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.
239 vfs_wait_for_freeze(XFS_MTOVFS(mp), SB_FREEZE_TRANS);
240 return _xfs_trans_alloc(mp, type);
250 atomic_inc(&mp->m_active_trans);
252 tp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
253 tp->t_magic = XFS_TRANS_MAGIC;
256 tp->t_items_free = XFS_LIC_NUM_SLOTS;
257 tp->t_busy_free = XFS_LBC_NUM_SLOTS;
258 XFS_LIC_INIT(&(tp->t_items));
259 XFS_LBC_INIT(&(tp->t_busy));
264 * This is called to create a new transaction which will share the
265 * permanent log reservation of the given transaction. The remaining
266 * unused block and rt extent reservations are also inherited. This
267 * implies that the original transaction is no longer allowed to allocate
268 * blocks. Locks and log items, however, are no inherited. They must
269 * be added to the new transaction explicitly.
277 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
280 * Initialize the new transaction structure.
282 ntp->t_magic = XFS_TRANS_MAGIC;
283 ntp->t_type = tp->t_type;
284 ntp->t_mountp = tp->t_mountp;
285 ntp->t_items_free = XFS_LIC_NUM_SLOTS;
286 ntp->t_busy_free = XFS_LBC_NUM_SLOTS;
287 XFS_LIC_INIT(&(ntp->t_items));
288 XFS_LBC_INIT(&(ntp->t_busy));
290 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
291 ASSERT(tp->t_ticket != NULL);
293 ntp->t_flags = XFS_TRANS_PERM_LOG_RES | (tp->t_flags & XFS_TRANS_RESERVE);
294 ntp->t_ticket = tp->t_ticket;
295 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
296 tp->t_blk_res = tp->t_blk_res_used;
297 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
298 tp->t_rtx_res = tp->t_rtx_res_used;
299 ntp->t_pflags = tp->t_pflags;
301 XFS_TRANS_DUP_DQINFO(tp->t_mountp, tp, ntp);
303 atomic_inc(&tp->t_mountp->m_active_trans);
308 * This is called to reserve free disk blocks and log space for the
309 * given transaction. This must be done before allocating any resources
310 * within the transaction.
312 * This will return ENOSPC if there are not enough blocks available.
313 * It will sleep waiting for available log space.
314 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
315 * is used by long running transactions. If any one of the reservations
316 * fails then they will all be backed out.
318 * This does not do quota reservations. That typically is done by the
332 int rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
334 /* Mark this thread as being in a transaction */
335 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
338 * Attempt to reserve the needed disk blocks by decrementing
339 * the number needed from the number available. This will
340 * fail if the count would go below zero.
343 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
346 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
347 return (XFS_ERROR(ENOSPC));
349 tp->t_blk_res += blocks;
353 * Reserve the log space needed for this transaction.
356 ASSERT((tp->t_log_res == 0) || (tp->t_log_res == logspace));
357 ASSERT((tp->t_log_count == 0) ||
358 (tp->t_log_count == logcount));
359 if (flags & XFS_TRANS_PERM_LOG_RES) {
360 log_flags = XFS_LOG_PERM_RESERV;
361 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
363 ASSERT(tp->t_ticket == NULL);
364 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
368 error = xfs_log_reserve(tp->t_mountp, logspace, logcount,
370 XFS_TRANSACTION, log_flags, tp->t_type);
374 tp->t_log_res = logspace;
375 tp->t_log_count = logcount;
379 * Attempt to reserve the needed realtime extents by decrementing
380 * the number needed from the number available. This will
381 * fail if the count would go below zero.
384 error = xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FREXTENTS,
387 error = XFS_ERROR(ENOSPC);
390 tp->t_rtx_res += rtextents;
396 * Error cases jump to one of these labels to undo any
397 * reservations which have already been performed.
401 if (flags & XFS_TRANS_PERM_LOG_RES) {
402 log_flags = XFS_LOG_REL_PERM_RESERV;
406 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
409 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
414 (void) xfs_mod_incore_sb(tp->t_mountp, XFS_SBS_FDBLOCKS,
419 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
426 * Record the indicated change to the given field for application
427 * to the file system's superblock when the transaction commits.
428 * For now, just store the change in the transaction structure.
430 * Mark the transaction structure to indicate that the superblock
431 * needs to be updated before committing.
441 case XFS_TRANS_SB_ICOUNT:
442 tp->t_icount_delta += delta;
444 case XFS_TRANS_SB_IFREE:
445 tp->t_ifree_delta += delta;
447 case XFS_TRANS_SB_FDBLOCKS:
449 * Track the number of blocks allocated in the
450 * transaction. Make sure it does not exceed the
454 tp->t_blk_res_used += (uint)-delta;
455 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
457 tp->t_fdblocks_delta += delta;
459 case XFS_TRANS_SB_RES_FDBLOCKS:
461 * The allocation has already been applied to the
462 * in-core superblock's counter. This should only
463 * be applied to the on-disk superblock.
466 tp->t_res_fdblocks_delta += delta;
468 case XFS_TRANS_SB_FREXTENTS:
470 * Track the number of blocks allocated in the
471 * transaction. Make sure it does not exceed the
475 tp->t_rtx_res_used += (uint)-delta;
476 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
478 tp->t_frextents_delta += delta;
480 case XFS_TRANS_SB_RES_FREXTENTS:
482 * The allocation has already been applied to the
483 * in-core superblock's counter. This should only
484 * be applied to the on-disk superblock.
487 tp->t_res_frextents_delta += delta;
489 case XFS_TRANS_SB_DBLOCKS:
491 tp->t_dblocks_delta += delta;
493 case XFS_TRANS_SB_AGCOUNT:
495 tp->t_agcount_delta += delta;
497 case XFS_TRANS_SB_IMAXPCT:
498 tp->t_imaxpct_delta += delta;
500 case XFS_TRANS_SB_REXTSIZE:
501 tp->t_rextsize_delta += delta;
503 case XFS_TRANS_SB_RBMBLOCKS:
504 tp->t_rbmblocks_delta += delta;
506 case XFS_TRANS_SB_RBLOCKS:
507 tp->t_rblocks_delta += delta;
509 case XFS_TRANS_SB_REXTENTS:
510 tp->t_rextents_delta += delta;
512 case XFS_TRANS_SB_REXTSLOG:
513 tp->t_rextslog_delta += delta;
520 tp->t_flags |= (XFS_TRANS_SB_DIRTY | XFS_TRANS_DIRTY);
524 * xfs_trans_apply_sb_deltas() is called from the commit code
525 * to bring the superblock buffer into the current transaction
526 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
528 * For now we just look at each field allowed to change and change
532 xfs_trans_apply_sb_deltas(
539 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
540 sbp = XFS_BUF_TO_SBP(bp);
543 * Check that superblock mods match the mods made to AGF counters.
545 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
546 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
547 tp->t_ag_btree_delta));
549 if (tp->t_icount_delta != 0) {
550 INT_MOD(sbp->sb_icount, ARCH_CONVERT, tp->t_icount_delta);
552 if (tp->t_ifree_delta != 0) {
553 INT_MOD(sbp->sb_ifree, ARCH_CONVERT, tp->t_ifree_delta);
556 if (tp->t_fdblocks_delta != 0) {
557 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_fdblocks_delta);
559 if (tp->t_res_fdblocks_delta != 0) {
560 INT_MOD(sbp->sb_fdblocks, ARCH_CONVERT, tp->t_res_fdblocks_delta);
563 if (tp->t_frextents_delta != 0) {
564 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_frextents_delta);
566 if (tp->t_res_frextents_delta != 0) {
567 INT_MOD(sbp->sb_frextents, ARCH_CONVERT, tp->t_res_frextents_delta);
569 if (tp->t_dblocks_delta != 0) {
570 INT_MOD(sbp->sb_dblocks, ARCH_CONVERT, tp->t_dblocks_delta);
573 if (tp->t_agcount_delta != 0) {
574 INT_MOD(sbp->sb_agcount, ARCH_CONVERT, tp->t_agcount_delta);
577 if (tp->t_imaxpct_delta != 0) {
578 INT_MOD(sbp->sb_imax_pct, ARCH_CONVERT, tp->t_imaxpct_delta);
581 if (tp->t_rextsize_delta != 0) {
582 INT_MOD(sbp->sb_rextsize, ARCH_CONVERT, tp->t_rextsize_delta);
585 if (tp->t_rbmblocks_delta != 0) {
586 INT_MOD(sbp->sb_rbmblocks, ARCH_CONVERT, tp->t_rbmblocks_delta);
589 if (tp->t_rblocks_delta != 0) {
590 INT_MOD(sbp->sb_rblocks, ARCH_CONVERT, tp->t_rblocks_delta);
593 if (tp->t_rextents_delta != 0) {
594 INT_MOD(sbp->sb_rextents, ARCH_CONVERT, tp->t_rextents_delta);
597 if (tp->t_rextslog_delta != 0) {
598 INT_MOD(sbp->sb_rextslog, ARCH_CONVERT, tp->t_rextslog_delta);
604 * Log the whole thing, the fields are noncontiguous.
606 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_sb_t) - 1);
609 * Since all the modifiable fields are contiguous, we
610 * can get away with this.
612 xfs_trans_log_buf(tp, bp, offsetof(xfs_sb_t, sb_icount),
613 offsetof(xfs_sb_t, sb_frextents) +
614 sizeof(sbp->sb_frextents) - 1);
616 XFS_MTOVFS(tp->t_mountp)->vfs_super->s_dirt = 1;
620 * xfs_trans_unreserve_and_mod_sb() is called to release unused
621 * reservations and apply superblock counter changes to the in-core
624 * This is done efficiently with a single call to xfs_mod_incore_sb_batch().
627 xfs_trans_unreserve_and_mod_sb(
630 xfs_mod_sb_t msb[14]; /* If you add cases, add entries */
637 rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
640 * Release any reserved blocks. Any that were allocated
641 * will be taken back again by fdblocks_delta below.
643 if (tp->t_blk_res > 0) {
644 msbp->msb_field = XFS_SBS_FDBLOCKS;
645 msbp->msb_delta = tp->t_blk_res;
650 * Release any reserved real time extents . Any that were
651 * allocated will be taken back again by frextents_delta below.
653 if (tp->t_rtx_res > 0) {
654 msbp->msb_field = XFS_SBS_FREXTENTS;
655 msbp->msb_delta = tp->t_rtx_res;
660 * Apply any superblock modifications to the in-core version.
661 * The t_res_fdblocks_delta and t_res_frextents_delta fields are
662 * explicitly NOT applied to the in-core superblock.
663 * The idea is that that has already been done.
665 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
666 if (tp->t_icount_delta != 0) {
667 msbp->msb_field = XFS_SBS_ICOUNT;
668 msbp->msb_delta = (int)tp->t_icount_delta;
671 if (tp->t_ifree_delta != 0) {
672 msbp->msb_field = XFS_SBS_IFREE;
673 msbp->msb_delta = (int)tp->t_ifree_delta;
676 if (tp->t_fdblocks_delta != 0) {
677 msbp->msb_field = XFS_SBS_FDBLOCKS;
678 msbp->msb_delta = (int)tp->t_fdblocks_delta;
681 if (tp->t_frextents_delta != 0) {
682 msbp->msb_field = XFS_SBS_FREXTENTS;
683 msbp->msb_delta = (int)tp->t_frextents_delta;
686 if (tp->t_dblocks_delta != 0) {
687 msbp->msb_field = XFS_SBS_DBLOCKS;
688 msbp->msb_delta = (int)tp->t_dblocks_delta;
691 if (tp->t_agcount_delta != 0) {
692 msbp->msb_field = XFS_SBS_AGCOUNT;
693 msbp->msb_delta = (int)tp->t_agcount_delta;
696 if (tp->t_imaxpct_delta != 0) {
697 msbp->msb_field = XFS_SBS_IMAX_PCT;
698 msbp->msb_delta = (int)tp->t_imaxpct_delta;
701 if (tp->t_rextsize_delta != 0) {
702 msbp->msb_field = XFS_SBS_REXTSIZE;
703 msbp->msb_delta = (int)tp->t_rextsize_delta;
706 if (tp->t_rbmblocks_delta != 0) {
707 msbp->msb_field = XFS_SBS_RBMBLOCKS;
708 msbp->msb_delta = (int)tp->t_rbmblocks_delta;
711 if (tp->t_rblocks_delta != 0) {
712 msbp->msb_field = XFS_SBS_RBLOCKS;
713 msbp->msb_delta = (int)tp->t_rblocks_delta;
716 if (tp->t_rextents_delta != 0) {
717 msbp->msb_field = XFS_SBS_REXTENTS;
718 msbp->msb_delta = (int)tp->t_rextents_delta;
721 if (tp->t_rextslog_delta != 0) {
722 msbp->msb_field = XFS_SBS_REXTSLOG;
723 msbp->msb_delta = (int)tp->t_rextslog_delta;
729 * If we need to change anything, do it.
732 error = xfs_mod_incore_sb_batch(tp->t_mountp, msb,
733 (uint)(msbp - msb), rsvd);
742 * Commit the given transaction to the log a/synchronously.
744 * XFS disk error handling mechanism is not based on a typical
745 * transaction abort mechanism. Logically after the filesystem
746 * gets marked 'SHUTDOWN', we can't let any new transactions
747 * be durable - ie. committed to disk - because some metadata might
748 * be inconsistent. In such cases, this returns an error, and the
749 * caller may assume that all locked objects joined to the transaction
750 * have already been unlocked as if the commit had succeeded.
751 * Do not reference the transaction structure after this call.
758 xfs_lsn_t *commit_lsn_p,
761 xfs_log_iovec_t *log_vector;
764 xfs_lsn_t commit_lsn;
769 #define XFS_TRANS_LOGVEC_COUNT 16
770 xfs_log_iovec_t log_vector_fast[XFS_TRANS_LOGVEC_COUNT];
777 * Determine whether this commit is releasing a permanent
778 * log reservation or not.
780 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
781 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
782 log_flags = XFS_LOG_REL_PERM_RESERV;
789 * If there is nothing to be logged by the transaction,
790 * then unlock all of the items associated with the
791 * transaction and free the transaction structure.
792 * Also make sure to return any reserved blocks to
796 shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
797 if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
798 xfs_trans_unreserve_and_mod_sb(tp);
800 * It is indeed possible for the transaction to be
801 * not dirty but the dqinfo portion to be. All that
802 * means is that we have some (non-persistent) quota
803 * reservations that need to be unreserved.
805 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
807 commit_lsn = xfs_log_done(mp, tp->t_ticket,
809 if (commit_lsn == -1 && !shutdown)
810 shutdown = XFS_ERROR(EIO);
812 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
813 xfs_trans_free_items(tp, shutdown? XFS_TRANS_ABORT : 0);
814 xfs_trans_free_busy(tp);
816 XFS_STATS_INC(xs_trans_empty);
818 *commit_lsn_p = commit_lsn;
821 ASSERT(tp->t_ticket != NULL);
824 * If we need to update the superblock, then do it now.
826 if (tp->t_flags & XFS_TRANS_SB_DIRTY) {
827 xfs_trans_apply_sb_deltas(tp);
829 XFS_TRANS_APPLY_DQUOT_DELTAS(mp, tp);
832 * Ask each log item how many log_vector entries it will
833 * need so we can figure out how many to allocate.
834 * Try to avoid the kmem_alloc() call in the common case
835 * by using a vector from the stack when it fits.
837 nvec = xfs_trans_count_vecs(tp);
839 xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
841 } else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
842 log_vector = log_vector_fast;
844 log_vector = (xfs_log_iovec_t *)kmem_alloc(nvec *
845 sizeof(xfs_log_iovec_t),
850 * Fill in the log_vector and pin the logged items, and
851 * then write the transaction to the log.
853 xfs_trans_fill_vecs(tp, log_vector);
855 error = xfs_log_write(mp, log_vector, nvec, tp->t_ticket, &(tp->t_lsn));
858 * The transaction is committed incore here, and can go out to disk
859 * at any time after this call. However, all the items associated
860 * with the transaction are still locked and pinned in memory.
862 commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
864 tp->t_commit_lsn = commit_lsn;
865 if (nvec > XFS_TRANS_LOGVEC_COUNT) {
866 kmem_free(log_vector, nvec * sizeof(xfs_log_iovec_t));
870 *commit_lsn_p = commit_lsn;
873 * If we got a log write error. Unpin the logitems that we
874 * had pinned, clean up, free trans structure, and return error.
876 if (error || commit_lsn == -1) {
877 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
878 xfs_trans_uncommit(tp, flags|XFS_TRANS_ABORT);
879 return XFS_ERROR(EIO);
883 * Once the transaction has committed, unused
884 * reservations need to be released and changes to
885 * the superblock need to be reflected in the in-core
886 * version. Do that now.
888 xfs_trans_unreserve_and_mod_sb(tp);
890 sync = tp->t_flags & XFS_TRANS_SYNC;
893 * Tell the LM to call the transaction completion routine
894 * when the log write with LSN commit_lsn completes (e.g.
895 * when the transaction commit really hits the on-disk log).
896 * After this call we cannot reference tp, because the call
897 * can happen at any time and the call will free the transaction
898 * structure pointed to by tp. The only case where we call
899 * the completion routine (xfs_trans_committed) directly is
900 * if the log is turned off on a debug kernel or we're
901 * running in simulation mode (the log is explicitly turned
904 tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
905 tp->t_logcb.cb_arg = tp;
908 * We need to pass the iclog buffer which was used for the
909 * transaction commit record into this function, and attach
910 * the callback to it. The callback must be attached before
911 * the items are unlocked to avoid racing with other threads
912 * waiting for an item to unlock.
914 shutdown = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
917 * Mark this thread as no longer being in a transaction
919 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
922 * Once all the items of the transaction have been copied
923 * to the in core log and the callback is attached, the
924 * items can be unlocked.
926 * This will free descriptors pointing to items which were
927 * not logged since there is nothing more to do with them.
928 * For items which were logged, we will keep pointers to them
929 * so they can be unpinned after the transaction commits to disk.
930 * This will also stamp each modified meta-data item with
931 * the commit lsn of this transaction for dependency tracking
934 xfs_trans_unlock_items(tp, commit_lsn);
937 * If we detected a log error earlier, finish committing
938 * the transaction now (unpin log items, etc).
940 * Order is critical here, to avoid using the transaction
941 * pointer after its been freed (by xfs_trans_committed
942 * either here now, or as a callback). We cannot do this
943 * step inside xfs_log_notify as was done earlier because
947 xfs_trans_committed(tp, XFS_LI_ABORTED);
950 * Now that the xfs_trans_committed callback has been attached,
951 * and the items are released we can finally allow the iclog to
954 error = xfs_log_release_iclog(mp, commit_iclog);
957 * If the transaction needs to be synchronous, then force the
958 * log out now and wait for it.
962 error = _xfs_log_force(mp, commit_lsn,
963 XFS_LOG_FORCE | XFS_LOG_SYNC,
966 XFS_STATS_INC(xs_trans_sync);
968 XFS_STATS_INC(xs_trans_async);
976 * Total up the number of log iovecs needed to commit this
977 * transaction. The transaction itself needs one for the
978 * transaction header. Ask each dirty item in turn how many
979 * it needs to get the total.
982 xfs_trans_count_vecs(
986 xfs_log_item_desc_t *lidp;
989 lidp = xfs_trans_first_item(tp);
990 ASSERT(lidp != NULL);
992 /* In the non-debug case we need to start bailing out if we
993 * didn't find a log_item here, return zero and let trans_commit
999 while (lidp != NULL) {
1001 * Skip items which aren't dirty in this transaction.
1003 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1004 lidp = xfs_trans_next_item(tp, lidp);
1007 lidp->lid_size = IOP_SIZE(lidp->lid_item);
1008 nvecs += lidp->lid_size;
1009 lidp = xfs_trans_next_item(tp, lidp);
1016 * Called from the trans_commit code when we notice that
1017 * the filesystem is in the middle of a forced shutdown.
1024 xfs_log_item_desc_t *lidp;
1026 for (lidp = xfs_trans_first_item(tp);
1028 lidp = xfs_trans_next_item(tp, lidp)) {
1030 * Unpin all but those that aren't dirty.
1032 if (lidp->lid_flags & XFS_LID_DIRTY)
1033 IOP_UNPIN_REMOVE(lidp->lid_item, tp);
1036 xfs_trans_unreserve_and_mod_sb(tp);
1037 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(tp->t_mountp, tp);
1039 xfs_trans_free_items(tp, flags);
1040 xfs_trans_free_busy(tp);
1045 * Fill in the vector with pointers to data to be logged
1046 * by this transaction. The transaction header takes
1047 * the first vector, and then each dirty item takes the
1048 * number of vectors it indicated it needed in xfs_trans_count_vecs().
1050 * As each item fills in the entries it needs, also pin the item
1051 * so that it cannot be flushed out until the log write completes.
1054 xfs_trans_fill_vecs(
1056 xfs_log_iovec_t *log_vector)
1058 xfs_log_item_desc_t *lidp;
1059 xfs_log_iovec_t *vecp;
1063 * Skip over the entry for the transaction header, we'll
1064 * fill that in at the end.
1066 vecp = log_vector + 1; /* pointer arithmetic */
1069 lidp = xfs_trans_first_item(tp);
1070 ASSERT(lidp != NULL);
1071 while (lidp != NULL) {
1073 * Skip items which aren't dirty in this transaction.
1075 if (!(lidp->lid_flags & XFS_LID_DIRTY)) {
1076 lidp = xfs_trans_next_item(tp, lidp);
1080 * The item may be marked dirty but not log anything.
1081 * This can be used to get called when a transaction
1084 if (lidp->lid_size) {
1087 IOP_FORMAT(lidp->lid_item, vecp);
1088 vecp += lidp->lid_size; /* pointer arithmetic */
1089 IOP_PIN(lidp->lid_item);
1090 lidp = xfs_trans_next_item(tp, lidp);
1094 * Now that we've counted the number of items in this
1095 * transaction, fill in the transaction header.
1097 tp->t_header.th_magic = XFS_TRANS_HEADER_MAGIC;
1098 tp->t_header.th_type = tp->t_type;
1099 tp->t_header.th_num_items = nitems;
1100 log_vector->i_addr = (xfs_caddr_t)&tp->t_header;
1101 log_vector->i_len = sizeof(xfs_trans_header_t);
1102 XLOG_VEC_SET_TYPE(log_vector, XLOG_REG_TYPE_TRANSHDR);
1107 * Unlock all of the transaction's items and free the transaction.
1108 * The transaction must not have modified any of its items, because
1109 * there is no way to restore them to their previous state.
1111 * If the transaction has made a log reservation, make sure to release
1121 xfs_log_item_chunk_t *licp;
1122 xfs_log_item_desc_t *lidp;
1123 xfs_log_item_t *lip;
1126 xfs_mount_t *mp = tp->t_mountp;
1129 * See if the caller is being too lazy to figure out if
1130 * the transaction really needs an abort.
1132 if ((flags & XFS_TRANS_ABORT) && !(tp->t_flags & XFS_TRANS_DIRTY))
1133 flags &= ~XFS_TRANS_ABORT;
1135 * See if the caller is relying on us to shut down the
1136 * filesystem. This happens in paths where we detect
1137 * corruption and decide to give up.
1139 if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
1140 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
1141 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1144 if (!(flags & XFS_TRANS_ABORT)) {
1145 licp = &(tp->t_items);
1146 while (licp != NULL) {
1147 lidp = licp->lic_descs;
1148 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1149 if (XFS_LIC_ISFREE(licp, i)) {
1153 lip = lidp->lid_item;
1154 if (!XFS_FORCED_SHUTDOWN(mp))
1155 ASSERT(!(lip->li_type == XFS_LI_EFD));
1157 licp = licp->lic_next;
1161 xfs_trans_unreserve_and_mod_sb(tp);
1162 XFS_TRANS_UNRESERVE_AND_MOD_DQUOTS(mp, tp);
1165 if (flags & XFS_TRANS_RELEASE_LOG_RES) {
1166 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
1167 log_flags = XFS_LOG_REL_PERM_RESERV;
1171 xfs_log_done(mp, tp->t_ticket, NULL, log_flags);
1174 /* mark this thread as no longer being in a transaction */
1175 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1177 xfs_trans_free_items(tp, flags);
1178 xfs_trans_free_busy(tp);
1184 * Free the transaction structure. If there is more clean up
1185 * to do when the structure is freed, add it here.
1191 atomic_dec(&tp->t_mountp->m_active_trans);
1192 XFS_TRANS_FREE_DQINFO(tp->t_mountp, tp);
1193 kmem_zone_free(xfs_trans_zone, tp);
1198 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item().
1200 * This is typically called by the LM when a transaction has been fully
1201 * committed to disk. It needs to unpin the items which have
1202 * been logged by the transaction and update their positions
1203 * in the AIL if necessary.
1204 * This also gets called when the transactions didn't get written out
1205 * because of an I/O error. Abortflag & XFS_LI_ABORTED is set then.
1207 * Call xfs_trans_chunk_committed() to process the items in
1211 xfs_trans_committed(
1215 xfs_log_item_chunk_t *licp;
1216 xfs_log_item_chunk_t *next_licp;
1217 xfs_log_busy_chunk_t *lbcp;
1218 xfs_log_busy_slot_t *lbsp;
1222 * Call the transaction's completion callback if there
1225 if (tp->t_callback != NULL) {
1226 tp->t_callback(tp, tp->t_callarg);
1230 * Special case the chunk embedded in the transaction.
1232 licp = &(tp->t_items);
1233 if (!(XFS_LIC_ARE_ALL_FREE(licp))) {
1234 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1238 * Process the items in each chunk in turn.
1240 licp = licp->lic_next;
1241 while (licp != NULL) {
1242 ASSERT(!XFS_LIC_ARE_ALL_FREE(licp));
1243 xfs_trans_chunk_committed(licp, tp->t_lsn, abortflag);
1244 next_licp = licp->lic_next;
1245 kmem_free(licp, sizeof(xfs_log_item_chunk_t));
1250 * Clear all the per-AG busy list items listed in this transaction
1253 while (lbcp != NULL) {
1254 for (i = 0, lbsp = lbcp->lbc_busy; i < lbcp->lbc_unused; i++, lbsp++) {
1255 if (!XFS_LBC_ISFREE(lbcp, i)) {
1256 xfs_alloc_clear_busy(tp, lbsp->lbc_ag,
1260 lbcp = lbcp->lbc_next;
1262 xfs_trans_free_busy(tp);
1265 * That's it for the transaction structure. Free it.
1271 * This is called to perform the commit processing for each
1272 * item described by the given chunk.
1274 * The commit processing consists of unlocking items which were
1275 * held locked with the SYNC_UNLOCK attribute, calling the committed
1276 * routine of each logged item, updating the item's position in the AIL
1277 * if necessary, and unpinning each item. If the committed routine
1278 * returns -1, then do nothing further with the item because it
1279 * may have been freed.
1281 * Since items are unlocked when they are copied to the incore
1282 * log, it is possible for two transactions to be completing
1283 * and manipulating the same item simultaneously. The AIL lock
1284 * will protect the lsn field of each item. The value of this
1285 * field can never go backwards.
1287 * We unpin the items after repositioning them in the AIL, because
1288 * otherwise they could be immediately flushed and we'd have to race
1289 * with the flusher trying to pull the item from the AIL as we add it.
1292 xfs_trans_chunk_committed(
1293 xfs_log_item_chunk_t *licp,
1297 xfs_log_item_desc_t *lidp;
1298 xfs_log_item_t *lip;
1300 struct xfs_mount *mp;
1304 lidp = licp->lic_descs;
1305 for (i = 0; i < licp->lic_unused; i++, lidp++) {
1306 if (XFS_LIC_ISFREE(licp, i)) {
1310 lip = lidp->lid_item;
1312 lip->li_flags |= XFS_LI_ABORTED;
1315 * Send in the ABORTED flag to the COMMITTED routine
1316 * so that it knows whether the transaction was aborted
1319 item_lsn = IOP_COMMITTED(lip, lsn);
1322 * If the committed routine returns -1, make
1323 * no more references to the item.
1325 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0) {
1330 * If the returned lsn is greater than what it
1331 * contained before, update the location of the
1332 * item in the AIL. If it is not, then do nothing.
1333 * Items can never move backwards in the AIL.
1335 * While the new lsn should usually be greater, it
1336 * is possible that a later transaction completing
1337 * simultaneously with an earlier one using the
1338 * same item could complete first with a higher lsn.
1339 * This would cause the earlier transaction to fail
1342 mp = lip->li_mountp;
1344 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0) {
1346 * This will set the item's lsn to item_lsn
1347 * and update the position of the item in
1350 * xfs_trans_update_ail() drops the AIL lock.
1352 xfs_trans_update_ail(mp, lip, item_lsn, s);
1358 * Now that we've repositioned the item in the AIL,
1359 * unpin it so it can be flushed. Pass information
1360 * about buffer stale state down from the log item
1361 * flags, if anyone else stales the buffer we do not
1362 * want to pay any attention to it.
1364 IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);