 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1215 * This is typically called by the LM when a transaction has been fully
1216 * committed to disk. It needs to unpin the items which have
1217 * been logged by the transaction and update their positions
1218 * in the AIL if necessary.
1219 * This also gets called when the transactions didn't get written out
1220 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1222 * Call xfs_trans_chunk_committed() to process the items in
1226 xfs_trans_committed(
1230 xfs_log_item_chunk_t *licp;
1231 xfs_log_item_chunk_t *next_licp;
1232 xfs_log_busy_chunk_t *lbcp;
1233 xfs_log_busy_slot_t *lbsp;
1237 * Call the transaction's completion callback if there
1240 if (tp->t_callback != NULL) {
1241 tp->t_callback(tp, tp->t_callarg);
1245 * Special case the chunk embedded in the transaction.
1247 licp = &(tp->t_items);
1248 if (!(XFS_LIC_ARE_ALL_FREE(licp))) {
1249 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1253 * Process the items in each chunk in turn.
1255 licp = licp->lic_next;
1256 while (licp != NULL) {
1257 ASSERT(!XFS_LIC_ARE_ALL_FREE(licp));
1258 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1259 next_licp = licp->lic_next;
1260 kmem_free(licp, sizeof(xfs_log_item_chunk_t));
1265 * Clear all the per-AG busy list items listed in this transaction
1268 while (lbcp != NULL) {
1269 for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1270 if (!XFS_LBC_ISFREE(lbcp, i)) {
1271 xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1275 lbcp = lbcp->lbc_next;
1277 xfs_trans_free_busy(tp);
1280 * That's it for the transaction structure. Free it.
1286 * This is called to perform the commit processing for each
1287 * item described by the given chunk.
1289 * The commit processing consists of unlocking items which were
1290 * held locked with the SYNC_UNLOCK attribute, calling the committed
1291 * routine of each logged item, updating the item's position in the AIL
1292 * if necessary, and unpinning each item. If the committed routine
1293 * returns -1, then do nothing further with the item because it
1294 * may have been freed.
1296 * Since items are unlocked when they are copied to the incore
1297 * log, it is possible for two transactions to be completing
1298 * and manipulating the same item simultaneously. The AIL lock
1299 * will protect the lsn field of each item. The value of this
1300 * field can never go backwards.
1302 * We unpin the items after repositioning them in the AIL, because
1303 * otherwise they could be immediately flushed and we'd have to race
1304 * with the flusher trying to pull the item from the AIL as we add it.
1307 xfs_trans_chunk_committed(
1308 xfs_log_item_chunk_t *licp,
1312 xfs_log_item_desc_t *lidp;
1313 xfs_log_item_t *lip;
1315 struct xfs_mount *mp;
1319 lidp = licp->lic_descs;
1320 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1321 if (XFS_LIC_ISFREE(licp, i)) {
1325 lip = lidp->lid_item;
1327 lip->li_flags |= XFS_LI_ABORTED;
1330 * Send in the ABORTED flag to the COMMITTED routine
1331 * so that it knows whether the transaction was aborted
1334 item_lsn = IOP_COMMITTED(lip, lsn);
1337 * If the committed routine returns -1, make
1338 * no more references to the item.
1340 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1345 * If the returned lsn is greater than what it
1346 * contained before, update the location of the
1347 * item in the AIL. If it is not, then do nothing.
1348 * Items can never move backwards in the AIL.
1350 * While the new lsn should usually be greater, it
1351 * is possible that a later transaction completing
1352 * simultaneously with an earlier one using the
1353 * same item could complete first with a higher lsn.
1354 * This would cause the earlier transaction to fail
1357 mp = lip->li_mountp;
1359 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1361 * This will set the item's lsn to item_lsn
1362 * and update the position of the item in
1365 * xfs_trans_update_ail() drops the AIL lock.
1367 xfs_trans_update_ail(mp, lip, item_lsn, s);
1373 * Now that we've repositioned the item in the AIL,
1374 * unpin it so it can be flushed. Pass information
1375 * about buffer stale state down from the log item
1376 * flags, if anyone else stales the buffer we do not
1377 * want to pay any attention to it.
1379 IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);