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 /* Set the map extent flags for this reverse mapping. */
230 xfs_trans_set_rmap_flags(
231 struct xfs_map_extent *map,
232 enum xfs_rmap_intent_type type,
237 if (state == XFS_EXT_UNWRITTEN)
238 map->me_flags |= XFS_RMAP_EXTENT_UNWRITTEN;
239 if (whichfork == XFS_ATTR_FORK)
240 map->me_flags |= XFS_RMAP_EXTENT_ATTR_FORK;
243 map->me_flags |= XFS_RMAP_EXTENT_MAP;
245 case XFS_RMAP_MAP_SHARED:
246 map->me_flags |= XFS_RMAP_EXTENT_MAP_SHARED;
249 map->me_flags |= XFS_RMAP_EXTENT_UNMAP;
251 case XFS_RMAP_UNMAP_SHARED:
252 map->me_flags |= XFS_RMAP_EXTENT_UNMAP_SHARED;
254 case XFS_RMAP_CONVERT:
255 map->me_flags |= XFS_RMAP_EXTENT_CONVERT;
257 case XFS_RMAP_CONVERT_SHARED:
258 map->me_flags |= XFS_RMAP_EXTENT_CONVERT_SHARED;
261 map->me_flags |= XFS_RMAP_EXTENT_ALLOC;
264 map->me_flags |= XFS_RMAP_EXTENT_FREE;
271 /* Sort rmap intents by AG. */
273 xfs_rmap_update_diff_items(
275 const struct list_head *a,
276 const struct list_head *b)
278 struct xfs_rmap_intent *ra;
279 struct xfs_rmap_intent *rb;
281 ra = container_of(a, struct xfs_rmap_intent, ri_list);
282 rb = container_of(b, struct xfs_rmap_intent, ri_list);
284 return ra->ri_pag->pag_agno - rb->ri_pag->pag_agno;
287 /* Log rmap updates in the intent item. */
289 xfs_rmap_update_log_item(
290 struct xfs_trans *tp,
291 struct xfs_rui_log_item *ruip,
292 struct xfs_rmap_intent *ri)
295 struct xfs_map_extent *map;
298 * atomic_inc_return gives us the value after the increment;
299 * we want to use it as an array index so we need to subtract 1 from
302 next_extent = atomic_inc_return(&ruip->rui_next_extent) - 1;
303 ASSERT(next_extent < ruip->rui_format.rui_nextents);
304 map = &ruip->rui_format.rui_extents[next_extent];
305 map->me_owner = ri->ri_owner;
306 map->me_startblock = ri->ri_bmap.br_startblock;
307 map->me_startoff = ri->ri_bmap.br_startoff;
308 map->me_len = ri->ri_bmap.br_blockcount;
309 xfs_trans_set_rmap_flags(map, ri->ri_type, ri->ri_whichfork,
310 ri->ri_bmap.br_state);
313 static struct xfs_log_item *
314 xfs_rmap_update_create_intent(
315 struct xfs_trans *tp,
316 struct list_head *items,
320 struct xfs_mount *mp = tp->t_mountp;
321 struct xfs_rui_log_item *ruip = xfs_rui_init(mp, count);
322 struct xfs_rmap_intent *ri;
327 list_sort(mp, items, xfs_rmap_update_diff_items);
328 list_for_each_entry(ri, items, ri_list)
329 xfs_rmap_update_log_item(tp, ruip, ri);
330 return &ruip->rui_item;
333 /* Get an RUD so we can process all the deferred rmap updates. */
334 static struct xfs_log_item *
335 xfs_rmap_update_create_done(
336 struct xfs_trans *tp,
337 struct xfs_log_item *intent,
340 struct xfs_rui_log_item *ruip = RUI_ITEM(intent);
341 struct xfs_rud_log_item *rudp;
343 rudp = kmem_cache_zalloc(xfs_rud_cache, GFP_KERNEL | __GFP_NOFAIL);
344 xfs_log_item_init(tp->t_mountp, &rudp->rud_item, XFS_LI_RUD,
346 rudp->rud_ruip = ruip;
347 rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id;
349 return &rudp->rud_item;
352 /* Take a passive ref to the AG containing the space we're rmapping. */
354 xfs_rmap_update_get_group(
355 struct xfs_mount *mp,
356 struct xfs_rmap_intent *ri)
360 agno = XFS_FSB_TO_AGNO(mp, ri->ri_bmap.br_startblock);
361 ri->ri_pag = xfs_perag_intent_get(mp, agno);
364 /* Release a passive AG ref after finishing rmapping work. */
366 xfs_rmap_update_put_group(
367 struct xfs_rmap_intent *ri)
369 xfs_perag_intent_put(ri->ri_pag);
372 /* Process a deferred rmap update. */
374 xfs_rmap_update_finish_item(
375 struct xfs_trans *tp,
376 struct xfs_log_item *done,
377 struct list_head *item,
378 struct xfs_btree_cur **state)
380 struct xfs_rmap_intent *ri;
383 ri = container_of(item, struct xfs_rmap_intent, ri_list);
385 error = xfs_rmap_finish_one(tp, ri, state);
387 xfs_rmap_update_put_group(ri);
388 kmem_cache_free(xfs_rmap_intent_cache, ri);
392 /* Abort all pending RUIs. */
394 xfs_rmap_update_abort_intent(
395 struct xfs_log_item *intent)
397 xfs_rui_release(RUI_ITEM(intent));
400 /* Cancel a deferred rmap update. */
402 xfs_rmap_update_cancel_item(
403 struct list_head *item)
405 struct xfs_rmap_intent *ri;
407 ri = container_of(item, struct xfs_rmap_intent, ri_list);
409 xfs_rmap_update_put_group(ri);
410 kmem_cache_free(xfs_rmap_intent_cache, ri);
413 /* Is this recovered RUI ok? */
415 xfs_rui_validate_map(
416 struct xfs_mount *mp,
417 struct xfs_map_extent *map)
419 if (!xfs_has_rmapbt(mp))
422 if (map->me_flags & ~XFS_RMAP_EXTENT_FLAGS)
425 switch (map->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
426 case XFS_RMAP_EXTENT_MAP:
427 case XFS_RMAP_EXTENT_MAP_SHARED:
428 case XFS_RMAP_EXTENT_UNMAP:
429 case XFS_RMAP_EXTENT_UNMAP_SHARED:
430 case XFS_RMAP_EXTENT_CONVERT:
431 case XFS_RMAP_EXTENT_CONVERT_SHARED:
432 case XFS_RMAP_EXTENT_ALLOC:
433 case XFS_RMAP_EXTENT_FREE:
439 if (!XFS_RMAP_NON_INODE_OWNER(map->me_owner) &&
440 !xfs_verify_ino(mp, map->me_owner))
443 if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
446 return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
450 xfs_rui_recover_work(
451 struct xfs_mount *mp,
452 struct xfs_defer_pending *dfp,
453 const struct xfs_map_extent *map)
455 struct xfs_rmap_intent *ri;
457 ri = kmem_cache_alloc(xfs_rmap_intent_cache, GFP_NOFS | __GFP_NOFAIL);
459 switch (map->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
460 case XFS_RMAP_EXTENT_MAP:
461 ri->ri_type = XFS_RMAP_MAP;
463 case XFS_RMAP_EXTENT_MAP_SHARED:
464 ri->ri_type = XFS_RMAP_MAP_SHARED;
466 case XFS_RMAP_EXTENT_UNMAP:
467 ri->ri_type = XFS_RMAP_UNMAP;
469 case XFS_RMAP_EXTENT_UNMAP_SHARED:
470 ri->ri_type = XFS_RMAP_UNMAP_SHARED;
472 case XFS_RMAP_EXTENT_CONVERT:
473 ri->ri_type = XFS_RMAP_CONVERT;
475 case XFS_RMAP_EXTENT_CONVERT_SHARED:
476 ri->ri_type = XFS_RMAP_CONVERT_SHARED;
478 case XFS_RMAP_EXTENT_ALLOC:
479 ri->ri_type = XFS_RMAP_ALLOC;
481 case XFS_RMAP_EXTENT_FREE:
482 ri->ri_type = XFS_RMAP_FREE;
489 ri->ri_owner = map->me_owner;
490 ri->ri_whichfork = (map->me_flags & XFS_RMAP_EXTENT_ATTR_FORK) ?
491 XFS_ATTR_FORK : XFS_DATA_FORK;
492 ri->ri_bmap.br_startblock = map->me_startblock;
493 ri->ri_bmap.br_startoff = map->me_startoff;
494 ri->ri_bmap.br_blockcount = map->me_len;
495 ri->ri_bmap.br_state = (map->me_flags & XFS_RMAP_EXTENT_UNWRITTEN) ?
496 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
497 xfs_rmap_update_get_group(mp, ri);
499 xfs_defer_add_item(dfp, &ri->ri_list);
503 * Process an rmap update intent item that was recovered from the log.
504 * We need to update the rmapbt.
507 xfs_rmap_recover_work(
508 struct xfs_defer_pending *dfp,
509 struct list_head *capture_list)
511 struct xfs_trans_res resv;
512 struct xfs_log_item *lip = dfp->dfp_intent;
513 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
514 struct xfs_trans *tp;
515 struct xfs_mount *mp = lip->li_log->l_mp;
520 * First check the validity of the extents described by the
521 * RUI. If any are bad, then assume that all are bad and
524 for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
525 if (!xfs_rui_validate_map(mp,
526 &ruip->rui_format.rui_extents[i])) {
527 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
529 sizeof(ruip->rui_format));
530 return -EFSCORRUPTED;
533 xfs_rui_recover_work(mp, dfp, &ruip->rui_format.rui_extents[i]);
536 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
537 error = xfs_trans_alloc(mp, &resv, mp->m_rmap_maxlevels, 0,
538 XFS_TRANS_RESERVE, &tp);
542 error = xlog_recover_finish_intent(tp, dfp);
543 if (error == -EFSCORRUPTED)
544 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
546 sizeof(ruip->rui_format));
550 return xfs_defer_ops_capture_and_commit(tp, capture_list);
553 xfs_trans_cancel(tp);
557 /* Relog an intent item to push the log tail forward. */
558 static struct xfs_log_item *
559 xfs_rmap_relog_intent(
560 struct xfs_trans *tp,
561 struct xfs_log_item *intent,
562 struct xfs_log_item *done_item)
564 struct xfs_rui_log_item *ruip;
565 struct xfs_map_extent *map;
568 count = RUI_ITEM(intent)->rui_format.rui_nextents;
569 map = RUI_ITEM(intent)->rui_format.rui_extents;
571 ruip = xfs_rui_init(tp->t_mountp, count);
572 memcpy(ruip->rui_format.rui_extents, map, count * sizeof(*map));
573 atomic_set(&ruip->rui_next_extent, count);
575 return &ruip->rui_item;
578 const struct xfs_defer_op_type xfs_rmap_update_defer_type = {
580 .max_items = XFS_RUI_MAX_FAST_EXTENTS,
581 .create_intent = xfs_rmap_update_create_intent,
582 .abort_intent = xfs_rmap_update_abort_intent,
583 .create_done = xfs_rmap_update_create_done,
584 .finish_item = xfs_rmap_update_finish_item,
585 .finish_cleanup = xfs_rmap_finish_one_cleanup,
586 .cancel_item = xfs_rmap_update_cancel_item,
587 .recover_work = xfs_rmap_recover_work,
588 .relog_intent = xfs_rmap_relog_intent,
593 struct xfs_log_item *lip,
596 return RUI_ITEM(lip)->rui_format.rui_id == intent_id;
599 static const struct xfs_item_ops xfs_rui_item_ops = {
600 .flags = XFS_ITEM_INTENT,
601 .iop_size = xfs_rui_item_size,
602 .iop_format = xfs_rui_item_format,
603 .iop_unpin = xfs_rui_item_unpin,
604 .iop_release = xfs_rui_item_release,
605 .iop_match = xfs_rui_item_match,
610 struct xfs_rui_log_format *dst,
611 const struct xfs_rui_log_format *src)
615 memcpy(dst, src, offsetof(struct xfs_rui_log_format, rui_extents));
617 for (i = 0; i < src->rui_nextents; i++)
618 memcpy(&dst->rui_extents[i], &src->rui_extents[i],
619 sizeof(struct xfs_map_extent));
623 * This routine is called to create an in-core extent rmap update
624 * item from the rui format structure which was logged on disk.
625 * It allocates an in-core rui, copies the extents from the format
626 * structure into it, and adds the rui to the AIL with the given
630 xlog_recover_rui_commit_pass2(
632 struct list_head *buffer_list,
633 struct xlog_recover_item *item,
636 struct xfs_mount *mp = log->l_mp;
637 struct xfs_rui_log_item *ruip;
638 struct xfs_rui_log_format *rui_formatp;
641 rui_formatp = item->ri_buf[0].i_addr;
643 if (item->ri_buf[0].i_len < xfs_rui_log_format_sizeof(0)) {
644 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
645 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
646 return -EFSCORRUPTED;
649 len = xfs_rui_log_format_sizeof(rui_formatp->rui_nextents);
650 if (item->ri_buf[0].i_len != len) {
651 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
652 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
653 return -EFSCORRUPTED;
656 ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
657 xfs_rui_copy_format(&ruip->rui_format, rui_formatp);
658 atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
660 xlog_recover_intent_item(log, &ruip->rui_item, lsn,
661 &xfs_rmap_update_defer_type);
665 const struct xlog_recover_item_ops xlog_rui_item_ops = {
666 .item_type = XFS_LI_RUI,
667 .commit_pass2 = xlog_recover_rui_commit_pass2,
671 * This routine is called when an RUD format structure is found in a committed
672 * transaction in the log. Its purpose is to cancel the corresponding RUI if it
673 * was still in the log. To do this it searches the AIL for the RUI with an id
674 * equal to that in the RUD format structure. If we find it we drop the RUD
675 * reference, which removes the RUI from the AIL and frees it.
678 xlog_recover_rud_commit_pass2(
680 struct list_head *buffer_list,
681 struct xlog_recover_item *item,
684 struct xfs_rud_log_format *rud_formatp;
686 rud_formatp = item->ri_buf[0].i_addr;
687 if (item->ri_buf[0].i_len != sizeof(struct xfs_rud_log_format)) {
688 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
689 rud_formatp, item->ri_buf[0].i_len);
690 return -EFSCORRUPTED;
693 xlog_recover_release_intent(log, XFS_LI_RUI, rud_formatp->rud_rui_id);
697 const struct xlog_recover_item_ops xlog_rud_item_ops = {
698 .item_type = XFS_LI_RUD,
699 .commit_pass2 = xlog_recover_rud_commit_pass2,