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"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_buf_item.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
34 kmem_zone_t *xfs_buf_item_zone;
36 static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
38 return container_of(lip, struct xfs_buf_log_item, bli_item);
42 #ifdef XFS_TRANS_DEBUG
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
53 xfs_buf_item_log_debug(
54 xfs_buf_log_item_t *bip,
67 ASSERT(bip->bli_logged != NULL);
69 nbytes = last - first + 1;
70 bfset(bip->bli_logged, first, nbytes);
71 for (x = 0; x < nbytes; x++) {
72 chunk_num = byte >> XFS_BLF_SHIFT;
73 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
74 bit_num = chunk_num & (NBWORD - 1);
75 wordp = &(bip->bli_format.blf_data_map[word_num]);
76 bit_set = *wordp & (1 << bit_num);
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
88 xfs_buf_item_flush_log_debug(
93 xfs_buf_log_item_t *bip;
96 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
97 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
101 ASSERT(bip->bli_logged != NULL);
102 nbytes = last - first + 1;
103 bfset(bip->bli_logged, first, nbytes);
107 * This function is called to verify that our callers have logged
108 * all the bytes that they changed.
110 * It does this by comparing the original copy of the buffer stored in
111 * the buf log item's bli_orig array to the current copy of the buffer
112 * and ensuring that all bytes which mismatch are set in the bli_logged
113 * array of the buf log item.
116 xfs_buf_item_log_check(
117 xfs_buf_log_item_t *bip)
124 ASSERT(bip->bli_orig != NULL);
125 ASSERT(bip->bli_logged != NULL);
128 ASSERT(XFS_BUF_COUNT(bp) > 0);
129 ASSERT(XFS_BUF_PTR(bp) != NULL);
130 orig = bip->bli_orig;
131 buffer = XFS_BUF_PTR(bp);
132 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
133 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
135 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
140 #define xfs_buf_item_log_debug(x,y,z)
141 #define xfs_buf_item_log_check(x)
144 STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
145 STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
148 * This returns the number of log iovecs needed to log the
149 * given buf log item.
151 * It calculates this as 1 iovec for the buf log format structure
152 * and 1 for each stretch of non-contiguous chunks to be logged.
153 * Contiguous chunks are logged in a single iovec.
155 * If the XFS_BLI_STALE flag has been set, then log nothing.
159 struct xfs_log_item *lip)
161 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
162 struct xfs_buf *bp = bip->bli_buf;
167 ASSERT(atomic_read(&bip->bli_refcount) > 0);
168 if (bip->bli_flags & XFS_BLI_STALE) {
170 * The buffer is stale, so all we need to log
171 * is the buf log format structure with the
174 trace_xfs_buf_item_size_stale(bip);
175 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
179 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
181 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
182 bip->bli_format.blf_map_size, 0);
183 ASSERT(last_bit != -1);
185 while (last_bit != -1) {
187 * This takes the bit number to start looking from and
188 * returns the next set bit from there. It returns -1
189 * if there are no more bits set or the start bit is
190 * beyond the end of the bitmap.
192 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
193 bip->bli_format.blf_map_size,
196 * If we run out of bits, leave the loop,
197 * else if we find a new set of bits bump the number of vecs,
198 * else keep scanning the current set of bits.
200 if (next_bit == -1) {
202 } else if (next_bit != last_bit + 1) {
205 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
206 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
215 trace_xfs_buf_item_size(bip);
220 * This is called to fill in the vector of log iovecs for the
221 * given log buf item. It fills the first entry with a buf log
222 * format structure, and the rest point to contiguous chunks
227 struct xfs_log_item *lip,
228 struct xfs_log_iovec *vecp)
230 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
231 struct xfs_buf *bp = bip->bli_buf;
240 ASSERT(atomic_read(&bip->bli_refcount) > 0);
241 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
242 (bip->bli_flags & XFS_BLI_STALE));
245 * The size of the base structure is the size of the
246 * declared structure plus the space for the extra words
247 * of the bitmap. We subtract one from the map size, because
248 * the first element of the bitmap is accounted for in the
249 * size of the base structure.
252 (uint)(sizeof(xfs_buf_log_format_t) +
253 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
254 vecp->i_addr = &bip->bli_format;
255 vecp->i_len = base_size;
256 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
261 * If it is an inode buffer, transfer the in-memory state to the
262 * format flags and clear the in-memory state. We do not transfer
263 * this state if the inode buffer allocation has not yet been committed
264 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
265 * correct replay of the inode allocation.
267 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
268 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
269 xfs_log_item_in_current_chkpt(lip)))
270 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
271 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
274 if (bip->bli_flags & XFS_BLI_STALE) {
276 * The buffer is stale, so all we need to log
277 * is the buf log format structure with the
280 trace_xfs_buf_item_format_stale(bip);
281 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
282 bip->bli_format.blf_size = nvecs;
287 * Fill in an iovec for each set of contiguous chunks.
289 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
290 bip->bli_format.blf_map_size, 0);
291 ASSERT(first_bit != -1);
292 last_bit = first_bit;
296 * This takes the bit number to start looking from and
297 * returns the next set bit from there. It returns -1
298 * if there are no more bits set or the start bit is
299 * beyond the end of the bitmap.
301 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
302 bip->bli_format.blf_map_size,
305 * If we run out of bits fill in the last iovec and get
307 * Else if we start a new set of bits then fill in the
308 * iovec for the series we were looking at and start
309 * counting the bits in the new one.
310 * Else we're still in the same set of bits so just
311 * keep counting and scanning.
313 if (next_bit == -1) {
314 buffer_offset = first_bit * XFS_BLF_CHUNK;
315 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
316 vecp->i_len = nbits * XFS_BLF_CHUNK;
317 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
320 } else if (next_bit != last_bit + 1) {
321 buffer_offset = first_bit * XFS_BLF_CHUNK;
322 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
323 vecp->i_len = nbits * XFS_BLF_CHUNK;
324 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
327 first_bit = next_bit;
330 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
331 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
333 buffer_offset = first_bit * XFS_BLF_CHUNK;
334 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
335 vecp->i_len = nbits * XFS_BLF_CHUNK;
336 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
337 /* You would think we need to bump the nvecs here too, but we do not
338 * this number is used by recovery, and it gets confused by the boundary
343 first_bit = next_bit;
351 bip->bli_format.blf_size = nvecs;
354 * Check to make sure everything is consistent.
356 trace_xfs_buf_item_format(bip);
357 xfs_buf_item_log_check(bip);
361 * This is called to pin the buffer associated with the buf log item in memory
362 * so it cannot be written out.
364 * We also always take a reference to the buffer log item here so that the bli
365 * is held while the item is pinned in memory. This means that we can
366 * unconditionally drop the reference count a transaction holds when the
367 * transaction is completed.
371 struct xfs_log_item *lip)
373 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
375 ASSERT(XFS_BUF_ISBUSY(bip->bli_buf));
376 ASSERT(atomic_read(&bip->bli_refcount) > 0);
377 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
378 (bip->bli_flags & XFS_BLI_STALE));
380 trace_xfs_buf_item_pin(bip);
382 atomic_inc(&bip->bli_refcount);
383 atomic_inc(&bip->bli_buf->b_pin_count);
387 * This is called to unpin the buffer associated with the buf log
388 * item which was previously pinned with a call to xfs_buf_item_pin().
390 * Also drop the reference to the buf item for the current transaction.
391 * If the XFS_BLI_STALE flag is set and we are the last reference,
392 * then free up the buf log item and unlock the buffer.
394 * If the remove flag is set we are called from uncommit in the
395 * forced-shutdown path. If that is true and the reference count on
396 * the log item is going to drop to zero we need to free the item's
397 * descriptor in the transaction.
401 struct xfs_log_item *lip,
404 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
405 xfs_buf_t *bp = bip->bli_buf;
406 struct xfs_ail *ailp = lip->li_ailp;
407 int stale = bip->bli_flags & XFS_BLI_STALE;
410 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
411 ASSERT(atomic_read(&bip->bli_refcount) > 0);
413 trace_xfs_buf_item_unpin(bip);
415 freed = atomic_dec_and_test(&bip->bli_refcount);
417 if (atomic_dec_and_test(&bp->b_pin_count))
418 wake_up_all(&bp->b_waiters);
420 if (freed && stale) {
421 ASSERT(bip->bli_flags & XFS_BLI_STALE);
422 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
423 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
424 ASSERT(XFS_BUF_ISSTALE(bp));
425 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
427 trace_xfs_buf_item_unpin_stale(bip);
431 * We have to remove the log item from the transaction
432 * as we are about to release our reference to the
433 * buffer. If we don't, the unlock that occurs later
434 * in xfs_trans_uncommit() will ry to reference the
435 * buffer which we no longer have a hold on.
437 xfs_trans_del_item(lip);
440 * Since the transaction no longer refers to the buffer,
441 * the buffer should no longer refer to the transaction.
443 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
447 * If we get called here because of an IO error, we may
448 * or may not have the item on the AIL. xfs_trans_ail_delete()
449 * will take care of that situation.
450 * xfs_trans_ail_delete() drops the AIL lock.
452 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
453 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
454 XFS_BUF_SET_FSPRIVATE(bp, NULL);
455 XFS_BUF_CLR_IODONE_FUNC(bp);
457 spin_lock(&ailp->xa_lock);
458 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
459 xfs_buf_item_relse(bp);
460 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
467 * This is called to attempt to lock the buffer associated with this
468 * buf log item. Don't sleep on the buffer lock. If we can't get
469 * the lock right away, return 0. If we can get the lock, take a
470 * reference to the buffer. If this is a delayed write buffer that
471 * needs AIL help to be written back, invoke the pushbuf routine
472 * rather than the normal success path.
475 xfs_buf_item_trylock(
476 struct xfs_log_item *lip)
478 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
479 struct xfs_buf *bp = bip->bli_buf;
481 if (XFS_BUF_ISPINNED(bp))
482 return XFS_ITEM_PINNED;
483 if (!XFS_BUF_CPSEMA(bp))
484 return XFS_ITEM_LOCKED;
486 /* take a reference to the buffer. */
489 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
490 trace_xfs_buf_item_trylock(bip);
491 if (XFS_BUF_ISDELAYWRITE(bp))
492 return XFS_ITEM_PUSHBUF;
493 return XFS_ITEM_SUCCESS;
497 * Release the buffer associated with the buf log item. If there is no dirty
498 * logged data associated with the buffer recorded in the buf log item, then
499 * free the buf log item and remove the reference to it in the buffer.
501 * This call ignores the recursion count. It is only called when the buffer
502 * should REALLY be unlocked, regardless of the recursion count.
504 * We unconditionally drop the transaction's reference to the log item. If the
505 * item was logged, then another reference was taken when it was pinned, so we
506 * can safely drop the transaction reference now. This also allows us to avoid
507 * potential races with the unpin code freeing the bli by not referencing the
508 * bli after we've dropped the reference count.
510 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
511 * if necessary but do not unlock the buffer. This is for support of
512 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
517 struct xfs_log_item *lip)
519 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
520 struct xfs_buf *bp = bip->bli_buf;
524 /* Clear the buffer's association with this transaction. */
525 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
528 * If this is a transaction abort, don't return early. Instead, allow
529 * the brelse to happen. Normally it would be done for stale
530 * (cancelled) buffers at unpin time, but we'll never go through the
531 * pin/unpin cycle if we abort inside commit.
533 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
536 * Before possibly freeing the buf item, determine if we should
537 * release the buffer at the end of this routine.
539 hold = bip->bli_flags & XFS_BLI_HOLD;
541 /* Clear the per transaction state. */
542 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
545 * If the buf item is marked stale, then don't do anything. We'll
546 * unlock the buffer and free the buf item when the buffer is unpinned
549 if (bip->bli_flags & XFS_BLI_STALE) {
550 trace_xfs_buf_item_unlock_stale(bip);
551 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
553 atomic_dec(&bip->bli_refcount);
558 trace_xfs_buf_item_unlock(bip);
561 * If the buf item isn't tracking any data, free it, otherwise drop the
562 * reference we hold to it.
564 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
565 bip->bli_format.blf_map_size))
566 xfs_buf_item_relse(bp);
568 atomic_dec(&bip->bli_refcount);
575 * This is called to find out where the oldest active copy of the
576 * buf log item in the on disk log resides now that the last log
577 * write of it completed at the given lsn.
578 * We always re-log all the dirty data in a buffer, so usually the
579 * latest copy in the on disk log is the only one that matters. For
580 * those cases we simply return the given lsn.
582 * The one exception to this is for buffers full of newly allocated
583 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
584 * flag set, indicating that only the di_next_unlinked fields from the
585 * inodes in the buffers will be replayed during recovery. If the
586 * original newly allocated inode images have not yet been flushed
587 * when the buffer is so relogged, then we need to make sure that we
588 * keep the old images in the 'active' portion of the log. We do this
589 * by returning the original lsn of that transaction here rather than
593 xfs_buf_item_committed(
594 struct xfs_log_item *lip,
597 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
599 trace_xfs_buf_item_committed(bip);
601 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
607 * The buffer is locked, but is not a delayed write buffer. This happens
608 * if we race with IO completion and hence we don't want to try to write it
609 * again. Just release the buffer.
613 struct xfs_log_item *lip)
615 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
616 struct xfs_buf *bp = bip->bli_buf;
618 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
619 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
621 trace_xfs_buf_item_push(bip);
627 * The buffer is locked and is a delayed write buffer. Promote the buffer
628 * in the delayed write queue as the caller knows that they must invoke
629 * the xfsbufd to get this buffer written. We have to unlock the buffer
630 * to allow the xfsbufd to write it, too.
633 xfs_buf_item_pushbuf(
634 struct xfs_log_item *lip)
636 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
637 struct xfs_buf *bp = bip->bli_buf;
639 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
640 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
642 trace_xfs_buf_item_pushbuf(bip);
644 xfs_buf_delwri_promote(bp);
649 xfs_buf_item_committing(
650 struct xfs_log_item *lip,
651 xfs_lsn_t commit_lsn)
656 * This is the ops vector shared by all buf log items.
658 static struct xfs_item_ops xfs_buf_item_ops = {
659 .iop_size = xfs_buf_item_size,
660 .iop_format = xfs_buf_item_format,
661 .iop_pin = xfs_buf_item_pin,
662 .iop_unpin = xfs_buf_item_unpin,
663 .iop_trylock = xfs_buf_item_trylock,
664 .iop_unlock = xfs_buf_item_unlock,
665 .iop_committed = xfs_buf_item_committed,
666 .iop_push = xfs_buf_item_push,
667 .iop_pushbuf = xfs_buf_item_pushbuf,
668 .iop_committing = xfs_buf_item_committing
673 * Allocate a new buf log item to go with the given buffer.
674 * Set the buffer's b_fsprivate field to point to the new
675 * buf log item. If there are other item's attached to the
676 * buffer (see xfs_buf_attach_iodone() below), then put the
677 * buf log item at the front.
685 xfs_buf_log_item_t *bip;
690 * Check to see if there is already a buf log item for
691 * this buffer. If there is, it is guaranteed to be
692 * the first. If we do already have one, there is
693 * nothing to do here so return.
695 if (bp->b_mount != mp)
697 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
698 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
699 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
700 if (lip->li_type == XFS_LI_BUF) {
706 * chunks is the number of XFS_BLF_CHUNK size pieces
707 * the buffer can be divided into. Make sure not to
708 * truncate any pieces. map_size is the size of the
709 * bitmap needed to describe the chunks of the buffer.
711 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
712 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
714 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
716 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
719 bip->bli_format.blf_type = XFS_LI_BUF;
720 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
721 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
722 bip->bli_format.blf_map_size = map_size;
724 #ifdef XFS_TRANS_DEBUG
726 * Allocate the arrays for tracking what needs to be logged
727 * and what our callers request to be logged. bli_orig
728 * holds a copy of the original, clean buffer for comparison
729 * against, and bli_logged keeps a 1 bit flag per byte in
730 * the buffer to indicate which bytes the callers have asked
733 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
734 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
735 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
739 * Put the buf item into the list of items attached to the
740 * buffer at the front.
742 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
743 bip->bli_item.li_bio_list =
744 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
746 XFS_BUF_SET_FSPRIVATE(bp, bip);
751 * Mark bytes first through last inclusive as dirty in the buf
756 xfs_buf_log_item_t *bip,
771 * Mark the item as having some dirty data for
772 * quick reference in xfs_buf_item_dirty.
774 bip->bli_flags |= XFS_BLI_DIRTY;
777 * Convert byte offsets to bit numbers.
779 first_bit = first >> XFS_BLF_SHIFT;
780 last_bit = last >> XFS_BLF_SHIFT;
783 * Calculate the total number of bits to be set.
785 bits_to_set = last_bit - first_bit + 1;
788 * Get a pointer to the first word in the bitmap
791 word_num = first_bit >> BIT_TO_WORD_SHIFT;
792 wordp = &(bip->bli_format.blf_data_map[word_num]);
795 * Calculate the starting bit in the first word.
797 bit = first_bit & (uint)(NBWORD - 1);
800 * First set any bits in the first word of our range.
801 * If it starts at bit 0 of the word, it will be
802 * set below rather than here. That is what the variable
803 * bit tells us. The variable bits_set tracks the number
804 * of bits that have been set so far. End_bit is the number
805 * of the last bit to be set in this word plus one.
808 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
809 mask = ((1 << (end_bit - bit)) - 1) << bit;
812 bits_set = end_bit - bit;
818 * Now set bits a whole word at a time that are between
819 * first_bit and last_bit.
821 while ((bits_to_set - bits_set) >= NBWORD) {
822 *wordp |= 0xffffffff;
828 * Finally, set any bits left to be set in one last partial word.
830 end_bit = bits_to_set - bits_set;
832 mask = (1 << end_bit) - 1;
836 xfs_buf_item_log_debug(bip, first, last);
841 * Return 1 if the buffer has some data that has been logged (at any
842 * point, not just the current transaction) and 0 if not.
846 xfs_buf_log_item_t *bip)
848 return (bip->bli_flags & XFS_BLI_DIRTY);
853 xfs_buf_log_item_t *bip)
855 #ifdef XFS_TRANS_DEBUG
856 kmem_free(bip->bli_orig);
857 kmem_free(bip->bli_logged);
858 #endif /* XFS_TRANS_DEBUG */
860 kmem_zone_free(xfs_buf_item_zone, bip);
864 * This is called when the buf log item is no longer needed. It should
865 * free the buf log item associated with the given buffer and clear
866 * the buffer's pointer to the buf log item. If there are no more
867 * items in the list, clear the b_iodone field of the buffer (see
868 * xfs_buf_attach_iodone() below).
874 xfs_buf_log_item_t *bip;
876 trace_xfs_buf_item_relse(bp, _RET_IP_);
878 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
879 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
880 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
881 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
882 XFS_BUF_CLR_IODONE_FUNC(bp);
885 xfs_buf_item_free(bip);
890 * Add the given log item with its callback to the list of callbacks
891 * to be called when the buffer's I/O completes. If it is not set
892 * already, set the buffer's b_iodone() routine to be
893 * xfs_buf_iodone_callbacks() and link the log item into the list of
894 * items rooted at b_fsprivate. Items are always added as the second
895 * entry in the list if there is a first, because the buf item code
896 * assumes that the buf log item is first.
899 xfs_buf_attach_iodone(
901 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
904 xfs_log_item_t *head_lip;
906 ASSERT(XFS_BUF_ISBUSY(bp));
907 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
910 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
911 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
912 lip->li_bio_list = head_lip->li_bio_list;
913 head_lip->li_bio_list = lip;
915 XFS_BUF_SET_FSPRIVATE(bp, lip);
918 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
919 (XFS_BUF_IODONE_FUNC(bp) == NULL));
920 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
924 xfs_buf_do_callbacks(
928 xfs_log_item_t *nlip;
930 while (lip != NULL) {
931 nlip = lip->li_bio_list;
932 ASSERT(lip->li_cb != NULL);
934 * Clear the next pointer so we don't have any
935 * confusion if the item is added to another buf.
936 * Don't touch the log item after calling its
937 * callback, because it could have freed itself.
939 lip->li_bio_list = NULL;
946 * This is the iodone() function for buffers which have had callbacks
947 * attached to them by xfs_buf_attach_iodone(). It should remove each
948 * log item from the buffer's list and call the callback of each in turn.
949 * When done, the buffer's fsprivate field is set to NULL and the buffer
950 * is unlocked with a call to iodone().
953 xfs_buf_iodone_callbacks(
957 static ulong lasttime;
958 static xfs_buftarg_t *lasttarg;
961 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
962 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
964 if (XFS_BUF_GETERROR(bp) != 0) {
966 * If we've already decided to shutdown the filesystem
967 * because of IO errors, there's no point in giving this
971 if (XFS_FORCED_SHUTDOWN(mp)) {
972 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
973 XFS_BUF_SUPER_STALE(bp);
974 trace_xfs_buf_item_iodone(bp, _RET_IP_);
975 xfs_buf_do_callbacks(bp, lip);
976 XFS_BUF_SET_FSPRIVATE(bp, NULL);
977 XFS_BUF_CLR_IODONE_FUNC(bp);
982 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
983 (time_after(jiffies, (lasttime + 5*HZ)))) {
985 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
986 " block 0x%llx in %s",
987 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
988 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
990 lasttarg = XFS_BUF_TARGET(bp);
992 if (XFS_BUF_ISASYNC(bp)) {
994 * If the write was asynchronous then noone will be
995 * looking for the error. Clear the error state
996 * and write the buffer out again delayed write.
998 * XXXsup This is OK, so long as we catch these
999 * before we start the umount; we don't want these
1000 * DELWRI metadata bufs to be hanging around.
1002 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1004 if (!(XFS_BUF_ISSTALE(bp))) {
1005 XFS_BUF_DELAYWRITE(bp);
1007 XFS_BUF_SET_START(bp);
1009 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1010 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1014 * If the write of the buffer was not asynchronous,
1015 * then we want to make sure to return the error
1016 * to the caller of bwrite(). Because of this we
1017 * cannot clear the B_ERROR state at this point.
1018 * Instead we install a callback function that
1019 * will be called when the buffer is released, and
1020 * that routine will clear the error state and
1021 * set the buffer to be written out again after
1024 /* We actually overwrite the existing b-relse
1025 function at times, but we're gonna be shutting down
1027 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1029 XFS_BUF_FINISH_IOWAIT(bp);
1034 xfs_buf_do_callbacks(bp, lip);
1035 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1036 XFS_BUF_CLR_IODONE_FUNC(bp);
1041 * This is a callback routine attached to a buffer which gets an error
1042 * when being written out synchronously.
1045 xfs_buf_error_relse(
1048 xfs_log_item_t *lip;
1051 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1052 mp = (xfs_mount_t *)lip->li_mountp;
1053 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1057 XFS_BUF_UNDELAYWRITE(bp);
1058 XFS_BUF_ERROR(bp,0);
1060 trace_xfs_buf_error_relse(bp, _RET_IP_);
1062 if (! XFS_FORCED_SHUTDOWN(mp))
1063 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1065 * We have to unpin the pinned buffers so do the
1068 xfs_buf_do_callbacks(bp, lip);
1069 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1070 XFS_BUF_CLR_IODONE_FUNC(bp);
1071 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1077 * This is the iodone() function for buffers which have been
1078 * logged. It is called when they are eventually flushed out.
1079 * It should remove the buf item from the AIL, and free the buf item.
1080 * It is called by xfs_buf_iodone_callbacks() above which will take
1081 * care of cleaning up the buffer itself.
1086 struct xfs_log_item *lip)
1088 struct xfs_ail *ailp = lip->li_ailp;
1090 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1095 * If we are forcibly shutting down, this may well be
1096 * off the AIL already. That's because we simulate the
1097 * log-committed callbacks to unpin these buffers. Or we may never
1098 * have put this item on AIL because of the transaction was
1099 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1101 * Either way, AIL is useless if we're forcing a shutdown.
1103 spin_lock(&ailp->xa_lock);
1104 xfs_trans_ail_delete(ailp, lip);
1105 xfs_buf_item_free(BUF_ITEM(lip));