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_trans.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_rmap_item.h"
20 #include "xfs_error.h"
21 #include "xfs_log_priv.h"
22 #include "xfs_log_recover.h"
25 struct kmem_cache *xfs_rui_cache;
26 struct kmem_cache *xfs_rud_cache;
28 static const struct xfs_item_ops xfs_rui_item_ops;
30 static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip)
32 return container_of(lip, struct xfs_rui_log_item, rui_item);
37 struct xfs_rui_log_item *ruip)
39 kmem_free(ruip->rui_item.li_lv_shadow);
40 if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS)
43 kmem_cache_free(xfs_rui_cache, ruip);
47 * Freeing the RUI requires that we remove it from the AIL if it has already
48 * been placed there. However, the RUI may not yet have been placed in the AIL
49 * when called by xfs_rui_release() from RUD 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 RUI.
55 struct xfs_rui_log_item *ruip)
57 ASSERT(atomic_read(&ruip->rui_refcount) > 0);
58 if (!atomic_dec_and_test(&ruip->rui_refcount))
61 xfs_trans_ail_delete(&ruip->rui_item, 0);
62 xfs_rui_item_free(ruip);
67 struct xfs_log_item *lip,
71 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
74 *nbytes += xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents);
78 * This is called to fill in the vector of log iovecs for the
79 * given rui log item. We use only 1 iovec, and we point that
80 * at the rui_log_format structure embedded in the rui item.
81 * It is at this point that we assert that all of the extent
82 * slots in the rui item have been filled.
86 struct xfs_log_item *lip,
87 struct xfs_log_vec *lv)
89 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
90 struct xfs_log_iovec *vecp = NULL;
92 ASSERT(atomic_read(&ruip->rui_next_extent) ==
93 ruip->rui_format.rui_nextents);
95 ruip->rui_format.rui_type = XFS_LI_RUI;
96 ruip->rui_format.rui_size = 1;
98 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format,
99 xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents));
103 * The unpin operation is the last place an RUI 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 RUI transaction has been successfully committed to make it
106 * this far. Therefore, we expect whoever committed the RUI to either construct
107 * and commit the RUD or drop the RUD's reference in the event of error. Simply
108 * drop the log's RUI reference now that the log is done with it.
112 struct xfs_log_item *lip,
115 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
117 xfs_rui_release(ruip);
121 * The RUI has been either committed or aborted if the transaction has been
122 * cancelled. If the transaction was cancelled, an RUD isn't going to be
123 * constructed and thus we free the RUI here directly.
126 xfs_rui_item_release(
127 struct xfs_log_item *lip)
129 xfs_rui_release(RUI_ITEM(lip));
133 * Allocate and initialize an rui item with the given number of extents.
135 STATIC struct xfs_rui_log_item *
137 struct xfs_mount *mp,
141 struct xfs_rui_log_item *ruip;
143 ASSERT(nextents > 0);
144 if (nextents > XFS_RUI_MAX_FAST_EXTENTS)
145 ruip = kmem_zalloc(xfs_rui_log_item_sizeof(nextents), 0);
147 ruip = kmem_cache_zalloc(xfs_rui_cache,
148 GFP_KERNEL | __GFP_NOFAIL);
150 xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops);
151 ruip->rui_format.rui_nextents = nextents;
152 ruip->rui_format.rui_id = (uintptr_t)(void *)ruip;
153 atomic_set(&ruip->rui_next_extent, 0);
154 atomic_set(&ruip->rui_refcount, 2);
159 static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip)
161 return container_of(lip, struct xfs_rud_log_item, rud_item);
166 struct xfs_log_item *lip,
171 *nbytes += sizeof(struct xfs_rud_log_format);
175 * This is called to fill in the vector of log iovecs for the
176 * given rud log item. We use only 1 iovec, and we point that
177 * at the rud_log_format structure embedded in the rud item.
178 * It is at this point that we assert that all of the extent
179 * slots in the rud item have been filled.
183 struct xfs_log_item *lip,
184 struct xfs_log_vec *lv)
186 struct xfs_rud_log_item *rudp = RUD_ITEM(lip);
187 struct xfs_log_iovec *vecp = NULL;
189 rudp->rud_format.rud_type = XFS_LI_RUD;
190 rudp->rud_format.rud_size = 1;
192 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format,
193 sizeof(struct xfs_rud_log_format));
197 * The RUD is either committed or aborted if the transaction is cancelled. If
198 * the transaction is cancelled, drop our reference to the RUI and free the
202 xfs_rud_item_release(
203 struct xfs_log_item *lip)
205 struct xfs_rud_log_item *rudp = RUD_ITEM(lip);
207 xfs_rui_release(rudp->rud_ruip);
208 kmem_free(rudp->rud_item.li_lv_shadow);
209 kmem_cache_free(xfs_rud_cache, rudp);
212 static struct xfs_log_item *
214 struct xfs_log_item *lip)
216 return &RUD_ITEM(lip)->rud_ruip->rui_item;
219 static const struct xfs_item_ops xfs_rud_item_ops = {
220 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
221 XFS_ITEM_INTENT_DONE,
222 .iop_size = xfs_rud_item_size,
223 .iop_format = xfs_rud_item_format,
224 .iop_release = xfs_rud_item_release,
225 .iop_intent = xfs_rud_item_intent,
228 static struct xfs_rud_log_item *
230 struct xfs_trans *tp,
231 struct xfs_rui_log_item *ruip)
233 struct xfs_rud_log_item *rudp;
235 rudp = kmem_cache_zalloc(xfs_rud_cache, GFP_KERNEL | __GFP_NOFAIL);
236 xfs_log_item_init(tp->t_mountp, &rudp->rud_item, XFS_LI_RUD,
238 rudp->rud_ruip = ruip;
239 rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id;
241 xfs_trans_add_item(tp, &rudp->rud_item);
245 /* Set the map extent flags for this reverse mapping. */
247 xfs_trans_set_rmap_flags(
248 struct xfs_map_extent *map,
249 enum xfs_rmap_intent_type type,
254 if (state == XFS_EXT_UNWRITTEN)
255 map->me_flags |= XFS_RMAP_EXTENT_UNWRITTEN;
256 if (whichfork == XFS_ATTR_FORK)
257 map->me_flags |= XFS_RMAP_EXTENT_ATTR_FORK;
260 map->me_flags |= XFS_RMAP_EXTENT_MAP;
262 case XFS_RMAP_MAP_SHARED:
263 map->me_flags |= XFS_RMAP_EXTENT_MAP_SHARED;
266 map->me_flags |= XFS_RMAP_EXTENT_UNMAP;
268 case XFS_RMAP_UNMAP_SHARED:
269 map->me_flags |= XFS_RMAP_EXTENT_UNMAP_SHARED;
271 case XFS_RMAP_CONVERT:
272 map->me_flags |= XFS_RMAP_EXTENT_CONVERT;
274 case XFS_RMAP_CONVERT_SHARED:
275 map->me_flags |= XFS_RMAP_EXTENT_CONVERT_SHARED;
278 map->me_flags |= XFS_RMAP_EXTENT_ALLOC;
281 map->me_flags |= XFS_RMAP_EXTENT_FREE;
289 * Finish an rmap update and log it to the RUD. Note that the transaction is
290 * marked dirty regardless of whether the rmap update succeeds or fails to
291 * support the RUI/RUD lifecycle rules.
294 xfs_trans_log_finish_rmap_update(
295 struct xfs_trans *tp,
296 struct xfs_rud_log_item *rudp,
297 struct xfs_rmap_intent *ri,
298 struct xfs_btree_cur **pcur)
302 error = xfs_rmap_finish_one(tp, ri, pcur);
305 * Mark the transaction dirty, even on error. This ensures the
306 * transaction is aborted, which:
308 * 1.) releases the RUI and frees the RUD
309 * 2.) shuts down the filesystem
311 tp->t_flags |= XFS_TRANS_DIRTY | XFS_TRANS_HAS_INTENT_DONE;
312 set_bit(XFS_LI_DIRTY, &rudp->rud_item.li_flags);
317 /* Sort rmap intents by AG. */
319 xfs_rmap_update_diff_items(
321 const struct list_head *a,
322 const struct list_head *b)
324 struct xfs_rmap_intent *ra;
325 struct xfs_rmap_intent *rb;
327 ra = container_of(a, struct xfs_rmap_intent, ri_list);
328 rb = container_of(b, struct xfs_rmap_intent, ri_list);
330 return ra->ri_pag->pag_agno - rb->ri_pag->pag_agno;
333 /* Log rmap updates in the intent item. */
335 xfs_rmap_update_log_item(
336 struct xfs_trans *tp,
337 struct xfs_rui_log_item *ruip,
338 struct xfs_rmap_intent *ri)
341 struct xfs_map_extent *map;
343 tp->t_flags |= XFS_TRANS_DIRTY;
344 set_bit(XFS_LI_DIRTY, &ruip->rui_item.li_flags);
347 * atomic_inc_return gives us the value after the increment;
348 * we want to use it as an array index so we need to subtract 1 from
351 next_extent = atomic_inc_return(&ruip->rui_next_extent) - 1;
352 ASSERT(next_extent < ruip->rui_format.rui_nextents);
353 map = &ruip->rui_format.rui_extents[next_extent];
354 map->me_owner = ri->ri_owner;
355 map->me_startblock = ri->ri_bmap.br_startblock;
356 map->me_startoff = ri->ri_bmap.br_startoff;
357 map->me_len = ri->ri_bmap.br_blockcount;
358 xfs_trans_set_rmap_flags(map, ri->ri_type, ri->ri_whichfork,
359 ri->ri_bmap.br_state);
362 static struct xfs_log_item *
363 xfs_rmap_update_create_intent(
364 struct xfs_trans *tp,
365 struct list_head *items,
369 struct xfs_mount *mp = tp->t_mountp;
370 struct xfs_rui_log_item *ruip = xfs_rui_init(mp, count);
371 struct xfs_rmap_intent *ri;
375 xfs_trans_add_item(tp, &ruip->rui_item);
377 list_sort(mp, items, xfs_rmap_update_diff_items);
378 list_for_each_entry(ri, items, ri_list)
379 xfs_rmap_update_log_item(tp, ruip, ri);
380 return &ruip->rui_item;
383 /* Get an RUD so we can process all the deferred rmap updates. */
384 static struct xfs_log_item *
385 xfs_rmap_update_create_done(
386 struct xfs_trans *tp,
387 struct xfs_log_item *intent,
390 return &xfs_trans_get_rud(tp, RUI_ITEM(intent))->rud_item;
393 /* Take a passive ref to the AG containing the space we're rmapping. */
395 xfs_rmap_update_get_group(
396 struct xfs_mount *mp,
397 struct xfs_rmap_intent *ri)
401 agno = XFS_FSB_TO_AGNO(mp, ri->ri_bmap.br_startblock);
402 ri->ri_pag = xfs_perag_intent_get(mp, agno);
405 /* Release a passive AG ref after finishing rmapping work. */
407 xfs_rmap_update_put_group(
408 struct xfs_rmap_intent *ri)
410 xfs_perag_intent_put(ri->ri_pag);
413 /* Process a deferred rmap update. */
415 xfs_rmap_update_finish_item(
416 struct xfs_trans *tp,
417 struct xfs_log_item *done,
418 struct list_head *item,
419 struct xfs_btree_cur **state)
421 struct xfs_rmap_intent *ri;
424 ri = container_of(item, struct xfs_rmap_intent, ri_list);
426 error = xfs_trans_log_finish_rmap_update(tp, RUD_ITEM(done), ri,
429 xfs_rmap_update_put_group(ri);
430 kmem_cache_free(xfs_rmap_intent_cache, ri);
434 /* Abort all pending RUIs. */
436 xfs_rmap_update_abort_intent(
437 struct xfs_log_item *intent)
439 xfs_rui_release(RUI_ITEM(intent));
442 /* Cancel a deferred rmap update. */
444 xfs_rmap_update_cancel_item(
445 struct list_head *item)
447 struct xfs_rmap_intent *ri;
449 ri = container_of(item, struct xfs_rmap_intent, ri_list);
451 xfs_rmap_update_put_group(ri);
452 kmem_cache_free(xfs_rmap_intent_cache, ri);
455 const struct xfs_defer_op_type xfs_rmap_update_defer_type = {
456 .max_items = XFS_RUI_MAX_FAST_EXTENTS,
457 .create_intent = xfs_rmap_update_create_intent,
458 .abort_intent = xfs_rmap_update_abort_intent,
459 .create_done = xfs_rmap_update_create_done,
460 .finish_item = xfs_rmap_update_finish_item,
461 .finish_cleanup = xfs_rmap_finish_one_cleanup,
462 .cancel_item = xfs_rmap_update_cancel_item,
465 /* Is this recovered RUI ok? */
467 xfs_rui_validate_map(
468 struct xfs_mount *mp,
469 struct xfs_map_extent *map)
471 if (!xfs_has_rmapbt(mp))
474 if (map->me_flags & ~XFS_RMAP_EXTENT_FLAGS)
477 switch (map->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
478 case XFS_RMAP_EXTENT_MAP:
479 case XFS_RMAP_EXTENT_MAP_SHARED:
480 case XFS_RMAP_EXTENT_UNMAP:
481 case XFS_RMAP_EXTENT_UNMAP_SHARED:
482 case XFS_RMAP_EXTENT_CONVERT:
483 case XFS_RMAP_EXTENT_CONVERT_SHARED:
484 case XFS_RMAP_EXTENT_ALLOC:
485 case XFS_RMAP_EXTENT_FREE:
491 if (!XFS_RMAP_NON_INODE_OWNER(map->me_owner) &&
492 !xfs_verify_ino(mp, map->me_owner))
495 if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
498 return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
502 * Process an rmap update intent item that was recovered from the log.
503 * We need to update the rmapbt.
506 xfs_rui_item_recover(
507 struct xfs_log_item *lip,
508 struct list_head *capture_list)
510 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
511 struct xfs_rud_log_item *rudp;
512 struct xfs_trans *tp;
513 struct xfs_btree_cur *rcur = NULL;
514 struct xfs_mount *mp = lip->li_log->l_mp;
519 * First check the validity of the extents described by the
520 * RUI. If any are bad, then assume that all are bad and
523 for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
524 if (!xfs_rui_validate_map(mp,
525 &ruip->rui_format.rui_extents[i])) {
526 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
528 sizeof(ruip->rui_format));
529 return -EFSCORRUPTED;
533 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
534 mp->m_rmap_maxlevels, 0, XFS_TRANS_RESERVE, &tp);
537 rudp = xfs_trans_get_rud(tp, ruip);
539 for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
540 struct xfs_rmap_intent fake = { };
541 struct xfs_map_extent *map;
543 map = &ruip->rui_format.rui_extents[i];
544 switch (map->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
545 case XFS_RMAP_EXTENT_MAP:
546 fake.ri_type = XFS_RMAP_MAP;
548 case XFS_RMAP_EXTENT_MAP_SHARED:
549 fake.ri_type = XFS_RMAP_MAP_SHARED;
551 case XFS_RMAP_EXTENT_UNMAP:
552 fake.ri_type = XFS_RMAP_UNMAP;
554 case XFS_RMAP_EXTENT_UNMAP_SHARED:
555 fake.ri_type = XFS_RMAP_UNMAP_SHARED;
557 case XFS_RMAP_EXTENT_CONVERT:
558 fake.ri_type = XFS_RMAP_CONVERT;
560 case XFS_RMAP_EXTENT_CONVERT_SHARED:
561 fake.ri_type = XFS_RMAP_CONVERT_SHARED;
563 case XFS_RMAP_EXTENT_ALLOC:
564 fake.ri_type = XFS_RMAP_ALLOC;
566 case XFS_RMAP_EXTENT_FREE:
567 fake.ri_type = XFS_RMAP_FREE;
570 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
572 sizeof(ruip->rui_format));
573 error = -EFSCORRUPTED;
577 fake.ri_owner = map->me_owner;
578 fake.ri_whichfork = (map->me_flags & XFS_RMAP_EXTENT_ATTR_FORK) ?
579 XFS_ATTR_FORK : XFS_DATA_FORK;
580 fake.ri_bmap.br_startblock = map->me_startblock;
581 fake.ri_bmap.br_startoff = map->me_startoff;
582 fake.ri_bmap.br_blockcount = map->me_len;
583 fake.ri_bmap.br_state = (map->me_flags & XFS_RMAP_EXTENT_UNWRITTEN) ?
584 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
586 xfs_rmap_update_get_group(mp, &fake);
587 error = xfs_trans_log_finish_rmap_update(tp, rudp, &fake,
589 if (error == -EFSCORRUPTED)
590 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
592 xfs_rmap_update_put_group(&fake);
598 xfs_rmap_finish_one_cleanup(tp, rcur, error);
599 return xfs_defer_ops_capture_and_commit(tp, capture_list);
602 xfs_rmap_finish_one_cleanup(tp, rcur, error);
603 xfs_trans_cancel(tp);
609 struct xfs_log_item *lip,
612 return RUI_ITEM(lip)->rui_format.rui_id == intent_id;
615 /* Relog an intent item to push the log tail forward. */
616 static struct xfs_log_item *
618 struct xfs_log_item *intent,
619 struct xfs_trans *tp)
621 struct xfs_rud_log_item *rudp;
622 struct xfs_rui_log_item *ruip;
623 struct xfs_map_extent *map;
626 count = RUI_ITEM(intent)->rui_format.rui_nextents;
627 map = RUI_ITEM(intent)->rui_format.rui_extents;
629 tp->t_flags |= XFS_TRANS_DIRTY;
630 rudp = xfs_trans_get_rud(tp, RUI_ITEM(intent));
631 set_bit(XFS_LI_DIRTY, &rudp->rud_item.li_flags);
633 ruip = xfs_rui_init(tp->t_mountp, count);
634 memcpy(ruip->rui_format.rui_extents, map, count * sizeof(*map));
635 atomic_set(&ruip->rui_next_extent, count);
636 xfs_trans_add_item(tp, &ruip->rui_item);
637 set_bit(XFS_LI_DIRTY, &ruip->rui_item.li_flags);
638 return &ruip->rui_item;
641 static const struct xfs_item_ops xfs_rui_item_ops = {
642 .flags = XFS_ITEM_INTENT,
643 .iop_size = xfs_rui_item_size,
644 .iop_format = xfs_rui_item_format,
645 .iop_unpin = xfs_rui_item_unpin,
646 .iop_release = xfs_rui_item_release,
647 .iop_recover = xfs_rui_item_recover,
648 .iop_match = xfs_rui_item_match,
649 .iop_relog = xfs_rui_item_relog,
654 struct xfs_rui_log_format *dst,
655 const struct xfs_rui_log_format *src)
659 memcpy(dst, src, offsetof(struct xfs_rui_log_format, rui_extents));
661 for (i = 0; i < src->rui_nextents; i++)
662 memcpy(&dst->rui_extents[i], &src->rui_extents[i],
663 sizeof(struct xfs_map_extent));
667 * This routine is called to create an in-core extent rmap update
668 * item from the rui format structure which was logged on disk.
669 * It allocates an in-core rui, copies the extents from the format
670 * structure into it, and adds the rui to the AIL with the given
674 xlog_recover_rui_commit_pass2(
676 struct list_head *buffer_list,
677 struct xlog_recover_item *item,
680 struct xfs_mount *mp = log->l_mp;
681 struct xfs_rui_log_item *ruip;
682 struct xfs_rui_log_format *rui_formatp;
685 rui_formatp = item->ri_buf[0].i_addr;
687 if (item->ri_buf[0].i_len < xfs_rui_log_format_sizeof(0)) {
688 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
689 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
690 return -EFSCORRUPTED;
693 len = xfs_rui_log_format_sizeof(rui_formatp->rui_nextents);
694 if (item->ri_buf[0].i_len != len) {
695 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
696 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
697 return -EFSCORRUPTED;
700 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
701 xfs_rui_copy_format(&ruip->rui_format, rui_formatp);
702 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
704 * Insert the intent into the AIL directly and drop one reference so
705 * that finishing or canceling the work will drop the other.
707 xfs_trans_ail_insert(log->l_ailp, &ruip->rui_item, lsn);
708 xfs_rui_release(ruip);
712 const struct xlog_recover_item_ops xlog_rui_item_ops = {
713 .item_type = XFS_LI_RUI,
714 .commit_pass2 = xlog_recover_rui_commit_pass2,
718 * This routine is called when an RUD format structure is found in a committed
719 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
720 * was still in the log. To do this it searches the AIL for the RUI with an id
721 * equal to that in the RUD format structure. If we find it we drop the RUD
722 * reference, which removes the RUI from the AIL and frees it.
725 xlog_recover_rud_commit_pass2(
727 struct list_head *buffer_list,
728 struct xlog_recover_item *item,
731 struct xfs_rud_log_format *rud_formatp;
733 rud_formatp = item->ri_buf[0].i_addr;
734 if (item->ri_buf[0].i_len != sizeof(struct xfs_rud_log_format)) {
735 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
736 rud_formatp, item->ri_buf[0].i_len);
737 return -EFSCORRUPTED;
740 xlog_recover_release_intent(log, XFS_LI_RUI, rud_formatp->rud_rui_id);
744 const struct xlog_recover_item_ops xlog_rud_item_ops = {
745 .item_type = XFS_LI_RUD,
746 .commit_pass2 = xlog_recover_rud_commit_pass2,