1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <darrick.wong@oracle.com>
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
12 #include "xfs_shared.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_inode.h"
16 #include "xfs_trans.h"
17 #include "xfs_trans_priv.h"
18 #include "xfs_bmap_item.h"
21 #include "xfs_icache.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_trans_space.h"
24 #include "xfs_error.h"
25 #include "xfs_log_priv.h"
26 #include "xfs_log_recover.h"
27 #include "xfs_quota.h"
29 kmem_zone_t *xfs_bui_zone;
30 kmem_zone_t *xfs_bud_zone;
32 static const struct xfs_item_ops xfs_bui_item_ops;
34 static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
36 return container_of(lip, struct xfs_bui_log_item, bui_item);
41 struct xfs_bui_log_item *buip)
43 kmem_cache_free(xfs_bui_zone, buip);
47 * Freeing the BUI requires that we remove it from the AIL if it has already
48 * been placed there. However, the BUI may not yet have been placed in the AIL
49 * when called by xfs_bui_release() from BUD processing due to the ordering of
50 * committed vs unpin operations in bulk insert operations. Hence the reference
51 * count to ensure only the last caller frees the BUI.
55 struct xfs_bui_log_item *buip)
57 ASSERT(atomic_read(&buip->bui_refcount) > 0);
58 if (atomic_dec_and_test(&buip->bui_refcount)) {
59 xfs_trans_ail_delete(&buip->bui_item, SHUTDOWN_LOG_IO_ERROR);
60 xfs_bui_item_free(buip);
67 struct xfs_log_item *lip,
71 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
74 *nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
78 * This is called to fill in the vector of log iovecs for the
79 * given bui log item. We use only 1 iovec, and we point that
80 * at the bui_log_format structure embedded in the bui item.
81 * It is at this point that we assert that all of the extent
82 * slots in the bui item have been filled.
86 struct xfs_log_item *lip,
87 struct xfs_log_vec *lv)
89 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
90 struct xfs_log_iovec *vecp = NULL;
92 ASSERT(atomic_read(&buip->bui_next_extent) ==
93 buip->bui_format.bui_nextents);
95 buip->bui_format.bui_type = XFS_LI_BUI;
96 buip->bui_format.bui_size = 1;
98 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
99 xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
103 * The unpin operation is the last place an BUI is manipulated in the log. It is
104 * either inserted in the AIL or aborted in the event of a log I/O error. In
105 * either case, the BUI transaction has been successfully committed to make it
106 * this far. Therefore, we expect whoever committed the BUI to either construct
107 * and commit the BUD or drop the BUD's reference in the event of error. Simply
108 * drop the log's BUI reference now that the log is done with it.
112 struct xfs_log_item *lip,
115 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
117 xfs_bui_release(buip);
121 * The BUI has been either committed or aborted if the transaction has been
122 * cancelled. If the transaction was cancelled, an BUD isn't going to be
123 * constructed and thus we free the BUI here directly.
126 xfs_bui_item_release(
127 struct xfs_log_item *lip)
129 xfs_bui_release(BUI_ITEM(lip));
133 * Allocate and initialize an bui item with the given number of extents.
135 STATIC struct xfs_bui_log_item *
137 struct xfs_mount *mp)
140 struct xfs_bui_log_item *buip;
142 buip = kmem_cache_zalloc(xfs_bui_zone, GFP_KERNEL | __GFP_NOFAIL);
144 xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
145 buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
146 buip->bui_format.bui_id = (uintptr_t)(void *)buip;
147 atomic_set(&buip->bui_next_extent, 0);
148 atomic_set(&buip->bui_refcount, 2);
153 static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
155 return container_of(lip, struct xfs_bud_log_item, bud_item);
160 struct xfs_log_item *lip,
165 *nbytes += sizeof(struct xfs_bud_log_format);
169 * This is called to fill in the vector of log iovecs for the
170 * given bud log item. We use only 1 iovec, and we point that
171 * at the bud_log_format structure embedded in the bud item.
172 * It is at this point that we assert that all of the extent
173 * slots in the bud item have been filled.
177 struct xfs_log_item *lip,
178 struct xfs_log_vec *lv)
180 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
181 struct xfs_log_iovec *vecp = NULL;
183 budp->bud_format.bud_type = XFS_LI_BUD;
184 budp->bud_format.bud_size = 1;
186 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
187 sizeof(struct xfs_bud_log_format));
191 * The BUD is either committed or aborted if the transaction is cancelled. If
192 * the transaction is cancelled, drop our reference to the BUI and free the
196 xfs_bud_item_release(
197 struct xfs_log_item *lip)
199 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
201 xfs_bui_release(budp->bud_buip);
202 kmem_cache_free(xfs_bud_zone, budp);
205 static const struct xfs_item_ops xfs_bud_item_ops = {
206 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
207 .iop_size = xfs_bud_item_size,
208 .iop_format = xfs_bud_item_format,
209 .iop_release = xfs_bud_item_release,
212 static struct xfs_bud_log_item *
214 struct xfs_trans *tp,
215 struct xfs_bui_log_item *buip)
217 struct xfs_bud_log_item *budp;
219 budp = kmem_cache_zalloc(xfs_bud_zone, GFP_KERNEL | __GFP_NOFAIL);
220 xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
222 budp->bud_buip = buip;
223 budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
225 xfs_trans_add_item(tp, &budp->bud_item);
230 * Finish an bmap update and log it to the BUD. Note that the
231 * transaction is marked dirty regardless of whether the bmap update
232 * succeeds or fails to support the BUI/BUD lifecycle rules.
235 xfs_trans_log_finish_bmap_update(
236 struct xfs_trans *tp,
237 struct xfs_bud_log_item *budp,
238 enum xfs_bmap_intent_type type,
239 struct xfs_inode *ip,
241 xfs_fileoff_t startoff,
242 xfs_fsblock_t startblock,
243 xfs_filblks_t *blockcount,
248 error = xfs_bmap_finish_one(tp, ip, type, whichfork, startoff,
249 startblock, blockcount, state);
252 * Mark the transaction dirty, even on error. This ensures the
253 * transaction is aborted, which:
255 * 1.) releases the BUI and frees the BUD
256 * 2.) shuts down the filesystem
258 tp->t_flags |= XFS_TRANS_DIRTY;
259 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
264 /* Sort bmap intents by inode. */
266 xfs_bmap_update_diff_items(
268 const struct list_head *a,
269 const struct list_head *b)
271 struct xfs_bmap_intent *ba;
272 struct xfs_bmap_intent *bb;
274 ba = container_of(a, struct xfs_bmap_intent, bi_list);
275 bb = container_of(b, struct xfs_bmap_intent, bi_list);
276 return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
279 /* Set the map extent flags for this mapping. */
281 xfs_trans_set_bmap_flags(
282 struct xfs_map_extent *bmap,
283 enum xfs_bmap_intent_type type,
291 bmap->me_flags = type;
296 if (state == XFS_EXT_UNWRITTEN)
297 bmap->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
298 if (whichfork == XFS_ATTR_FORK)
299 bmap->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
302 /* Log bmap updates in the intent item. */
304 xfs_bmap_update_log_item(
305 struct xfs_trans *tp,
306 struct xfs_bui_log_item *buip,
307 struct xfs_bmap_intent *bmap)
310 struct xfs_map_extent *map;
312 tp->t_flags |= XFS_TRANS_DIRTY;
313 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
316 * atomic_inc_return gives us the value after the increment;
317 * we want to use it as an array index so we need to subtract 1 from
320 next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
321 ASSERT(next_extent < buip->bui_format.bui_nextents);
322 map = &buip->bui_format.bui_extents[next_extent];
323 map->me_owner = bmap->bi_owner->i_ino;
324 map->me_startblock = bmap->bi_bmap.br_startblock;
325 map->me_startoff = bmap->bi_bmap.br_startoff;
326 map->me_len = bmap->bi_bmap.br_blockcount;
327 xfs_trans_set_bmap_flags(map, bmap->bi_type, bmap->bi_whichfork,
328 bmap->bi_bmap.br_state);
331 static struct xfs_log_item *
332 xfs_bmap_update_create_intent(
333 struct xfs_trans *tp,
334 struct list_head *items,
338 struct xfs_mount *mp = tp->t_mountp;
339 struct xfs_bui_log_item *buip = xfs_bui_init(mp);
340 struct xfs_bmap_intent *bmap;
342 ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
344 xfs_trans_add_item(tp, &buip->bui_item);
346 list_sort(mp, items, xfs_bmap_update_diff_items);
347 list_for_each_entry(bmap, items, bi_list)
348 xfs_bmap_update_log_item(tp, buip, bmap);
349 return &buip->bui_item;
352 /* Get an BUD so we can process all the deferred rmap updates. */
353 static struct xfs_log_item *
354 xfs_bmap_update_create_done(
355 struct xfs_trans *tp,
356 struct xfs_log_item *intent,
359 return &xfs_trans_get_bud(tp, BUI_ITEM(intent))->bud_item;
362 /* Process a deferred rmap update. */
364 xfs_bmap_update_finish_item(
365 struct xfs_trans *tp,
366 struct xfs_log_item *done,
367 struct list_head *item,
368 struct xfs_btree_cur **state)
370 struct xfs_bmap_intent *bmap;
374 bmap = container_of(item, struct xfs_bmap_intent, bi_list);
375 count = bmap->bi_bmap.br_blockcount;
376 error = xfs_trans_log_finish_bmap_update(tp, BUD_ITEM(done),
378 bmap->bi_owner, bmap->bi_whichfork,
379 bmap->bi_bmap.br_startoff,
380 bmap->bi_bmap.br_startblock,
382 bmap->bi_bmap.br_state);
383 if (!error && count > 0) {
384 ASSERT(bmap->bi_type == XFS_BMAP_UNMAP);
385 bmap->bi_bmap.br_blockcount = count;
392 /* Abort all pending BUIs. */
394 xfs_bmap_update_abort_intent(
395 struct xfs_log_item *intent)
397 xfs_bui_release(BUI_ITEM(intent));
400 /* Cancel a deferred rmap update. */
402 xfs_bmap_update_cancel_item(
403 struct list_head *item)
405 struct xfs_bmap_intent *bmap;
407 bmap = container_of(item, struct xfs_bmap_intent, bi_list);
411 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
412 .max_items = XFS_BUI_MAX_FAST_EXTENTS,
413 .create_intent = xfs_bmap_update_create_intent,
414 .abort_intent = xfs_bmap_update_abort_intent,
415 .create_done = xfs_bmap_update_create_done,
416 .finish_item = xfs_bmap_update_finish_item,
417 .cancel_item = xfs_bmap_update_cancel_item,
420 /* Is this recovered BUI ok? */
423 struct xfs_mount *mp,
424 struct xfs_bui_log_item *buip)
426 struct xfs_map_extent *bmap;
428 /* Only one mapping operation per BUI... */
429 if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
432 bmap = &buip->bui_format.bui_extents[0];
434 if (bmap->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
437 switch (bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
445 if (!xfs_verify_ino(mp, bmap->me_owner))
448 if (!xfs_verify_fileext(mp, bmap->me_startoff, bmap->me_len))
451 return xfs_verify_fsbext(mp, bmap->me_startblock, bmap->me_len);
455 * Process a bmap update intent item that was recovered from the log.
456 * We need to update some inode's bmbt.
459 xfs_bui_item_recover(
460 struct xfs_log_item *lip,
461 struct list_head *capture_list)
463 struct xfs_bmbt_irec irec;
464 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
465 struct xfs_trans *tp;
466 struct xfs_inode *ip = NULL;
467 struct xfs_mount *mp = lip->li_mountp;
468 struct xfs_map_extent *bmap;
469 struct xfs_bud_log_item *budp;
472 unsigned int bui_type;
477 if (!xfs_bui_validate(mp, buip)) {
478 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
479 &buip->bui_format, sizeof(buip->bui_format));
480 return -EFSCORRUPTED;
483 bmap = &buip->bui_format.bui_extents[0];
484 state = (bmap->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
485 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
486 whichfork = (bmap->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
487 XFS_ATTR_FORK : XFS_DATA_FORK;
488 bui_type = bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
490 /* Grab the inode. */
491 error = xfs_iget(mp, NULL, bmap->me_owner, 0, 0, &ip);
495 error = xfs_qm_dqattach(ip);
499 if (VFS_I(ip)->i_nlink == 0)
500 xfs_iflags_set(ip, XFS_IRECOVERY);
502 /* Allocate transaction and do the work. */
503 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
504 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
508 budp = xfs_trans_get_bud(tp, buip);
509 xfs_ilock(ip, XFS_ILOCK_EXCL);
510 xfs_trans_ijoin(tp, ip, 0);
512 if (bui_type == XFS_BMAP_MAP)
513 iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
515 iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
517 error = xfs_iext_count_may_overflow(ip, whichfork, iext_delta);
521 count = bmap->me_len;
522 error = xfs_trans_log_finish_bmap_update(tp, budp, bui_type, ip,
523 whichfork, bmap->me_startoff, bmap->me_startblock,
529 ASSERT(bui_type == XFS_BMAP_UNMAP);
530 irec.br_startblock = bmap->me_startblock;
531 irec.br_blockcount = count;
532 irec.br_startoff = bmap->me_startoff;
533 irec.br_state = state;
534 xfs_bmap_unmap_extent(tp, ip, &irec);
538 * Commit transaction, which frees the transaction and saves the inode
539 * for later replay activities.
541 error = xfs_defer_ops_capture_and_commit(tp, ip, capture_list);
545 xfs_iunlock(ip, XFS_ILOCK_EXCL);
550 xfs_trans_cancel(tp);
552 xfs_iunlock(ip, XFS_ILOCK_EXCL);
560 struct xfs_log_item *lip,
563 return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
566 /* Relog an intent item to push the log tail forward. */
567 static struct xfs_log_item *
569 struct xfs_log_item *intent,
570 struct xfs_trans *tp)
572 struct xfs_bud_log_item *budp;
573 struct xfs_bui_log_item *buip;
574 struct xfs_map_extent *extp;
577 count = BUI_ITEM(intent)->bui_format.bui_nextents;
578 extp = BUI_ITEM(intent)->bui_format.bui_extents;
580 tp->t_flags |= XFS_TRANS_DIRTY;
581 budp = xfs_trans_get_bud(tp, BUI_ITEM(intent));
582 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
584 buip = xfs_bui_init(tp->t_mountp);
585 memcpy(buip->bui_format.bui_extents, extp, count * sizeof(*extp));
586 atomic_set(&buip->bui_next_extent, count);
587 xfs_trans_add_item(tp, &buip->bui_item);
588 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
589 return &buip->bui_item;
592 static const struct xfs_item_ops xfs_bui_item_ops = {
593 .iop_size = xfs_bui_item_size,
594 .iop_format = xfs_bui_item_format,
595 .iop_unpin = xfs_bui_item_unpin,
596 .iop_release = xfs_bui_item_release,
597 .iop_recover = xfs_bui_item_recover,
598 .iop_match = xfs_bui_item_match,
599 .iop_relog = xfs_bui_item_relog,
603 * Copy an BUI format buffer from the given buf, and into the destination
604 * BUI format structure. The BUI/BUD items were designed not to need any
605 * special alignment handling.
609 struct xfs_log_iovec *buf,
610 struct xfs_bui_log_format *dst_bui_fmt)
612 struct xfs_bui_log_format *src_bui_fmt;
615 src_bui_fmt = buf->i_addr;
616 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
618 if (buf->i_len == len) {
619 memcpy(dst_bui_fmt, src_bui_fmt, len);
622 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
623 return -EFSCORRUPTED;
627 * This routine is called to create an in-core extent bmap update
628 * item from the bui format structure which was logged on disk.
629 * It allocates an in-core bui, copies the extents from the format
630 * structure into it, and adds the bui to the AIL with the given
634 xlog_recover_bui_commit_pass2(
636 struct list_head *buffer_list,
637 struct xlog_recover_item *item,
641 struct xfs_mount *mp = log->l_mp;
642 struct xfs_bui_log_item *buip;
643 struct xfs_bui_log_format *bui_formatp;
645 bui_formatp = item->ri_buf[0].i_addr;
647 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
648 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
649 return -EFSCORRUPTED;
651 buip = xfs_bui_init(mp);
652 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
654 xfs_bui_item_free(buip);
657 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
659 * Insert the intent into the AIL directly and drop one reference so
660 * that finishing or canceling the work will drop the other.
662 xfs_trans_ail_insert(log->l_ailp, &buip->bui_item, lsn);
663 xfs_bui_release(buip);
667 const struct xlog_recover_item_ops xlog_bui_item_ops = {
668 .item_type = XFS_LI_BUI,
669 .commit_pass2 = xlog_recover_bui_commit_pass2,
673 * This routine is called when an BUD format structure is found in a committed
674 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
675 * was still in the log. To do this it searches the AIL for the BUI with an id
676 * equal to that in the BUD format structure. If we find it we drop the BUD
677 * reference, which removes the BUI from the AIL and frees it.
680 xlog_recover_bud_commit_pass2(
682 struct list_head *buffer_list,
683 struct xlog_recover_item *item,
686 struct xfs_bud_log_format *bud_formatp;
688 bud_formatp = item->ri_buf[0].i_addr;
689 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
690 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
691 return -EFSCORRUPTED;
694 xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
698 const struct xlog_recover_item_ops xlog_bud_item_ops = {
699 .item_type = XFS_LI_BUD,
700 .commit_pass2 = xlog_recover_bud_commit_pass2,