2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
30 #include "xfs_trace.h"
33 kmem_zone_t *xfs_buf_item_zone;
35 static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
37 return container_of(lip, struct xfs_buf_log_item, bli_item);
40 STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
43 * This returns the number of log iovecs needed to log the
46 * It calculates this as 1 iovec for the buf log format structure
47 * and 1 for each stretch of non-contiguous chunks to be logged.
48 * Contiguous chunks are logged in a single iovec.
50 * If the XFS_BLI_STALE flag has been set, then log nothing.
53 xfs_buf_item_size_segment(
54 struct xfs_buf_log_item *bip,
55 struct xfs_buf_log_format *blfp)
57 struct xfs_buf *bp = bip->bli_buf;
62 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
67 * initial count for a dirty buffer is 2 vectors - the format structure
68 * and the first dirty region.
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) +
105 * This returns the number of log iovecs needed to log the given buf log item.
107 * It calculates this as 1 iovec for the buf log format structure and 1 for each
108 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
111 * Discontiguous buffers need a format structure per region that that is being
112 * logged. This makes the changes in the buffer appear to log recovery as though
113 * they came from separate buffers, just like would occur if multiple buffers
114 * were used instead of a single discontiguous buffer. This enables
115 * discontiguous buffers to be in-memory constructs, completely transparent to
116 * what ends up on disk.
118 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
123 struct xfs_log_item *lip)
125 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 return bip->bli_format_count;
141 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
144 * the vector count is based on the number of buffer vectors we have
145 * dirty bits in. This will only be greater than one when we have a
146 * compound buffer with more than one segment dirty. Hence for compound
147 * buffers we need to track which segment the dirty bits correspond to,
148 * and when we move from one segment to the next increment the vector
149 * count for the extra buf log format structure that will need to be
153 for (i = 0; i < bip->bli_format_count; i++) {
154 nvecs += xfs_buf_item_size_segment(bip, &bip->bli_formats[i]);
157 trace_xfs_buf_item_size(bip);
161 static struct xfs_log_iovec *
162 xfs_buf_item_format_segment(
163 struct xfs_buf_log_item *bip,
164 struct xfs_log_iovec *vecp,
166 struct xfs_buf_log_format *blfp)
168 struct xfs_buf *bp = bip->bli_buf;
177 /* copy the flags across from the base format item */
178 blfp->blf_flags = bip->__bli_format.blf_flags;
181 * Base size is the actual size of the ondisk structure - it reflects
182 * the actual size of the dirty bitmap rather than the size of the in
185 base_size = offsetof(struct xfs_buf_log_format, blf_data_map) +
186 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
189 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
190 if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
192 * If the map is not be dirty in the transaction, mark
193 * the size as zero and do not advance the vector pointer.
199 vecp->i_len = base_size;
200 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
204 if (bip->bli_flags & XFS_BLI_STALE) {
206 * The buffer is stale, so all we need to log
207 * is the buf log format structure with the
210 trace_xfs_buf_item_format_stale(bip);
211 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
216 * Fill in an iovec for each set of contiguous chunks.
219 last_bit = first_bit;
223 * This takes the bit number to start looking from and
224 * returns the next set bit from there. It returns -1
225 * if there are no more bits set or the start bit is
226 * beyond the end of the bitmap.
228 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
231 * If we run out of bits fill in the last iovec and get
233 * Else if we start a new set of bits then fill in the
234 * iovec for the series we were looking at and start
235 * counting the bits in the new one.
236 * Else we're still in the same set of bits so just
237 * keep counting and scanning.
239 if (next_bit == -1) {
240 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
241 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
242 vecp->i_len = nbits * XFS_BLF_CHUNK;
243 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
246 } else if (next_bit != last_bit + 1) {
247 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
248 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
249 vecp->i_len = nbits * XFS_BLF_CHUNK;
250 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
253 first_bit = next_bit;
256 } else if (xfs_buf_offset(bp, offset +
257 (next_bit << XFS_BLF_SHIFT)) !=
258 (xfs_buf_offset(bp, offset +
259 (last_bit << XFS_BLF_SHIFT)) +
261 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
262 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
263 vecp->i_len = nbits * XFS_BLF_CHUNK;
264 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
266 * You would think we need to bump the nvecs here too, but we do not
267 * this number is used by recovery, and it gets confused by the boundary
272 first_bit = next_bit;
281 blfp->blf_size = nvecs;
286 * This is called to fill in the vector of log iovecs for the
287 * given log buf item. It fills the first entry with a buf log
288 * format structure, and the rest point to contiguous chunks
293 struct xfs_log_item *lip,
294 struct xfs_log_iovec *vecp)
296 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
297 struct xfs_buf *bp = bip->bli_buf;
301 ASSERT(atomic_read(&bip->bli_refcount) > 0);
302 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
303 (bip->bli_flags & XFS_BLI_STALE));
306 * If it is an inode buffer, transfer the in-memory state to the
307 * format flags and clear the in-memory state. We do not transfer
308 * this state if the inode buffer allocation has not yet been committed
309 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
310 * correct replay of the inode allocation.
312 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
313 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
314 xfs_log_item_in_current_chkpt(lip)))
315 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
316 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
319 for (i = 0; i < bip->bli_format_count; i++) {
320 vecp = xfs_buf_item_format_segment(bip, vecp, offset,
321 &bip->bli_formats[i]);
322 offset += bp->b_maps[i].bm_len;
326 * Check to make sure everything is consistent.
328 trace_xfs_buf_item_format(bip);
332 * This is called to pin the buffer associated with the buf log item in memory
333 * so it cannot be written out.
335 * We also always take a reference to the buffer log item here so that the bli
336 * is held while the item is pinned in memory. This means that we can
337 * unconditionally drop the reference count a transaction holds when the
338 * transaction is completed.
342 struct xfs_log_item *lip)
344 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
346 ASSERT(atomic_read(&bip->bli_refcount) > 0);
347 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
348 (bip->bli_flags & XFS_BLI_STALE));
350 trace_xfs_buf_item_pin(bip);
352 atomic_inc(&bip->bli_refcount);
353 atomic_inc(&bip->bli_buf->b_pin_count);
357 * This is called to unpin the buffer associated with the buf log
358 * item which was previously pinned with a call to xfs_buf_item_pin().
360 * Also drop the reference to the buf item for the current transaction.
361 * If the XFS_BLI_STALE flag is set and we are the last reference,
362 * then free up the buf log item and unlock the buffer.
364 * If the remove flag is set we are called from uncommit in the
365 * forced-shutdown path. If that is true and the reference count on
366 * the log item is going to drop to zero we need to free the item's
367 * descriptor in the transaction.
371 struct xfs_log_item *lip,
374 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
375 xfs_buf_t *bp = bip->bli_buf;
376 struct xfs_ail *ailp = lip->li_ailp;
377 int stale = bip->bli_flags & XFS_BLI_STALE;
380 ASSERT(bp->b_fspriv == bip);
381 ASSERT(atomic_read(&bip->bli_refcount) > 0);
383 trace_xfs_buf_item_unpin(bip);
385 freed = atomic_dec_and_test(&bip->bli_refcount);
387 if (atomic_dec_and_test(&bp->b_pin_count))
388 wake_up_all(&bp->b_waiters);
390 if (freed && stale) {
391 ASSERT(bip->bli_flags & XFS_BLI_STALE);
392 ASSERT(xfs_buf_islocked(bp));
393 ASSERT(XFS_BUF_ISSTALE(bp));
394 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
396 trace_xfs_buf_item_unpin_stale(bip);
400 * If we are in a transaction context, we have to
401 * remove the log item from the transaction as we are
402 * about to release our reference to the buffer. If we
403 * don't, the unlock that occurs later in
404 * xfs_trans_uncommit() will try to reference the
405 * buffer which we no longer have a hold on.
408 xfs_trans_del_item(lip);
411 * Since the transaction no longer refers to the buffer,
412 * the buffer should no longer refer to the transaction.
418 * If we get called here because of an IO error, we may
419 * or may not have the item on the AIL. xfs_trans_ail_delete()
420 * will take care of that situation.
421 * xfs_trans_ail_delete() drops the AIL lock.
423 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
424 xfs_buf_do_callbacks(bp);
428 spin_lock(&ailp->xa_lock);
429 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
430 xfs_buf_item_relse(bp);
431 ASSERT(bp->b_fspriv == NULL);
434 } else if (freed && remove) {
436 * There are currently two references to the buffer - the active
437 * LRU reference and the buf log item. What we are about to do
438 * here - simulate a failed IO completion - requires 3
441 * The LRU reference is removed by the xfs_buf_stale() call. The
442 * buf item reference is removed by the xfs_buf_iodone()
443 * callback that is run by xfs_buf_do_callbacks() during ioend
444 * processing (via the bp->b_iodone callback), and then finally
445 * the ioend processing will drop the IO reference if the buffer
446 * is marked XBF_ASYNC.
448 * Hence we need to take an additional reference here so that IO
449 * completion processing doesn't free the buffer prematurely.
453 bp->b_flags |= XBF_ASYNC;
454 xfs_buf_ioerror(bp, EIO);
457 xfs_buf_ioend(bp, 0);
463 struct xfs_log_item *lip,
464 struct list_head *buffer_list)
466 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
467 struct xfs_buf *bp = bip->bli_buf;
468 uint rval = XFS_ITEM_SUCCESS;
470 if (xfs_buf_ispinned(bp))
471 return XFS_ITEM_PINNED;
472 if (!xfs_buf_trylock(bp)) {
474 * If we have just raced with a buffer being pinned and it has
475 * been marked stale, we could end up stalling until someone else
476 * issues a log force to unpin the stale buffer. Check for the
477 * race condition here so xfsaild recognizes the buffer is pinned
478 * and queues a log force to move it along.
480 if (xfs_buf_ispinned(bp))
481 return XFS_ITEM_PINNED;
482 return XFS_ITEM_LOCKED;
485 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
487 trace_xfs_buf_item_push(bip);
489 if (!xfs_buf_delwri_queue(bp, buffer_list))
490 rval = XFS_ITEM_FLUSHING;
496 * Release the buffer associated with the buf log item. If there is no dirty
497 * logged data associated with the buffer recorded in the buf log item, then
498 * free the buf log item and remove the reference to it in the buffer.
500 * This call ignores the recursion count. It is only called when the buffer
501 * should REALLY be unlocked, regardless of the recursion count.
503 * We unconditionally drop the transaction's reference to the log item. If the
504 * item was logged, then another reference was taken when it was pinned, so we
505 * can safely drop the transaction reference now. This also allows us to avoid
506 * potential races with the unpin code freeing the bli by not referencing the
507 * bli after we've dropped the reference count.
509 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
510 * if necessary but do not unlock the buffer. This is for support of
511 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
516 struct xfs_log_item *lip)
518 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
519 struct xfs_buf *bp = bip->bli_buf;
520 int aborted, clean, i;
523 /* Clear the buffer's association with this transaction. */
527 * If this is a transaction abort, don't return early. Instead, allow
528 * the brelse to happen. Normally it would be done for stale
529 * (cancelled) buffers at unpin time, but we'll never go through the
530 * pin/unpin cycle if we abort inside commit.
532 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
535 * Before possibly freeing the buf item, determine if we should
536 * release the buffer at the end of this routine.
538 hold = bip->bli_flags & XFS_BLI_HOLD;
540 /* Clear the per transaction state. */
541 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
544 * If the buf item is marked stale, then don't do anything. We'll
545 * unlock the buffer and free the buf item when the buffer is unpinned
548 if (bip->bli_flags & XFS_BLI_STALE) {
549 trace_xfs_buf_item_unlock_stale(bip);
550 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
552 atomic_dec(&bip->bli_refcount);
557 trace_xfs_buf_item_unlock(bip);
560 * If the buf item isn't tracking any data, free it, otherwise drop the
561 * reference we hold to it. If we are aborting the transaction, this may
562 * be the only reference to the buf item, so we free it anyway
563 * regardless of whether it is dirty or not. A dirty abort implies a
567 for (i = 0; i < bip->bli_format_count; i++) {
568 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
569 bip->bli_formats[i].blf_map_size)) {
575 xfs_buf_item_relse(bp);
577 if (atomic_dec_and_test(&bip->bli_refcount)) {
578 ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
579 xfs_buf_item_relse(bp);
582 atomic_dec(&bip->bli_refcount);
589 * This is called to find out where the oldest active copy of the
590 * buf log item in the on disk log resides now that the last log
591 * write of it completed at the given lsn.
592 * We always re-log all the dirty data in a buffer, so usually the
593 * latest copy in the on disk log is the only one that matters. For
594 * those cases we simply return the given lsn.
596 * The one exception to this is for buffers full of newly allocated
597 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
598 * flag set, indicating that only the di_next_unlinked fields from the
599 * inodes in the buffers will be replayed during recovery. If the
600 * original newly allocated inode images have not yet been flushed
601 * when the buffer is so relogged, then we need to make sure that we
602 * keep the old images in the 'active' portion of the log. We do this
603 * by returning the original lsn of that transaction here rather than
607 xfs_buf_item_committed(
608 struct xfs_log_item *lip,
611 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
613 trace_xfs_buf_item_committed(bip);
615 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
621 xfs_buf_item_committing(
622 struct xfs_log_item *lip,
623 xfs_lsn_t commit_lsn)
628 * This is the ops vector shared by all buf log items.
630 static const struct xfs_item_ops xfs_buf_item_ops = {
631 .iop_size = xfs_buf_item_size,
632 .iop_format = xfs_buf_item_format,
633 .iop_pin = xfs_buf_item_pin,
634 .iop_unpin = xfs_buf_item_unpin,
635 .iop_unlock = xfs_buf_item_unlock,
636 .iop_committed = xfs_buf_item_committed,
637 .iop_push = xfs_buf_item_push,
638 .iop_committing = xfs_buf_item_committing
642 xfs_buf_item_get_format(
643 struct xfs_buf_log_item *bip,
646 ASSERT(bip->bli_formats == NULL);
647 bip->bli_format_count = count;
650 bip->bli_formats = &bip->__bli_format;
654 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
656 if (!bip->bli_formats)
662 xfs_buf_item_free_format(
663 struct xfs_buf_log_item *bip)
665 if (bip->bli_formats != &bip->__bli_format) {
666 kmem_free(bip->bli_formats);
667 bip->bli_formats = NULL;
672 * Allocate a new buf log item to go with the given buffer.
673 * Set the buffer's b_fsprivate field to point to the new
674 * buf log item. If there are other item's attached to the
675 * buffer (see xfs_buf_attach_iodone() below), then put the
676 * buf log item at the front.
683 xfs_log_item_t *lip = bp->b_fspriv;
684 xfs_buf_log_item_t *bip;
691 * Check to see if there is already a buf log item for
692 * this buffer. If there is, it is guaranteed to be
693 * the first. If we do already have one, there is
694 * nothing to do here so return.
696 ASSERT(bp->b_target->bt_mount == mp);
697 if (lip != NULL && lip->li_type == XFS_LI_BUF)
700 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
701 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
706 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
707 * can be divided into. Make sure not to truncate any pieces.
708 * map_size is the size of the bitmap needed to describe the
709 * chunks of the buffer.
711 * Discontiguous buffer support follows the layout of the underlying
712 * buffer. This makes the implementation as simple as possible.
714 error = xfs_buf_item_get_format(bip, bp->b_map_count);
717 for (i = 0; i < bip->bli_format_count; i++) {
718 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
720 map_size = DIV_ROUND_UP(chunks, NBWORD);
722 bip->bli_formats[i].blf_type = XFS_LI_BUF;
723 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
724 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
725 bip->bli_formats[i].blf_map_size = map_size;
728 #ifdef XFS_TRANS_DEBUG
730 * Allocate the arrays for tracking what needs to be logged
731 * and what our callers request to be logged. bli_orig
732 * holds a copy of the original, clean buffer for comparison
733 * against, and bli_logged keeps a 1 bit flag per byte in
734 * the buffer to indicate which bytes the callers have asked
737 bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP);
738 memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length));
739 bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP);
743 * Put the buf item into the list of items attached to the
744 * buffer at the front.
747 bip->bli_item.li_bio_list = bp->b_fspriv;
753 * Mark bytes first through last inclusive as dirty in the buf
757 xfs_buf_item_log_segment(
758 struct xfs_buf_log_item *bip,
774 * Convert byte offsets to bit numbers.
776 first_bit = first >> XFS_BLF_SHIFT;
777 last_bit = last >> XFS_BLF_SHIFT;
780 * Calculate the total number of bits to be set.
782 bits_to_set = last_bit - first_bit + 1;
785 * Get a pointer to the first word in the bitmap
788 word_num = first_bit >> BIT_TO_WORD_SHIFT;
789 wordp = &map[word_num];
792 * Calculate the starting bit in the first word.
794 bit = first_bit & (uint)(NBWORD - 1);
797 * First set any bits in the first word of our range.
798 * If it starts at bit 0 of the word, it will be
799 * set below rather than here. That is what the variable
800 * bit tells us. The variable bits_set tracks the number
801 * of bits that have been set so far. End_bit is the number
802 * of the last bit to be set in this word plus one.
805 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
806 mask = ((1 << (end_bit - bit)) - 1) << bit;
809 bits_set = end_bit - bit;
815 * Now set bits a whole word at a time that are between
816 * first_bit and last_bit.
818 while ((bits_to_set - bits_set) >= NBWORD) {
819 *wordp |= 0xffffffff;
825 * Finally, set any bits left to be set in one last partial word.
827 end_bit = bits_to_set - bits_set;
829 mask = (1 << end_bit) - 1;
835 * Mark bytes first through last inclusive as dirty in the buf
840 xfs_buf_log_item_t *bip,
847 struct xfs_buf *bp = bip->bli_buf;
850 * Mark the item as having some dirty data for
851 * quick reference in xfs_buf_item_dirty.
853 bip->bli_flags |= XFS_BLI_DIRTY;
856 * walk each buffer segment and mark them dirty appropriately.
859 for (i = 0; i < bip->bli_format_count; i++) {
862 end = start + BBTOB(bp->b_maps[i].bm_len);
864 start += BBTOB(bp->b_maps[i].bm_len);
872 xfs_buf_item_log_segment(bip, first, end,
873 &bip->bli_formats[i].blf_data_map[0]);
875 start += bp->b_maps[i].bm_len;
881 * Return 1 if the buffer has some data that has been logged (at any
882 * point, not just the current transaction) and 0 if not.
886 xfs_buf_log_item_t *bip)
888 return (bip->bli_flags & XFS_BLI_DIRTY);
893 xfs_buf_log_item_t *bip)
895 #ifdef XFS_TRANS_DEBUG
896 kmem_free(bip->bli_orig);
897 kmem_free(bip->bli_logged);
898 #endif /* XFS_TRANS_DEBUG */
900 xfs_buf_item_free_format(bip);
901 kmem_zone_free(xfs_buf_item_zone, bip);
905 * This is called when the buf log item is no longer needed. It should
906 * free the buf log item associated with the given buffer and clear
907 * the buffer's pointer to the buf log item. If there are no more
908 * items in the list, clear the b_iodone field of the buffer (see
909 * xfs_buf_attach_iodone() below).
915 xfs_buf_log_item_t *bip;
917 trace_xfs_buf_item_relse(bp, _RET_IP_);
920 bp->b_fspriv = bip->bli_item.li_bio_list;
921 if (bp->b_fspriv == NULL)
925 xfs_buf_item_free(bip);
930 * Add the given log item with its callback to the list of callbacks
931 * to be called when the buffer's I/O completes. If it is not set
932 * already, set the buffer's b_iodone() routine to be
933 * xfs_buf_iodone_callbacks() and link the log item into the list of
934 * items rooted at b_fsprivate. Items are always added as the second
935 * entry in the list if there is a first, because the buf item code
936 * assumes that the buf log item is first.
939 xfs_buf_attach_iodone(
941 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
944 xfs_log_item_t *head_lip;
946 ASSERT(xfs_buf_islocked(bp));
949 head_lip = bp->b_fspriv;
951 lip->li_bio_list = head_lip->li_bio_list;
952 head_lip->li_bio_list = lip;
957 ASSERT(bp->b_iodone == NULL ||
958 bp->b_iodone == xfs_buf_iodone_callbacks);
959 bp->b_iodone = xfs_buf_iodone_callbacks;
963 * We can have many callbacks on a buffer. Running the callbacks individually
964 * can cause a lot of contention on the AIL lock, so we allow for a single
965 * callback to be able to scan the remaining lip->li_bio_list for other items
966 * of the same type and callback to be processed in the first call.
968 * As a result, the loop walking the callback list below will also modify the
969 * list. it removes the first item from the list and then runs the callback.
970 * The loop then restarts from the new head of the list. This allows the
971 * callback to scan and modify the list attached to the buffer and we don't
972 * have to care about maintaining a next item pointer.
975 xfs_buf_do_callbacks(
978 struct xfs_log_item *lip;
980 while ((lip = bp->b_fspriv) != NULL) {
981 bp->b_fspriv = lip->li_bio_list;
982 ASSERT(lip->li_cb != NULL);
984 * Clear the next pointer so we don't have any
985 * confusion if the item is added to another buf.
986 * Don't touch the log item after calling its
987 * callback, because it could have freed itself.
989 lip->li_bio_list = NULL;
995 * This is the iodone() function for buffers which have had callbacks
996 * attached to them by xfs_buf_attach_iodone(). It should remove each
997 * log item from the buffer's list and call the callback of each in turn.
998 * When done, the buffer's fsprivate field is set to NULL and the buffer
999 * is unlocked with a call to iodone().
1002 xfs_buf_iodone_callbacks(
1005 struct xfs_log_item *lip = bp->b_fspriv;
1006 struct xfs_mount *mp = lip->li_mountp;
1007 static ulong lasttime;
1008 static xfs_buftarg_t *lasttarg;
1010 if (likely(!xfs_buf_geterror(bp)))
1014 * If we've already decided to shutdown the filesystem because of
1015 * I/O errors, there's no point in giving this a retry.
1017 if (XFS_FORCED_SHUTDOWN(mp)) {
1020 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1024 if (bp->b_target != lasttarg ||
1025 time_after(jiffies, (lasttime + 5*HZ))) {
1027 xfs_buf_ioerror_alert(bp, __func__);
1029 lasttarg = bp->b_target;
1032 * If the write was asynchronous then no one will be looking for the
1033 * error. Clear the error state and write the buffer out again.
1035 * XXX: This helps against transient write errors, but we need to find
1036 * a way to shut the filesystem down if the writes keep failing.
1038 * In practice we'll shut the filesystem down soon as non-transient
1039 * erorrs tend to affect the whole device and a failing log write
1040 * will make us give up. But we really ought to do better here.
1042 if (XFS_BUF_ISASYNC(bp)) {
1043 ASSERT(bp->b_iodone != NULL);
1045 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1047 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
1049 if (!XFS_BUF_ISSTALE(bp)) {
1050 bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE;
1051 xfs_buf_iorequest(bp);
1060 * If the write of the buffer was synchronous, we want to make
1061 * sure to return the error to the caller of xfs_bwrite().
1066 trace_xfs_buf_error_relse(bp, _RET_IP_);
1069 xfs_buf_do_callbacks(bp);
1070 bp->b_fspriv = NULL;
1071 bp->b_iodone = NULL;
1072 xfs_buf_ioend(bp, 0);
1076 * This is the iodone() function for buffers which have been
1077 * logged. It is called when they are eventually flushed out.
1078 * It should remove the buf item from the AIL, and free the buf item.
1079 * It is called by xfs_buf_iodone_callbacks() above which will take
1080 * care of cleaning up the buffer itself.
1085 struct xfs_log_item *lip)
1087 struct xfs_ail *ailp = lip->li_ailp;
1089 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1094 * If we are forcibly shutting down, this may well be
1095 * off the AIL already. That's because we simulate the
1096 * log-committed callbacks to unpin these buffers. Or we may never
1097 * have put this item on AIL because of the transaction was
1098 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1100 * Either way, AIL is useless if we're forcing a shutdown.
1102 spin_lock(&ailp->xa_lock);
1103 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1104 xfs_buf_item_free(BUF_ITEM(lip));