1 // SPDX-License-Identifier: GPL-2.0+
3 * Copyright (C) 2016 Oracle. All Rights Reserved.
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 #include "xfs_trace.h"
30 struct kmem_cache *xfs_bui_cache;
31 struct kmem_cache *xfs_bud_cache;
33 static const struct xfs_item_ops xfs_bui_item_ops;
35 static inline struct xfs_bui_log_item *BUI_ITEM(struct xfs_log_item *lip)
37 return container_of(lip, struct xfs_bui_log_item, bui_item);
42 struct xfs_bui_log_item *buip)
44 kvfree(buip->bui_item.li_lv_shadow);
45 kmem_cache_free(xfs_bui_cache, buip);
49 * Freeing the BUI requires that we remove it from the AIL if it has already
50 * been placed there. However, the BUI may not yet have been placed in the AIL
51 * when called by xfs_bui_release() from BUD processing due to the ordering of
52 * committed vs unpin operations in bulk insert operations. Hence the reference
53 * count to ensure only the last caller frees the BUI.
57 struct xfs_bui_log_item *buip)
59 ASSERT(atomic_read(&buip->bui_refcount) > 0);
60 if (!atomic_dec_and_test(&buip->bui_refcount))
63 xfs_trans_ail_delete(&buip->bui_item, 0);
64 xfs_bui_item_free(buip);
70 struct xfs_log_item *lip,
74 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
77 *nbytes += xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents);
81 * This is called to fill in the vector of log iovecs for the
82 * given bui log item. We use only 1 iovec, and we point that
83 * at the bui_log_format structure embedded in the bui item.
84 * It is at this point that we assert that all of the extent
85 * slots in the bui item have been filled.
89 struct xfs_log_item *lip,
90 struct xfs_log_vec *lv)
92 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
93 struct xfs_log_iovec *vecp = NULL;
95 ASSERT(atomic_read(&buip->bui_next_extent) ==
96 buip->bui_format.bui_nextents);
98 buip->bui_format.bui_type = XFS_LI_BUI;
99 buip->bui_format.bui_size = 1;
101 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUI_FORMAT, &buip->bui_format,
102 xfs_bui_log_format_sizeof(buip->bui_format.bui_nextents));
106 * The unpin operation is the last place an BUI is manipulated in the log. It is
107 * either inserted in the AIL or aborted in the event of a log I/O error. In
108 * either case, the BUI transaction has been successfully committed to make it
109 * this far. Therefore, we expect whoever committed the BUI to either construct
110 * and commit the BUD or drop the BUD's reference in the event of error. Simply
111 * drop the log's BUI reference now that the log is done with it.
115 struct xfs_log_item *lip,
118 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
120 xfs_bui_release(buip);
124 * The BUI has been either committed or aborted if the transaction has been
125 * cancelled. If the transaction was cancelled, an BUD isn't going to be
126 * constructed and thus we free the BUI here directly.
129 xfs_bui_item_release(
130 struct xfs_log_item *lip)
132 xfs_bui_release(BUI_ITEM(lip));
136 * Allocate and initialize an bui item with the given number of extents.
138 STATIC struct xfs_bui_log_item *
140 struct xfs_mount *mp)
143 struct xfs_bui_log_item *buip;
145 buip = kmem_cache_zalloc(xfs_bui_cache, GFP_KERNEL | __GFP_NOFAIL);
147 xfs_log_item_init(mp, &buip->bui_item, XFS_LI_BUI, &xfs_bui_item_ops);
148 buip->bui_format.bui_nextents = XFS_BUI_MAX_FAST_EXTENTS;
149 buip->bui_format.bui_id = (uintptr_t)(void *)buip;
150 atomic_set(&buip->bui_next_extent, 0);
151 atomic_set(&buip->bui_refcount, 2);
156 static inline struct xfs_bud_log_item *BUD_ITEM(struct xfs_log_item *lip)
158 return container_of(lip, struct xfs_bud_log_item, bud_item);
163 struct xfs_log_item *lip,
168 *nbytes += sizeof(struct xfs_bud_log_format);
172 * This is called to fill in the vector of log iovecs for the
173 * given bud log item. We use only 1 iovec, and we point that
174 * at the bud_log_format structure embedded in the bud item.
175 * It is at this point that we assert that all of the extent
176 * slots in the bud item have been filled.
180 struct xfs_log_item *lip,
181 struct xfs_log_vec *lv)
183 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
184 struct xfs_log_iovec *vecp = NULL;
186 budp->bud_format.bud_type = XFS_LI_BUD;
187 budp->bud_format.bud_size = 1;
189 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_BUD_FORMAT, &budp->bud_format,
190 sizeof(struct xfs_bud_log_format));
194 * The BUD is either committed or aborted if the transaction is cancelled. If
195 * the transaction is cancelled, drop our reference to the BUI and free the
199 xfs_bud_item_release(
200 struct xfs_log_item *lip)
202 struct xfs_bud_log_item *budp = BUD_ITEM(lip);
204 xfs_bui_release(budp->bud_buip);
205 kvfree(budp->bud_item.li_lv_shadow);
206 kmem_cache_free(xfs_bud_cache, budp);
209 static struct xfs_log_item *
211 struct xfs_log_item *lip)
213 return &BUD_ITEM(lip)->bud_buip->bui_item;
216 static const struct xfs_item_ops xfs_bud_item_ops = {
217 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED |
218 XFS_ITEM_INTENT_DONE,
219 .iop_size = xfs_bud_item_size,
220 .iop_format = xfs_bud_item_format,
221 .iop_release = xfs_bud_item_release,
222 .iop_intent = xfs_bud_item_intent,
225 static inline struct xfs_bmap_intent *bi_entry(const struct list_head *e)
227 return list_entry(e, struct xfs_bmap_intent, bi_list);
230 /* Sort bmap intents by inode. */
232 xfs_bmap_update_diff_items(
234 const struct list_head *a,
235 const struct list_head *b)
237 struct xfs_bmap_intent *ba = bi_entry(a);
238 struct xfs_bmap_intent *bb = bi_entry(b);
240 return ba->bi_owner->i_ino - bb->bi_owner->i_ino;
243 /* Log bmap updates in the intent item. */
245 xfs_bmap_update_log_item(
246 struct xfs_trans *tp,
247 struct xfs_bui_log_item *buip,
248 struct xfs_bmap_intent *bi)
251 struct xfs_map_extent *map;
254 * atomic_inc_return gives us the value after the increment;
255 * we want to use it as an array index so we need to subtract 1 from
258 next_extent = atomic_inc_return(&buip->bui_next_extent) - 1;
259 ASSERT(next_extent < buip->bui_format.bui_nextents);
260 map = &buip->bui_format.bui_extents[next_extent];
261 map->me_owner = bi->bi_owner->i_ino;
262 map->me_startblock = bi->bi_bmap.br_startblock;
263 map->me_startoff = bi->bi_bmap.br_startoff;
264 map->me_len = bi->bi_bmap.br_blockcount;
266 switch (bi->bi_type) {
269 map->me_flags = bi->bi_type;
274 if (bi->bi_bmap.br_state == XFS_EXT_UNWRITTEN)
275 map->me_flags |= XFS_BMAP_EXTENT_UNWRITTEN;
276 if (bi->bi_whichfork == XFS_ATTR_FORK)
277 map->me_flags |= XFS_BMAP_EXTENT_ATTR_FORK;
278 if (xfs_ifork_is_realtime(bi->bi_owner, bi->bi_whichfork))
279 map->me_flags |= XFS_BMAP_EXTENT_REALTIME;
282 static struct xfs_log_item *
283 xfs_bmap_update_create_intent(
284 struct xfs_trans *tp,
285 struct list_head *items,
289 struct xfs_mount *mp = tp->t_mountp;
290 struct xfs_bui_log_item *buip = xfs_bui_init(mp);
291 struct xfs_bmap_intent *bi;
293 ASSERT(count == XFS_BUI_MAX_FAST_EXTENTS);
296 list_sort(mp, items, xfs_bmap_update_diff_items);
297 list_for_each_entry(bi, items, bi_list)
298 xfs_bmap_update_log_item(tp, buip, bi);
299 return &buip->bui_item;
302 /* Get an BUD so we can process all the deferred bmap updates. */
303 static struct xfs_log_item *
304 xfs_bmap_update_create_done(
305 struct xfs_trans *tp,
306 struct xfs_log_item *intent,
309 struct xfs_bui_log_item *buip = BUI_ITEM(intent);
310 struct xfs_bud_log_item *budp;
312 budp = kmem_cache_zalloc(xfs_bud_cache, GFP_KERNEL | __GFP_NOFAIL);
313 xfs_log_item_init(tp->t_mountp, &budp->bud_item, XFS_LI_BUD,
315 budp->bud_buip = buip;
316 budp->bud_format.bud_bui_id = buip->bui_format.bui_id;
318 return &budp->bud_item;
321 /* Take a passive ref to the AG containing the space we're mapping. */
323 xfs_bmap_update_get_group(
324 struct xfs_mount *mp,
325 struct xfs_bmap_intent *bi)
327 if (xfs_ifork_is_realtime(bi->bi_owner, bi->bi_whichfork))
331 * Bump the intent count on behalf of the deferred rmap and refcount
332 * intent items that that we can queue when we finish this bmap work.
333 * This new intent item will bump the intent count before the bmap
334 * intent drops the intent count, ensuring that the intent count
335 * remains nonzero across the transaction roll.
337 bi->bi_pag = xfs_perag_intent_get(mp, bi->bi_bmap.br_startblock);
340 /* Add this deferred BUI to the transaction. */
343 struct xfs_trans *tp,
344 struct xfs_bmap_intent *bi)
346 trace_xfs_bmap_defer(bi);
348 xfs_bmap_update_get_group(tp->t_mountp, bi);
351 * Ensure the deferred mapping is pre-recorded in i_delayed_blks.
353 * Otherwise stat can report zero blocks for an inode that actually has
354 * data when the entire mapping is in the process of being overwritten
355 * using the out of place write path. This is undone in xfs_bmapi_remap
356 * after it has incremented di_nblocks for a successful operation.
358 if (bi->bi_type == XFS_BMAP_MAP)
359 bi->bi_owner->i_delayed_blks += bi->bi_bmap.br_blockcount;
360 xfs_defer_add(tp, &bi->bi_list, &xfs_bmap_update_defer_type);
363 /* Release a passive AG ref after finishing mapping work. */
365 xfs_bmap_update_put_group(
366 struct xfs_bmap_intent *bi)
368 if (xfs_ifork_is_realtime(bi->bi_owner, bi->bi_whichfork))
371 xfs_perag_intent_put(bi->bi_pag);
374 /* Cancel a deferred bmap update. */
376 xfs_bmap_update_cancel_item(
377 struct list_head *item)
379 struct xfs_bmap_intent *bi = bi_entry(item);
381 if (bi->bi_type == XFS_BMAP_MAP)
382 bi->bi_owner->i_delayed_blks -= bi->bi_bmap.br_blockcount;
384 xfs_bmap_update_put_group(bi);
385 kmem_cache_free(xfs_bmap_intent_cache, bi);
388 /* Process a deferred bmap update. */
390 xfs_bmap_update_finish_item(
391 struct xfs_trans *tp,
392 struct xfs_log_item *done,
393 struct list_head *item,
394 struct xfs_btree_cur **state)
396 struct xfs_bmap_intent *bi = bi_entry(item);
399 error = xfs_bmap_finish_one(tp, bi);
400 if (!error && bi->bi_bmap.br_blockcount > 0) {
401 ASSERT(bi->bi_type == XFS_BMAP_UNMAP);
405 xfs_bmap_update_cancel_item(item);
409 /* Abort all pending BUIs. */
411 xfs_bmap_update_abort_intent(
412 struct xfs_log_item *intent)
414 xfs_bui_release(BUI_ITEM(intent));
417 /* Is this recovered BUI ok? */
420 struct xfs_mount *mp,
421 struct xfs_bui_log_item *buip)
423 struct xfs_map_extent *map;
425 /* Only one mapping operation per BUI... */
426 if (buip->bui_format.bui_nextents != XFS_BUI_MAX_FAST_EXTENTS)
429 map = &buip->bui_format.bui_extents[0];
431 if (map->me_flags & ~XFS_BMAP_EXTENT_FLAGS)
434 switch (map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK) {
442 if (!xfs_verify_ino(mp, map->me_owner))
445 if (!xfs_verify_fileext(mp, map->me_startoff, map->me_len))
448 if (map->me_flags & XFS_BMAP_EXTENT_REALTIME)
449 return xfs_verify_rtbext(mp, map->me_startblock, map->me_len);
451 return xfs_verify_fsbext(mp, map->me_startblock, map->me_len);
454 static inline struct xfs_bmap_intent *
455 xfs_bui_recover_work(
456 struct xfs_mount *mp,
457 struct xfs_defer_pending *dfp,
458 struct xfs_inode **ipp,
459 struct xfs_map_extent *map)
461 struct xfs_bmap_intent *bi;
464 error = xlog_recover_iget(mp, map->me_owner, ipp);
466 return ERR_PTR(error);
468 bi = kmem_cache_zalloc(xfs_bmap_intent_cache,
469 GFP_KERNEL | __GFP_NOFAIL);
470 bi->bi_whichfork = (map->me_flags & XFS_BMAP_EXTENT_ATTR_FORK) ?
471 XFS_ATTR_FORK : XFS_DATA_FORK;
472 bi->bi_type = map->me_flags & XFS_BMAP_EXTENT_TYPE_MASK;
473 bi->bi_bmap.br_startblock = map->me_startblock;
474 bi->bi_bmap.br_startoff = map->me_startoff;
475 bi->bi_bmap.br_blockcount = map->me_len;
476 bi->bi_bmap.br_state = (map->me_flags & XFS_BMAP_EXTENT_UNWRITTEN) ?
477 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
479 xfs_bmap_update_get_group(mp, bi);
481 /* see xfs_bmap_defer_add for details */
482 if (bi->bi_type == XFS_BMAP_MAP)
483 bi->bi_owner->i_delayed_blks += bi->bi_bmap.br_blockcount;
484 xfs_defer_add_item(dfp, &bi->bi_list);
489 * Process a bmap update intent item that was recovered from the log.
490 * We need to update some inode's bmbt.
493 xfs_bmap_recover_work(
494 struct xfs_defer_pending *dfp,
495 struct list_head *capture_list)
497 struct xfs_trans_res resv;
498 struct xfs_log_item *lip = dfp->dfp_intent;
499 struct xfs_bui_log_item *buip = BUI_ITEM(lip);
500 struct xfs_trans *tp;
501 struct xfs_inode *ip = NULL;
502 struct xfs_mount *mp = lip->li_log->l_mp;
503 struct xfs_map_extent *map;
504 struct xfs_bmap_intent *work;
508 if (!xfs_bui_validate(mp, buip)) {
509 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
510 &buip->bui_format, sizeof(buip->bui_format));
511 return -EFSCORRUPTED;
514 map = &buip->bui_format.bui_extents[0];
515 work = xfs_bui_recover_work(mp, dfp, &ip, map);
517 return PTR_ERR(work);
519 /* Allocate transaction and do the work. */
520 resv = xlog_recover_resv(&M_RES(mp)->tr_itruncate);
521 error = xfs_trans_alloc(mp, &resv,
522 XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK), 0, 0, &tp);
526 xfs_ilock(ip, XFS_ILOCK_EXCL);
527 xfs_trans_ijoin(tp, ip, 0);
529 if (!!(map->me_flags & XFS_BMAP_EXTENT_REALTIME) !=
530 xfs_ifork_is_realtime(ip, work->bi_whichfork)) {
531 error = -EFSCORRUPTED;
535 if (work->bi_type == XFS_BMAP_MAP)
536 iext_delta = XFS_IEXT_ADD_NOSPLIT_CNT;
538 iext_delta = XFS_IEXT_PUNCH_HOLE_CNT;
540 error = xfs_iext_count_extend(tp, ip, work->bi_whichfork, iext_delta);
544 error = xlog_recover_finish_intent(tp, dfp);
545 if (error == -EFSCORRUPTED)
546 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
547 &buip->bui_format, sizeof(buip->bui_format));
552 * Commit transaction, which frees the transaction and saves the inode
553 * for later replay activities.
555 error = xfs_defer_ops_capture_and_commit(tp, capture_list);
559 xfs_iunlock(ip, XFS_ILOCK_EXCL);
564 xfs_trans_cancel(tp);
566 xfs_iunlock(ip, XFS_ILOCK_EXCL);
572 /* Relog an intent item to push the log tail forward. */
573 static struct xfs_log_item *
574 xfs_bmap_relog_intent(
575 struct xfs_trans *tp,
576 struct xfs_log_item *intent,
577 struct xfs_log_item *done_item)
579 struct xfs_bui_log_item *buip;
580 struct xfs_map_extent *map;
583 count = BUI_ITEM(intent)->bui_format.bui_nextents;
584 map = BUI_ITEM(intent)->bui_format.bui_extents;
586 buip = xfs_bui_init(tp->t_mountp);
587 memcpy(buip->bui_format.bui_extents, map, count * sizeof(*map));
588 atomic_set(&buip->bui_next_extent, count);
590 return &buip->bui_item;
593 const struct xfs_defer_op_type xfs_bmap_update_defer_type = {
595 .max_items = XFS_BUI_MAX_FAST_EXTENTS,
596 .create_intent = xfs_bmap_update_create_intent,
597 .abort_intent = xfs_bmap_update_abort_intent,
598 .create_done = xfs_bmap_update_create_done,
599 .finish_item = xfs_bmap_update_finish_item,
600 .cancel_item = xfs_bmap_update_cancel_item,
601 .recover_work = xfs_bmap_recover_work,
602 .relog_intent = xfs_bmap_relog_intent,
607 struct xfs_log_item *lip,
610 return BUI_ITEM(lip)->bui_format.bui_id == intent_id;
613 static const struct xfs_item_ops xfs_bui_item_ops = {
614 .flags = XFS_ITEM_INTENT,
615 .iop_size = xfs_bui_item_size,
616 .iop_format = xfs_bui_item_format,
617 .iop_unpin = xfs_bui_item_unpin,
618 .iop_release = xfs_bui_item_release,
619 .iop_match = xfs_bui_item_match,
624 struct xfs_bui_log_format *dst,
625 const struct xfs_bui_log_format *src)
629 memcpy(dst, src, offsetof(struct xfs_bui_log_format, bui_extents));
631 for (i = 0; i < src->bui_nextents; i++)
632 memcpy(&dst->bui_extents[i], &src->bui_extents[i],
633 sizeof(struct xfs_map_extent));
637 * This routine is called to create an in-core extent bmap update
638 * item from the bui format structure which was logged on disk.
639 * It allocates an in-core bui, copies the extents from the format
640 * structure into it, and adds the bui to the AIL with the given
644 xlog_recover_bui_commit_pass2(
646 struct list_head *buffer_list,
647 struct xlog_recover_item *item,
650 struct xfs_mount *mp = log->l_mp;
651 struct xfs_bui_log_item *buip;
652 struct xfs_bui_log_format *bui_formatp;
655 bui_formatp = item->ri_buf[0].i_addr;
657 if (item->ri_buf[0].i_len < xfs_bui_log_format_sizeof(0)) {
658 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
659 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
660 return -EFSCORRUPTED;
663 if (bui_formatp->bui_nextents != XFS_BUI_MAX_FAST_EXTENTS) {
664 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
665 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
666 return -EFSCORRUPTED;
669 len = xfs_bui_log_format_sizeof(bui_formatp->bui_nextents);
670 if (item->ri_buf[0].i_len != len) {
671 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
672 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
673 return -EFSCORRUPTED;
676 buip = xfs_bui_init(mp);
677 xfs_bui_copy_format(&buip->bui_format, bui_formatp);
678 atomic_set(&buip->bui_next_extent, bui_formatp->bui_nextents);
680 xlog_recover_intent_item(log, &buip->bui_item, lsn,
681 &xfs_bmap_update_defer_type);
685 const struct xlog_recover_item_ops xlog_bui_item_ops = {
686 .item_type = XFS_LI_BUI,
687 .commit_pass2 = xlog_recover_bui_commit_pass2,
691 * This routine is called when an BUD format structure is found in a committed
692 * transaction in the log. Its purpose is to cancel the corresponding BUI if it
693 * was still in the log. To do this it searches the AIL for the BUI with an id
694 * equal to that in the BUD format structure. If we find it we drop the BUD
695 * reference, which removes the BUI from the AIL and frees it.
698 xlog_recover_bud_commit_pass2(
700 struct list_head *buffer_list,
701 struct xlog_recover_item *item,
704 struct xfs_bud_log_format *bud_formatp;
706 bud_formatp = item->ri_buf[0].i_addr;
707 if (item->ri_buf[0].i_len != sizeof(struct xfs_bud_log_format)) {
708 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, log->l_mp,
709 item->ri_buf[0].i_addr, item->ri_buf[0].i_len);
710 return -EFSCORRUPTED;
713 xlog_recover_release_intent(log, XFS_LI_BUI, bud_formatp->bud_bui_id);
717 const struct xlog_recover_item_ops xlog_bud_item_ops = {
718 .item_type = XFS_LI_BUD,
719 .commit_pass2 = xlog_recover_bud_commit_pass2,