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"
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
);
186 xfs_buf_item_copy_iovec(
187 struct xfs_log_vec
*lv
,
188 struct xfs_log_iovec
**vecp
,
194 offset
+= first_bit
* XFS_BLF_CHUNK
;
195 xlog_copy_iovec(lv
, vecp
, XLOG_REG_TYPE_BCHUNK
,
196 xfs_buf_offset(bp
, offset
),
197 nbits
* XFS_BLF_CHUNK
);
201 xfs_buf_item_straddle(
207 return xfs_buf_offset(bp
, offset
+ (next_bit
<< XFS_BLF_SHIFT
)) !=
208 (xfs_buf_offset(bp
, offset
+ (last_bit
<< XFS_BLF_SHIFT
)) +
213 xfs_buf_item_format_segment(
214 struct xfs_buf_log_item
*bip
,
215 struct xfs_log_vec
*lv
,
216 struct xfs_log_iovec
**vecp
,
218 struct xfs_buf_log_format
*blfp
)
220 struct xfs_buf
*bp
= bip
->bli_buf
;
227 /* copy the flags across from the base format item */
228 blfp
->blf_flags
= bip
->__bli_format
.blf_flags
;
231 * Base size is the actual size of the ondisk structure - it reflects
232 * the actual size of the dirty bitmap rather than the size of the in
235 base_size
= xfs_buf_log_format_size(blfp
);
237 first_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
238 if (!(bip
->bli_flags
& XFS_BLI_STALE
) && first_bit
== -1) {
240 * If the map is not be dirty in the transaction, mark
241 * the size as zero and do not advance the vector pointer.
246 blfp
= xlog_copy_iovec(lv
, vecp
, XLOG_REG_TYPE_BFORMAT
, blfp
, base_size
);
249 if (bip
->bli_flags
& XFS_BLI_STALE
) {
251 * The buffer is stale, so all we need to log
252 * is the buf log format structure with the
255 trace_xfs_buf_item_format_stale(bip
);
256 ASSERT(blfp
->blf_flags
& XFS_BLF_CANCEL
);
262 * Fill in an iovec for each set of contiguous chunks.
264 last_bit
= first_bit
;
268 * This takes the bit number to start looking from and
269 * returns the next set bit from there. It returns -1
270 * if there are no more bits set or the start bit is
271 * beyond the end of the bitmap.
273 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
276 * If we run out of bits fill in the last iovec and get out of
277 * the loop. Else if we start a new set of bits then fill in
278 * the iovec for the series we were looking at and start
279 * counting the bits in the new one. Else we're still in the
280 * same set of bits so just keep counting and scanning.
282 if (next_bit
== -1) {
283 xfs_buf_item_copy_iovec(lv
, vecp
, bp
, offset
,
287 } else if (next_bit
!= last_bit
+ 1 ||
288 xfs_buf_item_straddle(bp
, offset
, next_bit
, last_bit
)) {
289 xfs_buf_item_copy_iovec(lv
, vecp
, bp
, offset
,
292 first_bit
= next_bit
;
303 * This is called to fill in the vector of log iovecs for the
304 * given log buf item. It fills the first entry with a buf log
305 * format structure, and the rest point to contiguous chunks
310 struct xfs_log_item
*lip
,
311 struct xfs_log_vec
*lv
)
313 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
314 struct xfs_buf
*bp
= bip
->bli_buf
;
315 struct xfs_log_iovec
*vecp
= NULL
;
319 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
320 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
321 (bip
->bli_flags
& XFS_BLI_STALE
));
322 ASSERT((bip
->bli_flags
& XFS_BLI_STALE
) ||
323 (xfs_blft_from_flags(&bip
->__bli_format
) > XFS_BLFT_UNKNOWN_BUF
324 && xfs_blft_from_flags(&bip
->__bli_format
) < XFS_BLFT_MAX_BUF
));
328 * If it is an inode buffer, transfer the in-memory state to the
329 * format flags and clear the in-memory state.
331 * For buffer based inode allocation, we do not transfer
332 * this state if the inode buffer allocation has not yet been committed
333 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
334 * correct replay of the inode allocation.
336 * For icreate item based inode allocation, the buffers aren't written
337 * to the journal during allocation, and hence we should always tag the
338 * buffer as an inode buffer so that the correct unlinked list replay
339 * occurs during recovery.
341 if (bip
->bli_flags
& XFS_BLI_INODE_BUF
) {
342 if (xfs_sb_version_hascrc(&lip
->li_mountp
->m_sb
) ||
343 !((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
344 xfs_log_item_in_current_chkpt(lip
)))
345 bip
->__bli_format
.blf_flags
|= XFS_BLF_INODE_BUF
;
346 bip
->bli_flags
&= ~XFS_BLI_INODE_BUF
;
349 if ((bip
->bli_flags
& (XFS_BLI_ORDERED
|XFS_BLI_STALE
)) ==
352 * The buffer has been logged just to order it. It is not being
353 * included in the transaction commit, so don't format it.
355 trace_xfs_buf_item_format_ordered(bip
);
359 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
360 xfs_buf_item_format_segment(bip
, lv
, &vecp
, offset
,
361 &bip
->bli_formats
[i
]);
362 offset
+= BBTOB(bp
->b_maps
[i
].bm_len
);
366 * Check to make sure everything is consistent.
368 trace_xfs_buf_item_format(bip
);
372 * This is called to pin the buffer associated with the buf log item in memory
373 * so it cannot be written out.
375 * We also always take a reference to the buffer log item here so that the bli
376 * is held while the item is pinned in memory. This means that we can
377 * unconditionally drop the reference count a transaction holds when the
378 * transaction is completed.
382 struct xfs_log_item
*lip
)
384 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
386 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
387 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
388 (bip
->bli_flags
& XFS_BLI_ORDERED
) ||
389 (bip
->bli_flags
& XFS_BLI_STALE
));
391 trace_xfs_buf_item_pin(bip
);
393 atomic_inc(&bip
->bli_refcount
);
394 atomic_inc(&bip
->bli_buf
->b_pin_count
);
398 * This is called to unpin the buffer associated with the buf log
399 * item which was previously pinned with a call to xfs_buf_item_pin().
401 * Also drop the reference to the buf item for the current transaction.
402 * If the XFS_BLI_STALE flag is set and we are the last reference,
403 * then free up the buf log item and unlock the buffer.
405 * If the remove flag is set we are called from uncommit in the
406 * forced-shutdown path. If that is true and the reference count on
407 * the log item is going to drop to zero we need to free the item's
408 * descriptor in the transaction.
412 struct xfs_log_item
*lip
,
415 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
416 xfs_buf_t
*bp
= bip
->bli_buf
;
417 struct xfs_ail
*ailp
= lip
->li_ailp
;
418 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
421 ASSERT(bp
->b_fspriv
== bip
);
422 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
424 trace_xfs_buf_item_unpin(bip
);
426 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
428 if (atomic_dec_and_test(&bp
->b_pin_count
))
429 wake_up_all(&bp
->b_waiters
);
431 if (freed
&& stale
) {
432 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
433 ASSERT(xfs_buf_islocked(bp
));
434 ASSERT(bp
->b_flags
& XBF_STALE
);
435 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
437 trace_xfs_buf_item_unpin_stale(bip
);
441 * If we are in a transaction context, we have to
442 * remove the log item from the transaction as we are
443 * about to release our reference to the buffer. If we
444 * don't, the unlock that occurs later in
445 * xfs_trans_uncommit() will try to reference the
446 * buffer which we no longer have a hold on.
449 xfs_trans_del_item(lip
);
452 * Since the transaction no longer refers to the buffer,
453 * the buffer should no longer refer to the transaction.
459 * If we get called here because of an IO error, we may
460 * or may not have the item on the AIL. xfs_trans_ail_delete()
461 * will take care of that situation.
462 * xfs_trans_ail_delete() drops the AIL lock.
464 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
465 xfs_buf_do_callbacks(bp
);
469 spin_lock(&ailp
->xa_lock
);
470 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_LOG_IO_ERROR
);
471 xfs_buf_item_relse(bp
);
472 ASSERT(bp
->b_fspriv
== NULL
);
475 } else if (freed
&& remove
) {
477 * There are currently two references to the buffer - the active
478 * LRU reference and the buf log item. What we are about to do
479 * here - simulate a failed IO completion - requires 3
482 * The LRU reference is removed by the xfs_buf_stale() call. The
483 * buf item reference is removed by the xfs_buf_iodone()
484 * callback that is run by xfs_buf_do_callbacks() during ioend
485 * processing (via the bp->b_iodone callback), and then finally
486 * the ioend processing will drop the IO reference if the buffer
487 * is marked XBF_ASYNC.
489 * Hence we need to take an additional reference here so that IO
490 * completion processing doesn't free the buffer prematurely.
494 bp
->b_flags
|= XBF_ASYNC
;
495 xfs_buf_ioerror(bp
, -EIO
);
496 bp
->b_flags
&= ~XBF_DONE
;
503 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
504 * seconds so as to not spam logs too much on repeated detection of the same
508 static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state
, 30 * HZ
, 10);
512 struct xfs_log_item
*lip
,
513 struct list_head
*buffer_list
)
515 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
516 struct xfs_buf
*bp
= bip
->bli_buf
;
517 uint rval
= XFS_ITEM_SUCCESS
;
519 if (xfs_buf_ispinned(bp
))
520 return XFS_ITEM_PINNED
;
521 if (!xfs_buf_trylock(bp
)) {
523 * If we have just raced with a buffer being pinned and it has
524 * been marked stale, we could end up stalling until someone else
525 * issues a log force to unpin the stale buffer. Check for the
526 * race condition here so xfsaild recognizes the buffer is pinned
527 * and queues a log force to move it along.
529 if (xfs_buf_ispinned(bp
))
530 return XFS_ITEM_PINNED
;
531 return XFS_ITEM_LOCKED
;
534 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
536 trace_xfs_buf_item_push(bip
);
538 /* has a previous flush failed due to IO errors? */
539 if ((bp
->b_flags
& XBF_WRITE_FAIL
) &&
540 ___ratelimit(&xfs_buf_write_fail_rl_state
, "XFS: Failing async write")) {
541 xfs_warn(bp
->b_target
->bt_mount
,
542 "Failing async write on buffer block 0x%llx. Retrying async write.",
543 (long long)bp
->b_bn
);
546 if (!xfs_buf_delwri_queue(bp
, buffer_list
))
547 rval
= XFS_ITEM_FLUSHING
;
553 * Release the buffer associated with the buf log item. If there is no dirty
554 * logged data associated with the buffer recorded in the buf log item, then
555 * free the buf log item and remove the reference to it in the buffer.
557 * This call ignores the recursion count. It is only called when the buffer
558 * should REALLY be unlocked, regardless of the recursion count.
560 * We unconditionally drop the transaction's reference to the log item. If the
561 * item was logged, then another reference was taken when it was pinned, so we
562 * can safely drop the transaction reference now. This also allows us to avoid
563 * potential races with the unpin code freeing the bli by not referencing the
564 * bli after we've dropped the reference count.
566 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
567 * if necessary but do not unlock the buffer. This is for support of
568 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
573 struct xfs_log_item
*lip
)
575 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
576 struct xfs_buf
*bp
= bip
->bli_buf
;
581 /* Clear the buffer's association with this transaction. */
585 * If this is a transaction abort, don't return early. Instead, allow
586 * the brelse to happen. Normally it would be done for stale
587 * (cancelled) buffers at unpin time, but we'll never go through the
588 * pin/unpin cycle if we abort inside commit.
590 aborted
= (lip
->li_flags
& XFS_LI_ABORTED
) ? true : false;
592 * Before possibly freeing the buf item, copy the per-transaction state
593 * so we can reference it safely later after clearing it from the
596 flags
= bip
->bli_flags
;
597 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
| XFS_BLI_ORDERED
);
600 * If the buf item is marked stale, then don't do anything. We'll
601 * unlock the buffer and free the buf item when the buffer is unpinned
604 if (flags
& XFS_BLI_STALE
) {
605 trace_xfs_buf_item_unlock_stale(bip
);
606 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
608 atomic_dec(&bip
->bli_refcount
);
613 trace_xfs_buf_item_unlock(bip
);
616 * If the buf item isn't tracking any data, free it, otherwise drop the
617 * reference we hold to it. If we are aborting the transaction, this may
618 * be the only reference to the buf item, so we free it anyway
619 * regardless of whether it is dirty or not. A dirty abort implies a
622 * Ordered buffers are dirty but may have no recorded changes, so ensure
623 * we only release clean items here.
625 clean
= (flags
& XFS_BLI_DIRTY
) ? false : true;
628 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
629 if (!xfs_bitmap_empty(bip
->bli_formats
[i
].blf_data_map
,
630 bip
->bli_formats
[i
].blf_map_size
)) {
638 * Clean buffers, by definition, cannot be in the AIL. However, aborted
639 * buffers may be in the AIL regardless of dirty state. An aborted
640 * transaction that invalidates a buffer already in the AIL may have
641 * marked it stale and cleared the dirty state, for example.
643 * Therefore if we are aborting a buffer and we've just taken the last
644 * reference away, we have to check if it is in the AIL before freeing
645 * it. We need to free it in this case, because an aborted transaction
646 * has already shut the filesystem down and this is the last chance we
647 * will have to do so.
649 if (atomic_dec_and_test(&bip
->bli_refcount
)) {
651 ASSERT(XFS_FORCED_SHUTDOWN(lip
->li_mountp
));
652 xfs_trans_ail_remove(lip
, SHUTDOWN_LOG_IO_ERROR
);
653 xfs_buf_item_relse(bp
);
655 xfs_buf_item_relse(bp
);
658 if (!(flags
& XFS_BLI_HOLD
))
663 * This is called to find out where the oldest active copy of the
664 * buf log item in the on disk log resides now that the last log
665 * write of it completed at the given lsn.
666 * We always re-log all the dirty data in a buffer, so usually the
667 * latest copy in the on disk log is the only one that matters. For
668 * those cases we simply return the given lsn.
670 * The one exception to this is for buffers full of newly allocated
671 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
672 * flag set, indicating that only the di_next_unlinked fields from the
673 * inodes in the buffers will be replayed during recovery. If the
674 * original newly allocated inode images have not yet been flushed
675 * when the buffer is so relogged, then we need to make sure that we
676 * keep the old images in the 'active' portion of the log. We do this
677 * by returning the original lsn of that transaction here rather than
681 xfs_buf_item_committed(
682 struct xfs_log_item
*lip
,
685 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
687 trace_xfs_buf_item_committed(bip
);
689 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
695 xfs_buf_item_committing(
696 struct xfs_log_item
*lip
,
697 xfs_lsn_t commit_lsn
)
702 * This is the ops vector shared by all buf log items.
704 static const struct xfs_item_ops xfs_buf_item_ops
= {
705 .iop_size
= xfs_buf_item_size
,
706 .iop_format
= xfs_buf_item_format
,
707 .iop_pin
= xfs_buf_item_pin
,
708 .iop_unpin
= xfs_buf_item_unpin
,
709 .iop_unlock
= xfs_buf_item_unlock
,
710 .iop_committed
= xfs_buf_item_committed
,
711 .iop_push
= xfs_buf_item_push
,
712 .iop_committing
= xfs_buf_item_committing
716 xfs_buf_item_get_format(
717 struct xfs_buf_log_item
*bip
,
720 ASSERT(bip
->bli_formats
== NULL
);
721 bip
->bli_format_count
= count
;
724 bip
->bli_formats
= &bip
->__bli_format
;
728 bip
->bli_formats
= kmem_zalloc(count
* sizeof(struct xfs_buf_log_format
),
730 if (!bip
->bli_formats
)
736 xfs_buf_item_free_format(
737 struct xfs_buf_log_item
*bip
)
739 if (bip
->bli_formats
!= &bip
->__bli_format
) {
740 kmem_free(bip
->bli_formats
);
741 bip
->bli_formats
= NULL
;
746 * Allocate a new buf log item to go with the given buffer.
747 * Set the buffer's b_fsprivate field to point to the new
748 * buf log item. If there are other item's attached to the
749 * buffer (see xfs_buf_attach_iodone() below), then put the
750 * buf log item at the front.
755 struct xfs_mount
*mp
)
757 struct xfs_log_item
*lip
= bp
->b_fspriv
;
758 struct xfs_buf_log_item
*bip
;
765 * Check to see if there is already a buf log item for
766 * this buffer. If there is, it is guaranteed to be
767 * the first. If we do already have one, there is
768 * nothing to do here so return.
770 ASSERT(bp
->b_target
->bt_mount
== mp
);
771 if (lip
!= NULL
&& lip
->li_type
== XFS_LI_BUF
)
774 bip
= kmem_zone_zalloc(xfs_buf_item_zone
, KM_SLEEP
);
775 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
779 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
780 * can be divided into. Make sure not to truncate any pieces.
781 * map_size is the size of the bitmap needed to describe the
782 * chunks of the buffer.
784 * Discontiguous buffer support follows the layout of the underlying
785 * buffer. This makes the implementation as simple as possible.
787 error
= xfs_buf_item_get_format(bip
, bp
->b_map_count
);
789 if (error
) { /* to stop gcc throwing set-but-unused warnings */
790 kmem_zone_free(xfs_buf_item_zone
, bip
);
795 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
796 chunks
= DIV_ROUND_UP(BBTOB(bp
->b_maps
[i
].bm_len
),
798 map_size
= DIV_ROUND_UP(chunks
, NBWORD
);
800 bip
->bli_formats
[i
].blf_type
= XFS_LI_BUF
;
801 bip
->bli_formats
[i
].blf_blkno
= bp
->b_maps
[i
].bm_bn
;
802 bip
->bli_formats
[i
].blf_len
= bp
->b_maps
[i
].bm_len
;
803 bip
->bli_formats
[i
].blf_map_size
= map_size
;
807 * Put the buf item into the list of items attached to the
808 * buffer at the front.
811 bip
->bli_item
.li_bio_list
= bp
->b_fspriv
;
819 * Mark bytes first through last inclusive as dirty in the buf
823 xfs_buf_item_log_segment(
839 * Convert byte offsets to bit numbers.
841 first_bit
= first
>> XFS_BLF_SHIFT
;
842 last_bit
= last
>> XFS_BLF_SHIFT
;
845 * Calculate the total number of bits to be set.
847 bits_to_set
= last_bit
- first_bit
+ 1;
850 * Get a pointer to the first word in the bitmap
853 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
854 wordp
= &map
[word_num
];
857 * Calculate the starting bit in the first word.
859 bit
= first_bit
& (uint
)(NBWORD
- 1);
862 * First set any bits in the first word of our range.
863 * If it starts at bit 0 of the word, it will be
864 * set below rather than here. That is what the variable
865 * bit tells us. The variable bits_set tracks the number
866 * of bits that have been set so far. End_bit is the number
867 * of the last bit to be set in this word plus one.
870 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
871 mask
= ((1U << (end_bit
- bit
)) - 1) << bit
;
874 bits_set
= end_bit
- bit
;
880 * Now set bits a whole word at a time that are between
881 * first_bit and last_bit.
883 while ((bits_to_set
- bits_set
) >= NBWORD
) {
884 *wordp
|= 0xffffffff;
890 * Finally, set any bits left to be set in one last partial word.
892 end_bit
= bits_to_set
- bits_set
;
894 mask
= (1U << end_bit
) - 1;
900 * Mark bytes first through last inclusive as dirty in the buf
905 xfs_buf_log_item_t
*bip
,
912 struct xfs_buf
*bp
= bip
->bli_buf
;
915 * walk each buffer segment and mark them dirty appropriately.
918 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
921 end
= start
+ BBTOB(bp
->b_maps
[i
].bm_len
) - 1;
923 /* skip to the map that includes the first byte to log */
925 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
930 * Trim the range to this segment and mark it in the bitmap.
931 * Note that we must convert buffer offsets to segment relative
932 * offsets (e.g., the first byte of each segment is byte 0 of
939 xfs_buf_item_log_segment(first
- start
, end
- start
,
940 &bip
->bli_formats
[i
].blf_data_map
[0]);
942 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
948 * Return 1 if the buffer has been logged or ordered in a transaction (at any
949 * point, not just the current transaction) and 0 if not.
953 xfs_buf_log_item_t
*bip
)
955 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
960 xfs_buf_log_item_t
*bip
)
962 xfs_buf_item_free_format(bip
);
963 kmem_free(bip
->bli_item
.li_lv_shadow
);
964 kmem_zone_free(xfs_buf_item_zone
, bip
);
968 * This is called when the buf log item is no longer needed. It should
969 * free the buf log item associated with the given buffer and clear
970 * the buffer's pointer to the buf log item. If there are no more
971 * items in the list, clear the b_iodone field of the buffer (see
972 * xfs_buf_attach_iodone() below).
978 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
980 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
981 ASSERT(!(bip
->bli_item
.li_flags
& XFS_LI_IN_AIL
));
983 bp
->b_fspriv
= bip
->bli_item
.li_bio_list
;
984 if (bp
->b_fspriv
== NULL
)
988 xfs_buf_item_free(bip
);
993 * Add the given log item with its callback to the list of callbacks
994 * to be called when the buffer's I/O completes. If it is not set
995 * already, set the buffer's b_iodone() routine to be
996 * xfs_buf_iodone_callbacks() and link the log item into the list of
997 * items rooted at b_fsprivate. Items are always added as the second
998 * entry in the list if there is a first, because the buf item code
999 * assumes that the buf log item is first.
1002 xfs_buf_attach_iodone(
1004 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
1005 xfs_log_item_t
*lip
)
1007 xfs_log_item_t
*head_lip
;
1009 ASSERT(xfs_buf_islocked(bp
));
1012 head_lip
= bp
->b_fspriv
;
1014 lip
->li_bio_list
= head_lip
->li_bio_list
;
1015 head_lip
->li_bio_list
= lip
;
1020 ASSERT(bp
->b_iodone
== NULL
||
1021 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
1022 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
1026 * We can have many callbacks on a buffer. Running the callbacks individually
1027 * can cause a lot of contention on the AIL lock, so we allow for a single
1028 * callback to be able to scan the remaining lip->li_bio_list for other items
1029 * of the same type and callback to be processed in the first call.
1031 * As a result, the loop walking the callback list below will also modify the
1032 * list. it removes the first item from the list and then runs the callback.
1033 * The loop then restarts from the new head of the list. This allows the
1034 * callback to scan and modify the list attached to the buffer and we don't
1035 * have to care about maintaining a next item pointer.
1038 xfs_buf_do_callbacks(
1041 struct xfs_log_item
*lip
;
1043 while ((lip
= bp
->b_fspriv
) != NULL
) {
1044 bp
->b_fspriv
= lip
->li_bio_list
;
1045 ASSERT(lip
->li_cb
!= NULL
);
1047 * Clear the next pointer so we don't have any
1048 * confusion if the item is added to another buf.
1049 * Don't touch the log item after calling its
1050 * callback, because it could have freed itself.
1052 lip
->li_bio_list
= NULL
;
1053 lip
->li_cb(bp
, lip
);
1058 xfs_buf_iodone_callback_error(
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
;
1065 struct xfs_error_cfg
*cfg
;
1068 * If we've already decided to shutdown the filesystem because of
1069 * I/O errors, there's no point in giving this a retry.
1071 if (XFS_FORCED_SHUTDOWN(mp
))
1074 if (bp
->b_target
!= lasttarg
||
1075 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
1077 xfs_buf_ioerror_alert(bp
, __func__
);
1079 lasttarg
= bp
->b_target
;
1081 /* synchronous writes will have callers process the error */
1082 if (!(bp
->b_flags
& XBF_ASYNC
))
1085 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1086 ASSERT(bp
->b_iodone
!= NULL
);
1088 cfg
= xfs_error_get_cfg(mp
, XFS_ERR_METADATA
, bp
->b_error
);
1091 * If the write was asynchronous then no one will be looking for the
1092 * error. If this is the first failure of this type, clear the error
1093 * state and write the buffer out again. This means we always retry an
1094 * async write failure at least once, but we also need to set the buffer
1095 * up to behave correctly now for repeated failures.
1097 if (!(bp
->b_flags
& (XBF_STALE
| XBF_WRITE_FAIL
)) ||
1098 bp
->b_last_error
!= bp
->b_error
) {
1099 bp
->b_flags
|= (XBF_WRITE
| XBF_DONE
| XBF_WRITE_FAIL
);
1100 bp
->b_last_error
= bp
->b_error
;
1101 if (cfg
->retry_timeout
!= XFS_ERR_RETRY_FOREVER
&&
1102 !bp
->b_first_retry_time
)
1103 bp
->b_first_retry_time
= jiffies
;
1105 xfs_buf_ioerror(bp
, 0);
1111 * Repeated failure on an async write. Take action according to the
1112 * error configuration we have been set up to use.
1115 if (cfg
->max_retries
!= XFS_ERR_RETRY_FOREVER
&&
1116 ++bp
->b_retries
> cfg
->max_retries
)
1117 goto permanent_error
;
1118 if (cfg
->retry_timeout
!= XFS_ERR_RETRY_FOREVER
&&
1119 time_after(jiffies
, cfg
->retry_timeout
+ bp
->b_first_retry_time
))
1120 goto permanent_error
;
1122 /* At unmount we may treat errors differently */
1123 if ((mp
->m_flags
& XFS_MOUNT_UNMOUNTING
) && mp
->m_fail_unmount
)
1124 goto permanent_error
;
1126 /* still a transient error, higher layers will retry */
1127 xfs_buf_ioerror(bp
, 0);
1132 * Permanent error - we need to trigger a shutdown if we haven't already
1133 * to indicate that inconsistency will result from this action.
1136 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1139 bp
->b_flags
|= XBF_DONE
;
1140 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1145 * This is the iodone() function for buffers which have had callbacks attached
1146 * to them by xfs_buf_attach_iodone(). We need to iterate the items on the
1147 * callback list, mark the buffer as having no more callbacks and then push the
1148 * buffer through IO completion processing.
1151 xfs_buf_iodone_callbacks(
1155 * If there is an error, process it. Some errors require us
1156 * to run callbacks after failure processing is done so we
1157 * detect that and take appropriate action.
1159 if (bp
->b_error
&& xfs_buf_iodone_callback_error(bp
))
1163 * Successful IO or permanent error. Either way, we can clear the
1164 * retry state here in preparation for the next error that may occur.
1166 bp
->b_last_error
= 0;
1168 bp
->b_first_retry_time
= 0;
1170 xfs_buf_do_callbacks(bp
);
1171 bp
->b_fspriv
= NULL
;
1172 bp
->b_iodone
= NULL
;
1177 * This is the iodone() function for buffers which have been
1178 * logged. It is called when they are eventually flushed out.
1179 * It should remove the buf item from the AIL, and free the buf item.
1180 * It is called by xfs_buf_iodone_callbacks() above which will take
1181 * care of cleaning up the buffer itself.
1186 struct xfs_log_item
*lip
)
1188 struct xfs_ail
*ailp
= lip
->li_ailp
;
1190 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1195 * If we are forcibly shutting down, this may well be
1196 * off the AIL already. That's because we simulate the
1197 * log-committed callbacks to unpin these buffers. Or we may never
1198 * have put this item on AIL because of the transaction was
1199 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1201 * Either way, AIL is useless if we're forcing a shutdown.
1203 spin_lock(&ailp
->xa_lock
);
1204 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1205 xfs_buf_item_free(BUF_ITEM(lip
));