Loading...
1/*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (C) 2010 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_shared.h"
22#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_mount.h"
26#include "xfs_inode.h"
27#include "xfs_extent_busy.h"
28#include "xfs_quota.h"
29#include "xfs_trans.h"
30#include "xfs_trans_priv.h"
31#include "xfs_log.h"
32#include "xfs_trace.h"
33#include "xfs_error.h"
34
35kmem_zone_t *xfs_trans_zone;
36kmem_zone_t *xfs_log_item_desc_zone;
37
38/*
39 * Initialize the precomputed transaction reservation values
40 * in the mount structure.
41 */
42void
43xfs_trans_init(
44 struct xfs_mount *mp)
45{
46 xfs_trans_resv_calc(mp, M_RES(mp));
47}
48
49/*
50 * Free the transaction structure. If there is more clean up
51 * to do when the structure is freed, add it here.
52 */
53STATIC void
54xfs_trans_free(
55 struct xfs_trans *tp)
56{
57 xfs_extent_busy_sort(&tp->t_busy);
58 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
59
60 atomic_dec(&tp->t_mountp->m_active_trans);
61 if (!(tp->t_flags & XFS_TRANS_NO_WRITECOUNT))
62 sb_end_intwrite(tp->t_mountp->m_super);
63 xfs_trans_free_dqinfo(tp);
64 kmem_zone_free(xfs_trans_zone, tp);
65}
66
67/*
68 * This is called to create a new transaction which will share the
69 * permanent log reservation of the given transaction. The remaining
70 * unused block and rt extent reservations are also inherited. This
71 * implies that the original transaction is no longer allowed to allocate
72 * blocks. Locks and log items, however, are no inherited. They must
73 * be added to the new transaction explicitly.
74 */
75STATIC xfs_trans_t *
76xfs_trans_dup(
77 xfs_trans_t *tp)
78{
79 xfs_trans_t *ntp;
80
81 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
82
83 /*
84 * Initialize the new transaction structure.
85 */
86 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
87 ntp->t_mountp = tp->t_mountp;
88 INIT_LIST_HEAD(&ntp->t_items);
89 INIT_LIST_HEAD(&ntp->t_busy);
90
91 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
92 ASSERT(tp->t_ticket != NULL);
93
94 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
95 (tp->t_flags & XFS_TRANS_RESERVE) |
96 (tp->t_flags & XFS_TRANS_NO_WRITECOUNT);
97 /* We gave our writer reference to the new transaction */
98 tp->t_flags |= XFS_TRANS_NO_WRITECOUNT;
99 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
100 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
101 tp->t_blk_res = tp->t_blk_res_used;
102 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
103 tp->t_rtx_res = tp->t_rtx_res_used;
104 ntp->t_pflags = tp->t_pflags;
105
106 xfs_trans_dup_dqinfo(tp, ntp);
107
108 atomic_inc(&tp->t_mountp->m_active_trans);
109 return ntp;
110}
111
112/*
113 * This is called to reserve free disk blocks and log space for the
114 * given transaction. This must be done before allocating any resources
115 * within the transaction.
116 *
117 * This will return ENOSPC if there are not enough blocks available.
118 * It will sleep waiting for available log space.
119 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
120 * is used by long running transactions. If any one of the reservations
121 * fails then they will all be backed out.
122 *
123 * This does not do quota reservations. That typically is done by the
124 * caller afterwards.
125 */
126static int
127xfs_trans_reserve(
128 struct xfs_trans *tp,
129 struct xfs_trans_res *resp,
130 uint blocks,
131 uint rtextents)
132{
133 int error = 0;
134 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
135
136 /* Mark this thread as being in a transaction */
137 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
138
139 /*
140 * Attempt to reserve the needed disk blocks by decrementing
141 * the number needed from the number available. This will
142 * fail if the count would go below zero.
143 */
144 if (blocks > 0) {
145 error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
146 if (error != 0) {
147 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
148 return -ENOSPC;
149 }
150 tp->t_blk_res += blocks;
151 }
152
153 /*
154 * Reserve the log space needed for this transaction.
155 */
156 if (resp->tr_logres > 0) {
157 bool permanent = false;
158
159 ASSERT(tp->t_log_res == 0 ||
160 tp->t_log_res == resp->tr_logres);
161 ASSERT(tp->t_log_count == 0 ||
162 tp->t_log_count == resp->tr_logcount);
163
164 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
165 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
166 permanent = true;
167 } else {
168 ASSERT(tp->t_ticket == NULL);
169 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
170 }
171
172 if (tp->t_ticket != NULL) {
173 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
174 error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
175 } else {
176 error = xfs_log_reserve(tp->t_mountp,
177 resp->tr_logres,
178 resp->tr_logcount,
179 &tp->t_ticket, XFS_TRANSACTION,
180 permanent);
181 }
182
183 if (error)
184 goto undo_blocks;
185
186 tp->t_log_res = resp->tr_logres;
187 tp->t_log_count = resp->tr_logcount;
188 }
189
190 /*
191 * Attempt to reserve the needed realtime extents by decrementing
192 * the number needed from the number available. This will
193 * fail if the count would go below zero.
194 */
195 if (rtextents > 0) {
196 error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
197 if (error) {
198 error = -ENOSPC;
199 goto undo_log;
200 }
201 tp->t_rtx_res += rtextents;
202 }
203
204 return 0;
205
206 /*
207 * Error cases jump to one of these labels to undo any
208 * reservations which have already been performed.
209 */
210undo_log:
211 if (resp->tr_logres > 0) {
212 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
213 tp->t_ticket = NULL;
214 tp->t_log_res = 0;
215 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
216 }
217
218undo_blocks:
219 if (blocks > 0) {
220 xfs_mod_fdblocks(tp->t_mountp, (int64_t)blocks, rsvd);
221 tp->t_blk_res = 0;
222 }
223
224 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
225
226 return error;
227}
228
229int
230xfs_trans_alloc(
231 struct xfs_mount *mp,
232 struct xfs_trans_res *resp,
233 uint blocks,
234 uint rtextents,
235 uint flags,
236 struct xfs_trans **tpp)
237{
238 struct xfs_trans *tp;
239 int error;
240
241 if (!(flags & XFS_TRANS_NO_WRITECOUNT))
242 sb_start_intwrite(mp->m_super);
243
244 WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
245 atomic_inc(&mp->m_active_trans);
246
247 tp = kmem_zone_zalloc(xfs_trans_zone,
248 (flags & XFS_TRANS_NOFS) ? KM_NOFS : KM_SLEEP);
249 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
250 tp->t_flags = flags;
251 tp->t_mountp = mp;
252 INIT_LIST_HEAD(&tp->t_items);
253 INIT_LIST_HEAD(&tp->t_busy);
254
255 error = xfs_trans_reserve(tp, resp, blocks, rtextents);
256 if (error) {
257 xfs_trans_cancel(tp);
258 return error;
259 }
260
261 *tpp = tp;
262 return 0;
263}
264
265/*
266 * Record the indicated change to the given field for application
267 * to the file system's superblock when the transaction commits.
268 * For now, just store the change in the transaction structure.
269 *
270 * Mark the transaction structure to indicate that the superblock
271 * needs to be updated before committing.
272 *
273 * Because we may not be keeping track of allocated/free inodes and
274 * used filesystem blocks in the superblock, we do not mark the
275 * superblock dirty in this transaction if we modify these fields.
276 * We still need to update the transaction deltas so that they get
277 * applied to the incore superblock, but we don't want them to
278 * cause the superblock to get locked and logged if these are the
279 * only fields in the superblock that the transaction modifies.
280 */
281void
282xfs_trans_mod_sb(
283 xfs_trans_t *tp,
284 uint field,
285 int64_t delta)
286{
287 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
288 xfs_mount_t *mp = tp->t_mountp;
289
290 switch (field) {
291 case XFS_TRANS_SB_ICOUNT:
292 tp->t_icount_delta += delta;
293 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
294 flags &= ~XFS_TRANS_SB_DIRTY;
295 break;
296 case XFS_TRANS_SB_IFREE:
297 tp->t_ifree_delta += delta;
298 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
299 flags &= ~XFS_TRANS_SB_DIRTY;
300 break;
301 case XFS_TRANS_SB_FDBLOCKS:
302 /*
303 * Track the number of blocks allocated in the
304 * transaction. Make sure it does not exceed the
305 * number reserved.
306 */
307 if (delta < 0) {
308 tp->t_blk_res_used += (uint)-delta;
309 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
310 }
311 tp->t_fdblocks_delta += delta;
312 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
313 flags &= ~XFS_TRANS_SB_DIRTY;
314 break;
315 case XFS_TRANS_SB_RES_FDBLOCKS:
316 /*
317 * The allocation has already been applied to the
318 * in-core superblock's counter. This should only
319 * be applied to the on-disk superblock.
320 */
321 tp->t_res_fdblocks_delta += delta;
322 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
323 flags &= ~XFS_TRANS_SB_DIRTY;
324 break;
325 case XFS_TRANS_SB_FREXTENTS:
326 /*
327 * Track the number of blocks allocated in the
328 * transaction. Make sure it does not exceed the
329 * number reserved.
330 */
331 if (delta < 0) {
332 tp->t_rtx_res_used += (uint)-delta;
333 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
334 }
335 tp->t_frextents_delta += delta;
336 break;
337 case XFS_TRANS_SB_RES_FREXTENTS:
338 /*
339 * The allocation has already been applied to the
340 * in-core superblock's counter. This should only
341 * be applied to the on-disk superblock.
342 */
343 ASSERT(delta < 0);
344 tp->t_res_frextents_delta += delta;
345 break;
346 case XFS_TRANS_SB_DBLOCKS:
347 ASSERT(delta > 0);
348 tp->t_dblocks_delta += delta;
349 break;
350 case XFS_TRANS_SB_AGCOUNT:
351 ASSERT(delta > 0);
352 tp->t_agcount_delta += delta;
353 break;
354 case XFS_TRANS_SB_IMAXPCT:
355 tp->t_imaxpct_delta += delta;
356 break;
357 case XFS_TRANS_SB_REXTSIZE:
358 tp->t_rextsize_delta += delta;
359 break;
360 case XFS_TRANS_SB_RBMBLOCKS:
361 tp->t_rbmblocks_delta += delta;
362 break;
363 case XFS_TRANS_SB_RBLOCKS:
364 tp->t_rblocks_delta += delta;
365 break;
366 case XFS_TRANS_SB_REXTENTS:
367 tp->t_rextents_delta += delta;
368 break;
369 case XFS_TRANS_SB_REXTSLOG:
370 tp->t_rextslog_delta += delta;
371 break;
372 default:
373 ASSERT(0);
374 return;
375 }
376
377 tp->t_flags |= flags;
378}
379
380/*
381 * xfs_trans_apply_sb_deltas() is called from the commit code
382 * to bring the superblock buffer into the current transaction
383 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
384 *
385 * For now we just look at each field allowed to change and change
386 * it if necessary.
387 */
388STATIC void
389xfs_trans_apply_sb_deltas(
390 xfs_trans_t *tp)
391{
392 xfs_dsb_t *sbp;
393 xfs_buf_t *bp;
394 int whole = 0;
395
396 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
397 sbp = XFS_BUF_TO_SBP(bp);
398
399 /*
400 * Check that superblock mods match the mods made to AGF counters.
401 */
402 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
403 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
404 tp->t_ag_btree_delta));
405
406 /*
407 * Only update the superblock counters if we are logging them
408 */
409 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
410 if (tp->t_icount_delta)
411 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
412 if (tp->t_ifree_delta)
413 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
414 if (tp->t_fdblocks_delta)
415 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
416 if (tp->t_res_fdblocks_delta)
417 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
418 }
419
420 if (tp->t_frextents_delta)
421 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
422 if (tp->t_res_frextents_delta)
423 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
424
425 if (tp->t_dblocks_delta) {
426 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
427 whole = 1;
428 }
429 if (tp->t_agcount_delta) {
430 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
431 whole = 1;
432 }
433 if (tp->t_imaxpct_delta) {
434 sbp->sb_imax_pct += tp->t_imaxpct_delta;
435 whole = 1;
436 }
437 if (tp->t_rextsize_delta) {
438 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
439 whole = 1;
440 }
441 if (tp->t_rbmblocks_delta) {
442 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
443 whole = 1;
444 }
445 if (tp->t_rblocks_delta) {
446 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
447 whole = 1;
448 }
449 if (tp->t_rextents_delta) {
450 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
451 whole = 1;
452 }
453 if (tp->t_rextslog_delta) {
454 sbp->sb_rextslog += tp->t_rextslog_delta;
455 whole = 1;
456 }
457
458 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
459 if (whole)
460 /*
461 * Log the whole thing, the fields are noncontiguous.
462 */
463 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
464 else
465 /*
466 * Since all the modifiable fields are contiguous, we
467 * can get away with this.
468 */
469 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
470 offsetof(xfs_dsb_t, sb_frextents) +
471 sizeof(sbp->sb_frextents) - 1);
472}
473
474STATIC int
475xfs_sb_mod8(
476 uint8_t *field,
477 int8_t delta)
478{
479 int8_t counter = *field;
480
481 counter += delta;
482 if (counter < 0) {
483 ASSERT(0);
484 return -EINVAL;
485 }
486 *field = counter;
487 return 0;
488}
489
490STATIC int
491xfs_sb_mod32(
492 uint32_t *field,
493 int32_t delta)
494{
495 int32_t counter = *field;
496
497 counter += delta;
498 if (counter < 0) {
499 ASSERT(0);
500 return -EINVAL;
501 }
502 *field = counter;
503 return 0;
504}
505
506STATIC int
507xfs_sb_mod64(
508 uint64_t *field,
509 int64_t delta)
510{
511 int64_t counter = *field;
512
513 counter += delta;
514 if (counter < 0) {
515 ASSERT(0);
516 return -EINVAL;
517 }
518 *field = counter;
519 return 0;
520}
521
522/*
523 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
524 * and apply superblock counter changes to the in-core superblock. The
525 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
526 * applied to the in-core superblock. The idea is that that has already been
527 * done.
528 *
529 * If we are not logging superblock counters, then the inode allocated/free and
530 * used block counts are not updated in the on disk superblock. In this case,
531 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
532 * still need to update the incore superblock with the changes.
533 */
534void
535xfs_trans_unreserve_and_mod_sb(
536 struct xfs_trans *tp)
537{
538 struct xfs_mount *mp = tp->t_mountp;
539 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
540 int64_t blkdelta = 0;
541 int64_t rtxdelta = 0;
542 int64_t idelta = 0;
543 int64_t ifreedelta = 0;
544 int error;
545
546 /* calculate deltas */
547 if (tp->t_blk_res > 0)
548 blkdelta = tp->t_blk_res;
549 if ((tp->t_fdblocks_delta != 0) &&
550 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
551 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
552 blkdelta += tp->t_fdblocks_delta;
553
554 if (tp->t_rtx_res > 0)
555 rtxdelta = tp->t_rtx_res;
556 if ((tp->t_frextents_delta != 0) &&
557 (tp->t_flags & XFS_TRANS_SB_DIRTY))
558 rtxdelta += tp->t_frextents_delta;
559
560 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
561 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
562 idelta = tp->t_icount_delta;
563 ifreedelta = tp->t_ifree_delta;
564 }
565
566 /* apply the per-cpu counters */
567 if (blkdelta) {
568 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
569 if (error)
570 goto out;
571 }
572
573 if (idelta) {
574 error = xfs_mod_icount(mp, idelta);
575 if (error)
576 goto out_undo_fdblocks;
577 }
578
579 if (ifreedelta) {
580 error = xfs_mod_ifree(mp, ifreedelta);
581 if (error)
582 goto out_undo_icount;
583 }
584
585 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
586 return;
587
588 /* apply remaining deltas */
589 spin_lock(&mp->m_sb_lock);
590 if (rtxdelta) {
591 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
592 if (error)
593 goto out_undo_ifree;
594 }
595
596 if (tp->t_dblocks_delta != 0) {
597 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
598 if (error)
599 goto out_undo_frextents;
600 }
601 if (tp->t_agcount_delta != 0) {
602 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
603 if (error)
604 goto out_undo_dblocks;
605 }
606 if (tp->t_imaxpct_delta != 0) {
607 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
608 if (error)
609 goto out_undo_agcount;
610 }
611 if (tp->t_rextsize_delta != 0) {
612 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
613 tp->t_rextsize_delta);
614 if (error)
615 goto out_undo_imaxpct;
616 }
617 if (tp->t_rbmblocks_delta != 0) {
618 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
619 tp->t_rbmblocks_delta);
620 if (error)
621 goto out_undo_rextsize;
622 }
623 if (tp->t_rblocks_delta != 0) {
624 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
625 if (error)
626 goto out_undo_rbmblocks;
627 }
628 if (tp->t_rextents_delta != 0) {
629 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
630 tp->t_rextents_delta);
631 if (error)
632 goto out_undo_rblocks;
633 }
634 if (tp->t_rextslog_delta != 0) {
635 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
636 tp->t_rextslog_delta);
637 if (error)
638 goto out_undo_rextents;
639 }
640 spin_unlock(&mp->m_sb_lock);
641 return;
642
643out_undo_rextents:
644 if (tp->t_rextents_delta)
645 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
646out_undo_rblocks:
647 if (tp->t_rblocks_delta)
648 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
649out_undo_rbmblocks:
650 if (tp->t_rbmblocks_delta)
651 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
652out_undo_rextsize:
653 if (tp->t_rextsize_delta)
654 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
655out_undo_imaxpct:
656 if (tp->t_rextsize_delta)
657 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
658out_undo_agcount:
659 if (tp->t_agcount_delta)
660 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
661out_undo_dblocks:
662 if (tp->t_dblocks_delta)
663 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
664out_undo_frextents:
665 if (rtxdelta)
666 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
667out_undo_ifree:
668 spin_unlock(&mp->m_sb_lock);
669 if (ifreedelta)
670 xfs_mod_ifree(mp, -ifreedelta);
671out_undo_icount:
672 if (idelta)
673 xfs_mod_icount(mp, -idelta);
674out_undo_fdblocks:
675 if (blkdelta)
676 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
677out:
678 ASSERT(error == 0);
679 return;
680}
681
682/*
683 * Add the given log item to the transaction's list of log items.
684 *
685 * The log item will now point to its new descriptor with its li_desc field.
686 */
687void
688xfs_trans_add_item(
689 struct xfs_trans *tp,
690 struct xfs_log_item *lip)
691{
692 struct xfs_log_item_desc *lidp;
693
694 ASSERT(lip->li_mountp == tp->t_mountp);
695 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
696
697 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
698
699 lidp->lid_item = lip;
700 lidp->lid_flags = 0;
701 list_add_tail(&lidp->lid_trans, &tp->t_items);
702
703 lip->li_desc = lidp;
704}
705
706STATIC void
707xfs_trans_free_item_desc(
708 struct xfs_log_item_desc *lidp)
709{
710 list_del_init(&lidp->lid_trans);
711 kmem_zone_free(xfs_log_item_desc_zone, lidp);
712}
713
714/*
715 * Unlink and free the given descriptor.
716 */
717void
718xfs_trans_del_item(
719 struct xfs_log_item *lip)
720{
721 xfs_trans_free_item_desc(lip->li_desc);
722 lip->li_desc = NULL;
723}
724
725/*
726 * Unlock all of the items of a transaction and free all the descriptors
727 * of that transaction.
728 */
729void
730xfs_trans_free_items(
731 struct xfs_trans *tp,
732 xfs_lsn_t commit_lsn,
733 bool abort)
734{
735 struct xfs_log_item_desc *lidp, *next;
736
737 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
738 struct xfs_log_item *lip = lidp->lid_item;
739
740 lip->li_desc = NULL;
741
742 if (commit_lsn != NULLCOMMITLSN)
743 lip->li_ops->iop_committing(lip, commit_lsn);
744 if (abort)
745 lip->li_flags |= XFS_LI_ABORTED;
746 lip->li_ops->iop_unlock(lip);
747
748 xfs_trans_free_item_desc(lidp);
749 }
750}
751
752static inline void
753xfs_log_item_batch_insert(
754 struct xfs_ail *ailp,
755 struct xfs_ail_cursor *cur,
756 struct xfs_log_item **log_items,
757 int nr_items,
758 xfs_lsn_t commit_lsn)
759{
760 int i;
761
762 spin_lock(&ailp->xa_lock);
763 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
764 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
765
766 for (i = 0; i < nr_items; i++) {
767 struct xfs_log_item *lip = log_items[i];
768
769 lip->li_ops->iop_unpin(lip, 0);
770 }
771}
772
773/*
774 * Bulk operation version of xfs_trans_committed that takes a log vector of
775 * items to insert into the AIL. This uses bulk AIL insertion techniques to
776 * minimise lock traffic.
777 *
778 * If we are called with the aborted flag set, it is because a log write during
779 * a CIL checkpoint commit has failed. In this case, all the items in the
780 * checkpoint have already gone through iop_commited and iop_unlock, which
781 * means that checkpoint commit abort handling is treated exactly the same
782 * as an iclog write error even though we haven't started any IO yet. Hence in
783 * this case all we need to do is iop_committed processing, followed by an
784 * iop_unpin(aborted) call.
785 *
786 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
787 * at the end of the AIL, the insert cursor avoids the need to walk
788 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
789 * call. This saves a lot of needless list walking and is a net win, even
790 * though it slightly increases that amount of AIL lock traffic to set it up
791 * and tear it down.
792 */
793void
794xfs_trans_committed_bulk(
795 struct xfs_ail *ailp,
796 struct xfs_log_vec *log_vector,
797 xfs_lsn_t commit_lsn,
798 int aborted)
799{
800#define LOG_ITEM_BATCH_SIZE 32
801 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
802 struct xfs_log_vec *lv;
803 struct xfs_ail_cursor cur;
804 int i = 0;
805
806 spin_lock(&ailp->xa_lock);
807 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
808 spin_unlock(&ailp->xa_lock);
809
810 /* unpin all the log items */
811 for (lv = log_vector; lv; lv = lv->lv_next ) {
812 struct xfs_log_item *lip = lv->lv_item;
813 xfs_lsn_t item_lsn;
814
815 if (aborted)
816 lip->li_flags |= XFS_LI_ABORTED;
817 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
818
819 /* item_lsn of -1 means the item needs no further processing */
820 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
821 continue;
822
823 /*
824 * if we are aborting the operation, no point in inserting the
825 * object into the AIL as we are in a shutdown situation.
826 */
827 if (aborted) {
828 ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
829 lip->li_ops->iop_unpin(lip, 1);
830 continue;
831 }
832
833 if (item_lsn != commit_lsn) {
834
835 /*
836 * Not a bulk update option due to unusual item_lsn.
837 * Push into AIL immediately, rechecking the lsn once
838 * we have the ail lock. Then unpin the item. This does
839 * not affect the AIL cursor the bulk insert path is
840 * using.
841 */
842 spin_lock(&ailp->xa_lock);
843 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
844 xfs_trans_ail_update(ailp, lip, item_lsn);
845 else
846 spin_unlock(&ailp->xa_lock);
847 lip->li_ops->iop_unpin(lip, 0);
848 continue;
849 }
850
851 /* Item is a candidate for bulk AIL insert. */
852 log_items[i++] = lv->lv_item;
853 if (i >= LOG_ITEM_BATCH_SIZE) {
854 xfs_log_item_batch_insert(ailp, &cur, log_items,
855 LOG_ITEM_BATCH_SIZE, commit_lsn);
856 i = 0;
857 }
858 }
859
860 /* make sure we insert the remainder! */
861 if (i)
862 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
863
864 spin_lock(&ailp->xa_lock);
865 xfs_trans_ail_cursor_done(&cur);
866 spin_unlock(&ailp->xa_lock);
867}
868
869/*
870 * Commit the given transaction to the log.
871 *
872 * XFS disk error handling mechanism is not based on a typical
873 * transaction abort mechanism. Logically after the filesystem
874 * gets marked 'SHUTDOWN', we can't let any new transactions
875 * be durable - ie. committed to disk - because some metadata might
876 * be inconsistent. In such cases, this returns an error, and the
877 * caller may assume that all locked objects joined to the transaction
878 * have already been unlocked as if the commit had succeeded.
879 * Do not reference the transaction structure after this call.
880 */
881static int
882__xfs_trans_commit(
883 struct xfs_trans *tp,
884 bool regrant)
885{
886 struct xfs_mount *mp = tp->t_mountp;
887 xfs_lsn_t commit_lsn = -1;
888 int error = 0;
889 int sync = tp->t_flags & XFS_TRANS_SYNC;
890
891 /*
892 * If there is nothing to be logged by the transaction,
893 * then unlock all of the items associated with the
894 * transaction and free the transaction structure.
895 * Also make sure to return any reserved blocks to
896 * the free pool.
897 */
898 if (!(tp->t_flags & XFS_TRANS_DIRTY))
899 goto out_unreserve;
900
901 if (XFS_FORCED_SHUTDOWN(mp)) {
902 error = -EIO;
903 goto out_unreserve;
904 }
905
906 ASSERT(tp->t_ticket != NULL);
907
908 /*
909 * If we need to update the superblock, then do it now.
910 */
911 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
912 xfs_trans_apply_sb_deltas(tp);
913 xfs_trans_apply_dquot_deltas(tp);
914
915 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
916
917 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
918 xfs_trans_free(tp);
919
920 /*
921 * If the transaction needs to be synchronous, then force the
922 * log out now and wait for it.
923 */
924 if (sync) {
925 error = _xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
926 XFS_STATS_INC(mp, xs_trans_sync);
927 } else {
928 XFS_STATS_INC(mp, xs_trans_async);
929 }
930
931 return error;
932
933out_unreserve:
934 xfs_trans_unreserve_and_mod_sb(tp);
935
936 /*
937 * It is indeed possible for the transaction to be not dirty but
938 * the dqinfo portion to be. All that means is that we have some
939 * (non-persistent) quota reservations that need to be unreserved.
940 */
941 xfs_trans_unreserve_and_mod_dquots(tp);
942 if (tp->t_ticket) {
943 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
944 if (commit_lsn == -1 && !error)
945 error = -EIO;
946 }
947 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
948 xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
949 xfs_trans_free(tp);
950
951 XFS_STATS_INC(mp, xs_trans_empty);
952 return error;
953}
954
955int
956xfs_trans_commit(
957 struct xfs_trans *tp)
958{
959 return __xfs_trans_commit(tp, false);
960}
961
962/*
963 * Unlock all of the transaction's items and free the transaction.
964 * The transaction must not have modified any of its items, because
965 * there is no way to restore them to their previous state.
966 *
967 * If the transaction has made a log reservation, make sure to release
968 * it as well.
969 */
970void
971xfs_trans_cancel(
972 struct xfs_trans *tp)
973{
974 struct xfs_mount *mp = tp->t_mountp;
975 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
976
977 /*
978 * See if the caller is relying on us to shut down the
979 * filesystem. This happens in paths where we detect
980 * corruption and decide to give up.
981 */
982 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
983 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
984 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
985 }
986#ifdef DEBUG
987 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
988 struct xfs_log_item_desc *lidp;
989
990 list_for_each_entry(lidp, &tp->t_items, lid_trans)
991 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
992 }
993#endif
994 xfs_trans_unreserve_and_mod_sb(tp);
995 xfs_trans_unreserve_and_mod_dquots(tp);
996
997 if (tp->t_ticket)
998 xfs_log_done(mp, tp->t_ticket, NULL, false);
999
1000 /* mark this thread as no longer being in a transaction */
1001 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1002
1003 xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1004 xfs_trans_free(tp);
1005}
1006
1007/*
1008 * Roll from one trans in the sequence of PERMANENT transactions to
1009 * the next: permanent transactions are only flushed out when
1010 * committed with xfs_trans_commit(), but we still want as soon
1011 * as possible to let chunks of it go to the log. So we commit the
1012 * chunk we've been working on and get a new transaction to continue.
1013 */
1014int
1015__xfs_trans_roll(
1016 struct xfs_trans **tpp,
1017 struct xfs_inode *dp,
1018 int *committed)
1019{
1020 struct xfs_trans *trans;
1021 struct xfs_trans_res tres;
1022 int error;
1023
1024 *committed = 0;
1025
1026 /*
1027 * Ensure that the inode is always logged.
1028 */
1029 trans = *tpp;
1030 if (dp)
1031 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1032
1033 /*
1034 * Copy the critical parameters from one trans to the next.
1035 */
1036 tres.tr_logres = trans->t_log_res;
1037 tres.tr_logcount = trans->t_log_count;
1038 *tpp = xfs_trans_dup(trans);
1039
1040 /*
1041 * Commit the current transaction.
1042 * If this commit failed, then it'd just unlock those items that
1043 * are not marked ihold. That also means that a filesystem shutdown
1044 * is in progress. The caller takes the responsibility to cancel
1045 * the duplicate transaction that gets returned.
1046 */
1047 error = __xfs_trans_commit(trans, true);
1048 if (error)
1049 return error;
1050
1051 *committed = 1;
1052 trans = *tpp;
1053
1054 /*
1055 * Reserve space in the log for th next transaction.
1056 * This also pushes items in the "AIL", the list of logged items,
1057 * out to disk if they are taking up space at the tail of the log
1058 * that we want to use. This requires that either nothing be locked
1059 * across this call, or that anything that is locked be logged in
1060 * the prior and the next transactions.
1061 */
1062 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1063 error = xfs_trans_reserve(trans, &tres, 0, 0);
1064 /*
1065 * Ensure that the inode is in the new transaction and locked.
1066 */
1067 if (error)
1068 return error;
1069
1070 if (dp)
1071 xfs_trans_ijoin(trans, dp, 0);
1072 return 0;
1073}
1074
1075int
1076xfs_trans_roll(
1077 struct xfs_trans **tpp,
1078 struct xfs_inode *dp)
1079{
1080 int committed;
1081 return __xfs_trans_roll(tpp, dp, &committed);
1082}
1/*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (C) 2010 Red Hat, Inc.
4 * All Rights Reserved.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
21#include "xfs_shared.h"
22#include "xfs_format.h"
23#include "xfs_log_format.h"
24#include "xfs_trans_resv.h"
25#include "xfs_mount.h"
26#include "xfs_inode.h"
27#include "xfs_extent_busy.h"
28#include "xfs_quota.h"
29#include "xfs_trans.h"
30#include "xfs_trans_priv.h"
31#include "xfs_log.h"
32#include "xfs_trace.h"
33#include "xfs_error.h"
34
35kmem_zone_t *xfs_trans_zone;
36kmem_zone_t *xfs_log_item_desc_zone;
37
38/*
39 * Initialize the precomputed transaction reservation values
40 * in the mount structure.
41 */
42void
43xfs_trans_init(
44 struct xfs_mount *mp)
45{
46 xfs_trans_resv_calc(mp, M_RES(mp));
47}
48
49/*
50 * This routine is called to allocate a transaction structure.
51 * The type parameter indicates the type of the transaction. These
52 * are enumerated in xfs_trans.h.
53 *
54 * Dynamically allocate the transaction structure from the transaction
55 * zone, initialize it, and return it to the caller.
56 */
57xfs_trans_t *
58xfs_trans_alloc(
59 xfs_mount_t *mp,
60 uint type)
61{
62 xfs_trans_t *tp;
63
64 sb_start_intwrite(mp->m_super);
65 tp = _xfs_trans_alloc(mp, type, KM_SLEEP);
66 tp->t_flags |= XFS_TRANS_FREEZE_PROT;
67 return tp;
68}
69
70xfs_trans_t *
71_xfs_trans_alloc(
72 xfs_mount_t *mp,
73 uint type,
74 xfs_km_flags_t memflags)
75{
76 xfs_trans_t *tp;
77
78 WARN_ON(mp->m_super->s_writers.frozen == SB_FREEZE_COMPLETE);
79 atomic_inc(&mp->m_active_trans);
80
81 tp = kmem_zone_zalloc(xfs_trans_zone, memflags);
82 tp->t_magic = XFS_TRANS_HEADER_MAGIC;
83 tp->t_type = type;
84 tp->t_mountp = mp;
85 INIT_LIST_HEAD(&tp->t_items);
86 INIT_LIST_HEAD(&tp->t_busy);
87 return tp;
88}
89
90/*
91 * Free the transaction structure. If there is more clean up
92 * to do when the structure is freed, add it here.
93 */
94STATIC void
95xfs_trans_free(
96 struct xfs_trans *tp)
97{
98 xfs_extent_busy_sort(&tp->t_busy);
99 xfs_extent_busy_clear(tp->t_mountp, &tp->t_busy, false);
100
101 atomic_dec(&tp->t_mountp->m_active_trans);
102 if (tp->t_flags & XFS_TRANS_FREEZE_PROT)
103 sb_end_intwrite(tp->t_mountp->m_super);
104 xfs_trans_free_dqinfo(tp);
105 kmem_zone_free(xfs_trans_zone, tp);
106}
107
108/*
109 * This is called to create a new transaction which will share the
110 * permanent log reservation of the given transaction. The remaining
111 * unused block and rt extent reservations are also inherited. This
112 * implies that the original transaction is no longer allowed to allocate
113 * blocks. Locks and log items, however, are no inherited. They must
114 * be added to the new transaction explicitly.
115 */
116STATIC xfs_trans_t *
117xfs_trans_dup(
118 xfs_trans_t *tp)
119{
120 xfs_trans_t *ntp;
121
122 ntp = kmem_zone_zalloc(xfs_trans_zone, KM_SLEEP);
123
124 /*
125 * Initialize the new transaction structure.
126 */
127 ntp->t_magic = XFS_TRANS_HEADER_MAGIC;
128 ntp->t_type = tp->t_type;
129 ntp->t_mountp = tp->t_mountp;
130 INIT_LIST_HEAD(&ntp->t_items);
131 INIT_LIST_HEAD(&ntp->t_busy);
132
133 ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
134 ASSERT(tp->t_ticket != NULL);
135
136 ntp->t_flags = XFS_TRANS_PERM_LOG_RES |
137 (tp->t_flags & XFS_TRANS_RESERVE) |
138 (tp->t_flags & XFS_TRANS_FREEZE_PROT);
139 /* We gave our writer reference to the new transaction */
140 tp->t_flags &= ~XFS_TRANS_FREEZE_PROT;
141 ntp->t_ticket = xfs_log_ticket_get(tp->t_ticket);
142 ntp->t_blk_res = tp->t_blk_res - tp->t_blk_res_used;
143 tp->t_blk_res = tp->t_blk_res_used;
144 ntp->t_rtx_res = tp->t_rtx_res - tp->t_rtx_res_used;
145 tp->t_rtx_res = tp->t_rtx_res_used;
146 ntp->t_pflags = tp->t_pflags;
147
148 xfs_trans_dup_dqinfo(tp, ntp);
149
150 atomic_inc(&tp->t_mountp->m_active_trans);
151 return ntp;
152}
153
154/*
155 * This is called to reserve free disk blocks and log space for the
156 * given transaction. This must be done before allocating any resources
157 * within the transaction.
158 *
159 * This will return ENOSPC if there are not enough blocks available.
160 * It will sleep waiting for available log space.
161 * The only valid value for the flags parameter is XFS_RES_LOG_PERM, which
162 * is used by long running transactions. If any one of the reservations
163 * fails then they will all be backed out.
164 *
165 * This does not do quota reservations. That typically is done by the
166 * caller afterwards.
167 */
168int
169xfs_trans_reserve(
170 struct xfs_trans *tp,
171 struct xfs_trans_res *resp,
172 uint blocks,
173 uint rtextents)
174{
175 int error = 0;
176 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
177
178 /* Mark this thread as being in a transaction */
179 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
180
181 /*
182 * Attempt to reserve the needed disk blocks by decrementing
183 * the number needed from the number available. This will
184 * fail if the count would go below zero.
185 */
186 if (blocks > 0) {
187 error = xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
188 if (error != 0) {
189 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
190 return -ENOSPC;
191 }
192 tp->t_blk_res += blocks;
193 }
194
195 /*
196 * Reserve the log space needed for this transaction.
197 */
198 if (resp->tr_logres > 0) {
199 bool permanent = false;
200
201 ASSERT(tp->t_log_res == 0 ||
202 tp->t_log_res == resp->tr_logres);
203 ASSERT(tp->t_log_count == 0 ||
204 tp->t_log_count == resp->tr_logcount);
205
206 if (resp->tr_logflags & XFS_TRANS_PERM_LOG_RES) {
207 tp->t_flags |= XFS_TRANS_PERM_LOG_RES;
208 permanent = true;
209 } else {
210 ASSERT(tp->t_ticket == NULL);
211 ASSERT(!(tp->t_flags & XFS_TRANS_PERM_LOG_RES));
212 }
213
214 if (tp->t_ticket != NULL) {
215 ASSERT(resp->tr_logflags & XFS_TRANS_PERM_LOG_RES);
216 error = xfs_log_regrant(tp->t_mountp, tp->t_ticket);
217 } else {
218 error = xfs_log_reserve(tp->t_mountp,
219 resp->tr_logres,
220 resp->tr_logcount,
221 &tp->t_ticket, XFS_TRANSACTION,
222 permanent, tp->t_type);
223 }
224
225 if (error)
226 goto undo_blocks;
227
228 tp->t_log_res = resp->tr_logres;
229 tp->t_log_count = resp->tr_logcount;
230 }
231
232 /*
233 * Attempt to reserve the needed realtime extents by decrementing
234 * the number needed from the number available. This will
235 * fail if the count would go below zero.
236 */
237 if (rtextents > 0) {
238 error = xfs_mod_frextents(tp->t_mountp, -((int64_t)rtextents));
239 if (error) {
240 error = -ENOSPC;
241 goto undo_log;
242 }
243 tp->t_rtx_res += rtextents;
244 }
245
246 return 0;
247
248 /*
249 * Error cases jump to one of these labels to undo any
250 * reservations which have already been performed.
251 */
252undo_log:
253 if (resp->tr_logres > 0) {
254 xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, false);
255 tp->t_ticket = NULL;
256 tp->t_log_res = 0;
257 tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
258 }
259
260undo_blocks:
261 if (blocks > 0) {
262 xfs_mod_fdblocks(tp->t_mountp, -((int64_t)blocks), rsvd);
263 tp->t_blk_res = 0;
264 }
265
266 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
267
268 return error;
269}
270
271/*
272 * Record the indicated change to the given field for application
273 * to the file system's superblock when the transaction commits.
274 * For now, just store the change in the transaction structure.
275 *
276 * Mark the transaction structure to indicate that the superblock
277 * needs to be updated before committing.
278 *
279 * Because we may not be keeping track of allocated/free inodes and
280 * used filesystem blocks in the superblock, we do not mark the
281 * superblock dirty in this transaction if we modify these fields.
282 * We still need to update the transaction deltas so that they get
283 * applied to the incore superblock, but we don't want them to
284 * cause the superblock to get locked and logged if these are the
285 * only fields in the superblock that the transaction modifies.
286 */
287void
288xfs_trans_mod_sb(
289 xfs_trans_t *tp,
290 uint field,
291 int64_t delta)
292{
293 uint32_t flags = (XFS_TRANS_DIRTY|XFS_TRANS_SB_DIRTY);
294 xfs_mount_t *mp = tp->t_mountp;
295
296 switch (field) {
297 case XFS_TRANS_SB_ICOUNT:
298 tp->t_icount_delta += delta;
299 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
300 flags &= ~XFS_TRANS_SB_DIRTY;
301 break;
302 case XFS_TRANS_SB_IFREE:
303 tp->t_ifree_delta += delta;
304 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
305 flags &= ~XFS_TRANS_SB_DIRTY;
306 break;
307 case XFS_TRANS_SB_FDBLOCKS:
308 /*
309 * Track the number of blocks allocated in the
310 * transaction. Make sure it does not exceed the
311 * number reserved.
312 */
313 if (delta < 0) {
314 tp->t_blk_res_used += (uint)-delta;
315 ASSERT(tp->t_blk_res_used <= tp->t_blk_res);
316 }
317 tp->t_fdblocks_delta += delta;
318 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
319 flags &= ~XFS_TRANS_SB_DIRTY;
320 break;
321 case XFS_TRANS_SB_RES_FDBLOCKS:
322 /*
323 * The allocation has already been applied to the
324 * in-core superblock's counter. This should only
325 * be applied to the on-disk superblock.
326 */
327 ASSERT(delta < 0);
328 tp->t_res_fdblocks_delta += delta;
329 if (xfs_sb_version_haslazysbcount(&mp->m_sb))
330 flags &= ~XFS_TRANS_SB_DIRTY;
331 break;
332 case XFS_TRANS_SB_FREXTENTS:
333 /*
334 * Track the number of blocks allocated in the
335 * transaction. Make sure it does not exceed the
336 * number reserved.
337 */
338 if (delta < 0) {
339 tp->t_rtx_res_used += (uint)-delta;
340 ASSERT(tp->t_rtx_res_used <= tp->t_rtx_res);
341 }
342 tp->t_frextents_delta += delta;
343 break;
344 case XFS_TRANS_SB_RES_FREXTENTS:
345 /*
346 * The allocation has already been applied to the
347 * in-core superblock's counter. This should only
348 * be applied to the on-disk superblock.
349 */
350 ASSERT(delta < 0);
351 tp->t_res_frextents_delta += delta;
352 break;
353 case XFS_TRANS_SB_DBLOCKS:
354 ASSERT(delta > 0);
355 tp->t_dblocks_delta += delta;
356 break;
357 case XFS_TRANS_SB_AGCOUNT:
358 ASSERT(delta > 0);
359 tp->t_agcount_delta += delta;
360 break;
361 case XFS_TRANS_SB_IMAXPCT:
362 tp->t_imaxpct_delta += delta;
363 break;
364 case XFS_TRANS_SB_REXTSIZE:
365 tp->t_rextsize_delta += delta;
366 break;
367 case XFS_TRANS_SB_RBMBLOCKS:
368 tp->t_rbmblocks_delta += delta;
369 break;
370 case XFS_TRANS_SB_RBLOCKS:
371 tp->t_rblocks_delta += delta;
372 break;
373 case XFS_TRANS_SB_REXTENTS:
374 tp->t_rextents_delta += delta;
375 break;
376 case XFS_TRANS_SB_REXTSLOG:
377 tp->t_rextslog_delta += delta;
378 break;
379 default:
380 ASSERT(0);
381 return;
382 }
383
384 tp->t_flags |= flags;
385}
386
387/*
388 * xfs_trans_apply_sb_deltas() is called from the commit code
389 * to bring the superblock buffer into the current transaction
390 * and modify it as requested by earlier calls to xfs_trans_mod_sb().
391 *
392 * For now we just look at each field allowed to change and change
393 * it if necessary.
394 */
395STATIC void
396xfs_trans_apply_sb_deltas(
397 xfs_trans_t *tp)
398{
399 xfs_dsb_t *sbp;
400 xfs_buf_t *bp;
401 int whole = 0;
402
403 bp = xfs_trans_getsb(tp, tp->t_mountp, 0);
404 sbp = XFS_BUF_TO_SBP(bp);
405
406 /*
407 * Check that superblock mods match the mods made to AGF counters.
408 */
409 ASSERT((tp->t_fdblocks_delta + tp->t_res_fdblocks_delta) ==
410 (tp->t_ag_freeblks_delta + tp->t_ag_flist_delta +
411 tp->t_ag_btree_delta));
412
413 /*
414 * Only update the superblock counters if we are logging them
415 */
416 if (!xfs_sb_version_haslazysbcount(&(tp->t_mountp->m_sb))) {
417 if (tp->t_icount_delta)
418 be64_add_cpu(&sbp->sb_icount, tp->t_icount_delta);
419 if (tp->t_ifree_delta)
420 be64_add_cpu(&sbp->sb_ifree, tp->t_ifree_delta);
421 if (tp->t_fdblocks_delta)
422 be64_add_cpu(&sbp->sb_fdblocks, tp->t_fdblocks_delta);
423 if (tp->t_res_fdblocks_delta)
424 be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
425 }
426
427 if (tp->t_frextents_delta)
428 be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
429 if (tp->t_res_frextents_delta)
430 be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
431
432 if (tp->t_dblocks_delta) {
433 be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
434 whole = 1;
435 }
436 if (tp->t_agcount_delta) {
437 be32_add_cpu(&sbp->sb_agcount, tp->t_agcount_delta);
438 whole = 1;
439 }
440 if (tp->t_imaxpct_delta) {
441 sbp->sb_imax_pct += tp->t_imaxpct_delta;
442 whole = 1;
443 }
444 if (tp->t_rextsize_delta) {
445 be32_add_cpu(&sbp->sb_rextsize, tp->t_rextsize_delta);
446 whole = 1;
447 }
448 if (tp->t_rbmblocks_delta) {
449 be32_add_cpu(&sbp->sb_rbmblocks, tp->t_rbmblocks_delta);
450 whole = 1;
451 }
452 if (tp->t_rblocks_delta) {
453 be64_add_cpu(&sbp->sb_rblocks, tp->t_rblocks_delta);
454 whole = 1;
455 }
456 if (tp->t_rextents_delta) {
457 be64_add_cpu(&sbp->sb_rextents, tp->t_rextents_delta);
458 whole = 1;
459 }
460 if (tp->t_rextslog_delta) {
461 sbp->sb_rextslog += tp->t_rextslog_delta;
462 whole = 1;
463 }
464
465 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF);
466 if (whole)
467 /*
468 * Log the whole thing, the fields are noncontiguous.
469 */
470 xfs_trans_log_buf(tp, bp, 0, sizeof(xfs_dsb_t) - 1);
471 else
472 /*
473 * Since all the modifiable fields are contiguous, we
474 * can get away with this.
475 */
476 xfs_trans_log_buf(tp, bp, offsetof(xfs_dsb_t, sb_icount),
477 offsetof(xfs_dsb_t, sb_frextents) +
478 sizeof(sbp->sb_frextents) - 1);
479}
480
481STATIC int
482xfs_sb_mod8(
483 uint8_t *field,
484 int8_t delta)
485{
486 int8_t counter = *field;
487
488 counter += delta;
489 if (counter < 0) {
490 ASSERT(0);
491 return -EINVAL;
492 }
493 *field = counter;
494 return 0;
495}
496
497STATIC int
498xfs_sb_mod32(
499 uint32_t *field,
500 int32_t delta)
501{
502 int32_t counter = *field;
503
504 counter += delta;
505 if (counter < 0) {
506 ASSERT(0);
507 return -EINVAL;
508 }
509 *field = counter;
510 return 0;
511}
512
513STATIC int
514xfs_sb_mod64(
515 uint64_t *field,
516 int64_t delta)
517{
518 int64_t counter = *field;
519
520 counter += delta;
521 if (counter < 0) {
522 ASSERT(0);
523 return -EINVAL;
524 }
525 *field = counter;
526 return 0;
527}
528
529/*
530 * xfs_trans_unreserve_and_mod_sb() is called to release unused reservations
531 * and apply superblock counter changes to the in-core superblock. The
532 * t_res_fdblocks_delta and t_res_frextents_delta fields are explicitly NOT
533 * applied to the in-core superblock. The idea is that that has already been
534 * done.
535 *
536 * If we are not logging superblock counters, then the inode allocated/free and
537 * used block counts are not updated in the on disk superblock. In this case,
538 * XFS_TRANS_SB_DIRTY will not be set when the transaction is updated but we
539 * still need to update the incore superblock with the changes.
540 */
541void
542xfs_trans_unreserve_and_mod_sb(
543 struct xfs_trans *tp)
544{
545 struct xfs_mount *mp = tp->t_mountp;
546 bool rsvd = (tp->t_flags & XFS_TRANS_RESERVE) != 0;
547 int64_t blkdelta = 0;
548 int64_t rtxdelta = 0;
549 int64_t idelta = 0;
550 int64_t ifreedelta = 0;
551 int error;
552
553 /* calculate deltas */
554 if (tp->t_blk_res > 0)
555 blkdelta = tp->t_blk_res;
556 if ((tp->t_fdblocks_delta != 0) &&
557 (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
558 (tp->t_flags & XFS_TRANS_SB_DIRTY)))
559 blkdelta += tp->t_fdblocks_delta;
560
561 if (tp->t_rtx_res > 0)
562 rtxdelta = tp->t_rtx_res;
563 if ((tp->t_frextents_delta != 0) &&
564 (tp->t_flags & XFS_TRANS_SB_DIRTY))
565 rtxdelta += tp->t_frextents_delta;
566
567 if (xfs_sb_version_haslazysbcount(&mp->m_sb) ||
568 (tp->t_flags & XFS_TRANS_SB_DIRTY)) {
569 idelta = tp->t_icount_delta;
570 ifreedelta = tp->t_ifree_delta;
571 }
572
573 /* apply the per-cpu counters */
574 if (blkdelta) {
575 error = xfs_mod_fdblocks(mp, blkdelta, rsvd);
576 if (error)
577 goto out;
578 }
579
580 if (idelta) {
581 error = xfs_mod_icount(mp, idelta);
582 if (error)
583 goto out_undo_fdblocks;
584 }
585
586 if (ifreedelta) {
587 error = xfs_mod_ifree(mp, ifreedelta);
588 if (error)
589 goto out_undo_icount;
590 }
591
592 if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
593 return;
594
595 /* apply remaining deltas */
596 spin_lock(&mp->m_sb_lock);
597 if (rtxdelta) {
598 error = xfs_sb_mod64(&mp->m_sb.sb_frextents, rtxdelta);
599 if (error)
600 goto out_undo_ifree;
601 }
602
603 if (tp->t_dblocks_delta != 0) {
604 error = xfs_sb_mod64(&mp->m_sb.sb_dblocks, tp->t_dblocks_delta);
605 if (error)
606 goto out_undo_frextents;
607 }
608 if (tp->t_agcount_delta != 0) {
609 error = xfs_sb_mod32(&mp->m_sb.sb_agcount, tp->t_agcount_delta);
610 if (error)
611 goto out_undo_dblocks;
612 }
613 if (tp->t_imaxpct_delta != 0) {
614 error = xfs_sb_mod8(&mp->m_sb.sb_imax_pct, tp->t_imaxpct_delta);
615 if (error)
616 goto out_undo_agcount;
617 }
618 if (tp->t_rextsize_delta != 0) {
619 error = xfs_sb_mod32(&mp->m_sb.sb_rextsize,
620 tp->t_rextsize_delta);
621 if (error)
622 goto out_undo_imaxpct;
623 }
624 if (tp->t_rbmblocks_delta != 0) {
625 error = xfs_sb_mod32(&mp->m_sb.sb_rbmblocks,
626 tp->t_rbmblocks_delta);
627 if (error)
628 goto out_undo_rextsize;
629 }
630 if (tp->t_rblocks_delta != 0) {
631 error = xfs_sb_mod64(&mp->m_sb.sb_rblocks, tp->t_rblocks_delta);
632 if (error)
633 goto out_undo_rbmblocks;
634 }
635 if (tp->t_rextents_delta != 0) {
636 error = xfs_sb_mod64(&mp->m_sb.sb_rextents,
637 tp->t_rextents_delta);
638 if (error)
639 goto out_undo_rblocks;
640 }
641 if (tp->t_rextslog_delta != 0) {
642 error = xfs_sb_mod8(&mp->m_sb.sb_rextslog,
643 tp->t_rextslog_delta);
644 if (error)
645 goto out_undo_rextents;
646 }
647 spin_unlock(&mp->m_sb_lock);
648 return;
649
650out_undo_rextents:
651 if (tp->t_rextents_delta)
652 xfs_sb_mod64(&mp->m_sb.sb_rextents, -tp->t_rextents_delta);
653out_undo_rblocks:
654 if (tp->t_rblocks_delta)
655 xfs_sb_mod64(&mp->m_sb.sb_rblocks, -tp->t_rblocks_delta);
656out_undo_rbmblocks:
657 if (tp->t_rbmblocks_delta)
658 xfs_sb_mod32(&mp->m_sb.sb_rbmblocks, -tp->t_rbmblocks_delta);
659out_undo_rextsize:
660 if (tp->t_rextsize_delta)
661 xfs_sb_mod32(&mp->m_sb.sb_rextsize, -tp->t_rextsize_delta);
662out_undo_imaxpct:
663 if (tp->t_rextsize_delta)
664 xfs_sb_mod8(&mp->m_sb.sb_imax_pct, -tp->t_imaxpct_delta);
665out_undo_agcount:
666 if (tp->t_agcount_delta)
667 xfs_sb_mod32(&mp->m_sb.sb_agcount, -tp->t_agcount_delta);
668out_undo_dblocks:
669 if (tp->t_dblocks_delta)
670 xfs_sb_mod64(&mp->m_sb.sb_dblocks, -tp->t_dblocks_delta);
671out_undo_frextents:
672 if (rtxdelta)
673 xfs_sb_mod64(&mp->m_sb.sb_frextents, -rtxdelta);
674out_undo_ifree:
675 spin_unlock(&mp->m_sb_lock);
676 if (ifreedelta)
677 xfs_mod_ifree(mp, -ifreedelta);
678out_undo_icount:
679 if (idelta)
680 xfs_mod_icount(mp, -idelta);
681out_undo_fdblocks:
682 if (blkdelta)
683 xfs_mod_fdblocks(mp, -blkdelta, rsvd);
684out:
685 ASSERT(error == 0);
686 return;
687}
688
689/*
690 * Add the given log item to the transaction's list of log items.
691 *
692 * The log item will now point to its new descriptor with its li_desc field.
693 */
694void
695xfs_trans_add_item(
696 struct xfs_trans *tp,
697 struct xfs_log_item *lip)
698{
699 struct xfs_log_item_desc *lidp;
700
701 ASSERT(lip->li_mountp == tp->t_mountp);
702 ASSERT(lip->li_ailp == tp->t_mountp->m_ail);
703
704 lidp = kmem_zone_zalloc(xfs_log_item_desc_zone, KM_SLEEP | KM_NOFS);
705
706 lidp->lid_item = lip;
707 lidp->lid_flags = 0;
708 list_add_tail(&lidp->lid_trans, &tp->t_items);
709
710 lip->li_desc = lidp;
711}
712
713STATIC void
714xfs_trans_free_item_desc(
715 struct xfs_log_item_desc *lidp)
716{
717 list_del_init(&lidp->lid_trans);
718 kmem_zone_free(xfs_log_item_desc_zone, lidp);
719}
720
721/*
722 * Unlink and free the given descriptor.
723 */
724void
725xfs_trans_del_item(
726 struct xfs_log_item *lip)
727{
728 xfs_trans_free_item_desc(lip->li_desc);
729 lip->li_desc = NULL;
730}
731
732/*
733 * Unlock all of the items of a transaction and free all the descriptors
734 * of that transaction.
735 */
736void
737xfs_trans_free_items(
738 struct xfs_trans *tp,
739 xfs_lsn_t commit_lsn,
740 bool abort)
741{
742 struct xfs_log_item_desc *lidp, *next;
743
744 list_for_each_entry_safe(lidp, next, &tp->t_items, lid_trans) {
745 struct xfs_log_item *lip = lidp->lid_item;
746
747 lip->li_desc = NULL;
748
749 if (commit_lsn != NULLCOMMITLSN)
750 lip->li_ops->iop_committing(lip, commit_lsn);
751 if (abort)
752 lip->li_flags |= XFS_LI_ABORTED;
753 lip->li_ops->iop_unlock(lip);
754
755 xfs_trans_free_item_desc(lidp);
756 }
757}
758
759static inline void
760xfs_log_item_batch_insert(
761 struct xfs_ail *ailp,
762 struct xfs_ail_cursor *cur,
763 struct xfs_log_item **log_items,
764 int nr_items,
765 xfs_lsn_t commit_lsn)
766{
767 int i;
768
769 spin_lock(&ailp->xa_lock);
770 /* xfs_trans_ail_update_bulk drops ailp->xa_lock */
771 xfs_trans_ail_update_bulk(ailp, cur, log_items, nr_items, commit_lsn);
772
773 for (i = 0; i < nr_items; i++) {
774 struct xfs_log_item *lip = log_items[i];
775
776 lip->li_ops->iop_unpin(lip, 0);
777 }
778}
779
780/*
781 * Bulk operation version of xfs_trans_committed that takes a log vector of
782 * items to insert into the AIL. This uses bulk AIL insertion techniques to
783 * minimise lock traffic.
784 *
785 * If we are called with the aborted flag set, it is because a log write during
786 * a CIL checkpoint commit has failed. In this case, all the items in the
787 * checkpoint have already gone through iop_commited and iop_unlock, which
788 * means that checkpoint commit abort handling is treated exactly the same
789 * as an iclog write error even though we haven't started any IO yet. Hence in
790 * this case all we need to do is iop_committed processing, followed by an
791 * iop_unpin(aborted) call.
792 *
793 * The AIL cursor is used to optimise the insert process. If commit_lsn is not
794 * at the end of the AIL, the insert cursor avoids the need to walk
795 * the AIL to find the insertion point on every xfs_log_item_batch_insert()
796 * call. This saves a lot of needless list walking and is a net win, even
797 * though it slightly increases that amount of AIL lock traffic to set it up
798 * and tear it down.
799 */
800void
801xfs_trans_committed_bulk(
802 struct xfs_ail *ailp,
803 struct xfs_log_vec *log_vector,
804 xfs_lsn_t commit_lsn,
805 int aborted)
806{
807#define LOG_ITEM_BATCH_SIZE 32
808 struct xfs_log_item *log_items[LOG_ITEM_BATCH_SIZE];
809 struct xfs_log_vec *lv;
810 struct xfs_ail_cursor cur;
811 int i = 0;
812
813 spin_lock(&ailp->xa_lock);
814 xfs_trans_ail_cursor_last(ailp, &cur, commit_lsn);
815 spin_unlock(&ailp->xa_lock);
816
817 /* unpin all the log items */
818 for (lv = log_vector; lv; lv = lv->lv_next ) {
819 struct xfs_log_item *lip = lv->lv_item;
820 xfs_lsn_t item_lsn;
821
822 if (aborted)
823 lip->li_flags |= XFS_LI_ABORTED;
824 item_lsn = lip->li_ops->iop_committed(lip, commit_lsn);
825
826 /* item_lsn of -1 means the item needs no further processing */
827 if (XFS_LSN_CMP(item_lsn, (xfs_lsn_t)-1) == 0)
828 continue;
829
830 /*
831 * if we are aborting the operation, no point in inserting the
832 * object into the AIL as we are in a shutdown situation.
833 */
834 if (aborted) {
835 ASSERT(XFS_FORCED_SHUTDOWN(ailp->xa_mount));
836 lip->li_ops->iop_unpin(lip, 1);
837 continue;
838 }
839
840 if (item_lsn != commit_lsn) {
841
842 /*
843 * Not a bulk update option due to unusual item_lsn.
844 * Push into AIL immediately, rechecking the lsn once
845 * we have the ail lock. Then unpin the item. This does
846 * not affect the AIL cursor the bulk insert path is
847 * using.
848 */
849 spin_lock(&ailp->xa_lock);
850 if (XFS_LSN_CMP(item_lsn, lip->li_lsn) > 0)
851 xfs_trans_ail_update(ailp, lip, item_lsn);
852 else
853 spin_unlock(&ailp->xa_lock);
854 lip->li_ops->iop_unpin(lip, 0);
855 continue;
856 }
857
858 /* Item is a candidate for bulk AIL insert. */
859 log_items[i++] = lv->lv_item;
860 if (i >= LOG_ITEM_BATCH_SIZE) {
861 xfs_log_item_batch_insert(ailp, &cur, log_items,
862 LOG_ITEM_BATCH_SIZE, commit_lsn);
863 i = 0;
864 }
865 }
866
867 /* make sure we insert the remainder! */
868 if (i)
869 xfs_log_item_batch_insert(ailp, &cur, log_items, i, commit_lsn);
870
871 spin_lock(&ailp->xa_lock);
872 xfs_trans_ail_cursor_done(&cur);
873 spin_unlock(&ailp->xa_lock);
874}
875
876/*
877 * Commit the given transaction to the log.
878 *
879 * XFS disk error handling mechanism is not based on a typical
880 * transaction abort mechanism. Logically after the filesystem
881 * gets marked 'SHUTDOWN', we can't let any new transactions
882 * be durable - ie. committed to disk - because some metadata might
883 * be inconsistent. In such cases, this returns an error, and the
884 * caller may assume that all locked objects joined to the transaction
885 * have already been unlocked as if the commit had succeeded.
886 * Do not reference the transaction structure after this call.
887 */
888static int
889__xfs_trans_commit(
890 struct xfs_trans *tp,
891 bool regrant)
892{
893 struct xfs_mount *mp = tp->t_mountp;
894 xfs_lsn_t commit_lsn = -1;
895 int error = 0;
896 int sync = tp->t_flags & XFS_TRANS_SYNC;
897
898 /*
899 * If there is nothing to be logged by the transaction,
900 * then unlock all of the items associated with the
901 * transaction and free the transaction structure.
902 * Also make sure to return any reserved blocks to
903 * the free pool.
904 */
905 if (!(tp->t_flags & XFS_TRANS_DIRTY))
906 goto out_unreserve;
907
908 if (XFS_FORCED_SHUTDOWN(mp)) {
909 error = -EIO;
910 goto out_unreserve;
911 }
912
913 ASSERT(tp->t_ticket != NULL);
914
915 /*
916 * If we need to update the superblock, then do it now.
917 */
918 if (tp->t_flags & XFS_TRANS_SB_DIRTY)
919 xfs_trans_apply_sb_deltas(tp);
920 xfs_trans_apply_dquot_deltas(tp);
921
922 xfs_log_commit_cil(mp, tp, &commit_lsn, regrant);
923
924 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
925 xfs_trans_free(tp);
926
927 /*
928 * If the transaction needs to be synchronous, then force the
929 * log out now and wait for it.
930 */
931 if (sync) {
932 error = _xfs_log_force_lsn(mp, commit_lsn, XFS_LOG_SYNC, NULL);
933 XFS_STATS_INC(mp, xs_trans_sync);
934 } else {
935 XFS_STATS_INC(mp, xs_trans_async);
936 }
937
938 return error;
939
940out_unreserve:
941 xfs_trans_unreserve_and_mod_sb(tp);
942
943 /*
944 * It is indeed possible for the transaction to be not dirty but
945 * the dqinfo portion to be. All that means is that we have some
946 * (non-persistent) quota reservations that need to be unreserved.
947 */
948 xfs_trans_unreserve_and_mod_dquots(tp);
949 if (tp->t_ticket) {
950 commit_lsn = xfs_log_done(mp, tp->t_ticket, NULL, regrant);
951 if (commit_lsn == -1 && !error)
952 error = -EIO;
953 }
954 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
955 xfs_trans_free_items(tp, NULLCOMMITLSN, !!error);
956 xfs_trans_free(tp);
957
958 XFS_STATS_INC(mp, xs_trans_empty);
959 return error;
960}
961
962int
963xfs_trans_commit(
964 struct xfs_trans *tp)
965{
966 return __xfs_trans_commit(tp, false);
967}
968
969/*
970 * Unlock all of the transaction's items and free the transaction.
971 * The transaction must not have modified any of its items, because
972 * there is no way to restore them to their previous state.
973 *
974 * If the transaction has made a log reservation, make sure to release
975 * it as well.
976 */
977void
978xfs_trans_cancel(
979 struct xfs_trans *tp)
980{
981 struct xfs_mount *mp = tp->t_mountp;
982 bool dirty = (tp->t_flags & XFS_TRANS_DIRTY);
983
984 /*
985 * See if the caller is relying on us to shut down the
986 * filesystem. This happens in paths where we detect
987 * corruption and decide to give up.
988 */
989 if (dirty && !XFS_FORCED_SHUTDOWN(mp)) {
990 XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
991 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
992 }
993#ifdef DEBUG
994 if (!dirty && !XFS_FORCED_SHUTDOWN(mp)) {
995 struct xfs_log_item_desc *lidp;
996
997 list_for_each_entry(lidp, &tp->t_items, lid_trans)
998 ASSERT(!(lidp->lid_item->li_type == XFS_LI_EFD));
999 }
1000#endif
1001 xfs_trans_unreserve_and_mod_sb(tp);
1002 xfs_trans_unreserve_and_mod_dquots(tp);
1003
1004 if (tp->t_ticket)
1005 xfs_log_done(mp, tp->t_ticket, NULL, false);
1006
1007 /* mark this thread as no longer being in a transaction */
1008 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
1009
1010 xfs_trans_free_items(tp, NULLCOMMITLSN, dirty);
1011 xfs_trans_free(tp);
1012}
1013
1014/*
1015 * Roll from one trans in the sequence of PERMANENT transactions to
1016 * the next: permanent transactions are only flushed out when
1017 * committed with xfs_trans_commit(), but we still want as soon
1018 * as possible to let chunks of it go to the log. So we commit the
1019 * chunk we've been working on and get a new transaction to continue.
1020 */
1021int
1022__xfs_trans_roll(
1023 struct xfs_trans **tpp,
1024 struct xfs_inode *dp,
1025 int *committed)
1026{
1027 struct xfs_trans *trans;
1028 struct xfs_trans_res tres;
1029 int error;
1030
1031 *committed = 0;
1032
1033 /*
1034 * Ensure that the inode is always logged.
1035 */
1036 trans = *tpp;
1037 if (dp)
1038 xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
1039
1040 /*
1041 * Copy the critical parameters from one trans to the next.
1042 */
1043 tres.tr_logres = trans->t_log_res;
1044 tres.tr_logcount = trans->t_log_count;
1045 *tpp = xfs_trans_dup(trans);
1046
1047 /*
1048 * Commit the current transaction.
1049 * If this commit failed, then it'd just unlock those items that
1050 * are not marked ihold. That also means that a filesystem shutdown
1051 * is in progress. The caller takes the responsibility to cancel
1052 * the duplicate transaction that gets returned.
1053 */
1054 error = __xfs_trans_commit(trans, true);
1055 if (error)
1056 return error;
1057
1058 *committed = 1;
1059 trans = *tpp;
1060
1061 /*
1062 * Reserve space in the log for th next transaction.
1063 * This also pushes items in the "AIL", the list of logged items,
1064 * out to disk if they are taking up space at the tail of the log
1065 * that we want to use. This requires that either nothing be locked
1066 * across this call, or that anything that is locked be logged in
1067 * the prior and the next transactions.
1068 */
1069 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
1070 error = xfs_trans_reserve(trans, &tres, 0, 0);
1071 /*
1072 * Ensure that the inode is in the new transaction and locked.
1073 */
1074 if (error)
1075 return error;
1076
1077 if (dp)
1078 xfs_trans_ijoin(trans, dp, 0);
1079 return 0;
1080}
1081
1082int
1083xfs_trans_roll(
1084 struct xfs_trans **tpp,
1085 struct xfs_inode *dp)
1086{
1087 int committed;
1088 return __xfs_trans_roll(tpp, dp, &committed);
1089}