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"
22 kmem_zone_t *xfs_rui_zone;
23 kmem_zone_t *xfs_rud_zone;
25 static inline struct xfs_rui_log_item *RUI_ITEM(struct xfs_log_item *lip)
27 return container_of(lip, struct xfs_rui_log_item, rui_item);
32 struct xfs_rui_log_item *ruip)
34 if (ruip->rui_format.rui_nextents > XFS_RUI_MAX_FAST_EXTENTS)
37 kmem_zone_free(xfs_rui_zone, ruip);
41 * Freeing the RUI requires that we remove it from the AIL if it has already
42 * been placed there. However, the RUI may not yet have been placed in the AIL
43 * when called by xfs_rui_release() from RUD processing due to the ordering of
44 * committed vs unpin operations in bulk insert operations. Hence the reference
45 * count to ensure only the last caller frees the RUI.
49 struct xfs_rui_log_item *ruip)
51 ASSERT(atomic_read(&ruip->rui_refcount) > 0);
52 if (atomic_dec_and_test(&ruip->rui_refcount)) {
53 xfs_trans_ail_remove(&ruip->rui_item, SHUTDOWN_LOG_IO_ERROR);
54 xfs_rui_item_free(ruip);
60 struct xfs_log_item *lip,
64 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
67 *nbytes += xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents);
71 * This is called to fill in the vector of log iovecs for the
72 * given rui log item. We use only 1 iovec, and we point that
73 * at the rui_log_format structure embedded in the rui item.
74 * It is at this point that we assert that all of the extent
75 * slots in the rui item have been filled.
79 struct xfs_log_item *lip,
80 struct xfs_log_vec *lv)
82 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
83 struct xfs_log_iovec *vecp = NULL;
85 ASSERT(atomic_read(&ruip->rui_next_extent) ==
86 ruip->rui_format.rui_nextents);
88 ruip->rui_format.rui_type = XFS_LI_RUI;
89 ruip->rui_format.rui_size = 1;
91 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUI_FORMAT, &ruip->rui_format,
92 xfs_rui_log_format_sizeof(ruip->rui_format.rui_nextents));
96 * The unpin operation is the last place an RUI is manipulated in the log. It is
97 * either inserted in the AIL or aborted in the event of a log I/O error. In
98 * either case, the RUI transaction has been successfully committed to make it
99 * this far. Therefore, we expect whoever committed the RUI to either construct
100 * and commit the RUD or drop the RUD's reference in the event of error. Simply
101 * drop the log's RUI reference now that the log is done with it.
105 struct xfs_log_item *lip,
108 struct xfs_rui_log_item *ruip = RUI_ITEM(lip);
110 xfs_rui_release(ruip);
114 * The RUI has been either committed or aborted if the transaction has been
115 * cancelled. If the transaction was cancelled, an RUD isn't going to be
116 * constructed and thus we free the RUI here directly.
119 xfs_rui_item_release(
120 struct xfs_log_item *lip)
122 xfs_rui_release(RUI_ITEM(lip));
125 static const struct xfs_item_ops xfs_rui_item_ops = {
126 .iop_size = xfs_rui_item_size,
127 .iop_format = xfs_rui_item_format,
128 .iop_unpin = xfs_rui_item_unpin,
129 .iop_release = xfs_rui_item_release,
133 * Allocate and initialize an rui item with the given number of extents.
135 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_zone_zalloc(xfs_rui_zone, 0);
149 xfs_log_item_init(mp, &ruip->rui_item, XFS_LI_RUI, &xfs_rui_item_ops);
150 ruip->rui_format.rui_nextents = nextents;
151 ruip->rui_format.rui_id = (uintptr_t)(void *)ruip;
152 atomic_set(&ruip->rui_next_extent, 0);
153 atomic_set(&ruip->rui_refcount, 2);
159 * Copy an RUI format buffer from the given buf, and into the destination
160 * RUI format structure. The RUI/RUD items were designed not to need any
161 * special alignment handling.
165 struct xfs_log_iovec *buf,
166 struct xfs_rui_log_format *dst_rui_fmt)
168 struct xfs_rui_log_format *src_rui_fmt;
171 src_rui_fmt = buf->i_addr;
172 len = xfs_rui_log_format_sizeof(src_rui_fmt->rui_nextents);
174 if (buf->i_len != len)
175 return -EFSCORRUPTED;
177 memcpy(dst_rui_fmt, src_rui_fmt, len);
181 static inline struct xfs_rud_log_item *RUD_ITEM(struct xfs_log_item *lip)
183 return container_of(lip, struct xfs_rud_log_item, rud_item);
188 struct xfs_log_item *lip,
193 *nbytes += sizeof(struct xfs_rud_log_format);
197 * This is called to fill in the vector of log iovecs for the
198 * given rud log item. We use only 1 iovec, and we point that
199 * at the rud_log_format structure embedded in the rud item.
200 * It is at this point that we assert that all of the extent
201 * slots in the rud item have been filled.
205 struct xfs_log_item *lip,
206 struct xfs_log_vec *lv)
208 struct xfs_rud_log_item *rudp = RUD_ITEM(lip);
209 struct xfs_log_iovec *vecp = NULL;
211 rudp->rud_format.rud_type = XFS_LI_RUD;
212 rudp->rud_format.rud_size = 1;
214 xlog_copy_iovec(lv, &vecp, XLOG_REG_TYPE_RUD_FORMAT, &rudp->rud_format,
215 sizeof(struct xfs_rud_log_format));
219 * The RUD is either committed or aborted if the transaction is cancelled. If
220 * the transaction is cancelled, drop our reference to the RUI and free the
224 xfs_rud_item_release(
225 struct xfs_log_item *lip)
227 struct xfs_rud_log_item *rudp = RUD_ITEM(lip);
229 xfs_rui_release(rudp->rud_ruip);
230 kmem_zone_free(xfs_rud_zone, rudp);
233 static const struct xfs_item_ops xfs_rud_item_ops = {
234 .flags = XFS_ITEM_RELEASE_WHEN_COMMITTED,
235 .iop_size = xfs_rud_item_size,
236 .iop_format = xfs_rud_item_format,
237 .iop_release = xfs_rud_item_release,
240 static struct xfs_rud_log_item *
242 struct xfs_trans *tp,
243 struct xfs_rui_log_item *ruip)
245 struct xfs_rud_log_item *rudp;
247 rudp = kmem_zone_zalloc(xfs_rud_zone, 0);
248 xfs_log_item_init(tp->t_mountp, &rudp->rud_item, XFS_LI_RUD,
250 rudp->rud_ruip = ruip;
251 rudp->rud_format.rud_rui_id = ruip->rui_format.rui_id;
253 xfs_trans_add_item(tp, &rudp->rud_item);
257 /* Set the map extent flags for this reverse mapping. */
259 xfs_trans_set_rmap_flags(
260 struct xfs_map_extent *rmap,
261 enum xfs_rmap_intent_type type,
266 if (state == XFS_EXT_UNWRITTEN)
267 rmap->me_flags |= XFS_RMAP_EXTENT_UNWRITTEN;
268 if (whichfork == XFS_ATTR_FORK)
269 rmap->me_flags |= XFS_RMAP_EXTENT_ATTR_FORK;
272 rmap->me_flags |= XFS_RMAP_EXTENT_MAP;
274 case XFS_RMAP_MAP_SHARED:
275 rmap->me_flags |= XFS_RMAP_EXTENT_MAP_SHARED;
278 rmap->me_flags |= XFS_RMAP_EXTENT_UNMAP;
280 case XFS_RMAP_UNMAP_SHARED:
281 rmap->me_flags |= XFS_RMAP_EXTENT_UNMAP_SHARED;
283 case XFS_RMAP_CONVERT:
284 rmap->me_flags |= XFS_RMAP_EXTENT_CONVERT;
286 case XFS_RMAP_CONVERT_SHARED:
287 rmap->me_flags |= XFS_RMAP_EXTENT_CONVERT_SHARED;
290 rmap->me_flags |= XFS_RMAP_EXTENT_ALLOC;
293 rmap->me_flags |= XFS_RMAP_EXTENT_FREE;
301 * Finish an rmap update and log it to the RUD. Note that the transaction is
302 * marked dirty regardless of whether the rmap update succeeds or fails to
303 * support the RUI/RUD lifecycle rules.
306 xfs_trans_log_finish_rmap_update(
307 struct xfs_trans *tp,
308 struct xfs_rud_log_item *rudp,
309 enum xfs_rmap_intent_type type,
312 xfs_fileoff_t startoff,
313 xfs_fsblock_t startblock,
314 xfs_filblks_t blockcount,
316 struct xfs_btree_cur **pcur)
320 error = xfs_rmap_finish_one(tp, type, owner, whichfork, startoff,
321 startblock, blockcount, state, pcur);
324 * Mark the transaction dirty, even on error. This ensures the
325 * transaction is aborted, which:
327 * 1.) releases the RUI and frees the RUD
328 * 2.) shuts down the filesystem
330 tp->t_flags |= XFS_TRANS_DIRTY;
331 set_bit(XFS_LI_DIRTY, &rudp->rud_item.li_flags);
336 /* Sort rmap intents by AG. */
338 xfs_rmap_update_diff_items(
343 struct xfs_mount *mp = priv;
344 struct xfs_rmap_intent *ra;
345 struct xfs_rmap_intent *rb;
347 ra = container_of(a, struct xfs_rmap_intent, ri_list);
348 rb = container_of(b, struct xfs_rmap_intent, ri_list);
349 return XFS_FSB_TO_AGNO(mp, ra->ri_bmap.br_startblock) -
350 XFS_FSB_TO_AGNO(mp, rb->ri_bmap.br_startblock);
355 xfs_rmap_update_create_intent(
356 struct xfs_trans *tp,
359 struct xfs_rui_log_item *ruip;
364 ruip = xfs_rui_init(tp->t_mountp, count);
365 ASSERT(ruip != NULL);
368 * Get a log_item_desc to point at the new item.
370 xfs_trans_add_item(tp, &ruip->rui_item);
374 /* Log rmap updates in the intent item. */
376 xfs_rmap_update_log_item(
377 struct xfs_trans *tp,
379 struct list_head *item)
381 struct xfs_rui_log_item *ruip = intent;
382 struct xfs_rmap_intent *rmap;
384 struct xfs_map_extent *map;
386 rmap = container_of(item, struct xfs_rmap_intent, ri_list);
388 tp->t_flags |= XFS_TRANS_DIRTY;
389 set_bit(XFS_LI_DIRTY, &ruip->rui_item.li_flags);
392 * atomic_inc_return gives us the value after the increment;
393 * we want to use it as an array index so we need to subtract 1 from
396 next_extent = atomic_inc_return(&ruip->rui_next_extent) - 1;
397 ASSERT(next_extent < ruip->rui_format.rui_nextents);
398 map = &ruip->rui_format.rui_extents[next_extent];
399 map->me_owner = rmap->ri_owner;
400 map->me_startblock = rmap->ri_bmap.br_startblock;
401 map->me_startoff = rmap->ri_bmap.br_startoff;
402 map->me_len = rmap->ri_bmap.br_blockcount;
403 xfs_trans_set_rmap_flags(map, rmap->ri_type, rmap->ri_whichfork,
404 rmap->ri_bmap.br_state);
407 /* Get an RUD so we can process all the deferred rmap updates. */
409 xfs_rmap_update_create_done(
410 struct xfs_trans *tp,
414 return xfs_trans_get_rud(tp, intent);
417 /* Process a deferred rmap update. */
419 xfs_rmap_update_finish_item(
420 struct xfs_trans *tp,
421 struct list_head *item,
425 struct xfs_rmap_intent *rmap;
428 rmap = container_of(item, struct xfs_rmap_intent, ri_list);
429 error = xfs_trans_log_finish_rmap_update(tp, done_item,
431 rmap->ri_owner, rmap->ri_whichfork,
432 rmap->ri_bmap.br_startoff,
433 rmap->ri_bmap.br_startblock,
434 rmap->ri_bmap.br_blockcount,
435 rmap->ri_bmap.br_state,
436 (struct xfs_btree_cur **)state);
441 /* Clean up after processing deferred rmaps. */
443 xfs_rmap_update_finish_cleanup(
444 struct xfs_trans *tp,
448 struct xfs_btree_cur *rcur = state;
450 xfs_rmap_finish_one_cleanup(tp, rcur, error);
453 /* Abort all pending RUIs. */
455 xfs_rmap_update_abort_intent(
458 xfs_rui_release(intent);
461 /* Cancel a deferred rmap update. */
463 xfs_rmap_update_cancel_item(
464 struct list_head *item)
466 struct xfs_rmap_intent *rmap;
468 rmap = container_of(item, struct xfs_rmap_intent, ri_list);
472 const struct xfs_defer_op_type xfs_rmap_update_defer_type = {
473 .max_items = XFS_RUI_MAX_FAST_EXTENTS,
474 .diff_items = xfs_rmap_update_diff_items,
475 .create_intent = xfs_rmap_update_create_intent,
476 .abort_intent = xfs_rmap_update_abort_intent,
477 .log_item = xfs_rmap_update_log_item,
478 .create_done = xfs_rmap_update_create_done,
479 .finish_item = xfs_rmap_update_finish_item,
480 .finish_cleanup = xfs_rmap_update_finish_cleanup,
481 .cancel_item = xfs_rmap_update_cancel_item,
485 * Process an rmap update intent item that was recovered from the log.
486 * We need to update the rmapbt.
490 struct xfs_mount *mp,
491 struct xfs_rui_log_item *ruip)
495 struct xfs_map_extent *rmap;
496 xfs_fsblock_t startblock_fsb;
498 struct xfs_rud_log_item *rudp;
499 enum xfs_rmap_intent_type type;
502 struct xfs_trans *tp;
503 struct xfs_btree_cur *rcur = NULL;
505 ASSERT(!test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags));
508 * First check the validity of the extents described by the
509 * RUI. If any are bad, then assume that all are bad and
512 for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
513 rmap = &ruip->rui_format.rui_extents[i];
514 startblock_fsb = XFS_BB_TO_FSB(mp,
515 XFS_FSB_TO_DADDR(mp, rmap->me_startblock));
516 switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
517 case XFS_RMAP_EXTENT_MAP:
518 case XFS_RMAP_EXTENT_MAP_SHARED:
519 case XFS_RMAP_EXTENT_UNMAP:
520 case XFS_RMAP_EXTENT_UNMAP_SHARED:
521 case XFS_RMAP_EXTENT_CONVERT:
522 case XFS_RMAP_EXTENT_CONVERT_SHARED:
523 case XFS_RMAP_EXTENT_ALLOC:
524 case XFS_RMAP_EXTENT_FREE:
531 if (!op_ok || startblock_fsb == 0 ||
533 startblock_fsb >= mp->m_sb.sb_dblocks ||
534 rmap->me_len >= mp->m_sb.sb_agblocks ||
535 (rmap->me_flags & ~XFS_RMAP_EXTENT_FLAGS)) {
537 * This will pull the RUI from the AIL and
538 * free the memory associated with it.
540 set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags);
541 xfs_rui_release(ruip);
546 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate,
547 mp->m_rmap_maxlevels, 0, XFS_TRANS_RESERVE, &tp);
550 rudp = xfs_trans_get_rud(tp, ruip);
552 for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
553 rmap = &ruip->rui_format.rui_extents[i];
554 state = (rmap->me_flags & XFS_RMAP_EXTENT_UNWRITTEN) ?
555 XFS_EXT_UNWRITTEN : XFS_EXT_NORM;
556 whichfork = (rmap->me_flags & XFS_RMAP_EXTENT_ATTR_FORK) ?
557 XFS_ATTR_FORK : XFS_DATA_FORK;
558 switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
559 case XFS_RMAP_EXTENT_MAP:
562 case XFS_RMAP_EXTENT_MAP_SHARED:
563 type = XFS_RMAP_MAP_SHARED;
565 case XFS_RMAP_EXTENT_UNMAP:
566 type = XFS_RMAP_UNMAP;
568 case XFS_RMAP_EXTENT_UNMAP_SHARED:
569 type = XFS_RMAP_UNMAP_SHARED;
571 case XFS_RMAP_EXTENT_CONVERT:
572 type = XFS_RMAP_CONVERT;
574 case XFS_RMAP_EXTENT_CONVERT_SHARED:
575 type = XFS_RMAP_CONVERT_SHARED;
577 case XFS_RMAP_EXTENT_ALLOC:
578 type = XFS_RMAP_ALLOC;
580 case XFS_RMAP_EXTENT_FREE:
581 type = XFS_RMAP_FREE;
584 error = -EFSCORRUPTED;
587 error = xfs_trans_log_finish_rmap_update(tp, rudp, type,
588 rmap->me_owner, whichfork,
589 rmap->me_startoff, rmap->me_startblock,
590 rmap->me_len, state, &rcur);
596 xfs_rmap_finish_one_cleanup(tp, rcur, error);
597 set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags);
598 error = xfs_trans_commit(tp);
602 xfs_rmap_finish_one_cleanup(tp, rcur, error);
603 xfs_trans_cancel(tp);