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_format.h"
21 #include "xfs_log_format.h"
22 #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"
32 #include "xfs_inode.h"
35 kmem_zone_t
*xfs_buf_item_zone
;
37 static inline struct xfs_buf_log_item
*BUF_ITEM(struct xfs_log_item
*lip
)
39 return container_of(lip
, struct xfs_buf_log_item
, bli_item
);
42 STATIC
void xfs_buf_do_callbacks(struct xfs_buf
*bp
);
45 xfs_buf_log_format_size(
46 struct xfs_buf_log_format
*blfp
)
48 return offsetof(struct xfs_buf_log_format
, blf_data_map
) +
49 (blfp
->blf_map_size
* sizeof(blfp
->blf_data_map
[0]));
53 * This returns the number of log iovecs needed to log the
56 * It calculates this as 1 iovec for the buf log format structure
57 * and 1 for each stretch of non-contiguous chunks to be logged.
58 * Contiguous chunks are logged in a single iovec.
60 * If the XFS_BLI_STALE flag has been set, then log nothing.
63 xfs_buf_item_size_segment(
64 struct xfs_buf_log_item
*bip
,
65 struct xfs_buf_log_format
*blfp
,
69 struct xfs_buf
*bp
= bip
->bli_buf
;
73 last_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
78 * initial count for a dirty buffer is 2 vectors - the format structure
79 * and the first dirty region.
82 *nbytes
+= xfs_buf_log_format_size(blfp
) + XFS_BLF_CHUNK
;
84 while (last_bit
!= -1) {
86 * This takes the bit number to start looking from and
87 * returns the next set bit from there. It returns -1
88 * if there are no more bits set or the start bit is
89 * beyond the end of the bitmap.
91 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
94 * If we run out of bits, leave the loop,
95 * else if we find a new set of bits bump the number of vecs,
96 * else keep scanning the current set of bits.
100 } else if (next_bit
!= last_bit
+ 1) {
103 } else if (xfs_buf_offset(bp
, next_bit
* XFS_BLF_CHUNK
) !=
104 (xfs_buf_offset(bp
, last_bit
* XFS_BLF_CHUNK
) +
111 *nbytes
+= XFS_BLF_CHUNK
;
116 * This returns the number of log iovecs needed to log the given buf log item.
118 * It calculates this as 1 iovec for the buf log format structure and 1 for each
119 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
122 * Discontiguous buffers need a format structure per region that that is being
123 * logged. This makes the changes in the buffer appear to log recovery as though
124 * they came from separate buffers, just like would occur if multiple buffers
125 * were used instead of a single discontiguous buffer. This enables
126 * discontiguous buffers to be in-memory constructs, completely transparent to
127 * what ends up on disk.
129 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
134 struct xfs_log_item
*lip
,
138 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
141 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
142 if (bip
->bli_flags
& XFS_BLI_STALE
) {
144 * The buffer is stale, so all we need to log
145 * is the buf log format structure with the
148 trace_xfs_buf_item_size_stale(bip
);
149 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
150 *nvecs
+= bip
->bli_format_count
;
151 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
152 *nbytes
+= xfs_buf_log_format_size(&bip
->bli_formats
[i
]);
157 ASSERT(bip
->bli_flags
& XFS_BLI_LOGGED
);
159 if (bip
->bli_flags
& XFS_BLI_ORDERED
) {
161 * The buffer has been logged just to order it.
162 * It is not being included in the transaction
163 * commit, so no vectors are used at all.
165 trace_xfs_buf_item_size_ordered(bip
);
166 *nvecs
= XFS_LOG_VEC_ORDERED
;
171 * the vector count is based on the number of buffer vectors we have
172 * dirty bits in. This will only be greater than one when we have a
173 * compound buffer with more than one segment dirty. Hence for compound
174 * buffers we need to track which segment the dirty bits correspond to,
175 * and when we move from one segment to the next increment the vector
176 * count for the extra buf log format structure that will need to be
179 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
180 xfs_buf_item_size_segment(bip
, &bip
->bli_formats
[i
],
183 trace_xfs_buf_item_size(bip
);
187 xfs_buf_item_copy_iovec(
188 struct xfs_log_vec
*lv
,
189 struct xfs_log_iovec
**vecp
,
195 offset
+= first_bit
* XFS_BLF_CHUNK
;
196 xlog_copy_iovec(lv
, vecp
, XLOG_REG_TYPE_BCHUNK
,
197 xfs_buf_offset(bp
, offset
),
198 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
)) +
214 xfs_buf_item_format_segment(
215 struct xfs_buf_log_item
*bip
,
216 struct xfs_log_vec
*lv
,
217 struct xfs_log_iovec
**vecp
,
219 struct xfs_buf_log_format
*blfp
)
221 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
);
238 first_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
239 if (!(bip
->bli_flags
& XFS_BLI_STALE
) && first_bit
== -1) {
241 * If the map is not be dirty in the transaction, mark
242 * the size as zero and do not advance the vector pointer.
247 blfp
= xlog_copy_iovec(lv
, vecp
, XLOG_REG_TYPE_BFORMAT
, blfp
, base_size
);
250 if (bip
->bli_flags
& XFS_BLI_STALE
) {
252 * The buffer is stale, so all we need to log
253 * is the buf log format structure with the
256 trace_xfs_buf_item_format_stale(bip
);
257 ASSERT(blfp
->blf_flags
& XFS_BLF_CANCEL
);
263 * Fill in an iovec for each set of contiguous chunks.
265 last_bit
= first_bit
;
269 * This takes the bit number to start looking from and
270 * returns the next set bit from there. It returns -1
271 * if there are no more bits set or the start bit is
272 * beyond the end of the bitmap.
274 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
277 * If we run out of bits fill in the last iovec and get out of
278 * the loop. Else if we start a new set of bits then fill in
279 * the iovec for the series we were looking at and start
280 * counting the bits in the new one. Else we're still in the
281 * same set of bits so just keep counting and scanning.
283 if (next_bit
== -1) {
284 xfs_buf_item_copy_iovec(lv
, vecp
, bp
, offset
,
288 } else if (next_bit
!= last_bit
+ 1 ||
289 xfs_buf_item_straddle(bp
, offset
, next_bit
, last_bit
)) {
290 xfs_buf_item_copy_iovec(lv
, vecp
, bp
, offset
,
293 first_bit
= next_bit
;
304 * This is called to fill in the vector of log iovecs for the
305 * given log buf item. It fills the first entry with a buf log
306 * format structure, and the rest point to contiguous chunks
311 struct xfs_log_item
*lip
,
312 struct xfs_log_vec
*lv
)
314 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
315 struct xfs_buf
*bp
= bip
->bli_buf
;
316 struct xfs_log_iovec
*vecp
= NULL
;
320 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
321 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
322 (bip
->bli_flags
& XFS_BLI_STALE
));
323 ASSERT((bip
->bli_flags
& XFS_BLI_STALE
) ||
324 (xfs_blft_from_flags(&bip
->__bli_format
) > XFS_BLFT_UNKNOWN_BUF
325 && xfs_blft_from_flags(&bip
->__bli_format
) < XFS_BLFT_MAX_BUF
));
329 * If it is an inode buffer, transfer the in-memory state to the
330 * format flags and clear the in-memory state.
332 * For buffer based inode allocation, we do not transfer
333 * this state if the inode buffer allocation has not yet been committed
334 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
335 * correct replay of the inode allocation.
337 * For icreate item based inode allocation, the buffers aren't written
338 * to the journal during allocation, and hence we should always tag the
339 * buffer as an inode buffer so that the correct unlinked list replay
340 * occurs during recovery.
342 if (bip
->bli_flags
& XFS_BLI_INODE_BUF
) {
343 if (xfs_sb_version_hascrc(&lip
->li_mountp
->m_sb
) ||
344 !((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
345 xfs_log_item_in_current_chkpt(lip
)))
346 bip
->__bli_format
.blf_flags
|= XFS_BLF_INODE_BUF
;
347 bip
->bli_flags
&= ~XFS_BLI_INODE_BUF
;
350 if ((bip
->bli_flags
& (XFS_BLI_ORDERED
|XFS_BLI_STALE
)) ==
353 * The buffer has been logged just to order it. It is not being
354 * included in the transaction commit, so don't format it.
356 trace_xfs_buf_item_format_ordered(bip
);
360 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
361 xfs_buf_item_format_segment(bip
, lv
, &vecp
, offset
,
362 &bip
->bli_formats
[i
]);
363 offset
+= BBTOB(bp
->b_maps
[i
].bm_len
);
367 * Check to make sure everything is consistent.
369 trace_xfs_buf_item_format(bip
);
373 * This is called to pin the buffer associated with the buf log item in memory
374 * so it cannot be written out.
376 * We also always take a reference to the buffer log item here so that the bli
377 * is held while the item is pinned in memory. This means that we can
378 * unconditionally drop the reference count a transaction holds when the
379 * transaction is completed.
383 struct xfs_log_item
*lip
)
385 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
387 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
388 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
389 (bip
->bli_flags
& XFS_BLI_ORDERED
) ||
390 (bip
->bli_flags
& XFS_BLI_STALE
));
392 trace_xfs_buf_item_pin(bip
);
394 atomic_inc(&bip
->bli_refcount
);
395 atomic_inc(&bip
->bli_buf
->b_pin_count
);
399 * This is called to unpin the buffer associated with the buf log
400 * item which was previously pinned with a call to xfs_buf_item_pin().
402 * Also drop the reference to the buf item for the current transaction.
403 * If the XFS_BLI_STALE flag is set and we are the last reference,
404 * then free up the buf log item and unlock the buffer.
406 * If the remove flag is set we are called from uncommit in the
407 * forced-shutdown path. If that is true and the reference count on
408 * the log item is going to drop to zero we need to free the item's
409 * descriptor in the transaction.
413 struct xfs_log_item
*lip
,
416 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
417 xfs_buf_t
*bp
= bip
->bli_buf
;
418 struct xfs_ail
*ailp
= lip
->li_ailp
;
419 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
422 ASSERT(bp
->b_fspriv
== bip
);
423 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
425 trace_xfs_buf_item_unpin(bip
);
427 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
429 if (atomic_dec_and_test(&bp
->b_pin_count
))
430 wake_up_all(&bp
->b_waiters
);
432 if (freed
&& stale
) {
433 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
434 ASSERT(xfs_buf_islocked(bp
));
435 ASSERT(bp
->b_flags
& XBF_STALE
);
436 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
438 trace_xfs_buf_item_unpin_stale(bip
);
442 * If we are in a transaction context, we have to
443 * remove the log item from the transaction as we are
444 * about to release our reference to the buffer. If we
445 * don't, the unlock that occurs later in
446 * xfs_trans_uncommit() will try to reference the
447 * buffer which we no longer have a hold on.
450 xfs_trans_del_item(lip
);
453 * Since the transaction no longer refers to the buffer,
454 * the buffer should no longer refer to the transaction.
460 * If we get called here because of an IO error, we may
461 * or may not have the item on the AIL. xfs_trans_ail_delete()
462 * will take care of that situation.
463 * xfs_trans_ail_delete() drops the AIL lock.
465 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
466 xfs_buf_do_callbacks(bp
);
470 spin_lock(&ailp
->xa_lock
);
471 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_LOG_IO_ERROR
);
472 xfs_buf_item_relse(bp
);
473 ASSERT(bp
->b_fspriv
== NULL
);
476 } else if (freed
&& remove
) {
478 * There are currently two references to the buffer - the active
479 * LRU reference and the buf log item. What we are about to do
480 * here - simulate a failed IO completion - requires 3
483 * The LRU reference is removed by the xfs_buf_stale() call. The
484 * buf item reference is removed by the xfs_buf_iodone()
485 * callback that is run by xfs_buf_do_callbacks() during ioend
486 * processing (via the bp->b_iodone callback), and then finally
487 * the ioend processing will drop the IO reference if the buffer
488 * is marked XBF_ASYNC.
490 * Hence we need to take an additional reference here so that IO
491 * completion processing doesn't free the buffer prematurely.
495 bp
->b_flags
|= XBF_ASYNC
;
496 xfs_buf_ioerror(bp
, -EIO
);
497 bp
->b_flags
&= ~XBF_DONE
;
504 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
505 * seconds so as to not spam logs too much on repeated detection of the same
509 static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state
, 30 * HZ
, 10);
513 struct xfs_log_item
*lip
,
514 struct list_head
*buffer_list
)
516 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
517 struct xfs_buf
*bp
= bip
->bli_buf
;
518 uint rval
= XFS_ITEM_SUCCESS
;
520 if (xfs_buf_ispinned(bp
))
521 return XFS_ITEM_PINNED
;
522 if (!xfs_buf_trylock(bp
)) {
524 * If we have just raced with a buffer being pinned and it has
525 * been marked stale, we could end up stalling until someone else
526 * issues a log force to unpin the stale buffer. Check for the
527 * race condition here so xfsaild recognizes the buffer is pinned
528 * and queues a log force to move it along.
530 if (xfs_buf_ispinned(bp
))
531 return XFS_ITEM_PINNED
;
532 return XFS_ITEM_LOCKED
;
535 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
537 trace_xfs_buf_item_push(bip
);
539 /* has a previous flush failed due to IO errors? */
540 if ((bp
->b_flags
& XBF_WRITE_FAIL
) &&
541 ___ratelimit(&xfs_buf_write_fail_rl_state
, "XFS: Failing async write")) {
542 xfs_warn(bp
->b_target
->bt_mount
,
543 "Failing async write on buffer block 0x%llx. Retrying async write.",
544 (long long)bp
->b_bn
);
547 if (!xfs_buf_delwri_queue(bp
, buffer_list
))
548 rval
= XFS_ITEM_FLUSHING
;
554 * Release the buffer associated with the buf log item. If there is no dirty
555 * logged data associated with the buffer recorded in the buf log item, then
556 * free the buf log item and remove the reference to it in the buffer.
558 * This call ignores the recursion count. It is only called when the buffer
559 * should REALLY be unlocked, regardless of the recursion count.
561 * We unconditionally drop the transaction's reference to the log item. If the
562 * item was logged, then another reference was taken when it was pinned, so we
563 * can safely drop the transaction reference now. This also allows us to avoid
564 * potential races with the unpin code freeing the bli by not referencing the
565 * bli after we've dropped the reference count.
567 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
568 * if necessary but do not unlock the buffer. This is for support of
569 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
574 struct xfs_log_item
*lip
)
576 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
577 struct xfs_buf
*bp
= bip
->bli_buf
;
578 bool aborted
= !!(lip
->li_flags
& XFS_LI_ABORTED
);
579 bool hold
= !!(bip
->bli_flags
& XFS_BLI_HOLD
);
580 bool dirty
= !!(bip
->bli_flags
& XFS_BLI_DIRTY
);
581 bool ordered
= !!(bip
->bli_flags
& XFS_BLI_ORDERED
);
583 /* Clear the buffer's association with this transaction. */
587 * The per-transaction state has been copied above so clear it from the
590 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
| XFS_BLI_ORDERED
);
593 * If the buf item is marked stale, then don't do anything. We'll
594 * unlock the buffer and free the buf item when the buffer is unpinned
597 if (bip
->bli_flags
& XFS_BLI_STALE
) {
598 trace_xfs_buf_item_unlock_stale(bip
);
599 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
601 atomic_dec(&bip
->bli_refcount
);
606 trace_xfs_buf_item_unlock(bip
);
609 * If the buf item isn't tracking any data, free it, otherwise drop the
610 * reference we hold to it. If we are aborting the transaction, this may
611 * be the only reference to the buf item, so we free it anyway
612 * regardless of whether it is dirty or not. A dirty abort implies a
615 * The bli dirty state should match whether the blf has logged segments
616 * except for ordered buffers, where only the bli should be dirty.
618 ASSERT((!ordered
&& dirty
== xfs_buf_item_dirty_format(bip
)) ||
619 (ordered
&& dirty
&& !xfs_buf_item_dirty_format(bip
)));
622 * Clean buffers, by definition, cannot be in the AIL. However, aborted
623 * buffers may be in the AIL regardless of dirty state. An aborted
624 * transaction that invalidates a buffer already in the AIL may have
625 * marked it stale and cleared the dirty state, for example.
627 * Therefore if we are aborting a buffer and we've just taken the last
628 * reference away, we have to check if it is in the AIL before freeing
629 * it. We need to free it in this case, because an aborted transaction
630 * has already shut the filesystem down and this is the last chance we
631 * will have to do so.
633 if (atomic_dec_and_test(&bip
->bli_refcount
)) {
635 ASSERT(XFS_FORCED_SHUTDOWN(lip
->li_mountp
));
636 xfs_trans_ail_remove(lip
, SHUTDOWN_LOG_IO_ERROR
);
637 xfs_buf_item_relse(bp
);
639 xfs_buf_item_relse(bp
);
647 * This is called to find out where the oldest active copy of the
648 * buf log item in the on disk log resides now that the last log
649 * write of it completed at the given lsn.
650 * We always re-log all the dirty data in a buffer, so usually the
651 * latest copy in the on disk log is the only one that matters. For
652 * those cases we simply return the given lsn.
654 * The one exception to this is for buffers full of newly allocated
655 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
656 * flag set, indicating that only the di_next_unlinked fields from the
657 * inodes in the buffers will be replayed during recovery. If the
658 * original newly allocated inode images have not yet been flushed
659 * when the buffer is so relogged, then we need to make sure that we
660 * keep the old images in the 'active' portion of the log. We do this
661 * by returning the original lsn of that transaction here rather than
665 xfs_buf_item_committed(
666 struct xfs_log_item
*lip
,
669 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
671 trace_xfs_buf_item_committed(bip
);
673 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
679 xfs_buf_item_committing(
680 struct xfs_log_item
*lip
,
681 xfs_lsn_t commit_lsn
)
686 * This is the ops vector shared by all buf log items.
688 static const struct xfs_item_ops xfs_buf_item_ops
= {
689 .iop_size
= xfs_buf_item_size
,
690 .iop_format
= xfs_buf_item_format
,
691 .iop_pin
= xfs_buf_item_pin
,
692 .iop_unpin
= xfs_buf_item_unpin
,
693 .iop_unlock
= xfs_buf_item_unlock
,
694 .iop_committed
= xfs_buf_item_committed
,
695 .iop_push
= xfs_buf_item_push
,
696 .iop_committing
= xfs_buf_item_committing
700 xfs_buf_item_get_format(
701 struct xfs_buf_log_item
*bip
,
704 ASSERT(bip
->bli_formats
== NULL
);
705 bip
->bli_format_count
= count
;
708 bip
->bli_formats
= &bip
->__bli_format
;
712 bip
->bli_formats
= kmem_zalloc(count
* sizeof(struct xfs_buf_log_format
),
714 if (!bip
->bli_formats
)
720 xfs_buf_item_free_format(
721 struct xfs_buf_log_item
*bip
)
723 if (bip
->bli_formats
!= &bip
->__bli_format
) {
724 kmem_free(bip
->bli_formats
);
725 bip
->bli_formats
= NULL
;
730 * Allocate a new buf log item to go with the given buffer.
731 * Set the buffer's b_fsprivate field to point to the new
732 * buf log item. If there are other item's attached to the
733 * buffer (see xfs_buf_attach_iodone() below), then put the
734 * buf log item at the front.
739 struct xfs_mount
*mp
)
741 struct xfs_log_item
*lip
= bp
->b_fspriv
;
742 struct xfs_buf_log_item
*bip
;
749 * Check to see if there is already a buf log item for
750 * this buffer. If there is, it is guaranteed to be
751 * the first. If we do already have one, there is
752 * nothing to do here so return.
754 ASSERT(bp
->b_target
->bt_mount
== mp
);
755 if (lip
!= NULL
&& lip
->li_type
== XFS_LI_BUF
)
758 bip
= kmem_zone_zalloc(xfs_buf_item_zone
, KM_SLEEP
);
759 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
763 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
764 * can be divided into. Make sure not to truncate any pieces.
765 * map_size is the size of the bitmap needed to describe the
766 * chunks of the buffer.
768 * Discontiguous buffer support follows the layout of the underlying
769 * buffer. This makes the implementation as simple as possible.
771 error
= xfs_buf_item_get_format(bip
, bp
->b_map_count
);
773 if (error
) { /* to stop gcc throwing set-but-unused warnings */
774 kmem_zone_free(xfs_buf_item_zone
, bip
);
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
;
791 * Put the buf item into the list of items attached to the
792 * buffer at the front.
795 bip
->bli_item
.li_bio_list
= bp
->b_fspriv
;
803 * Mark bytes first through last inclusive as dirty in the buf
807 xfs_buf_item_log_segment(
823 * Convert byte offsets to bit numbers.
825 first_bit
= first
>> XFS_BLF_SHIFT
;
826 last_bit
= last
>> XFS_BLF_SHIFT
;
829 * Calculate the total number of bits to be set.
831 bits_to_set
= last_bit
- first_bit
+ 1;
834 * Get a pointer to the first word in the bitmap
837 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
838 wordp
= &map
[word_num
];
841 * Calculate the starting bit in the first word.
843 bit
= first_bit
& (uint
)(NBWORD
- 1);
846 * First set any bits in the first word of our range.
847 * If it starts at bit 0 of the word, it will be
848 * set below rather than here. That is what the variable
849 * bit tells us. The variable bits_set tracks the number
850 * of bits that have been set so far. End_bit is the number
851 * of the last bit to be set in this word plus one.
854 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
855 mask
= ((1U << (end_bit
- bit
)) - 1) << bit
;
858 bits_set
= end_bit
- bit
;
864 * Now set bits a whole word at a time that are between
865 * first_bit and last_bit.
867 while ((bits_to_set
- bits_set
) >= NBWORD
) {
868 *wordp
|= 0xffffffff;
874 * Finally, set any bits left to be set in one last partial word.
876 end_bit
= bits_to_set
- bits_set
;
878 mask
= (1U << end_bit
) - 1;
884 * Mark bytes first through last inclusive as dirty in the buf
889 xfs_buf_log_item_t
*bip
,
896 struct xfs_buf
*bp
= bip
->bli_buf
;
899 * walk each buffer segment and mark them dirty appropriately.
902 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
905 end
= start
+ BBTOB(bp
->b_maps
[i
].bm_len
) - 1;
907 /* skip to the map that includes the first byte to log */
909 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
914 * Trim the range to this segment and mark it in the bitmap.
915 * Note that we must convert buffer offsets to segment relative
916 * offsets (e.g., the first byte of each segment is byte 0 of
923 xfs_buf_item_log_segment(first
- start
, end
- start
,
924 &bip
->bli_formats
[i
].blf_data_map
[0]);
926 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
932 * Return true if the buffer has any ranges logged/dirtied by a transaction,
936 xfs_buf_item_dirty_format(
937 struct xfs_buf_log_item
*bip
)
941 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
942 if (!xfs_bitmap_empty(bip
->bli_formats
[i
].blf_data_map
,
943 bip
->bli_formats
[i
].blf_map_size
))
952 xfs_buf_log_item_t
*bip
)
954 xfs_buf_item_free_format(bip
);
955 kmem_free(bip
->bli_item
.li_lv_shadow
);
956 kmem_zone_free(xfs_buf_item_zone
, bip
);
960 * This is called when the buf log item is no longer needed. It should
961 * free the buf log item associated with the given buffer and clear
962 * the buffer's pointer to the buf log item. If there are no more
963 * items in the list, clear the b_iodone field of the buffer (see
964 * xfs_buf_attach_iodone() below).
970 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
972 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
973 ASSERT(!(bip
->bli_item
.li_flags
& XFS_LI_IN_AIL
));
975 bp
->b_fspriv
= bip
->bli_item
.li_bio_list
;
976 if (bp
->b_fspriv
== NULL
)
980 xfs_buf_item_free(bip
);
985 * Add the given log item with its callback to the list of callbacks
986 * to be called when the buffer's I/O completes. If it is not set
987 * already, set the buffer's b_iodone() routine to be
988 * xfs_buf_iodone_callbacks() and link the log item into the list of
989 * items rooted at b_fsprivate. Items are always added as the second
990 * entry in the list if there is a first, because the buf item code
991 * assumes that the buf log item is first.
994 xfs_buf_attach_iodone(
996 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
999 xfs_log_item_t
*head_lip
;
1001 ASSERT(xfs_buf_islocked(bp
));
1004 head_lip
= bp
->b_fspriv
;
1006 lip
->li_bio_list
= head_lip
->li_bio_list
;
1007 head_lip
->li_bio_list
= lip
;
1012 ASSERT(bp
->b_iodone
== NULL
||
1013 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
1014 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
1018 * We can have many callbacks on a buffer. Running the callbacks individually
1019 * can cause a lot of contention on the AIL lock, so we allow for a single
1020 * callback to be able to scan the remaining lip->li_bio_list for other items
1021 * of the same type and callback to be processed in the first call.
1023 * As a result, the loop walking the callback list below will also modify the
1024 * list. it removes the first item from the list and then runs the callback.
1025 * The loop then restarts from the new head of the list. This allows the
1026 * callback to scan and modify the list attached to the buffer and we don't
1027 * have to care about maintaining a next item pointer.
1030 xfs_buf_do_callbacks(
1033 struct xfs_log_item
*lip
;
1035 while ((lip
= bp
->b_fspriv
) != NULL
) {
1036 bp
->b_fspriv
= lip
->li_bio_list
;
1037 ASSERT(lip
->li_cb
!= NULL
);
1039 * Clear the next pointer so we don't have any
1040 * confusion if the item is added to another buf.
1041 * Don't touch the log item after calling its
1042 * callback, because it could have freed itself.
1044 lip
->li_bio_list
= NULL
;
1045 lip
->li_cb(bp
, lip
);
1050 * Invoke the error state callback for each log item affected by the failed I/O.
1052 * If a metadata buffer write fails with a non-permanent error, the buffer is
1053 * eventually resubmitted and so the completion callbacks are not run. The error
1054 * state may need to be propagated to the log items attached to the buffer,
1055 * however, so the next AIL push of the item knows hot to handle it correctly.
1058 xfs_buf_do_callbacks_fail(
1061 struct xfs_log_item
*next
;
1062 struct xfs_log_item
*lip
= bp
->b_fspriv
;
1063 struct xfs_ail
*ailp
= lip
->li_ailp
;
1065 spin_lock(&ailp
->xa_lock
);
1066 for (; lip
; lip
= next
) {
1067 next
= lip
->li_bio_list
;
1068 if (lip
->li_ops
->iop_error
)
1069 lip
->li_ops
->iop_error(lip
, bp
);
1071 spin_unlock(&ailp
->xa_lock
);
1075 xfs_buf_iodone_callback_error(
1078 struct xfs_log_item
*lip
= bp
->b_fspriv
;
1079 struct xfs_mount
*mp
= lip
->li_mountp
;
1080 static ulong lasttime
;
1081 static xfs_buftarg_t
*lasttarg
;
1082 struct xfs_error_cfg
*cfg
;
1085 * If we've already decided to shutdown the filesystem because of
1086 * I/O errors, there's no point in giving this a retry.
1088 if (XFS_FORCED_SHUTDOWN(mp
))
1091 if (bp
->b_target
!= lasttarg
||
1092 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
1094 xfs_buf_ioerror_alert(bp
, __func__
);
1096 lasttarg
= bp
->b_target
;
1098 /* synchronous writes will have callers process the error */
1099 if (!(bp
->b_flags
& XBF_ASYNC
))
1102 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1103 ASSERT(bp
->b_iodone
!= NULL
);
1105 cfg
= xfs_error_get_cfg(mp
, XFS_ERR_METADATA
, bp
->b_error
);
1108 * If the write was asynchronous then no one will be looking for the
1109 * error. If this is the first failure of this type, clear the error
1110 * state and write the buffer out again. This means we always retry an
1111 * async write failure at least once, but we also need to set the buffer
1112 * up to behave correctly now for repeated failures.
1114 if (!(bp
->b_flags
& (XBF_STALE
| XBF_WRITE_FAIL
)) ||
1115 bp
->b_last_error
!= bp
->b_error
) {
1116 bp
->b_flags
|= (XBF_WRITE
| XBF_DONE
| XBF_WRITE_FAIL
);
1117 bp
->b_last_error
= bp
->b_error
;
1118 if (cfg
->retry_timeout
!= XFS_ERR_RETRY_FOREVER
&&
1119 !bp
->b_first_retry_time
)
1120 bp
->b_first_retry_time
= jiffies
;
1122 xfs_buf_ioerror(bp
, 0);
1128 * Repeated failure on an async write. Take action according to the
1129 * error configuration we have been set up to use.
1132 if (cfg
->max_retries
!= XFS_ERR_RETRY_FOREVER
&&
1133 ++bp
->b_retries
> cfg
->max_retries
)
1134 goto permanent_error
;
1135 if (cfg
->retry_timeout
!= XFS_ERR_RETRY_FOREVER
&&
1136 time_after(jiffies
, cfg
->retry_timeout
+ bp
->b_first_retry_time
))
1137 goto permanent_error
;
1139 /* At unmount we may treat errors differently */
1140 if ((mp
->m_flags
& XFS_MOUNT_UNMOUNTING
) && mp
->m_fail_unmount
)
1141 goto permanent_error
;
1144 * Still a transient error, run IO completion failure callbacks and let
1145 * the higher layers retry the buffer.
1147 xfs_buf_do_callbacks_fail(bp
);
1148 xfs_buf_ioerror(bp
, 0);
1153 * Permanent error - we need to trigger a shutdown if we haven't already
1154 * to indicate that inconsistency will result from this action.
1157 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1160 bp
->b_flags
|= XBF_DONE
;
1161 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1166 * This is the iodone() function for buffers which have had callbacks attached
1167 * to them by xfs_buf_attach_iodone(). We need to iterate the items on the
1168 * callback list, mark the buffer as having no more callbacks and then push the
1169 * buffer through IO completion processing.
1172 xfs_buf_iodone_callbacks(
1176 * If there is an error, process it. Some errors require us
1177 * to run callbacks after failure processing is done so we
1178 * detect that and take appropriate action.
1180 if (bp
->b_error
&& xfs_buf_iodone_callback_error(bp
))
1184 * Successful IO or permanent error. Either way, we can clear the
1185 * retry state here in preparation for the next error that may occur.
1187 bp
->b_last_error
= 0;
1189 bp
->b_first_retry_time
= 0;
1191 xfs_buf_do_callbacks(bp
);
1192 bp
->b_fspriv
= NULL
;
1193 bp
->b_iodone
= NULL
;
1198 * This is the iodone() function for buffers which have been
1199 * logged. It is called when they are eventually flushed out.
1200 * It should remove the buf item from the AIL, and free the buf item.
1201 * It is called by xfs_buf_iodone_callbacks() above which will take
1202 * care of cleaning up the buffer itself.
1207 struct xfs_log_item
*lip
)
1209 struct xfs_ail
*ailp
= lip
->li_ailp
;
1211 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1216 * If we are forcibly shutting down, this may well be
1217 * off the AIL already. That's because we simulate the
1218 * log-committed callbacks to unpin these buffers. Or we may never
1219 * have put this item on AIL because of the transaction was
1220 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1222 * Either way, AIL is useless if we're forcing a shutdown.
1224 spin_lock(&ailp
->xa_lock
);
1225 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1226 xfs_buf_item_free(BUF_ITEM(lip
));
1230 * Requeue a failed buffer for writeback
1232 * Return true if the buffer has been re-queued properly, false otherwise
1235 xfs_buf_resubmit_failed_buffers(
1237 struct xfs_log_item
*lip
,
1238 struct list_head
*buffer_list
)
1240 struct xfs_log_item
*next
;
1243 * Clear XFS_LI_FAILED flag from all items before resubmit
1245 * XFS_LI_FAILED set/clear is protected by xa_lock, caller this
1246 * function already have it acquired
1248 for (; lip
; lip
= next
) {
1249 next
= lip
->li_bio_list
;
1250 xfs_clear_li_failed(lip
);
1253 /* Add this buffer back to the delayed write list */
1254 return xfs_buf_delwri_queue(bp
, buffer_list
);