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_log_format.h"
21 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_trans.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
30 #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
);
41 STATIC
void xfs_buf_do_callbacks(struct xfs_buf
*bp
);
44 xfs_buf_log_format_size(
45 struct xfs_buf_log_format
*blfp
)
47 return offsetof(struct xfs_buf_log_format
, blf_data_map
) +
48 (blfp
->blf_map_size
* sizeof(blfp
->blf_data_map
[0]));
52 * This returns the number of log iovecs needed to log the
55 * It calculates this as 1 iovec for the buf log format structure
56 * and 1 for each stretch of non-contiguous chunks to be logged.
57 * Contiguous chunks are logged in a single iovec.
59 * If the XFS_BLI_STALE flag has been set, then log nothing.
62 xfs_buf_item_size_segment(
63 struct xfs_buf_log_item
*bip
,
64 struct xfs_buf_log_format
*blfp
,
68 struct xfs_buf
*bp
= bip
->bli_buf
;
72 last_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
77 * initial count for a dirty buffer is 2 vectors - the format structure
78 * and the first dirty region.
81 *nbytes
+= xfs_buf_log_format_size(blfp
) + XFS_BLF_CHUNK
;
83 while (last_bit
!= -1) {
85 * This takes the bit number to start looking from and
86 * returns the next set bit from there. It returns -1
87 * if there are no more bits set or the start bit is
88 * beyond the end of the bitmap.
90 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
93 * If we run out of bits, leave the loop,
94 * else if we find a new set of bits bump the number of vecs,
95 * else keep scanning the current set of bits.
99 } else if (next_bit
!= last_bit
+ 1) {
102 } else if (xfs_buf_offset(bp
, next_bit
* XFS_BLF_CHUNK
) !=
103 (xfs_buf_offset(bp
, last_bit
* XFS_BLF_CHUNK
) +
110 *nbytes
+= XFS_BLF_CHUNK
;
115 * This returns the number of log iovecs needed to log the given buf log item.
117 * It calculates this as 1 iovec for the buf log format structure and 1 for each
118 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
121 * Discontiguous buffers need a format structure per region that that is being
122 * logged. This makes the changes in the buffer appear to log recovery as though
123 * they came from separate buffers, just like would occur if multiple buffers
124 * were used instead of a single discontiguous buffer. This enables
125 * discontiguous buffers to be in-memory constructs, completely transparent to
126 * what ends up on disk.
128 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
133 struct xfs_log_item
*lip
,
137 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
140 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
141 if (bip
->bli_flags
& XFS_BLI_STALE
) {
143 * The buffer is stale, so all we need to log
144 * is the buf log format structure with the
147 trace_xfs_buf_item_size_stale(bip
);
148 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
149 *nvecs
+= bip
->bli_format_count
;
150 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
151 *nbytes
+= xfs_buf_log_format_size(&bip
->bli_formats
[i
]);
156 ASSERT(bip
->bli_flags
& XFS_BLI_LOGGED
);
158 if (bip
->bli_flags
& XFS_BLI_ORDERED
) {
160 * The buffer has been logged just to order it.
161 * It is not being included in the transaction
162 * commit, so no vectors are used at all.
164 trace_xfs_buf_item_size_ordered(bip
);
165 *nvecs
= XFS_LOG_VEC_ORDERED
;
170 * the vector count is based on the number of buffer vectors we have
171 * dirty bits in. This will only be greater than one when we have a
172 * compound buffer with more than one segment dirty. Hence for compound
173 * buffers we need to track which segment the dirty bits correspond to,
174 * and when we move from one segment to the next increment the vector
175 * count for the extra buf log format structure that will need to be
178 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
179 xfs_buf_item_size_segment(bip
, &bip
->bli_formats
[i
],
182 trace_xfs_buf_item_size(bip
);
185 static inline struct xfs_log_iovec
*
186 xfs_buf_item_copy_iovec(
187 struct xfs_log_iovec
*vecp
,
193 offset
+= first_bit
* XFS_BLF_CHUNK
;
195 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
196 vecp
->i_addr
= xfs_buf_offset(bp
, offset
);
197 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
202 xfs_buf_item_straddle(
208 return xfs_buf_offset(bp
, offset
+ (next_bit
<< XFS_BLF_SHIFT
)) !=
209 (xfs_buf_offset(bp
, offset
+ (last_bit
<< XFS_BLF_SHIFT
)) +
213 static struct xfs_log_iovec
*
214 xfs_buf_item_format_segment(
215 struct xfs_buf_log_item
*bip
,
216 struct xfs_log_iovec
*vecp
,
218 struct xfs_buf_log_format
*blfp
)
220 struct xfs_buf
*bp
= bip
->bli_buf
;
228 /* copy the flags across from the base format item */
229 blfp
->blf_flags
= bip
->__bli_format
.blf_flags
;
232 * Base size is the actual size of the ondisk structure - it reflects
233 * the actual size of the dirty bitmap rather than the size of the in
236 base_size
= xfs_buf_log_format_size(blfp
);
239 first_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
240 if (!(bip
->bli_flags
& XFS_BLI_STALE
) && first_bit
== -1) {
242 * If the map is not be dirty in the transaction, mark
243 * the size as zero and do not advance the vector pointer.
249 vecp
->i_len
= base_size
;
250 vecp
->i_type
= XLOG_REG_TYPE_BFORMAT
;
254 if (bip
->bli_flags
& XFS_BLI_STALE
) {
256 * The buffer is stale, so all we need to log
257 * is the buf log format structure with the
260 trace_xfs_buf_item_format_stale(bip
);
261 ASSERT(blfp
->blf_flags
& XFS_BLF_CANCEL
);
267 * Fill in an iovec for each set of contiguous chunks.
269 last_bit
= first_bit
;
273 * This takes the bit number to start looking from and
274 * returns the next set bit from there. It returns -1
275 * if there are no more bits set or the start bit is
276 * beyond the end of the bitmap.
278 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
281 * If we run out of bits fill in the last iovec and get out of
282 * the loop. Else if we start a new set of bits then fill in
283 * the iovec for the series we were looking at and start
284 * counting the bits in the new one. Else we're still in the
285 * same set of bits so just keep counting and scanning.
287 if (next_bit
== -1) {
288 xfs_buf_item_copy_iovec(vecp
, bp
, offset
,
292 } else if (next_bit
!= last_bit
+ 1 ||
293 xfs_buf_item_straddle(bp
, offset
, next_bit
, last_bit
)) {
294 vecp
= xfs_buf_item_copy_iovec(vecp
, bp
, offset
,
297 first_bit
= next_bit
;
306 blfp
->blf_size
= nvecs
;
311 * This is called to fill in the vector of log iovecs for the
312 * given log buf item. It fills the first entry with a buf log
313 * format structure, and the rest point to contiguous chunks
318 struct xfs_log_item
*lip
,
319 struct xfs_log_iovec
*vecp
)
321 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
322 struct xfs_buf
*bp
= bip
->bli_buf
;
326 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
327 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
328 (bip
->bli_flags
& XFS_BLI_STALE
));
331 * If it is an inode buffer, transfer the in-memory state to the
332 * format flags and clear the in-memory state.
334 * For buffer based inode allocation, we do not transfer
335 * this state if the inode buffer allocation has not yet been committed
336 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
337 * correct replay of the inode allocation.
339 * For icreate item based inode allocation, the buffers aren't written
340 * to the journal during allocation, and hence we should always tag the
341 * buffer as an inode buffer so that the correct unlinked list replay
342 * occurs during recovery.
344 if (bip
->bli_flags
& XFS_BLI_INODE_BUF
) {
345 if (xfs_sb_version_hascrc(&lip
->li_mountp
->m_sb
) ||
346 !((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
347 xfs_log_item_in_current_chkpt(lip
)))
348 bip
->__bli_format
.blf_flags
|= XFS_BLF_INODE_BUF
;
349 bip
->bli_flags
&= ~XFS_BLI_INODE_BUF
;
352 if ((bip
->bli_flags
& (XFS_BLI_ORDERED
|XFS_BLI_STALE
)) ==
355 * The buffer has been logged just to order it. It is not being
356 * included in the transaction commit, so don't format it.
358 trace_xfs_buf_item_format_ordered(bip
);
362 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
363 vecp
= xfs_buf_item_format_segment(bip
, vecp
, offset
,
364 &bip
->bli_formats
[i
]);
365 offset
+= bp
->b_maps
[i
].bm_len
;
369 * Check to make sure everything is consistent.
371 trace_xfs_buf_item_format(bip
);
375 * This is called to pin the buffer associated with the buf log item in memory
376 * so it cannot be written out.
378 * We also always take a reference to the buffer log item here so that the bli
379 * is held while the item is pinned in memory. This means that we can
380 * unconditionally drop the reference count a transaction holds when the
381 * transaction is completed.
385 struct xfs_log_item
*lip
)
387 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
389 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
390 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
391 (bip
->bli_flags
& XFS_BLI_ORDERED
) ||
392 (bip
->bli_flags
& XFS_BLI_STALE
));
394 trace_xfs_buf_item_pin(bip
);
396 atomic_inc(&bip
->bli_refcount
);
397 atomic_inc(&bip
->bli_buf
->b_pin_count
);
401 * This is called to unpin the buffer associated with the buf log
402 * item which was previously pinned with a call to xfs_buf_item_pin().
404 * Also drop the reference to the buf item for the current transaction.
405 * If the XFS_BLI_STALE flag is set and we are the last reference,
406 * then free up the buf log item and unlock the buffer.
408 * If the remove flag is set we are called from uncommit in the
409 * forced-shutdown path. If that is true and the reference count on
410 * the log item is going to drop to zero we need to free the item's
411 * descriptor in the transaction.
415 struct xfs_log_item
*lip
,
418 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
419 xfs_buf_t
*bp
= bip
->bli_buf
;
420 struct xfs_ail
*ailp
= lip
->li_ailp
;
421 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
424 ASSERT(bp
->b_fspriv
== bip
);
425 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
427 trace_xfs_buf_item_unpin(bip
);
429 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
431 if (atomic_dec_and_test(&bp
->b_pin_count
))
432 wake_up_all(&bp
->b_waiters
);
434 if (freed
&& stale
) {
435 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
436 ASSERT(xfs_buf_islocked(bp
));
437 ASSERT(XFS_BUF_ISSTALE(bp
));
438 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
440 trace_xfs_buf_item_unpin_stale(bip
);
444 * If we are in a transaction context, we have to
445 * remove the log item from the transaction as we are
446 * about to release our reference to the buffer. If we
447 * don't, the unlock that occurs later in
448 * xfs_trans_uncommit() will try to reference the
449 * buffer which we no longer have a hold on.
452 xfs_trans_del_item(lip
);
455 * Since the transaction no longer refers to the buffer,
456 * the buffer should no longer refer to the transaction.
462 * If we get called here because of an IO error, we may
463 * or may not have the item on the AIL. xfs_trans_ail_delete()
464 * will take care of that situation.
465 * xfs_trans_ail_delete() drops the AIL lock.
467 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
468 xfs_buf_do_callbacks(bp
);
472 spin_lock(&ailp
->xa_lock
);
473 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_LOG_IO_ERROR
);
474 xfs_buf_item_relse(bp
);
475 ASSERT(bp
->b_fspriv
== NULL
);
478 } else if (freed
&& remove
) {
480 * There are currently two references to the buffer - the active
481 * LRU reference and the buf log item. What we are about to do
482 * here - simulate a failed IO completion - requires 3
485 * The LRU reference is removed by the xfs_buf_stale() call. The
486 * buf item reference is removed by the xfs_buf_iodone()
487 * callback that is run by xfs_buf_do_callbacks() during ioend
488 * processing (via the bp->b_iodone callback), and then finally
489 * the ioend processing will drop the IO reference if the buffer
490 * is marked XBF_ASYNC.
492 * Hence we need to take an additional reference here so that IO
493 * completion processing doesn't free the buffer prematurely.
497 bp
->b_flags
|= XBF_ASYNC
;
498 xfs_buf_ioerror(bp
, EIO
);
501 xfs_buf_ioend(bp
, 0);
507 struct xfs_log_item
*lip
,
508 struct list_head
*buffer_list
)
510 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
511 struct xfs_buf
*bp
= bip
->bli_buf
;
512 uint rval
= XFS_ITEM_SUCCESS
;
514 if (xfs_buf_ispinned(bp
))
515 return XFS_ITEM_PINNED
;
516 if (!xfs_buf_trylock(bp
)) {
518 * If we have just raced with a buffer being pinned and it has
519 * been marked stale, we could end up stalling until someone else
520 * issues a log force to unpin the stale buffer. Check for the
521 * race condition here so xfsaild recognizes the buffer is pinned
522 * and queues a log force to move it along.
524 if (xfs_buf_ispinned(bp
))
525 return XFS_ITEM_PINNED
;
526 return XFS_ITEM_LOCKED
;
529 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
531 trace_xfs_buf_item_push(bip
);
533 if (!xfs_buf_delwri_queue(bp
, buffer_list
))
534 rval
= XFS_ITEM_FLUSHING
;
540 * Release the buffer associated with the buf log item. If there is no dirty
541 * logged data associated with the buffer recorded in the buf log item, then
542 * free the buf log item and remove the reference to it in the buffer.
544 * This call ignores the recursion count. It is only called when the buffer
545 * should REALLY be unlocked, regardless of the recursion count.
547 * We unconditionally drop the transaction's reference to the log item. If the
548 * item was logged, then another reference was taken when it was pinned, so we
549 * can safely drop the transaction reference now. This also allows us to avoid
550 * potential races with the unpin code freeing the bli by not referencing the
551 * bli after we've dropped the reference count.
553 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
554 * if necessary but do not unlock the buffer. This is for support of
555 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
560 struct xfs_log_item
*lip
)
562 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
563 struct xfs_buf
*bp
= bip
->bli_buf
;
568 /* Clear the buffer's association with this transaction. */
572 * If this is a transaction abort, don't return early. Instead, allow
573 * the brelse to happen. Normally it would be done for stale
574 * (cancelled) buffers at unpin time, but we'll never go through the
575 * pin/unpin cycle if we abort inside commit.
577 aborted
= (lip
->li_flags
& XFS_LI_ABORTED
) ? true : false;
579 * Before possibly freeing the buf item, copy the per-transaction state
580 * so we can reference it safely later after clearing it from the
583 flags
= bip
->bli_flags
;
584 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
| XFS_BLI_ORDERED
);
587 * If the buf item is marked stale, then don't do anything. We'll
588 * unlock the buffer and free the buf item when the buffer is unpinned
591 if (flags
& XFS_BLI_STALE
) {
592 trace_xfs_buf_item_unlock_stale(bip
);
593 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
595 atomic_dec(&bip
->bli_refcount
);
600 trace_xfs_buf_item_unlock(bip
);
603 * If the buf item isn't tracking any data, free it, otherwise drop the
604 * reference we hold to it. If we are aborting the transaction, this may
605 * be the only reference to the buf item, so we free it anyway
606 * regardless of whether it is dirty or not. A dirty abort implies a
609 * Ordered buffers are dirty but may have no recorded changes, so ensure
610 * we only release clean items here.
612 clean
= (flags
& XFS_BLI_DIRTY
) ? false : true;
615 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
616 if (!xfs_bitmap_empty(bip
->bli_formats
[i
].blf_data_map
,
617 bip
->bli_formats
[i
].blf_map_size
)) {
625 * Clean buffers, by definition, cannot be in the AIL. However, aborted
626 * buffers may be dirty and hence in the AIL. Therefore if we are
627 * aborting a buffer and we've just taken the last refernce away, we
628 * have to check if it is in the AIL before freeing it. We need to free
629 * it in this case, because an aborted transaction has already shut the
630 * filesystem down and this is the last chance we will have to do so.
632 if (atomic_dec_and_test(&bip
->bli_refcount
)) {
634 xfs_buf_item_relse(bp
);
636 ASSERT(XFS_FORCED_SHUTDOWN(lip
->li_mountp
));
637 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
638 spin_lock(&lip
->li_ailp
->xa_lock
);
639 xfs_trans_ail_delete(lip
->li_ailp
, lip
,
640 SHUTDOWN_LOG_IO_ERROR
);
642 xfs_buf_item_relse(bp
);
646 if (!(flags
& XFS_BLI_HOLD
))
651 * This is called to find out where the oldest active copy of the
652 * buf log item in the on disk log resides now that the last log
653 * write of it completed at the given lsn.
654 * We always re-log all the dirty data in a buffer, so usually the
655 * latest copy in the on disk log is the only one that matters. For
656 * those cases we simply return the given lsn.
658 * The one exception to this is for buffers full of newly allocated
659 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
660 * flag set, indicating that only the di_next_unlinked fields from the
661 * inodes in the buffers will be replayed during recovery. If the
662 * original newly allocated inode images have not yet been flushed
663 * when the buffer is so relogged, then we need to make sure that we
664 * keep the old images in the 'active' portion of the log. We do this
665 * by returning the original lsn of that transaction here rather than
669 xfs_buf_item_committed(
670 struct xfs_log_item
*lip
,
673 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
675 trace_xfs_buf_item_committed(bip
);
677 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
683 xfs_buf_item_committing(
684 struct xfs_log_item
*lip
,
685 xfs_lsn_t commit_lsn
)
690 * This is the ops vector shared by all buf log items.
692 static const struct xfs_item_ops xfs_buf_item_ops
= {
693 .iop_size
= xfs_buf_item_size
,
694 .iop_format
= xfs_buf_item_format
,
695 .iop_pin
= xfs_buf_item_pin
,
696 .iop_unpin
= xfs_buf_item_unpin
,
697 .iop_unlock
= xfs_buf_item_unlock
,
698 .iop_committed
= xfs_buf_item_committed
,
699 .iop_push
= xfs_buf_item_push
,
700 .iop_committing
= xfs_buf_item_committing
704 xfs_buf_item_get_format(
705 struct xfs_buf_log_item
*bip
,
708 ASSERT(bip
->bli_formats
== NULL
);
709 bip
->bli_format_count
= count
;
712 bip
->bli_formats
= &bip
->__bli_format
;
716 bip
->bli_formats
= kmem_zalloc(count
* sizeof(struct xfs_buf_log_format
),
718 if (!bip
->bli_formats
)
724 xfs_buf_item_free_format(
725 struct xfs_buf_log_item
*bip
)
727 if (bip
->bli_formats
!= &bip
->__bli_format
) {
728 kmem_free(bip
->bli_formats
);
729 bip
->bli_formats
= NULL
;
734 * Allocate a new buf log item to go with the given buffer.
735 * Set the buffer's b_fsprivate field to point to the new
736 * buf log item. If there are other item's attached to the
737 * buffer (see xfs_buf_attach_iodone() below), then put the
738 * buf log item at the front.
745 xfs_log_item_t
*lip
= bp
->b_fspriv
;
746 xfs_buf_log_item_t
*bip
;
753 * Check to see if there is already a buf log item for
754 * this buffer. If there is, it is guaranteed to be
755 * the first. If we do already have one, there is
756 * nothing to do here so return.
758 ASSERT(bp
->b_target
->bt_mount
== mp
);
759 if (lip
!= NULL
&& lip
->li_type
== XFS_LI_BUF
)
762 bip
= kmem_zone_zalloc(xfs_buf_item_zone
, KM_SLEEP
);
763 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
768 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
769 * can be divided into. Make sure not to truncate any pieces.
770 * map_size is the size of the bitmap needed to describe the
771 * chunks of the buffer.
773 * Discontiguous buffer support follows the layout of the underlying
774 * buffer. This makes the implementation as simple as possible.
776 error
= xfs_buf_item_get_format(bip
, bp
->b_map_count
);
779 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
780 chunks
= DIV_ROUND_UP(BBTOB(bp
->b_maps
[i
].bm_len
),
782 map_size
= DIV_ROUND_UP(chunks
, NBWORD
);
784 bip
->bli_formats
[i
].blf_type
= XFS_LI_BUF
;
785 bip
->bli_formats
[i
].blf_blkno
= bp
->b_maps
[i
].bm_bn
;
786 bip
->bli_formats
[i
].blf_len
= bp
->b_maps
[i
].bm_len
;
787 bip
->bli_formats
[i
].blf_map_size
= map_size
;
790 #ifdef XFS_TRANS_DEBUG
792 * Allocate the arrays for tracking what needs to be logged
793 * and what our callers request to be logged. bli_orig
794 * holds a copy of the original, clean buffer for comparison
795 * against, and bli_logged keeps a 1 bit flag per byte in
796 * the buffer to indicate which bytes the callers have asked
799 bip
->bli_orig
= kmem_alloc(BBTOB(bp
->b_length
), KM_SLEEP
);
800 memcpy(bip
->bli_orig
, bp
->b_addr
, BBTOB(bp
->b_length
));
801 bip
->bli_logged
= kmem_zalloc(BBTOB(bp
->b_length
) / NBBY
, KM_SLEEP
);
805 * Put the buf item into the list of items attached to the
806 * buffer at the front.
809 bip
->bli_item
.li_bio_list
= bp
->b_fspriv
;
815 * Mark bytes first through last inclusive as dirty in the buf
819 xfs_buf_item_log_segment(
820 struct xfs_buf_log_item
*bip
,
836 * Convert byte offsets to bit numbers.
838 first_bit
= first
>> XFS_BLF_SHIFT
;
839 last_bit
= last
>> XFS_BLF_SHIFT
;
842 * Calculate the total number of bits to be set.
844 bits_to_set
= last_bit
- first_bit
+ 1;
847 * Get a pointer to the first word in the bitmap
850 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
851 wordp
= &map
[word_num
];
854 * Calculate the starting bit in the first word.
856 bit
= first_bit
& (uint
)(NBWORD
- 1);
859 * First set any bits in the first word of our range.
860 * If it starts at bit 0 of the word, it will be
861 * set below rather than here. That is what the variable
862 * bit tells us. The variable bits_set tracks the number
863 * of bits that have been set so far. End_bit is the number
864 * of the last bit to be set in this word plus one.
867 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
868 mask
= ((1 << (end_bit
- bit
)) - 1) << bit
;
871 bits_set
= end_bit
- bit
;
877 * Now set bits a whole word at a time that are between
878 * first_bit and last_bit.
880 while ((bits_to_set
- bits_set
) >= NBWORD
) {
881 *wordp
|= 0xffffffff;
887 * Finally, set any bits left to be set in one last partial word.
889 end_bit
= bits_to_set
- bits_set
;
891 mask
= (1 << end_bit
) - 1;
897 * Mark bytes first through last inclusive as dirty in the buf
902 xfs_buf_log_item_t
*bip
,
909 struct xfs_buf
*bp
= bip
->bli_buf
;
912 * walk each buffer segment and mark them dirty appropriately.
915 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
918 end
= start
+ BBTOB(bp
->b_maps
[i
].bm_len
);
920 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
928 xfs_buf_item_log_segment(bip
, first
, end
,
929 &bip
->bli_formats
[i
].blf_data_map
[0]);
931 start
+= bp
->b_maps
[i
].bm_len
;
937 * Return 1 if the buffer has been logged or ordered in a transaction (at any
938 * point, not just the current transaction) and 0 if not.
942 xfs_buf_log_item_t
*bip
)
944 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
949 xfs_buf_log_item_t
*bip
)
951 #ifdef XFS_TRANS_DEBUG
952 kmem_free(bip
->bli_orig
);
953 kmem_free(bip
->bli_logged
);
954 #endif /* XFS_TRANS_DEBUG */
956 xfs_buf_item_free_format(bip
);
957 kmem_zone_free(xfs_buf_item_zone
, bip
);
961 * This is called when the buf log item is no longer needed. It should
962 * free the buf log item associated with the given buffer and clear
963 * the buffer's pointer to the buf log item. If there are no more
964 * items in the list, clear the b_iodone field of the buffer (see
965 * xfs_buf_attach_iodone() below).
971 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
973 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
974 ASSERT(!(bip
->bli_item
.li_flags
& XFS_LI_IN_AIL
));
976 bp
->b_fspriv
= bip
->bli_item
.li_bio_list
;
977 if (bp
->b_fspriv
== NULL
)
981 xfs_buf_item_free(bip
);
986 * Add the given log item with its callback to the list of callbacks
987 * to be called when the buffer's I/O completes. If it is not set
988 * already, set the buffer's b_iodone() routine to be
989 * xfs_buf_iodone_callbacks() and link the log item into the list of
990 * items rooted at b_fsprivate. Items are always added as the second
991 * entry in the list if there is a first, because the buf item code
992 * assumes that the buf log item is first.
995 xfs_buf_attach_iodone(
997 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
1000 xfs_log_item_t
*head_lip
;
1002 ASSERT(xfs_buf_islocked(bp
));
1005 head_lip
= bp
->b_fspriv
;
1007 lip
->li_bio_list
= head_lip
->li_bio_list
;
1008 head_lip
->li_bio_list
= lip
;
1013 ASSERT(bp
->b_iodone
== NULL
||
1014 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
1015 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
1019 * We can have many callbacks on a buffer. Running the callbacks individually
1020 * can cause a lot of contention on the AIL lock, so we allow for a single
1021 * callback to be able to scan the remaining lip->li_bio_list for other items
1022 * of the same type and callback to be processed in the first call.
1024 * As a result, the loop walking the callback list below will also modify the
1025 * list. it removes the first item from the list and then runs the callback.
1026 * The loop then restarts from the new head of the list. This allows the
1027 * callback to scan and modify the list attached to the buffer and we don't
1028 * have to care about maintaining a next item pointer.
1031 xfs_buf_do_callbacks(
1034 struct xfs_log_item
*lip
;
1036 while ((lip
= bp
->b_fspriv
) != NULL
) {
1037 bp
->b_fspriv
= lip
->li_bio_list
;
1038 ASSERT(lip
->li_cb
!= NULL
);
1040 * Clear the next pointer so we don't have any
1041 * confusion if the item is added to another buf.
1042 * Don't touch the log item after calling its
1043 * callback, because it could have freed itself.
1045 lip
->li_bio_list
= NULL
;
1046 lip
->li_cb(bp
, lip
);
1051 * This is the iodone() function for buffers which have had callbacks
1052 * attached to them by xfs_buf_attach_iodone(). It should remove each
1053 * log item from the buffer's list and call the callback of each in turn.
1054 * When done, the buffer's fsprivate field is set to NULL and the buffer
1055 * is unlocked with a call to iodone().
1058 xfs_buf_iodone_callbacks(
1061 struct xfs_log_item
*lip
= bp
->b_fspriv
;
1062 struct xfs_mount
*mp
= lip
->li_mountp
;
1063 static ulong lasttime
;
1064 static xfs_buftarg_t
*lasttarg
;
1066 if (likely(!xfs_buf_geterror(bp
)))
1070 * If we've already decided to shutdown the filesystem because of
1071 * I/O errors, there's no point in giving this a retry.
1073 if (XFS_FORCED_SHUTDOWN(mp
)) {
1076 trace_xfs_buf_item_iodone(bp
, _RET_IP_
);
1080 if (bp
->b_target
!= lasttarg
||
1081 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
1083 xfs_buf_ioerror_alert(bp
, __func__
);
1085 lasttarg
= bp
->b_target
;
1088 * If the write was asynchronous then no one will be looking for the
1089 * error. Clear the error state and write the buffer out again.
1091 * XXX: This helps against transient write errors, but we need to find
1092 * a way to shut the filesystem down if the writes keep failing.
1094 * In practice we'll shut the filesystem down soon as non-transient
1095 * erorrs tend to affect the whole device and a failing log write
1096 * will make us give up. But we really ought to do better here.
1098 if (XFS_BUF_ISASYNC(bp
)) {
1099 ASSERT(bp
->b_iodone
!= NULL
);
1101 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1103 xfs_buf_ioerror(bp
, 0); /* errno of 0 unsets the flag */
1105 if (!XFS_BUF_ISSTALE(bp
)) {
1106 bp
->b_flags
|= XBF_WRITE
| XBF_ASYNC
| XBF_DONE
;
1107 xfs_buf_iorequest(bp
);
1116 * If the write of the buffer was synchronous, we want to make
1117 * sure to return the error to the caller of xfs_bwrite().
1122 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1125 xfs_buf_do_callbacks(bp
);
1126 bp
->b_fspriv
= NULL
;
1127 bp
->b_iodone
= NULL
;
1128 xfs_buf_ioend(bp
, 0);
1132 * This is the iodone() function for buffers which have been
1133 * logged. It is called when they are eventually flushed out.
1134 * It should remove the buf item from the AIL, and free the buf item.
1135 * It is called by xfs_buf_iodone_callbacks() above which will take
1136 * care of cleaning up the buffer itself.
1141 struct xfs_log_item
*lip
)
1143 struct xfs_ail
*ailp
= lip
->li_ailp
;
1145 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1150 * If we are forcibly shutting down, this may well be
1151 * off the AIL already. That's because we simulate the
1152 * log-committed callbacks to unpin these buffers. Or we may never
1153 * have put this item on AIL because of the transaction was
1154 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1156 * Either way, AIL is useless if we're forcing a shutdown.
1158 spin_lock(&ailp
->xa_lock
);
1159 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1160 xfs_buf_item_free(BUF_ITEM(lip
));