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"
28 struct kmem_cache *xfs_bui_cache;
29 struct kmem_cache *xfs_bud_cache;
31 static const struct xfs_item_ops xfs_bui_item_ops;
33 static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
35 return container_of(lip, struct xfs_bui_log_item, bui_item);
40 struct xfs_bui_log_item *buip)
42 kmem_cache_free(xfs_bui_cache, buip);
46 * Freeing the BUI requires that we remove it from the AIL if it has already
47 * been placed there. However, the BUI may not yet have been placed in the AIL
48 * when called by xfs_bui_release() from BUD processing due to the ordering of
49 * committed vs unpin operations in bulk insert operations. Hence the reference
50 * count to ensure only the last caller frees the BUI.
54 struct xfs_bui_log_item *buip)
56 ASSERT(atomic_read(&buip->bui_refcount) > 0);
57 if (atomic_dec_and_test(&buip->bui_refcount)) {
58 xfs_trans_ail_delete(&buip->bui_item, SHUTDOWN_LOG_IO_ERROR);
59 xfs_bui_item_free(buip);
66 struct xfs_log_item *lip,
70 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
73 *nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
77 * This is called to fill in the vector of log iovecs for the
78 * given bui log item. We use only 1 iovec, and we point that
79 * at the bui_log_format structure embedded in the bui item.
80 * It is at this point that we assert that all of the extent
81 * slots in the bui item have been filled.
85 struct xfs_log_item *lip,
86 struct xfs_log_vec *lv)
88 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
89 struct xfs_log_iovec *vecp = NULL;
91 ASSERT(atomic_read(&buip->bui_next_extent) ==
92 buip->bui_format.bui_nextents);
94 buip->bui_format.bui_type = XFS_LI_BUI;
95 buip->bui_format.bui_size = 1;
97 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
98 xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
102 * The unpin operation is the last place an BUI is manipulated in the log. It is
103 * either inserted in the AIL or aborted in the event of a log I/O error. In
104 * either case, the BUI transaction has been successfully committed to make it
105 * this far. Therefore, we expect whoever committed the BUI to either construct
106 * and commit the BUD or drop the BUD's reference in the event of error. Simply
107 * drop the log's BUI reference now that the log is done with it.
111 struct xfs_log_item *lip,
114 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
116 xfs_bui_release(buip);
120 * The BUI has been either committed or aborted if the transaction has been
121 * cancelled. If the transaction was cancelled, an BUD isn't going to be
122 * constructed and thus we free the BUI here directly.
125 xfs_bui_item_release(
126 struct xfs_log_item *lip)
128 xfs_bui_release(BUI_ITEM(lip));
132 * Allocate and initialize an bui item with the given number of extents.
134 STATIC struct xfs_bui_log_item *
136 struct xfs_mount *mp)
139 struct xfs_bui_log_item *buip;
141 buip = kmem_cache_zalloc(xfs_bui_cache, GFP_KERNEL | __GFP_NOFAIL);
143 xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
144 buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
145 buip->bui_format.bui_id = (uintptr_t)(void *)buip;
146 atomic_set(&buip->bui_next_extent, 0);
147 atomic_set(&buip->bui_refcount, 2);
152 static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
154 return container_of(lip, struct xfs_bud_log_item, bud_item);
159 struct xfs_log_item *lip,
164 *nbytes += sizeof(struct xfs_bud_log_format);
168 * This is called to fill in the vector of log iovecs for the
169 * given bud log item. We use only 1 iovec, and we point that
170 * at the bud_log_format structure embedded in the bud item.
171 * It is at this point that we assert that all of the extent
172 * slots in the bud item have been filled.
176 struct xfs_log_item *lip,
177 struct xfs_log_vec *lv)
179 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
180 struct xfs_log_iovec *vecp = NULL;
182 budp->bud_format.bud_type = XFS_LI_BUD;
183 budp->bud_format.bud_size = 1;
185 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
186 sizeof(struct xfs_bud_log_format));
190 * The BUD is either committed or aborted if the transaction is cancelled. If
191 * the transaction is cancelled, drop our reference to the BUI and free the
195 xfs_bud_item_release(
196 struct xfs_log_item *lip)
198 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
200 xfs_bui_release(budp->bud_buip);
201 kmem_cache_free(xfs_bud_cache, budp);
204 static const struct xfs_item_ops xfs_bud_item_ops = {
205 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
206 .iop_size = xfs_bud_item_size,
207 .iop_format = xfs_bud_item_format,
208 .iop_release = xfs_bud_item_release,
211 static struct xfs_bud_log_item *
213 struct xfs_trans *tp,
214 struct xfs_bui_log_item *buip)
216 struct xfs_bud_log_item *budp;
218 budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
219 xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
221 budp->bud_buip = buip;
222 budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
224 xfs_trans_add_item(tp, &budp->bud_item);
229 * Finish an bmap update and log it to the BUD. Note that the
230 * transaction is marked dirty regardless of whether the bmap update
231 * succeeds or fails to support the BUI/BUD lifecycle rules.
234 xfs_trans_log_finish_bmap_update(
235 struct xfs_trans *tp,
236 struct xfs_bud_log_item *budp,
237 enum xfs_bmap_intent_type type,
238 struct xfs_inode *ip,
240 xfs_fileoff_t startoff,
241 xfs_fsblock_t startblock,
242 xfs_filblks_t *blockcount,
247 error = xfs_bmap_finish_one(tp, ip, type, whichfork, startoff,
248 startblock, blockcount, state);
251 * Mark the transaction dirty, even on error. This ensures the
252 * transaction is aborted, which:
254 * 1.) releases the BUI and frees the BUD
255 * 2.) shuts down the filesystem
257 tp->t_flags |= XFS_TRANS_DIRTY;
258 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
263 /* Sort bmap intents by inode. */
265 xfs_bmap_update_diff_items(
267 const struct list_head *a,
268 const struct list_head *b)
270 struct xfs_bmap_intent *ba;
271 struct xfs_bmap_intent *bb;
273 ba = container_of(a, struct xfs_bmap_intent, bi_list);
274 bb = container_of(b, struct xfs_bmap_intent, bi_list);
275 return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
278 /* Set the map extent flags for this mapping. */
280 xfs_trans_set_bmap_flags(
281 struct xfs_map_extent *bmap,
282 enum xfs_bmap_intent_type type,
290 bmap->me_flags = type;
295 if (state == XFS_EXT_UNWRITTEN)
296 bmap->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
297 if (whichfork == XFS_ATTR_FORK)
298 bmap->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
301 /* Log bmap updates in the intent item. */
303 xfs_bmap_update_log_item(
304 struct xfs_trans *tp,
305 struct xfs_bui_log_item *buip,
306 struct xfs_bmap_intent *bmap)
309 struct xfs_map_extent *map;
311 tp->t_flags |= XFS_TRANS_DIRTY;
312 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
315 * atomic_inc_return gives us the value after the increment;
316 * we want to use it as an array index so we need to subtract 1 from
319 next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
320 ASSERT(next_extent < buip->bui_format.bui_nextents);
321 map = &buip->bui_format.bui_extents[next_extent];
322 map->me_owner = bmap->bi_owner->i_ino;
323 map->me_startblock = bmap->bi_bmap.br_startblock;
324 map->me_startoff = bmap->bi_bmap.br_startoff;
325 map->me_len = bmap->bi_bmap.br_blockcount;
326 xfs_trans_set_bmap_flags(map, bmap->bi_type, bmap->bi_whichfork,
327 bmap->bi_bmap.br_state);
330 static struct xfs_log_item *
331 xfs_bmap_update_create_intent(
332 struct xfs_trans *tp,
333 struct list_head *items,
337 struct xfs_mount *mp = tp->t_mountp;
338 struct xfs_bui_log_item *buip = xfs_bui_init(mp);
339 struct xfs_bmap_intent *bmap;
341 ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
343 xfs_trans_add_item(tp, &buip->bui_item);
345 list_sort(mp, items, xfs_bmap_update_diff_items);
346 list_for_each_entry(bmap, items, bi_list)
347 xfs_bmap_update_log_item(tp, buip, bmap);
348 return &buip->bui_item;
351 /* Get an BUD so we can process all the deferred rmap updates. */
352 static struct xfs_log_item *
353 xfs_bmap_update_create_done(
354 struct xfs_trans *tp,
355 struct xfs_log_item *intent,
358 return &xfs_trans_get_bud(tp, BUI_ITEM(intent))->bud_item;
361 /* Process a deferred rmap update. */
363 xfs_bmap_update_finish_item(
364 struct xfs_trans *tp,
365 struct xfs_log_item *done,
366 struct list_head *item,
367 struct xfs_btree_cur **state)
369 struct xfs_bmap_intent *bmap;
373 bmap = container_of(item, struct xfs_bmap_intent, bi_list);
374 count = bmap->bi_bmap.br_blockcount;
375 error = xfs_trans_log_finish_bmap_update(tp, BUD_ITEM(done),
377 bmap->bi_owner, bmap->bi_whichfork,
378 bmap->bi_bmap.br_startoff,
379 bmap->bi_bmap.br_startblock,
381 bmap->bi_bmap.br_state);
382 if (!error && count > 0) {
383 ASSERT(bmap->bi_type == XFS_BMAP_UNMAP);
384 bmap->bi_bmap.br_blockcount = count;
387 kmem_cache_free(xfs_bmap_intent_cache, bmap);
391 /* Abort all pending BUIs. */
393 xfs_bmap_update_abort_intent(
394 struct xfs_log_item *intent)
396 xfs_bui_release(BUI_ITEM(intent));
399 /* Cancel a deferred rmap update. */
401 xfs_bmap_update_cancel_item(
402 struct list_head *item)
404 struct xfs_bmap_intent *bmap;
406 bmap = container_of(item, struct xfs_bmap_intent, bi_list);
407 kmem_cache_free(xfs_bmap_intent_cache, bmap);
410 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
411 .max_items = XFS_BUI_MAX_FAST_EXTENTS,
412 .create_intent = xfs_bmap_update_create_intent,
413 .abort_intent = xfs_bmap_update_abort_intent,
414 .create_done = xfs_bmap_update_create_done,
415 .finish_item = xfs_bmap_update_finish_item,
416 .cancel_item = xfs_bmap_update_cancel_item,
419 /* Is this recovered BUI ok? */
422 struct xfs_mount *mp,
423 struct xfs_bui_log_item *buip)
425 struct xfs_map_extent *bmap;
427 /* Only one mapping operation per BUI... */
428 if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
431 bmap = &buip->bui_format.bui_extents[0];
433 if (bmap->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
436 switch (bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
444 if (!xfs_verify_ino(mp, bmap->me_owner))
447 if (!xfs_verify_fileext(mp, bmap->me_startoff, bmap->me_len))
450 return xfs_verify_fsbext(mp, bmap->me_startblock, bmap->me_len);
454 * Process a bmap update intent item that was recovered from the log.
455 * We need to update some inode's bmbt.
458 xfs_bui_item_recover(
459 struct xfs_log_item *lip,
460 struct list_head *capture_list)
462 struct xfs_bmbt_irec irec;
463 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
464 struct xfs_trans *tp;
465 struct xfs_inode *ip = NULL;
466 struct xfs_mount *mp = lip->li_log->l_mp;
467 struct xfs_map_extent *bmap;
468 struct xfs_bud_log_item *budp;
471 unsigned int bui_type;
476 if (!xfs_bui_validate(mp, buip)) {
477 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
478 &buip->bui_format, sizeof(buip->bui_format));
479 return -EFSCORRUPTED;
482 bmap = &buip->bui_format.bui_extents[0];
483 state = (bmap->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
484 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
485 whichfork = (bmap->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
486 XFS_ATTR_FORK : XFS_DATA_FORK;
487 bui_type = bmap->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
489 error = xlog_recover_iget(mp, bmap->me_owner, &ip);
493 /* Allocate transaction and do the work. */
494 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
495 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
499 budp = xfs_trans_get_bud(tp, buip);
500 xfs_ilock(ip, XFS_ILOCK_EXCL);
501 xfs_trans_ijoin(tp, ip, 0);
503 if (bui_type == XFS_BMAP_MAP)
504 iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
506 iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
508 error = xfs_iext_count_may_overflow(ip, whichfork, iext_delta);
512 count = bmap->me_len;
513 error = xfs_trans_log_finish_bmap_update(tp, budp, bui_type, ip,
514 whichfork, bmap->me_startoff, bmap->me_startblock,
516 if (error == -EFSCORRUPTED)
517 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bmap,
523 ASSERT(bui_type == XFS_BMAP_UNMAP);
524 irec.br_startblock = bmap->me_startblock;
525 irec.br_blockcount = count;
526 irec.br_startoff = bmap->me_startoff;
527 irec.br_state = state;
528 xfs_bmap_unmap_extent(tp, ip, &irec);
532 * Commit transaction, which frees the transaction and saves the inode
533 * for later replay activities.
535 error = xfs_defer_ops_capture_and_commit(tp, capture_list);
539 xfs_iunlock(ip, XFS_ILOCK_EXCL);
544 xfs_trans_cancel(tp);
546 xfs_iunlock(ip, XFS_ILOCK_EXCL);
554 struct xfs_log_item *lip,
557 return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
560 /* Relog an intent item to push the log tail forward. */
561 static struct xfs_log_item *
563 struct xfs_log_item *intent,
564 struct xfs_trans *tp)
566 struct xfs_bud_log_item *budp;
567 struct xfs_bui_log_item *buip;
568 struct xfs_map_extent *extp;
571 count = BUI_ITEM(intent)->bui_format.bui_nextents;
572 extp = BUI_ITEM(intent)->bui_format.bui_extents;
574 tp->t_flags |= XFS_TRANS_DIRTY;
575 budp = xfs_trans_get_bud(tp, BUI_ITEM(intent));
576 set_bit(XFS_LI_DIRTY, &budp->bud_item.li_flags);
578 buip = xfs_bui_init(tp->t_mountp);
579 memcpy(buip->bui_format.bui_extents, extp, count * sizeof(*extp));
580 atomic_set(&buip->bui_next_extent, count);
581 xfs_trans_add_item(tp, &buip->bui_item);
582 set_bit(XFS_LI_DIRTY, &buip->bui_item.li_flags);
583 return &buip->bui_item;
586 static const struct xfs_item_ops xfs_bui_item_ops = {
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_recover = xfs_bui_item_recover,
592 .iop_match = xfs_bui_item_match,
593 .iop_relog = xfs_bui_item_relog,
597 * Copy an BUI format buffer from the given buf, and into the destination
598 * BUI format structure. The BUI/BUD items were designed not to need any
599 * special alignment handling.
603 struct xfs_log_iovec *buf,
604 struct xfs_bui_log_format *dst_bui_fmt)
606 struct xfs_bui_log_format *src_bui_fmt;
609 src_bui_fmt = buf->i_addr;
610 len = xfs_bui_log_format_sizeof(src_bui_fmt->bui_nextents);
612 if (buf->i_len == len) {
613 memcpy(dst_bui_fmt, src_bui_fmt, len);
616 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);
617 return -EFSCORRUPTED;
621 * This routine is called to create an in-core extent bmap update
622 * item from the bui format structure which was logged on disk.
623 * It allocates an in-core bui, copies the extents from the format
624 * structure into it, and adds the bui to the AIL with the given
628 xlog_recover_bui_commit_pass2(
630 struct list_head *buffer_list,
631 struct xlog_recover_item *item,
635 struct xfs_mount *mp = log->l_mp;
636 struct xfs_bui_log_item *buip;
637 struct xfs_bui_log_format *bui_formatp;
639 bui_formatp = item->ri_buf[0].i_addr;
641 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
642 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
643 return -EFSCORRUPTED;
645 buip = xfs_bui_init(mp);
646 error = xfs_bui_copy_format(&item->ri_buf[0], &buip->bui_format);
648 xfs_bui_item_free(buip);
651 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
653 * Insert the intent into the AIL directly and drop one reference so
654 * that finishing or canceling the work will drop the other.
656 xfs_trans_ail_insert(log->l_ailp, &buip->bui_item, lsn);
657 xfs_bui_release(buip);
661 const struct xlog_recover_item_ops xlog_bui_item_ops = {
662 .item_type = XFS_LI_BUI,
663 .commit_pass2 = xlog_recover_bui_commit_pass2,
667 * This routine is called when an BUD format structure is found in a committed
668 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
669 * was still in the log. To do this it searches the AIL for the BUI with an id
670 * equal to that in the BUD format structure. If we find it we drop the BUD
671 * reference, which removes the BUI from the AIL and frees it.
674 xlog_recover_bud_commit_pass2(
676 struct list_head *buffer_list,
677 struct xlog_recover_item *item,
680 struct xfs_bud_log_format *bud_formatp;
682 bud_formatp = item->ri_buf[0].i_addr;
683 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
684 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, log->l_mp);
685 return -EFSCORRUPTED;
688 xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
692 const struct xlog_recover_item_ops xlog_bud_item_ops = {
693 .item_type = XFS_LI_BUD,
694 .commit_pass2 = xlog_recover_bud_commit_pass2,