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"
29 struct kmem_cache *xfs_bui_cache;
30 struct kmem_cache *xfs_bud_cache;
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_free(buip->bui_item.li_lv_shadow);
44 kmem_cache_free(xfs_bui_cache, buip);
48 * Freeing the BUI requires that we remove it from the AIL if it has already
49 * been placed there. However, the BUI may not yet have been placed in the AIL
50 * when called by xfs_bui_release() from BUD processing due to the ordering of
51 * committed vs unpin operations in bulk insert operations. Hence the reference
52 * count to ensure only the last caller frees the BUI.
56 struct xfs_bui_log_item *buip)
58 ASSERT(atomic_read(&buip->bui_refcount) > 0);
59 if (!atomic_dec_and_test(&buip->bui_refcount))
62 xfs_trans_ail_delete(&buip->bui_item, 0);
63 xfs_bui_item_free(buip);
69 struct xfs_log_item *lip,
73 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
76 *nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
80 * This is called to fill in the vector of log iovecs for the
81 * given bui log item. We use only 1 iovec, and we point that
82 * at the bui_log_format structure embedded in the bui item.
83 * It is at this point that we assert that all of the extent
84 * slots in the bui item have been filled.
88 struct xfs_log_item *lip,
89 struct xfs_log_vec *lv)
91 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
92 struct xfs_log_iovec *vecp = NULL;
94 ASSERT(atomic_read(&buip->bui_next_extent) ==
95 buip->bui_format.bui_nextents);
97 buip->bui_format.bui_type = XFS_LI_BUI;
98 buip->bui_format.bui_size = 1;
100 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
101 xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
105 * The unpin operation is the last place an BUI is manipulated in the log. It is
106 * either inserted in the AIL or aborted in the event of a log I/O error. In
107 * either case, the BUI transaction has been successfully committed to make it
108 * this far. Therefore, we expect whoever committed the BUI to either construct
109 * and commit the BUD or drop the BUD's reference in the event of error. Simply
110 * drop the log's BUI reference now that the log is done with it.
114 struct xfs_log_item *lip,
117 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
119 xfs_bui_release(buip);
123 * The BUI has been either committed or aborted if the transaction has been
124 * cancelled. If the transaction was cancelled, an BUD isn't going to be
125 * constructed and thus we free the BUI here directly.
128 xfs_bui_item_release(
129 struct xfs_log_item *lip)
131 xfs_bui_release(BUI_ITEM(lip));
135 * Allocate and initialize an bui item with the given number of extents.
137 STATIC struct xfs_bui_log_item *
139 struct xfs_mount *mp)
142 struct xfs_bui_log_item *buip;
144 buip = kmem_cache_zalloc(xfs_bui_cache, GFP_KERNEL | __GFP_NOFAIL);
146 xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
147 buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
148 buip->bui_format.bui_id = (uintptr_t)(void *)buip;
149 atomic_set(&buip->bui_next_extent, 0);
150 atomic_set(&buip->bui_refcount, 2);
155 static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
157 return container_of(lip, struct xfs_bud_log_item, bud_item);
162 struct xfs_log_item *lip,
167 *nbytes += sizeof(struct xfs_bud_log_format);
171 * This is called to fill in the vector of log iovecs for the
172 * given bud log item. We use only 1 iovec, and we point that
173 * at the bud_log_format structure embedded in the bud item.
174 * It is at this point that we assert that all of the extent
175 * slots in the bud item have been filled.
179 struct xfs_log_item *lip,
180 struct xfs_log_vec *lv)
182 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
183 struct xfs_log_iovec *vecp = NULL;
185 budp->bud_format.bud_type = XFS_LI_BUD;
186 budp->bud_format.bud_size = 1;
188 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
189 sizeof(struct xfs_bud_log_format));
193 * The BUD is either committed or aborted if the transaction is cancelled. If
194 * the transaction is cancelled, drop our reference to the BUI and free the
198 xfs_bud_item_release(
199 struct xfs_log_item *lip)
201 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
203 xfs_bui_release(budp->bud_buip);
204 kmem_free(budp->bud_item.li_lv_shadow);
205 kmem_cache_free(xfs_bud_cache, budp);
208 static struct xfs_log_item *
210 struct xfs_log_item *lip)
212 return &BUD_ITEM(lip)->bud_buip->bui_item;
215 static const struct xfs_item_ops xfs_bud_item_ops = {
216 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
217 XFS_ITEM_INTENT_DONE,
218 .iop_size = xfs_bud_item_size,
219 .iop_format = xfs_bud_item_format,
220 .iop_release = xfs_bud_item_release,
221 .iop_intent = xfs_bud_item_intent,
224 static struct xfs_bud_log_item *
226 struct xfs_trans *tp,
227 struct xfs_bui_log_item *buip)
229 struct xfs_bud_log_item *budp;
231 budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
232 xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
234 budp->bud_buip = buip;
235 budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
237 xfs_trans_add_item(tp, &budp->bud_item);
242 * Finish an bmap update and log it to the BUD. Note that the
243 * transaction is marked dirty regardless of whether the bmap update
244 * succeeds or fails to support the BUI/BUD lifecycle rules.
247 xfs_trans_log_finish_bmap_update(
248 struct xfs_trans *tp,
249 struct xfs_bud_log_item *budp,
250 struct xfs_bmap_intent *bi)
254 error = xfs_bmap_finish_one(tp, bi);
257 * Mark the transaction dirty, even on error. This ensures the
258 * transaction is aborted, which:
260 * 1.) releases the BUI and frees the BUD
261 * 2.) shuts down the filesystem
263 tp->t_flags |= XFS_TRANS_DIRTY | XFS_TRANS_HAS_INTENT_DONE;
264 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
269 /* Sort bmap intents by inode. */
271 xfs_bmap_update_diff_items(
273 const struct list_head *a,
274 const struct list_head *b)
276 struct xfs_bmap_intent *ba;
277 struct xfs_bmap_intent *bb;
279 ba = container_of(a, struct xfs_bmap_intent, bi_list);
280 bb = container_of(b, struct xfs_bmap_intent, bi_list);
281 return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
284 /* Set the map extent flags for this mapping. */
286 xfs_trans_set_bmap_flags(
287 struct xfs_map_extent *map,
288 enum xfs_bmap_intent_type type,
296 map->me_flags = type;
301 if (state == XFS_EXT_UNWRITTEN)
302 map->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
303 if (whichfork == XFS_ATTR_FORK)
304 map->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
307 /* Log bmap updates in the intent item. */
309 xfs_bmap_update_log_item(
310 struct xfs_trans *tp,
311 struct xfs_bui_log_item *buip,
312 struct xfs_bmap_intent *bi)
315 struct xfs_map_extent *map;
317 tp->t_flags |= XFS_TRANS_DIRTY;
318 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
321 * atomic_inc_return gives us the value after the increment;
322 * we want to use it as an array index so we need to subtract 1 from
325 next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
326 ASSERT(next_extent < buip->bui_format.bui_nextents);
327 map = &buip->bui_format.bui_extents[next_extent];
328 map->me_owner = bi->bi_owner->i_ino;
329 map->me_startblock = bi->bi_bmap.br_startblock;
330 map->me_startoff = bi->bi_bmap.br_startoff;
331 map->me_len = bi->bi_bmap.br_blockcount;
332 xfs_trans_set_bmap_flags(map, bi->bi_type, bi->bi_whichfork,
333 bi->bi_bmap.br_state);
336 static struct xfs_log_item *
337 xfs_bmap_update_create_intent(
338 struct xfs_trans *tp,
339 struct list_head *items,
343 struct xfs_mount *mp = tp->t_mountp;
344 struct xfs_bui_log_item *buip = xfs_bui_init(mp);
345 struct xfs_bmap_intent *bi;
347 ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
349 xfs_trans_add_item(tp, &buip->bui_item);
351 list_sort(mp, items, xfs_bmap_update_diff_items);
352 list_for_each_entry(bi, items, bi_list)
353 xfs_bmap_update_log_item(tp, buip, bi);
354 return &buip->bui_item;
357 /* Get an BUD so we can process all the deferred rmap updates. */
358 static struct xfs_log_item *
359 xfs_bmap_update_create_done(
360 struct xfs_trans *tp,
361 struct xfs_log_item *intent,
364 return &xfs_trans_get_bud(tp, BUI_ITEM(intent))->bud_item;
367 /* Take a passive ref to the AG containing the space we're mapping. */
369 xfs_bmap_update_get_group(
370 struct xfs_mount *mp,
371 struct xfs_bmap_intent *bi)
375 agno = XFS_FSB_TO_AGNO(mp, bi->bi_bmap.br_startblock);
378 * Bump the intent count on behalf of the deferred rmap and refcount
379 * intent items that that we can queue when we finish this bmap work.
380 * This new intent item will bump the intent count before the bmap
381 * intent drops the intent count, ensuring that the intent count
382 * remains nonzero across the transaction roll.
384 bi->bi_pag = xfs_perag_intent_get(mp, agno);
387 /* Release a passive AG ref after finishing mapping work. */
389 xfs_bmap_update_put_group(
390 struct xfs_bmap_intent *bi)
392 xfs_perag_intent_put(bi->bi_pag);
395 /* Process a deferred rmap update. */
397 xfs_bmap_update_finish_item(
398 struct xfs_trans *tp,
399 struct xfs_log_item *done,
400 struct list_head *item,
401 struct xfs_btree_cur **state)
403 struct xfs_bmap_intent *bi;
406 bi = container_of(item, struct xfs_bmap_intent, bi_list);
408 error = xfs_trans_log_finish_bmap_update(tp, BUD_ITEM(done), bi);
409 if (!error && bi->bi_bmap.br_blockcount > 0) {
410 ASSERT(bi->bi_type == XFS_BMAP_UNMAP);
414 xfs_bmap_update_put_group(bi);
415 kmem_cache_free(xfs_bmap_intent_cache, bi);
419 /* Abort all pending BUIs. */
421 xfs_bmap_update_abort_intent(
422 struct xfs_log_item *intent)
424 xfs_bui_release(BUI_ITEM(intent));
427 /* Cancel a deferred bmap update. */
429 xfs_bmap_update_cancel_item(
430 struct list_head *item)
432 struct xfs_bmap_intent *bi;
434 bi = container_of(item, struct xfs_bmap_intent, bi_list);
436 xfs_bmap_update_put_group(bi);
437 kmem_cache_free(xfs_bmap_intent_cache, bi);
440 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
441 .max_items = XFS_BUI_MAX_FAST_EXTENTS,
442 .create_intent = xfs_bmap_update_create_intent,
443 .abort_intent = xfs_bmap_update_abort_intent,
444 .create_done = xfs_bmap_update_create_done,
445 .finish_item = xfs_bmap_update_finish_item,
446 .cancel_item = xfs_bmap_update_cancel_item,
449 /* Is this recovered BUI ok? */
452 struct xfs_mount *mp,
453 struct xfs_bui_log_item *buip)
455 struct xfs_map_extent *map;
457 /* Only one mapping operation per BUI... */
458 if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
461 map = &buip->bui_format.bui_extents[0];
463 if (map->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
466 switch (map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
474 if (!xfs_verify_ino(mp, map->me_owner))
477 if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
480 return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
484 * Process a bmap update intent item that was recovered from the log.
485 * We need to update some inode's bmbt.
488 xfs_bui_item_recover(
489 struct xfs_log_item *lip,
490 struct list_head *capture_list)
492 struct xfs_bmap_intent fake = { };
493 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
494 struct xfs_trans *tp;
495 struct xfs_inode *ip = NULL;
496 struct xfs_mount *mp = lip->li_log->l_mp;
497 struct xfs_map_extent *map;
498 struct xfs_bud_log_item *budp;
502 if (!xfs_bui_validate(mp, buip)) {
503 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
504 &buip->bui_format, sizeof(buip->bui_format));
505 return -EFSCORRUPTED;
508 map = &buip->bui_format.bui_extents[0];
509 fake.bi_whichfork = (map->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
510 XFS_ATTR_FORK : XFS_DATA_FORK;
511 fake.bi_type = map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
513 error = xlog_recover_iget(mp, map->me_owner, &ip);
517 /* Allocate transaction and do the work. */
518 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
519 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
523 budp = xfs_trans_get_bud(tp, buip);
524 xfs_ilock(ip, XFS_ILOCK_EXCL);
525 xfs_trans_ijoin(tp, ip, 0);
527 if (fake.bi_type == XFS_BMAP_MAP)
528 iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
530 iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
532 error = xfs_iext_count_may_overflow(ip, fake.bi_whichfork, iext_delta);
534 error = xfs_iext_count_upgrade(tp, ip, iext_delta);
539 fake.bi_bmap.br_startblock = map->me_startblock;
540 fake.bi_bmap.br_startoff = map->me_startoff;
541 fake.bi_bmap.br_blockcount = map->me_len;
542 fake.bi_bmap.br_state = (map->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
543 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
545 xfs_bmap_update_get_group(mp, &fake);
546 error = xfs_trans_log_finish_bmap_update(tp, budp, &fake);
547 if (error == -EFSCORRUPTED)
548 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, map,
550 xfs_bmap_update_put_group(&fake);
554 if (fake.bi_bmap.br_blockcount > 0) {
555 ASSERT(fake.bi_type == XFS_BMAP_UNMAP);
556 xfs_bmap_unmap_extent(tp, ip, &fake.bi_bmap);
560 * Commit transaction, which frees the transaction and saves the inode
561 * for later replay activities.
563 error = xfs_defer_ops_capture_and_commit(tp, capture_list);
567 xfs_iunlock(ip, XFS_ILOCK_EXCL);
572 xfs_trans_cancel(tp);
574 xfs_iunlock(ip, XFS_ILOCK_EXCL);
582 struct xfs_log_item *lip,
585 return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
588 /* Relog an intent item to push the log tail forward. */
589 static struct xfs_log_item *
591 struct xfs_log_item *intent,
592 struct xfs_trans *tp)
594 struct xfs_bud_log_item *budp;
595 struct xfs_bui_log_item *buip;
596 struct xfs_map_extent *map;
599 count = BUI_ITEM(intent)->bui_format.bui_nextents;
600 map = BUI_ITEM(intent)->bui_format.bui_extents;
602 tp->t_flags |= XFS_TRANS_DIRTY;
603 budp = xfs_trans_get_bud(tp, BUI_ITEM(intent));
604 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
606 buip = xfs_bui_init(tp->t_mountp);
607 memcpy(buip->bui_format.bui_extents, map, count * sizeof(*map));
608 atomic_set(&buip->bui_next_extent, count);
609 xfs_trans_add_item(tp, &buip->bui_item);
610 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
611 return &buip->bui_item;
614 static const struct xfs_item_ops xfs_bui_item_ops = {
615 .flags = XFS_ITEM_INTENT,
616 .iop_size = xfs_bui_item_size,
617 .iop_format = xfs_bui_item_format,
618 .iop_unpin = xfs_bui_item_unpin,
619 .iop_release = xfs_bui_item_release,
620 .iop_recover = xfs_bui_item_recover,
621 .iop_match = xfs_bui_item_match,
622 .iop_relog = xfs_bui_item_relog,
627 struct xfs_bui_log_format *dst,
628 const struct xfs_bui_log_format *src)
632 memcpy(dst, src, offsetof(struct xfs_bui_log_format, bui_extents));
634 for (i = 0; i < src->bui_nextents; i++)
635 memcpy(&dst->bui_extents[i], &src->bui_extents[i],
636 sizeof(struct xfs_map_extent));
640 * This routine is called to create an in-core extent bmap update
641 * item from the bui format structure which was logged on disk.
642 * It allocates an in-core bui, copies the extents from the format
643 * structure into it, and adds the bui to the AIL with the given
647 xlog_recover_bui_commit_pass2(
649 struct list_head *buffer_list,
650 struct xlog_recover_item *item,
653 struct xfs_mount *mp = log->l_mp;
654 struct xfs_bui_log_item *buip;
655 struct xfs_bui_log_format *bui_formatp;
658 bui_formatp = item->ri_buf[0].i_addr;
660 if (item->ri_buf[0].i_len < xfs_bui_log_format_sizeof(0)) {
661 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
662 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
663 return -EFSCORRUPTED;
666 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
667 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
668 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
669 return -EFSCORRUPTED;
672 len = xfs_bui_log_format_sizeof(bui_formatp->bui_nextents);
673 if (item->ri_buf[0].i_len != len) {
674 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
675 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
676 return -EFSCORRUPTED;
679 buip = xfs_bui_init(mp);
680 xfs_bui_copy_format(&buip->bui_format, bui_formatp);
681 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
683 * Insert the intent into the AIL directly and drop one reference so
684 * that finishing or canceling the work will drop the other.
686 xfs_trans_ail_insert(log->l_ailp, &buip->bui_item, lsn);
687 xfs_bui_release(buip);
691 const struct xlog_recover_item_ops xlog_bui_item_ops = {
692 .item_type = XFS_LI_BUI,
693 .commit_pass2 = xlog_recover_bui_commit_pass2,
697 * This routine is called when an BUD format structure is found in a committed
698 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
699 * was still in the log. To do this it searches the AIL for the BUI with an id
700 * equal to that in the BUD format structure. If we find it we drop the BUD
701 * reference, which removes the BUI from the AIL and frees it.
704 xlog_recover_bud_commit_pass2(
706 struct list_head *buffer_list,
707 struct xlog_recover_item *item,
710 struct xfs_bud_log_format *bud_formatp;
712 bud_formatp = item->ri_buf[0].i_addr;
713 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
714 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
715 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
716 return -EFSCORRUPTED;
719 xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
723 const struct xlog_recover_item_ops xlog_bud_item_ops = {
724 .item_type = XFS_LI_BUD,
725 .commit_pass2 = xlog_recover_bud_commit_pass2,