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 /* Sort bmap intents by inode. */
226 xfs_bmap_update_diff_items(
228 const struct list_head *a,
229 const struct list_head *b)
231 struct xfs_bmap_intent *ba;
232 struct xfs_bmap_intent *bb;
234 ba = container_of(a, struct xfs_bmap_intent, bi_list);
235 bb = container_of(b, struct xfs_bmap_intent, bi_list);
236 return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
239 /* Set the map extent flags for this mapping. */
241 xfs_trans_set_bmap_flags(
242 struct xfs_map_extent *map,
243 enum xfs_bmap_intent_type type,
251 map->me_flags = type;
256 if (state == XFS_EXT_UNWRITTEN)
257 map->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
258 if (whichfork == XFS_ATTR_FORK)
259 map->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
262 /* Log bmap updates in the intent item. */
264 xfs_bmap_update_log_item(
265 struct xfs_trans *tp,
266 struct xfs_bui_log_item *buip,
267 struct xfs_bmap_intent *bi)
270 struct xfs_map_extent *map;
273 * atomic_inc_return gives us the value after the increment;
274 * we want to use it as an array index so we need to subtract 1 from
277 next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
278 ASSERT(next_extent < buip->bui_format.bui_nextents);
279 map = &buip->bui_format.bui_extents[next_extent];
280 map->me_owner = bi->bi_owner->i_ino;
281 map->me_startblock = bi->bi_bmap.br_startblock;
282 map->me_startoff = bi->bi_bmap.br_startoff;
283 map->me_len = bi->bi_bmap.br_blockcount;
284 xfs_trans_set_bmap_flags(map, bi->bi_type, bi->bi_whichfork,
285 bi->bi_bmap.br_state);
288 static struct xfs_log_item *
289 xfs_bmap_update_create_intent(
290 struct xfs_trans *tp,
291 struct list_head *items,
295 struct xfs_mount *mp = tp->t_mountp;
296 struct xfs_bui_log_item *buip = xfs_bui_init(mp);
297 struct xfs_bmap_intent *bi;
299 ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
302 list_sort(mp, items, xfs_bmap_update_diff_items);
303 list_for_each_entry(bi, items, bi_list)
304 xfs_bmap_update_log_item(tp, buip, bi);
305 return &buip->bui_item;
308 /* Get an BUD so we can process all the deferred bmap updates. */
309 static struct xfs_log_item *
310 xfs_bmap_update_create_done(
311 struct xfs_trans *tp,
312 struct xfs_log_item *intent,
315 struct xfs_bui_log_item *buip = BUI_ITEM(intent);
316 struct xfs_bud_log_item *budp;
318 budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
319 xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
321 budp->bud_buip = buip;
322 budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
324 return &budp->bud_item;
327 /* Take a passive ref to the AG containing the space we're mapping. */
329 xfs_bmap_update_get_group(
330 struct xfs_mount *mp,
331 struct xfs_bmap_intent *bi)
335 agno = XFS_FSB_TO_AGNO(mp, bi->bi_bmap.br_startblock);
338 * Bump the intent count on behalf of the deferred rmap and refcount
339 * intent items that that we can queue when we finish this bmap work.
340 * This new intent item will bump the intent count before the bmap
341 * intent drops the intent count, ensuring that the intent count
342 * remains nonzero across the transaction roll.
344 bi->bi_pag = xfs_perag_intent_get(mp, agno);
347 /* Release a passive AG ref after finishing mapping work. */
349 xfs_bmap_update_put_group(
350 struct xfs_bmap_intent *bi)
352 xfs_perag_intent_put(bi->bi_pag);
355 /* Process a deferred bmap update. */
357 xfs_bmap_update_finish_item(
358 struct xfs_trans *tp,
359 struct xfs_log_item *done,
360 struct list_head *item,
361 struct xfs_btree_cur **state)
363 struct xfs_bmap_intent *bi;
366 bi = container_of(item, struct xfs_bmap_intent, bi_list);
368 error = xfs_bmap_finish_one(tp, bi);
369 if (!error && bi->bi_bmap.br_blockcount > 0) {
370 ASSERT(bi->bi_type == XFS_BMAP_UNMAP);
374 xfs_bmap_update_put_group(bi);
375 kmem_cache_free(xfs_bmap_intent_cache, bi);
379 /* Abort all pending BUIs. */
381 xfs_bmap_update_abort_intent(
382 struct xfs_log_item *intent)
384 xfs_bui_release(BUI_ITEM(intent));
387 /* Cancel a deferred bmap update. */
389 xfs_bmap_update_cancel_item(
390 struct list_head *item)
392 struct xfs_bmap_intent *bi;
394 bi = container_of(item, struct xfs_bmap_intent, bi_list);
396 xfs_bmap_update_put_group(bi);
397 kmem_cache_free(xfs_bmap_intent_cache, bi);
400 /* Is this recovered BUI ok? */
403 struct xfs_mount *mp,
404 struct xfs_bui_log_item *buip)
406 struct xfs_map_extent *map;
408 /* Only one mapping operation per BUI... */
409 if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
412 map = &buip->bui_format.bui_extents[0];
414 if (map->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
417 switch (map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
425 if (!xfs_verify_ino(mp, map->me_owner))
428 if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
431 return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
434 static inline struct xfs_bmap_intent *
435 xfs_bui_recover_work(
436 struct xfs_mount *mp,
437 struct xfs_defer_pending *dfp,
438 struct xfs_inode **ipp,
439 struct xfs_map_extent *map)
441 struct xfs_bmap_intent *bi;
444 error = xlog_recover_iget(mp, map->me_owner, ipp);
446 return ERR_PTR(error);
448 bi = kmem_cache_zalloc(xfs_bmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
449 bi->bi_whichfork = (map->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
450 XFS_ATTR_FORK : XFS_DATA_FORK;
451 bi->bi_type = map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
452 bi->bi_bmap.br_startblock = map->me_startblock;
453 bi->bi_bmap.br_startoff = map->me_startoff;
454 bi->bi_bmap.br_blockcount = map->me_len;
455 bi->bi_bmap.br_state = (map->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
456 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
458 xfs_bmap_update_get_group(mp, bi);
460 xfs_defer_add_item(dfp, &bi->bi_list);
465 * Process a bmap update intent item that was recovered from the log.
466 * We need to update some inode's bmbt.
469 xfs_bmap_recover_work(
470 struct xfs_defer_pending *dfp,
471 struct list_head *capture_list)
473 struct xfs_trans_res resv;
474 struct xfs_log_item *lip = dfp->dfp_intent;
475 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
476 struct xfs_trans *tp;
477 struct xfs_inode *ip = NULL;
478 struct xfs_mount *mp = lip->li_log->l_mp;
479 struct xfs_map_extent *map;
480 struct xfs_bmap_intent *work;
484 if (!xfs_bui_validate(mp, buip)) {
485 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
486 &buip->bui_format, sizeof(buip->bui_format));
487 return -EFSCORRUPTED;
490 map = &buip->bui_format.bui_extents[0];
491 work = xfs_bui_recover_work(mp, dfp, &ip, map);
493 return PTR_ERR(work);
495 /* Allocate transaction and do the work. */
496 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
497 error = xfs_trans_alloc(mp, &resv,
498 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
502 xfs_ilock(ip, XFS_ILOCK_EXCL);
503 xfs_trans_ijoin(tp, ip, 0);
505 if (work->bi_type == XFS_BMAP_MAP)
506 iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
508 iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
510 error = xfs_iext_count_may_overflow(ip, work->bi_whichfork, iext_delta);
512 error = xfs_iext_count_upgrade(tp, ip, iext_delta);
516 error = xlog_recover_finish_intent(tp, dfp);
517 if (error == -EFSCORRUPTED)
518 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
519 &buip->bui_format, sizeof(buip->bui_format));
524 * Commit transaction, which frees the transaction and saves the inode
525 * for later replay activities.
527 error = xfs_defer_ops_capture_and_commit(tp, capture_list);
531 xfs_iunlock(ip, XFS_ILOCK_EXCL);
536 xfs_trans_cancel(tp);
538 xfs_iunlock(ip, XFS_ILOCK_EXCL);
544 /* Relog an intent item to push the log tail forward. */
545 static struct xfs_log_item *
546 xfs_bmap_relog_intent(
547 struct xfs_trans *tp,
548 struct xfs_log_item *intent,
549 struct xfs_log_item *done_item)
551 struct xfs_bui_log_item *buip;
552 struct xfs_map_extent *map;
555 count = BUI_ITEM(intent)->bui_format.bui_nextents;
556 map = BUI_ITEM(intent)->bui_format.bui_extents;
558 buip = xfs_bui_init(tp->t_mountp);
559 memcpy(buip->bui_format.bui_extents, map, count * sizeof(*map));
560 atomic_set(&buip->bui_next_extent, count);
562 return &buip->bui_item;
565 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
567 .max_items = XFS_BUI_MAX_FAST_EXTENTS,
568 .create_intent = xfs_bmap_update_create_intent,
569 .abort_intent = xfs_bmap_update_abort_intent,
570 .create_done = xfs_bmap_update_create_done,
571 .finish_item = xfs_bmap_update_finish_item,
572 .cancel_item = xfs_bmap_update_cancel_item,
573 .recover_work = xfs_bmap_recover_work,
574 .relog_intent = xfs_bmap_relog_intent,
579 struct xfs_log_item *lip,
582 return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
585 static const struct xfs_item_ops xfs_bui_item_ops = {
586 .flags = XFS_ITEM_INTENT,
587 .iop_size = xfs_bui_item_size,
588 .iop_format = xfs_bui_item_format,
589 .iop_unpin = xfs_bui_item_unpin,
590 .iop_release = xfs_bui_item_release,
591 .iop_match = xfs_bui_item_match,
596 struct xfs_bui_log_format *dst,
597 const struct xfs_bui_log_format *src)
601 memcpy(dst, src, offsetof(struct xfs_bui_log_format, bui_extents));
603 for (i = 0; i < src->bui_nextents; i++)
604 memcpy(&dst->bui_extents[i], &src->bui_extents[i],
605 sizeof(struct xfs_map_extent));
609 * This routine is called to create an in-core extent bmap update
610 * item from the bui format structure which was logged on disk.
611 * It allocates an in-core bui, copies the extents from the format
612 * structure into it, and adds the bui to the AIL with the given
616 xlog_recover_bui_commit_pass2(
618 struct list_head *buffer_list,
619 struct xlog_recover_item *item,
622 struct xfs_mount *mp = log->l_mp;
623 struct xfs_bui_log_item *buip;
624 struct xfs_bui_log_format *bui_formatp;
627 bui_formatp = item->ri_buf[0].i_addr;
629 if (item->ri_buf[0].i_len < xfs_bui_log_format_sizeof(0)) {
630 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
631 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
632 return -EFSCORRUPTED;
635 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
636 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
637 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
638 return -EFSCORRUPTED;
641 len = xfs_bui_log_format_sizeof(bui_formatp->bui_nextents);
642 if (item->ri_buf[0].i_len != len) {
643 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
644 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
645 return -EFSCORRUPTED;
648 buip = xfs_bui_init(mp);
649 xfs_bui_copy_format(&buip->bui_format, bui_formatp);
650 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
652 xlog_recover_intent_item(log, &buip->bui_item, lsn,
653 &xfs_bmap_update_defer_type);
657 const struct xlog_recover_item_ops xlog_bui_item_ops = {
658 .item_type = XFS_LI_BUI,
659 .commit_pass2 = xlog_recover_bui_commit_pass2,
663 * This routine is called when an BUD format structure is found in a committed
664 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
665 * was still in the log. To do this it searches the AIL for the BUI with an id
666 * equal to that in the BUD format structure. If we find it we drop the BUD
667 * reference, which removes the BUI from the AIL and frees it.
670 xlog_recover_bud_commit_pass2(
672 struct list_head *buffer_list,
673 struct xlog_recover_item *item,
676 struct xfs_bud_log_format *bud_formatp;
678 bud_formatp = item->ri_buf[0].i_addr;
679 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
680 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
681 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
682 return -EFSCORRUPTED;
685 xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
689 const struct xlog_recover_item_ops xlog_bud_item_ops = {
690 .item_type = XFS_LI_BUD,
691 .commit_pass2 = xlog_recover_bud_commit_pass2,