1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_trans.h"
15 #include "xfs_buf_item.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_error.h"
18 #include "xfs_trace.h"
20 #include "xfs_inode.h"
23 kmem_zone_t *xfs_buf_item_zone;
25 static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
27 return container_of(lip, struct xfs_buf_log_item, bli_item);
30 STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
33 xfs_buf_log_format_size(
34 struct xfs_buf_log_format *blfp)
36 return offsetof(struct xfs_buf_log_format, blf_data_map) +
37 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
41 * This returns the number of log iovecs needed to log the
44 * It calculates this as 1 iovec for the buf log format structure
45 * and 1 for each stretch of non-contiguous chunks to be logged.
46 * Contiguous chunks are logged in a single iovec.
48 * If the XFS_BLI_STALE flag has been set, then log nothing.
51 xfs_buf_item_size_segment(
52 struct xfs_buf_log_item *bip,
53 struct xfs_buf_log_format *blfp,
57 struct xfs_buf *bp = bip->bli_buf;
61 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
66 * initial count for a dirty buffer is 2 vectors - the format structure
67 * and the first dirty region.
70 *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK;
72 while (last_bit != -1) {
74 * This takes the bit number to start looking from and
75 * returns the next set bit from there. It returns -1
76 * if there are no more bits set or the start bit is
77 * beyond the end of the bitmap.
79 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
82 * If we run out of bits, leave the loop,
83 * else if we find a new set of bits bump the number of vecs,
84 * else keep scanning the current set of bits.
88 } else if (next_bit != last_bit + 1) {
91 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
92 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
99 *nbytes += XFS_BLF_CHUNK;
104 * This returns the number of log iovecs needed to log the given buf log item.
106 * It calculates this as 1 iovec for the buf log format structure and 1 for each
107 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
110 * Discontiguous buffers need a format structure per region that that is being
111 * logged. This makes the changes in the buffer appear to log recovery as though
112 * they came from separate buffers, just like would occur if multiple buffers
113 * were used instead of a single discontiguous buffer. This enables
114 * discontiguous buffers to be in-memory constructs, completely transparent to
115 * what ends up on disk.
117 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
122 struct xfs_log_item *lip,
126 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
129 ASSERT(atomic_read(&bip->bli_refcount) > 0);
130 if (bip->bli_flags & XFS_BLI_STALE) {
132 * The buffer is stale, so all we need to log
133 * is the buf log format structure with the
136 trace_xfs_buf_item_size_stale(bip);
137 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
138 *nvecs += bip->bli_format_count;
139 for (i = 0; i < bip->bli_format_count; i++) {
140 *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
145 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
147 if (bip->bli_flags & XFS_BLI_ORDERED) {
149 * The buffer has been logged just to order it.
150 * It is not being included in the transaction
151 * commit, so no vectors are used at all.
153 trace_xfs_buf_item_size_ordered(bip);
154 *nvecs = XFS_LOG_VEC_ORDERED;
159 * the vector count is based on the number of buffer vectors we have
160 * dirty bits in. This will only be greater than one when we have a
161 * compound buffer with more than one segment dirty. Hence for compound
162 * buffers we need to track which segment the dirty bits correspond to,
163 * and when we move from one segment to the next increment the vector
164 * count for the extra buf log format structure that will need to be
167 for (i = 0; i < bip->bli_format_count; i++) {
168 xfs_buf_item_size_segment(bip, &bip->bli_formats[i],
171 trace_xfs_buf_item_size(bip);
175 xfs_buf_item_copy_iovec(
176 struct xfs_log_vec *lv,
177 struct xfs_log_iovec **vecp,
183 offset += first_bit * XFS_BLF_CHUNK;
184 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
185 xfs_buf_offset(bp, offset),
186 nbits * XFS_BLF_CHUNK);
190 xfs_buf_item_straddle(
196 return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) !=
197 (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) +
202 xfs_buf_item_format_segment(
203 struct xfs_buf_log_item *bip,
204 struct xfs_log_vec *lv,
205 struct xfs_log_iovec **vecp,
207 struct xfs_buf_log_format *blfp)
209 struct xfs_buf *bp = bip->bli_buf;
216 /* copy the flags across from the base format item */
217 blfp->blf_flags = bip->__bli_format.blf_flags;
220 * Base size is the actual size of the ondisk structure - it reflects
221 * the actual size of the dirty bitmap rather than the size of the in
224 base_size = xfs_buf_log_format_size(blfp);
226 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
227 if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
229 * If the map is not be dirty in the transaction, mark
230 * the size as zero and do not advance the vector pointer.
235 blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
238 if (bip->bli_flags & XFS_BLI_STALE) {
240 * The buffer is stale, so all we need to log
241 * is the buf log format structure with the
244 trace_xfs_buf_item_format_stale(bip);
245 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
251 * Fill in an iovec for each set of contiguous chunks.
253 last_bit = first_bit;
257 * This takes the bit number to start looking from and
258 * returns the next set bit from there. It returns -1
259 * if there are no more bits set or the start bit is
260 * beyond the end of the bitmap.
262 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
265 * If we run out of bits fill in the last iovec and get out of
266 * the loop. Else if we start a new set of bits then fill in
267 * the iovec for the series we were looking at and start
268 * counting the bits in the new one. Else we're still in the
269 * same set of bits so just keep counting and scanning.
271 if (next_bit == -1) {
272 xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
276 } else if (next_bit != last_bit + 1 ||
277 xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) {
278 xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
281 first_bit = next_bit;
292 * This is called to fill in the vector of log iovecs for the
293 * given log buf item. It fills the first entry with a buf log
294 * format structure, and the rest point to contiguous chunks
299 struct xfs_log_item *lip,
300 struct xfs_log_vec *lv)
302 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
303 struct xfs_buf *bp = bip->bli_buf;
304 struct xfs_log_iovec *vecp = NULL;
308 ASSERT(atomic_read(&bip->bli_refcount) > 0);
309 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
310 (bip->bli_flags & XFS_BLI_STALE));
311 ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
312 (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
313 && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
314 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) ||
315 (bip->bli_flags & XFS_BLI_STALE));
319 * If it is an inode buffer, transfer the in-memory state to the
320 * format flags and clear the in-memory state.
322 * For buffer based inode allocation, we do not transfer
323 * this state if the inode buffer allocation has not yet been committed
324 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
325 * correct replay of the inode allocation.
327 * For icreate item based inode allocation, the buffers aren't written
328 * to the journal during allocation, and hence we should always tag the
329 * buffer as an inode buffer so that the correct unlinked list replay
330 * occurs during recovery.
332 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
333 if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
334 !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
335 xfs_log_item_in_current_chkpt(lip)))
336 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
337 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
340 for (i = 0; i < bip->bli_format_count; i++) {
341 xfs_buf_item_format_segment(bip, lv, &vecp, offset,
342 &bip->bli_formats[i]);
343 offset += BBTOB(bp->b_maps[i].bm_len);
347 * Check to make sure everything is consistent.
349 trace_xfs_buf_item_format(bip);
353 * This is called to pin the buffer associated with the buf log item in memory
354 * so it cannot be written out.
356 * We also always take a reference to the buffer log item here so that the bli
357 * is held while the item is pinned in memory. This means that we can
358 * unconditionally drop the reference count a transaction holds when the
359 * transaction is completed.
363 struct xfs_log_item *lip)
365 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
367 ASSERT(atomic_read(&bip->bli_refcount) > 0);
368 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
369 (bip->bli_flags & XFS_BLI_ORDERED) ||
370 (bip->bli_flags & XFS_BLI_STALE));
372 trace_xfs_buf_item_pin(bip);
374 atomic_inc(&bip->bli_refcount);
375 atomic_inc(&bip->bli_buf->b_pin_count);
379 * This is called to unpin the buffer associated with the buf log
380 * item which was previously pinned with a call to xfs_buf_item_pin().
382 * Also drop the reference to the buf item for the current transaction.
383 * If the XFS_BLI_STALE flag is set and we are the last reference,
384 * then free up the buf log item and unlock the buffer.
386 * If the remove flag is set we are called from uncommit in the
387 * forced-shutdown path. If that is true and the reference count on
388 * the log item is going to drop to zero we need to free the item's
389 * descriptor in the transaction.
393 struct xfs_log_item *lip,
396 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
397 xfs_buf_t *bp = bip->bli_buf;
398 struct xfs_ail *ailp = lip->li_ailp;
399 int stale = bip->bli_flags & XFS_BLI_STALE;
402 ASSERT(bp->b_log_item == bip);
403 ASSERT(atomic_read(&bip->bli_refcount) > 0);
405 trace_xfs_buf_item_unpin(bip);
407 freed = atomic_dec_and_test(&bip->bli_refcount);
409 if (atomic_dec_and_test(&bp->b_pin_count))
410 wake_up_all(&bp->b_waiters);
412 if (freed && stale) {
413 ASSERT(bip->bli_flags & XFS_BLI_STALE);
414 ASSERT(xfs_buf_islocked(bp));
415 ASSERT(bp->b_flags & XBF_STALE);
416 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
418 trace_xfs_buf_item_unpin_stale(bip);
422 * If we are in a transaction context, we have to
423 * remove the log item from the transaction as we are
424 * about to release our reference to the buffer. If we
425 * don't, the unlock that occurs later in
426 * xfs_trans_uncommit() will try to reference the
427 * buffer which we no longer have a hold on.
429 if (!list_empty(&lip->li_trans))
430 xfs_trans_del_item(lip);
433 * Since the transaction no longer refers to the buffer,
434 * the buffer should no longer refer to the transaction.
440 * If we get called here because of an IO error, we may
441 * or may not have the item on the AIL. xfs_trans_ail_delete()
442 * will take care of that situation.
443 * xfs_trans_ail_delete() drops the AIL lock.
445 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
446 xfs_buf_do_callbacks(bp);
447 bp->b_log_item = NULL;
448 list_del_init(&bp->b_li_list);
451 spin_lock(&ailp->ail_lock);
452 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
453 xfs_buf_item_relse(bp);
454 ASSERT(bp->b_log_item == NULL);
457 } else if (freed && remove) {
459 * There are currently two references to the buffer - the active
460 * LRU reference and the buf log item. What we are about to do
461 * here - simulate a failed IO completion - requires 3
464 * The LRU reference is removed by the xfs_buf_stale() call. The
465 * buf item reference is removed by the xfs_buf_iodone()
466 * callback that is run by xfs_buf_do_callbacks() during ioend
467 * processing (via the bp->b_iodone callback), and then finally
468 * the ioend processing will drop the IO reference if the buffer
469 * is marked XBF_ASYNC.
471 * Hence we need to take an additional reference here so that IO
472 * completion processing doesn't free the buffer prematurely.
476 bp->b_flags |= XBF_ASYNC;
477 xfs_buf_ioerror(bp, -EIO);
478 bp->b_flags &= ~XBF_DONE;
485 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
486 * seconds so as to not spam logs too much on repeated detection of the same
490 static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10);
494 struct xfs_log_item *lip,
495 struct list_head *buffer_list)
497 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
498 struct xfs_buf *bp = bip->bli_buf;
499 uint rval = XFS_ITEM_SUCCESS;
501 if (xfs_buf_ispinned(bp))
502 return XFS_ITEM_PINNED;
503 if (!xfs_buf_trylock(bp)) {
505 * If we have just raced with a buffer being pinned and it has
506 * been marked stale, we could end up stalling until someone else
507 * issues a log force to unpin the stale buffer. Check for the
508 * race condition here so xfsaild recognizes the buffer is pinned
509 * and queues a log force to move it along.
511 if (xfs_buf_ispinned(bp))
512 return XFS_ITEM_PINNED;
513 return XFS_ITEM_LOCKED;
516 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
518 trace_xfs_buf_item_push(bip);
520 /* has a previous flush failed due to IO errors? */
521 if ((bp->b_flags & XBF_WRITE_FAIL) &&
522 ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) {
523 xfs_warn(bp->b_target->bt_mount,
524 "Failing async write on buffer block 0x%llx. Retrying async write.",
525 (long long)bp->b_bn);
528 if (!xfs_buf_delwri_queue(bp, buffer_list))
529 rval = XFS_ITEM_FLUSHING;
535 * Drop the buffer log item refcount and take appropriate action. This helper
536 * determines whether the bli must be freed or not, since a decrement to zero
537 * does not necessarily mean the bli is unused.
539 * Return true if the bli is freed, false otherwise.
543 struct xfs_buf_log_item *bip)
545 struct xfs_log_item *lip = &bip->bli_item;
549 /* drop the bli ref and return if it wasn't the last one */
550 if (!atomic_dec_and_test(&bip->bli_refcount))
554 * We dropped the last ref and must free the item if clean or aborted.
555 * If the bli is dirty and non-aborted, the buffer was clean in the
556 * transaction but still awaiting writeback from previous changes. In
557 * that case, the bli is freed on buffer writeback completion.
559 aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) ||
560 XFS_FORCED_SHUTDOWN(lip->li_mountp);
561 dirty = bip->bli_flags & XFS_BLI_DIRTY;
562 if (dirty && !aborted)
566 * The bli is aborted or clean. An aborted item may be in the AIL
567 * regardless of dirty state. For example, consider an aborted
568 * transaction that invalidated a dirty bli and cleared the dirty
572 xfs_trans_ail_remove(lip, SHUTDOWN_LOG_IO_ERROR);
573 xfs_buf_item_relse(bip->bli_buf);
578 * Release the buffer associated with the buf log item. If there is no dirty
579 * logged data associated with the buffer recorded in the buf log item, then
580 * free the buf log item and remove the reference to it in the buffer.
582 * This call ignores the recursion count. It is only called when the buffer
583 * should REALLY be unlocked, regardless of the recursion count.
585 * We unconditionally drop the transaction's reference to the log item. If the
586 * item was logged, then another reference was taken when it was pinned, so we
587 * can safely drop the transaction reference now. This also allows us to avoid
588 * potential races with the unpin code freeing the bli by not referencing the
589 * bli after we've dropped the reference count.
591 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
592 * if necessary but do not unlock the buffer. This is for support of
593 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
598 struct xfs_log_item *lip)
600 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
601 struct xfs_buf *bp = bip->bli_buf;
603 bool hold = bip->bli_flags & XFS_BLI_HOLD;
604 bool stale = bip->bli_flags & XFS_BLI_STALE;
605 #if defined(DEBUG) || defined(XFS_WARN)
606 bool ordered = bip->bli_flags & XFS_BLI_ORDERED;
607 bool dirty = bip->bli_flags & XFS_BLI_DIRTY;
608 bool aborted = test_bit(XFS_LI_ABORTED,
612 trace_xfs_buf_item_unlock(bip);
615 * The bli dirty state should match whether the blf has logged segments
616 * except for ordered buffers, where only the bli should be dirty.
618 ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) ||
619 (ordered && dirty && !xfs_buf_item_dirty_format(bip)));
620 ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
623 * Clear the buffer's association with this transaction and
624 * per-transaction state from the bli, which has been copied above.
627 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
630 * Unref the item and unlock the buffer unless held or stale. Stale
631 * buffers remain locked until final unpin unless the bli is freed by
632 * the unref call. The latter implies shutdown because buffer
633 * invalidation dirties the bli and transaction.
635 released = xfs_buf_item_put(bip);
636 if (hold || (stale && !released))
638 ASSERT(!stale || aborted);
643 * This is called to find out where the oldest active copy of the
644 * buf log item in the on disk log resides now that the last log
645 * write of it completed at the given lsn.
646 * We always re-log all the dirty data in a buffer, so usually the
647 * latest copy in the on disk log is the only one that matters. For
648 * those cases we simply return the given lsn.
650 * The one exception to this is for buffers full of newly allocated
651 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
652 * flag set, indicating that only the di_next_unlinked fields from the
653 * inodes in the buffers will be replayed during recovery. If the
654 * original newly allocated inode images have not yet been flushed
655 * when the buffer is so relogged, then we need to make sure that we
656 * keep the old images in the 'active' portion of the log. We do this
657 * by returning the original lsn of that transaction here rather than
661 xfs_buf_item_committed(
662 struct xfs_log_item *lip,
665 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
667 trace_xfs_buf_item_committed(bip);
669 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
675 xfs_buf_item_committing(
676 struct xfs_log_item *lip,
677 xfs_lsn_t commit_lsn)
682 * This is the ops vector shared by all buf log items.
684 static const struct xfs_item_ops xfs_buf_item_ops = {
685 .iop_size = xfs_buf_item_size,
686 .iop_format = xfs_buf_item_format,
687 .iop_pin = xfs_buf_item_pin,
688 .iop_unpin = xfs_buf_item_unpin,
689 .iop_unlock = xfs_buf_item_unlock,
690 .iop_committed = xfs_buf_item_committed,
691 .iop_push = xfs_buf_item_push,
692 .iop_committing = xfs_buf_item_committing
696 xfs_buf_item_get_format(
697 struct xfs_buf_log_item *bip,
700 ASSERT(bip->bli_formats == NULL);
701 bip->bli_format_count = count;
704 bip->bli_formats = &bip->__bli_format;
708 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
710 if (!bip->bli_formats)
716 xfs_buf_item_free_format(
717 struct xfs_buf_log_item *bip)
719 if (bip->bli_formats != &bip->__bli_format) {
720 kmem_free(bip->bli_formats);
721 bip->bli_formats = NULL;
726 * Allocate a new buf log item to go with the given buffer.
727 * Set the buffer's b_log_item field to point to the new
733 struct xfs_mount *mp)
735 struct xfs_buf_log_item *bip = bp->b_log_item;
742 * Check to see if there is already a buf log item for
743 * this buffer. If we do already have one, there is
744 * nothing to do here so return.
746 ASSERT(bp->b_target->bt_mount == mp);
748 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
749 ASSERT(!bp->b_transp);
750 ASSERT(bip->bli_buf == bp);
754 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
755 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
759 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
760 * can be divided into. Make sure not to truncate any pieces.
761 * map_size is the size of the bitmap needed to describe the
762 * chunks of the buffer.
764 * Discontiguous buffer support follows the layout of the underlying
765 * buffer. This makes the implementation as simple as possible.
767 error = xfs_buf_item_get_format(bip, bp->b_map_count);
769 if (error) { /* to stop gcc throwing set-but-unused warnings */
770 kmem_zone_free(xfs_buf_item_zone, bip);
775 for (i = 0; i < bip->bli_format_count; i++) {
776 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
778 map_size = DIV_ROUND_UP(chunks, NBWORD);
780 bip->bli_formats[i].blf_type = XFS_LI_BUF;
781 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
782 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
783 bip->bli_formats[i].blf_map_size = map_size;
786 bp->b_log_item = bip;
793 * Mark bytes first through last inclusive as dirty in the buf
797 xfs_buf_item_log_segment(
813 * Convert byte offsets to bit numbers.
815 first_bit = first >> XFS_BLF_SHIFT;
816 last_bit = last >> XFS_BLF_SHIFT;
819 * Calculate the total number of bits to be set.
821 bits_to_set = last_bit - first_bit + 1;
824 * Get a pointer to the first word in the bitmap
827 word_num = first_bit >> BIT_TO_WORD_SHIFT;
828 wordp = &map[word_num];
831 * Calculate the starting bit in the first word.
833 bit = first_bit & (uint)(NBWORD - 1);
836 * First set any bits in the first word of our range.
837 * If it starts at bit 0 of the word, it will be
838 * set below rather than here. That is what the variable
839 * bit tells us. The variable bits_set tracks the number
840 * of bits that have been set so far. End_bit is the number
841 * of the last bit to be set in this word plus one.
844 end_bit = min(bit + bits_to_set, (uint)NBWORD);
845 mask = ((1U << (end_bit - bit)) - 1) << bit;
848 bits_set = end_bit - bit;
854 * Now set bits a whole word at a time that are between
855 * first_bit and last_bit.
857 while ((bits_to_set - bits_set) >= NBWORD) {
858 *wordp |= 0xffffffff;
864 * Finally, set any bits left to be set in one last partial word.
866 end_bit = bits_to_set - bits_set;
868 mask = (1U << end_bit) - 1;
874 * Mark bytes first through last inclusive as dirty in the buf
879 struct xfs_buf_log_item *bip,
886 struct xfs_buf *bp = bip->bli_buf;
889 * walk each buffer segment and mark them dirty appropriately.
892 for (i = 0; i < bip->bli_format_count; i++) {
895 end = start + BBTOB(bp->b_maps[i].bm_len) - 1;
897 /* skip to the map that includes the first byte to log */
899 start += BBTOB(bp->b_maps[i].bm_len);
904 * Trim the range to this segment and mark it in the bitmap.
905 * Note that we must convert buffer offsets to segment relative
906 * offsets (e.g., the first byte of each segment is byte 0 of
913 xfs_buf_item_log_segment(first - start, end - start,
914 &bip->bli_formats[i].blf_data_map[0]);
916 start += BBTOB(bp->b_maps[i].bm_len);
922 * Return true if the buffer has any ranges logged/dirtied by a transaction,
926 xfs_buf_item_dirty_format(
927 struct xfs_buf_log_item *bip)
931 for (i = 0; i < bip->bli_format_count; i++) {
932 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
933 bip->bli_formats[i].blf_map_size))
942 struct xfs_buf_log_item *bip)
944 xfs_buf_item_free_format(bip);
945 kmem_free(bip->bli_item.li_lv_shadow);
946 kmem_zone_free(xfs_buf_item_zone, bip);
950 * This is called when the buf log item is no longer needed. It should
951 * free the buf log item associated with the given buffer and clear
952 * the buffer's pointer to the buf log item. If there are no more
953 * items in the list, clear the b_iodone field of the buffer (see
954 * xfs_buf_attach_iodone() below).
960 struct xfs_buf_log_item *bip = bp->b_log_item;
962 trace_xfs_buf_item_relse(bp, _RET_IP_);
963 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
965 bp->b_log_item = NULL;
966 if (list_empty(&bp->b_li_list))
970 xfs_buf_item_free(bip);
975 * Add the given log item with its callback to the list of callbacks
976 * to be called when the buffer's I/O completes. If it is not set
977 * already, set the buffer's b_iodone() routine to be
978 * xfs_buf_iodone_callbacks() and link the log item into the list of
979 * items rooted at b_li_list.
982 xfs_buf_attach_iodone(
984 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
987 ASSERT(xfs_buf_islocked(bp));
990 list_add_tail(&lip->li_bio_list, &bp->b_li_list);
992 ASSERT(bp->b_iodone == NULL ||
993 bp->b_iodone == xfs_buf_iodone_callbacks);
994 bp->b_iodone = xfs_buf_iodone_callbacks;
998 * We can have many callbacks on a buffer. Running the callbacks individually
999 * can cause a lot of contention on the AIL lock, so we allow for a single
1000 * callback to be able to scan the remaining items in bp->b_li_list for other
1001 * items of the same type and callback to be processed in the first call.
1003 * As a result, the loop walking the callback list below will also modify the
1004 * list. it removes the first item from the list and then runs the callback.
1005 * The loop then restarts from the new first item int the list. This allows the
1006 * callback to scan and modify the list attached to the buffer and we don't
1007 * have to care about maintaining a next item pointer.
1010 xfs_buf_do_callbacks(
1013 struct xfs_buf_log_item *blip = bp->b_log_item;
1014 struct xfs_log_item *lip;
1016 /* If there is a buf_log_item attached, run its callback */
1018 lip = &blip->bli_item;
1019 lip->li_cb(bp, lip);
1022 while (!list_empty(&bp->b_li_list)) {
1023 lip = list_first_entry(&bp->b_li_list, struct xfs_log_item,
1027 * Remove the item from the list, so we don't have any
1028 * confusion if the item is added to another buf.
1029 * Don't touch the log item after calling its
1030 * callback, because it could have freed itself.
1032 list_del_init(&lip->li_bio_list);
1033 lip->li_cb(bp, lip);
1038 * Invoke the error state callback for each log item affected by the failed I/O.
1040 * If a metadata buffer write fails with a non-permanent error, the buffer is
1041 * eventually resubmitted and so the completion callbacks are not run. The error
1042 * state may need to be propagated to the log items attached to the buffer,
1043 * however, so the next AIL push of the item knows hot to handle it correctly.
1046 xfs_buf_do_callbacks_fail(
1049 struct xfs_log_item *lip;
1050 struct xfs_ail *ailp;
1053 * Buffer log item errors are handled directly by xfs_buf_item_push()
1054 * and xfs_buf_iodone_callback_error, and they have no IO error
1055 * callbacks. Check only for items in b_li_list.
1057 if (list_empty(&bp->b_li_list))
1060 lip = list_first_entry(&bp->b_li_list, struct xfs_log_item,
1062 ailp = lip->li_ailp;
1063 spin_lock(&ailp->ail_lock);
1064 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
1065 if (lip->li_ops->iop_error)
1066 lip->li_ops->iop_error(lip, bp);
1068 spin_unlock(&ailp->ail_lock);
1072 xfs_buf_iodone_callback_error(
1075 struct xfs_buf_log_item *bip = bp->b_log_item;
1076 struct xfs_log_item *lip;
1077 struct xfs_mount *mp;
1078 static ulong lasttime;
1079 static xfs_buftarg_t *lasttarg;
1080 struct xfs_error_cfg *cfg;
1083 * The failed buffer might not have a buf_log_item attached or the
1084 * log_item list might be empty. Get the mp from the available
1087 lip = list_first_entry_or_null(&bp->b_li_list, struct xfs_log_item,
1089 mp = lip ? lip->li_mountp : bip->bli_item.li_mountp;
1092 * If we've already decided to shutdown the filesystem because of
1093 * I/O errors, there's no point in giving this a retry.
1095 if (XFS_FORCED_SHUTDOWN(mp))
1098 if (bp->b_target != lasttarg ||
1099 time_after(jiffies, (lasttime + 5*HZ))) {
1101 xfs_buf_ioerror_alert(bp, __func__);
1103 lasttarg = bp->b_target;
1105 /* synchronous writes will have callers process the error */
1106 if (!(bp->b_flags & XBF_ASYNC))
1109 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1110 ASSERT(bp->b_iodone != NULL);
1112 cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error);
1115 * If the write was asynchronous then no one will be looking for the
1116 * error. If this is the first failure of this type, clear the error
1117 * state and write the buffer out again. This means we always retry an
1118 * async write failure at least once, but we also need to set the buffer
1119 * up to behave correctly now for repeated failures.
1121 if (!(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL)) ||
1122 bp->b_last_error != bp->b_error) {
1123 bp->b_flags |= (XBF_WRITE | XBF_DONE | XBF_WRITE_FAIL);
1124 bp->b_last_error = bp->b_error;
1125 if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER &&
1126 !bp->b_first_retry_time)
1127 bp->b_first_retry_time = jiffies;
1129 xfs_buf_ioerror(bp, 0);
1135 * Repeated failure on an async write. Take action according to the
1136 * error configuration we have been set up to use.
1139 if (cfg->max_retries != XFS_ERR_RETRY_FOREVER &&
1140 ++bp->b_retries > cfg->max_retries)
1141 goto permanent_error;
1142 if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER &&
1143 time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time))
1144 goto permanent_error;
1146 /* At unmount we may treat errors differently */
1147 if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount)
1148 goto permanent_error;
1151 * Still a transient error, run IO completion failure callbacks and let
1152 * the higher layers retry the buffer.
1154 xfs_buf_do_callbacks_fail(bp);
1155 xfs_buf_ioerror(bp, 0);
1160 * Permanent error - we need to trigger a shutdown if we haven't already
1161 * to indicate that inconsistency will result from this action.
1164 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1167 bp->b_flags |= XBF_DONE;
1168 trace_xfs_buf_error_relse(bp, _RET_IP_);
1173 * This is the iodone() function for buffers which have had callbacks attached
1174 * to them by xfs_buf_attach_iodone(). We need to iterate the items on the
1175 * callback list, mark the buffer as having no more callbacks and then push the
1176 * buffer through IO completion processing.
1179 xfs_buf_iodone_callbacks(
1183 * If there is an error, process it. Some errors require us
1184 * to run callbacks after failure processing is done so we
1185 * detect that and take appropriate action.
1187 if (bp->b_error && xfs_buf_iodone_callback_error(bp))
1191 * Successful IO or permanent error. Either way, we can clear the
1192 * retry state here in preparation for the next error that may occur.
1194 bp->b_last_error = 0;
1196 bp->b_first_retry_time = 0;
1198 xfs_buf_do_callbacks(bp);
1199 bp->b_log_item = NULL;
1200 list_del_init(&bp->b_li_list);
1201 bp->b_iodone = NULL;
1206 * This is the iodone() function for buffers which have been
1207 * logged. It is called when they are eventually flushed out.
1208 * It should remove the buf item from the AIL, and free the buf item.
1209 * It is called by xfs_buf_iodone_callbacks() above which will take
1210 * care of cleaning up the buffer itself.
1215 struct xfs_log_item *lip)
1217 struct xfs_ail *ailp = lip->li_ailp;
1219 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1224 * If we are forcibly shutting down, this may well be
1225 * off the AIL already. That's because we simulate the
1226 * log-committed callbacks to unpin these buffers. Or we may never
1227 * have put this item on AIL because of the transaction was
1228 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1230 * Either way, AIL is useless if we're forcing a shutdown.
1232 spin_lock(&ailp->ail_lock);
1233 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1234 xfs_buf_item_free(BUF_ITEM(lip));
1238 * Requeue a failed buffer for writeback.
1240 * We clear the log item failed state here as well, but we have to be careful
1241 * about reference counts because the only active reference counts on the buffer
1242 * may be the failed log items. Hence if we clear the log item failed state
1243 * before queuing the buffer for IO we can release all active references to
1244 * the buffer and free it, leading to use after free problems in
1245 * xfs_buf_delwri_queue. It makes no difference to the buffer or log items which
1246 * order we process them in - the buffer is locked, and we own the buffer list
1247 * so nothing on them is going to change while we are performing this action.
1249 * Hence we can safely queue the buffer for IO before we clear the failed log
1250 * item state, therefore always having an active reference to the buffer and
1251 * avoiding the transient zero-reference state that leads to use-after-free.
1253 * Return true if the buffer was added to the buffer list, false if it was
1254 * already on the buffer list.
1257 xfs_buf_resubmit_failed_buffers(
1259 struct list_head *buffer_list)
1261 struct xfs_log_item *lip;
1264 ret = xfs_buf_delwri_queue(bp, buffer_list);
1267 * XFS_LI_FAILED set/clear is protected by ail_lock, caller of this
1268 * function already have it acquired
1270 list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1271 xfs_clear_li_failed(lip);