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);
41 #ifdef XFS_TRANS_DEBUG
43 * This function uses an alternate strategy for tracking the bytes
44 * that the user requests to be logged. This can then be used
45 * in conjunction with the bli_orig array in the buf log item to
46 * catch bugs in our callers' code.
48 * We also double check the bits set in xfs_buf_item_log using a
49 * simple algorithm to check that every byte is accounted for.
52 xfs_buf_item_log_debug(
53 xfs_buf_log_item_t *bip,
66 ASSERT(bip->bli_logged != NULL);
68 nbytes = last - first + 1;
69 bfset(bip->bli_logged, first, nbytes);
70 for (x = 0; x < nbytes; x++) {
71 chunk_num = byte >> XFS_BLF_SHIFT;
72 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
73 bit_num = chunk_num & (NBWORD - 1);
74 wordp = &(bip->__bli_format.blf_data_map[word_num]);
75 bit_set = *wordp & (1 << bit_num);
82 * This function is called when we flush something into a buffer without
83 * logging it. This happens for things like inodes which are logged
84 * separately from the buffer.
87 xfs_buf_item_flush_log_debug(
92 xfs_buf_log_item_t *bip = bp->b_fspriv;
95 if (bip == NULL || (bip->bli_item.li_type != XFS_LI_BUF))
98 ASSERT(bip->bli_logged != NULL);
99 nbytes = last - first + 1;
100 bfset(bip->bli_logged, first, nbytes);
104 * This function is called to verify that our callers have logged
105 * all the bytes that they changed.
107 * It does this by comparing the original copy of the buffer stored in
108 * the buf log item's bli_orig array to the current copy of the buffer
109 * and ensuring that all bytes which mismatch are set in the bli_logged
110 * array of the buf log item.
113 xfs_buf_item_log_check(
114 xfs_buf_log_item_t *bip)
121 ASSERT(bip->bli_orig != NULL);
122 ASSERT(bip->bli_logged != NULL);
125 ASSERT(bp->b_length > 0);
126 ASSERT(bp->b_addr != NULL);
127 orig = bip->bli_orig;
129 for (x = 0; x < BBTOB(bp->b_length); x++) {
130 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
131 xfs_emerg(bp->b_mount,
132 "%s: bip %x buffer %x orig %x index %d",
133 __func__, bip, bp, orig, x);
139 #define xfs_buf_item_log_debug(x,y,z)
140 #define xfs_buf_item_log_check(x)
143 STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
146 * This returns the number of log iovecs needed to log the
147 * given buf log item.
149 * It calculates this as 1 iovec for the buf log format structure
150 * and 1 for each stretch of non-contiguous chunks to be logged.
151 * Contiguous chunks are logged in a single iovec.
153 * If the XFS_BLI_STALE flag has been set, then log nothing.
156 xfs_buf_item_size_segment(
157 struct xfs_buf_log_item *bip,
158 struct xfs_buf_log_format *blfp)
160 struct xfs_buf *bp = bip->bli_buf;
165 last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
170 * initial count for a dirty buffer is 2 vectors - the format structure
171 * and the first dirty region.
175 while (last_bit != -1) {
177 * This takes the bit number to start looking from and
178 * returns the next set bit from there. It returns -1
179 * if there are no more bits set or the start bit is
180 * beyond the end of the bitmap.
182 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
185 * If we run out of bits, leave the loop,
186 * else if we find a new set of bits bump the number of vecs,
187 * else keep scanning the current set of bits.
189 if (next_bit == -1) {
191 } else if (next_bit != last_bit + 1) {
194 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
195 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
208 * This returns the number of log iovecs needed to log the given buf log item.
210 * It calculates this as 1 iovec for the buf log format structure and 1 for each
211 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
214 * Discontiguous buffers need a format structure per region that that is being
215 * logged. This makes the changes in the buffer appear to log recovery as though
216 * they came from separate buffers, just like would occur if multiple buffers
217 * were used instead of a single discontiguous buffer. This enables
218 * discontiguous buffers to be in-memory constructs, completely transparent to
219 * what ends up on disk.
221 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
226 struct xfs_log_item *lip)
228 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
232 ASSERT(atomic_read(&bip->bli_refcount) > 0);
233 if (bip->bli_flags & XFS_BLI_STALE) {
235 * The buffer is stale, so all we need to log
236 * is the buf log format structure with the
239 trace_xfs_buf_item_size_stale(bip);
240 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
241 return bip->bli_format_count;
244 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
247 * the vector count is based on the number of buffer vectors we have
248 * dirty bits in. This will only be greater than one when we have a
249 * compound buffer with more than one segment dirty. Hence for compound
250 * buffers we need to track which segment the dirty bits correspond to,
251 * and when we move from one segment to the next increment the vector
252 * count for the extra buf log format structure that will need to be
256 for (i = 0; i < bip->bli_format_count; i++) {
257 nvecs += xfs_buf_item_size_segment(bip, &bip->bli_formats[i]);
260 trace_xfs_buf_item_size(bip);
264 static struct xfs_log_iovec *
265 xfs_buf_item_format_segment(
266 struct xfs_buf_log_item *bip,
267 struct xfs_log_iovec *vecp,
269 struct xfs_buf_log_format *blfp)
271 struct xfs_buf *bp = bip->bli_buf;
280 /* copy the flags across from the base format item */
281 blfp->blf_flags = bip->__bli_format.blf_flags;
284 * Base size is the actual size of the ondisk structure - it reflects
285 * the actual size of the dirty bitmap rather than the size of the in
288 base_size = offsetof(struct xfs_buf_log_format, blf_data_map) +
289 (blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
292 first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
293 if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
295 * If the map is not be dirty in the transaction, mark
296 * the size as zero and do not advance the vector pointer.
302 vecp->i_len = base_size;
303 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
307 if (bip->bli_flags & XFS_BLI_STALE) {
309 * The buffer is stale, so all we need to log
310 * is the buf log format structure with the
313 trace_xfs_buf_item_format_stale(bip);
314 ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
319 * Fill in an iovec for each set of contiguous chunks.
322 last_bit = first_bit;
326 * This takes the bit number to start looking from and
327 * returns the next set bit from there. It returns -1
328 * if there are no more bits set or the start bit is
329 * beyond the end of the bitmap.
331 next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
334 * If we run out of bits fill in the last iovec and get
336 * Else if we start a new set of bits then fill in the
337 * iovec for the series we were looking at and start
338 * counting the bits in the new one.
339 * Else we're still in the same set of bits so just
340 * keep counting and scanning.
342 if (next_bit == -1) {
343 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
344 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
345 vecp->i_len = nbits * XFS_BLF_CHUNK;
346 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
349 } else if (next_bit != last_bit + 1) {
350 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
351 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
352 vecp->i_len = nbits * XFS_BLF_CHUNK;
353 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
356 first_bit = next_bit;
359 } else if (xfs_buf_offset(bp, offset +
360 (next_bit << XFS_BLF_SHIFT)) !=
361 (xfs_buf_offset(bp, offset +
362 (last_bit << XFS_BLF_SHIFT)) +
364 buffer_offset = offset + first_bit * XFS_BLF_CHUNK;
365 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
366 vecp->i_len = nbits * XFS_BLF_CHUNK;
367 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
369 * You would think we need to bump the nvecs here too, but we do not
370 * this number is used by recovery, and it gets confused by the boundary
375 first_bit = next_bit;
384 blfp->blf_size = nvecs;
389 * This is called to fill in the vector of log iovecs for the
390 * given log buf item. It fills the first entry with a buf log
391 * format structure, and the rest point to contiguous chunks
396 struct xfs_log_item *lip,
397 struct xfs_log_iovec *vecp)
399 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
400 struct xfs_buf *bp = bip->bli_buf;
404 ASSERT(atomic_read(&bip->bli_refcount) > 0);
405 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
406 (bip->bli_flags & XFS_BLI_STALE));
409 * If it is an inode buffer, transfer the in-memory state to the
410 * format flags and clear the in-memory state. We do not transfer
411 * this state if the inode buffer allocation has not yet been committed
412 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
413 * correct replay of the inode allocation.
415 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
416 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
417 xfs_log_item_in_current_chkpt(lip)))
418 bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
419 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
422 for (i = 0; i < bip->bli_format_count; i++) {
423 vecp = xfs_buf_item_format_segment(bip, vecp, offset,
424 &bip->bli_formats[i]);
425 offset += bp->b_maps[i].bm_len;
429 * Check to make sure everything is consistent.
431 trace_xfs_buf_item_format(bip);
432 xfs_buf_item_log_check(bip);
436 * This is called to pin the buffer associated with the buf log item in memory
437 * so it cannot be written out.
439 * We also always take a reference to the buffer log item here so that the bli
440 * is held while the item is pinned in memory. This means that we can
441 * unconditionally drop the reference count a transaction holds when the
442 * transaction is completed.
446 struct xfs_log_item *lip)
448 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
450 ASSERT(atomic_read(&bip->bli_refcount) > 0);
451 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
452 (bip->bli_flags & XFS_BLI_STALE));
454 trace_xfs_buf_item_pin(bip);
456 atomic_inc(&bip->bli_refcount);
457 atomic_inc(&bip->bli_buf->b_pin_count);
461 * This is called to unpin the buffer associated with the buf log
462 * item which was previously pinned with a call to xfs_buf_item_pin().
464 * Also drop the reference to the buf item for the current transaction.
465 * If the XFS_BLI_STALE flag is set and we are the last reference,
466 * then free up the buf log item and unlock the buffer.
468 * If the remove flag is set we are called from uncommit in the
469 * forced-shutdown path. If that is true and the reference count on
470 * the log item is going to drop to zero we need to free the item's
471 * descriptor in the transaction.
475 struct xfs_log_item *lip,
478 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
479 xfs_buf_t *bp = bip->bli_buf;
480 struct xfs_ail *ailp = lip->li_ailp;
481 int stale = bip->bli_flags & XFS_BLI_STALE;
484 ASSERT(bp->b_fspriv == bip);
485 ASSERT(atomic_read(&bip->bli_refcount) > 0);
487 trace_xfs_buf_item_unpin(bip);
489 freed = atomic_dec_and_test(&bip->bli_refcount);
491 if (atomic_dec_and_test(&bp->b_pin_count))
492 wake_up_all(&bp->b_waiters);
494 if (freed && stale) {
495 ASSERT(bip->bli_flags & XFS_BLI_STALE);
496 ASSERT(xfs_buf_islocked(bp));
497 ASSERT(XFS_BUF_ISSTALE(bp));
498 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
500 trace_xfs_buf_item_unpin_stale(bip);
504 * If we are in a transaction context, we have to
505 * remove the log item from the transaction as we are
506 * about to release our reference to the buffer. If we
507 * don't, the unlock that occurs later in
508 * xfs_trans_uncommit() will try to reference the
509 * buffer which we no longer have a hold on.
512 xfs_trans_del_item(lip);
515 * Since the transaction no longer refers to the buffer,
516 * the buffer should no longer refer to the transaction.
522 * If we get called here because of an IO error, we may
523 * or may not have the item on the AIL. xfs_trans_ail_delete()
524 * will take care of that situation.
525 * xfs_trans_ail_delete() drops the AIL lock.
527 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
528 xfs_buf_do_callbacks(bp);
532 spin_lock(&ailp->xa_lock);
533 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
534 xfs_buf_item_relse(bp);
535 ASSERT(bp->b_fspriv == NULL);
538 } else if (freed && remove) {
540 * There are currently two references to the buffer - the active
541 * LRU reference and the buf log item. What we are about to do
542 * here - simulate a failed IO completion - requires 3
545 * The LRU reference is removed by the xfs_buf_stale() call. The
546 * buf item reference is removed by the xfs_buf_iodone()
547 * callback that is run by xfs_buf_do_callbacks() during ioend
548 * processing (via the bp->b_iodone callback), and then finally
549 * the ioend processing will drop the IO reference if the buffer
550 * is marked XBF_ASYNC.
552 * Hence we need to take an additional reference here so that IO
553 * completion processing doesn't free the buffer prematurely.
557 bp->b_flags |= XBF_ASYNC;
558 xfs_buf_ioerror(bp, EIO);
561 xfs_buf_ioend(bp, 0);
567 struct xfs_log_item *lip,
568 struct list_head *buffer_list)
570 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
571 struct xfs_buf *bp = bip->bli_buf;
572 uint rval = XFS_ITEM_SUCCESS;
574 if (xfs_buf_ispinned(bp))
575 return XFS_ITEM_PINNED;
576 if (!xfs_buf_trylock(bp))
577 return XFS_ITEM_LOCKED;
579 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
581 trace_xfs_buf_item_push(bip);
583 if (!xfs_buf_delwri_queue(bp, buffer_list))
584 rval = XFS_ITEM_FLUSHING;
590 * Release the buffer associated with the buf log item. If there is no dirty
591 * logged data associated with the buffer recorded in the buf log item, then
592 * free the buf log item and remove the reference to it in the buffer.
594 * This call ignores the recursion count. It is only called when the buffer
595 * should REALLY be unlocked, regardless of the recursion count.
597 * We unconditionally drop the transaction's reference to the log item. If the
598 * item was logged, then another reference was taken when it was pinned, so we
599 * can safely drop the transaction reference now. This also allows us to avoid
600 * potential races with the unpin code freeing the bli by not referencing the
601 * bli after we've dropped the reference count.
603 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
604 * if necessary but do not unlock the buffer. This is for support of
605 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
610 struct xfs_log_item *lip)
612 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
613 struct xfs_buf *bp = bip->bli_buf;
614 int aborted, clean, i;
617 /* Clear the buffer's association with this transaction. */
621 * If this is a transaction abort, don't return early. Instead, allow
622 * the brelse to happen. Normally it would be done for stale
623 * (cancelled) buffers at unpin time, but we'll never go through the
624 * pin/unpin cycle if we abort inside commit.
626 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
629 * Before possibly freeing the buf item, determine if we should
630 * release the buffer at the end of this routine.
632 hold = bip->bli_flags & XFS_BLI_HOLD;
634 /* Clear the per transaction state. */
635 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
638 * If the buf item is marked stale, then don't do anything. We'll
639 * unlock the buffer and free the buf item when the buffer is unpinned
642 if (bip->bli_flags & XFS_BLI_STALE) {
643 trace_xfs_buf_item_unlock_stale(bip);
644 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
646 atomic_dec(&bip->bli_refcount);
651 trace_xfs_buf_item_unlock(bip);
654 * If the buf item isn't tracking any data, free it, otherwise drop the
655 * reference we hold to it.
658 for (i = 0; i < bip->bli_format_count; i++) {
659 if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
660 bip->bli_formats[i].blf_map_size)) {
666 xfs_buf_item_relse(bp);
668 atomic_dec(&bip->bli_refcount);
675 * This is called to find out where the oldest active copy of the
676 * buf log item in the on disk log resides now that the last log
677 * write of it completed at the given lsn.
678 * We always re-log all the dirty data in a buffer, so usually the
679 * latest copy in the on disk log is the only one that matters. For
680 * those cases we simply return the given lsn.
682 * The one exception to this is for buffers full of newly allocated
683 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
684 * flag set, indicating that only the di_next_unlinked fields from the
685 * inodes in the buffers will be replayed during recovery. If the
686 * original newly allocated inode images have not yet been flushed
687 * when the buffer is so relogged, then we need to make sure that we
688 * keep the old images in the 'active' portion of the log. We do this
689 * by returning the original lsn of that transaction here rather than
693 xfs_buf_item_committed(
694 struct xfs_log_item *lip,
697 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
699 trace_xfs_buf_item_committed(bip);
701 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
707 xfs_buf_item_committing(
708 struct xfs_log_item *lip,
709 xfs_lsn_t commit_lsn)
714 * This is the ops vector shared by all buf log items.
716 static const struct xfs_item_ops xfs_buf_item_ops = {
717 .iop_size = xfs_buf_item_size,
718 .iop_format = xfs_buf_item_format,
719 .iop_pin = xfs_buf_item_pin,
720 .iop_unpin = xfs_buf_item_unpin,
721 .iop_unlock = xfs_buf_item_unlock,
722 .iop_committed = xfs_buf_item_committed,
723 .iop_push = xfs_buf_item_push,
724 .iop_committing = xfs_buf_item_committing
728 xfs_buf_item_get_format(
729 struct xfs_buf_log_item *bip,
732 ASSERT(bip->bli_formats == NULL);
733 bip->bli_format_count = count;
736 bip->bli_formats = &bip->__bli_format;
740 bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
742 if (!bip->bli_formats)
748 xfs_buf_item_free_format(
749 struct xfs_buf_log_item *bip)
751 if (bip->bli_formats != &bip->__bli_format) {
752 kmem_free(bip->bli_formats);
753 bip->bli_formats = NULL;
758 * Allocate a new buf log item to go with the given buffer.
759 * Set the buffer's b_fsprivate field to point to the new
760 * buf log item. If there are other item's attached to the
761 * buffer (see xfs_buf_attach_iodone() below), then put the
762 * buf log item at the front.
769 xfs_log_item_t *lip = bp->b_fspriv;
770 xfs_buf_log_item_t *bip;
777 * Check to see if there is already a buf log item for
778 * this buffer. If there is, it is guaranteed to be
779 * the first. If we do already have one, there is
780 * nothing to do here so return.
782 ASSERT(bp->b_target->bt_mount == mp);
783 if (lip != NULL && lip->li_type == XFS_LI_BUF)
786 bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
787 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
792 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
793 * can be divided into. Make sure not to truncate any pieces.
794 * map_size is the size of the bitmap needed to describe the
795 * chunks of the buffer.
797 * Discontiguous buffer support follows the layout of the underlying
798 * buffer. This makes the implementation as simple as possible.
800 error = xfs_buf_item_get_format(bip, bp->b_map_count);
803 for (i = 0; i < bip->bli_format_count; i++) {
804 chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
806 map_size = DIV_ROUND_UP(chunks, NBWORD);
808 bip->bli_formats[i].blf_type = XFS_LI_BUF;
809 bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
810 bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
811 bip->bli_formats[i].blf_map_size = map_size;
814 #ifdef XFS_TRANS_DEBUG
816 * Allocate the arrays for tracking what needs to be logged
817 * and what our callers request to be logged. bli_orig
818 * holds a copy of the original, clean buffer for comparison
819 * against, and bli_logged keeps a 1 bit flag per byte in
820 * the buffer to indicate which bytes the callers have asked
823 bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP);
824 memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length));
825 bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP);
829 * Put the buf item into the list of items attached to the
830 * buffer at the front.
833 bip->bli_item.li_bio_list = bp->b_fspriv;
839 * Mark bytes first through last inclusive as dirty in the buf
843 xfs_buf_item_log_segment(
844 struct xfs_buf_log_item *bip,
860 * Convert byte offsets to bit numbers.
862 first_bit = first >> XFS_BLF_SHIFT;
863 last_bit = last >> XFS_BLF_SHIFT;
866 * Calculate the total number of bits to be set.
868 bits_to_set = last_bit - first_bit + 1;
871 * Get a pointer to the first word in the bitmap
874 word_num = first_bit >> BIT_TO_WORD_SHIFT;
875 wordp = &map[word_num];
878 * Calculate the starting bit in the first word.
880 bit = first_bit & (uint)(NBWORD - 1);
883 * First set any bits in the first word of our range.
884 * If it starts at bit 0 of the word, it will be
885 * set below rather than here. That is what the variable
886 * bit tells us. The variable bits_set tracks the number
887 * of bits that have been set so far. End_bit is the number
888 * of the last bit to be set in this word plus one.
891 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
892 mask = ((1 << (end_bit - bit)) - 1) << bit;
895 bits_set = end_bit - bit;
901 * Now set bits a whole word at a time that are between
902 * first_bit and last_bit.
904 while ((bits_to_set - bits_set) >= NBWORD) {
905 *wordp |= 0xffffffff;
911 * Finally, set any bits left to be set in one last partial word.
913 end_bit = bits_to_set - bits_set;
915 mask = (1 << end_bit) - 1;
919 xfs_buf_item_log_debug(bip, first, last);
923 * Mark bytes first through last inclusive as dirty in the buf
928 xfs_buf_log_item_t *bip,
935 struct xfs_buf *bp = bip->bli_buf;
938 * Mark the item as having some dirty data for
939 * quick reference in xfs_buf_item_dirty.
941 bip->bli_flags |= XFS_BLI_DIRTY;
944 * walk each buffer segment and mark them dirty appropriately.
947 for (i = 0; i < bip->bli_format_count; i++) {
950 end = start + BBTOB(bp->b_maps[i].bm_len);
952 start += BBTOB(bp->b_maps[i].bm_len);
960 xfs_buf_item_log_segment(bip, first, end,
961 &bip->bli_formats[i].blf_data_map[0]);
963 start += bp->b_maps[i].bm_len;
969 * Return 1 if the buffer has some data that has been logged (at any
970 * point, not just the current transaction) and 0 if not.
974 xfs_buf_log_item_t *bip)
976 return (bip->bli_flags & XFS_BLI_DIRTY);
981 xfs_buf_log_item_t *bip)
983 #ifdef XFS_TRANS_DEBUG
984 kmem_free(bip->bli_orig);
985 kmem_free(bip->bli_logged);
986 #endif /* XFS_TRANS_DEBUG */
988 xfs_buf_item_free_format(bip);
989 kmem_zone_free(xfs_buf_item_zone, bip);
993 * This is called when the buf log item is no longer needed. It should
994 * free the buf log item associated with the given buffer and clear
995 * the buffer's pointer to the buf log item. If there are no more
996 * items in the list, clear the b_iodone field of the buffer (see
997 * xfs_buf_attach_iodone() below).
1003 xfs_buf_log_item_t *bip;
1005 trace_xfs_buf_item_relse(bp, _RET_IP_);
1008 bp->b_fspriv = bip->bli_item.li_bio_list;
1009 if (bp->b_fspriv == NULL)
1010 bp->b_iodone = NULL;
1013 xfs_buf_item_free(bip);
1018 * Add the given log item with its callback to the list of callbacks
1019 * to be called when the buffer's I/O completes. If it is not set
1020 * already, set the buffer's b_iodone() routine to be
1021 * xfs_buf_iodone_callbacks() and link the log item into the list of
1022 * items rooted at b_fsprivate. Items are always added as the second
1023 * entry in the list if there is a first, because the buf item code
1024 * assumes that the buf log item is first.
1027 xfs_buf_attach_iodone(
1029 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
1030 xfs_log_item_t *lip)
1032 xfs_log_item_t *head_lip;
1034 ASSERT(xfs_buf_islocked(bp));
1037 head_lip = bp->b_fspriv;
1039 lip->li_bio_list = head_lip->li_bio_list;
1040 head_lip->li_bio_list = lip;
1045 ASSERT(bp->b_iodone == NULL ||
1046 bp->b_iodone == xfs_buf_iodone_callbacks);
1047 bp->b_iodone = xfs_buf_iodone_callbacks;
1051 * We can have many callbacks on a buffer. Running the callbacks individually
1052 * can cause a lot of contention on the AIL lock, so we allow for a single
1053 * callback to be able to scan the remaining lip->li_bio_list for other items
1054 * of the same type and callback to be processed in the first call.
1056 * As a result, the loop walking the callback list below will also modify the
1057 * list. it removes the first item from the list and then runs the callback.
1058 * The loop then restarts from the new head of the list. This allows the
1059 * callback to scan and modify the list attached to the buffer and we don't
1060 * have to care about maintaining a next item pointer.
1063 xfs_buf_do_callbacks(
1066 struct xfs_log_item *lip;
1068 while ((lip = bp->b_fspriv) != NULL) {
1069 bp->b_fspriv = lip->li_bio_list;
1070 ASSERT(lip->li_cb != NULL);
1072 * Clear the next pointer so we don't have any
1073 * confusion if the item is added to another buf.
1074 * Don't touch the log item after calling its
1075 * callback, because it could have freed itself.
1077 lip->li_bio_list = NULL;
1078 lip->li_cb(bp, lip);
1083 * This is the iodone() function for buffers which have had callbacks
1084 * attached to them by xfs_buf_attach_iodone(). It should remove each
1085 * log item from the buffer's list and call the callback of each in turn.
1086 * When done, the buffer's fsprivate field is set to NULL and the buffer
1087 * is unlocked with a call to iodone().
1090 xfs_buf_iodone_callbacks(
1093 struct xfs_log_item *lip = bp->b_fspriv;
1094 struct xfs_mount *mp = lip->li_mountp;
1095 static ulong lasttime;
1096 static xfs_buftarg_t *lasttarg;
1098 if (likely(!xfs_buf_geterror(bp)))
1102 * If we've already decided to shutdown the filesystem because of
1103 * I/O errors, there's no point in giving this a retry.
1105 if (XFS_FORCED_SHUTDOWN(mp)) {
1108 trace_xfs_buf_item_iodone(bp, _RET_IP_);
1112 if (bp->b_target != lasttarg ||
1113 time_after(jiffies, (lasttime + 5*HZ))) {
1115 xfs_buf_ioerror_alert(bp, __func__);
1117 lasttarg = bp->b_target;
1120 * If the write was asynchronous then no one will be looking for the
1121 * error. Clear the error state and write the buffer out again.
1123 * XXX: This helps against transient write errors, but we need to find
1124 * a way to shut the filesystem down if the writes keep failing.
1126 * In practice we'll shut the filesystem down soon as non-transient
1127 * erorrs tend to affect the whole device and a failing log write
1128 * will make us give up. But we really ought to do better here.
1130 if (XFS_BUF_ISASYNC(bp)) {
1131 ASSERT(bp->b_iodone != NULL);
1133 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1135 xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */
1137 if (!XFS_BUF_ISSTALE(bp)) {
1138 bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE;
1139 xfs_buf_iorequest(bp);
1148 * If the write of the buffer was synchronous, we want to make
1149 * sure to return the error to the caller of xfs_bwrite().
1154 trace_xfs_buf_error_relse(bp, _RET_IP_);
1157 xfs_buf_do_callbacks(bp);
1158 bp->b_fspriv = NULL;
1159 bp->b_iodone = NULL;
1160 xfs_buf_ioend(bp, 0);
1164 * This is the iodone() function for buffers which have been
1165 * logged. It is called when they are eventually flushed out.
1166 * It should remove the buf item from the AIL, and free the buf item.
1167 * It is called by xfs_buf_iodone_callbacks() above which will take
1168 * care of cleaning up the buffer itself.
1173 struct xfs_log_item *lip)
1175 struct xfs_ail *ailp = lip->li_ailp;
1177 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1182 * If we are forcibly shutting down, this may well be
1183 * off the AIL already. That's because we simulate the
1184 * log-committed callbacks to unpin these buffers. Or we may never
1185 * have put this item on AIL because of the transaction was
1186 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1188 * Either way, AIL is useless if we're forcing a shutdown.
1190 spin_lock(&ailp->xa_lock);
1191 xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1192 xfs_buf_item_free(BUF_ITEM(lip));