2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_buf_item.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
34 kmem_zone_t
*xfs_buf_item_zone
;
36 static inline struct xfs_buf_log_item
*BUF_ITEM(struct xfs_log_item
*lip
)
38 return container_of(lip
, struct xfs_buf_log_item
, bli_item
);
42 #ifdef XFS_TRANS_DEBUG
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
53 xfs_buf_item_log_debug(
54 xfs_buf_log_item_t
*bip
,
67 ASSERT(bip
->bli_logged
!= NULL
);
69 nbytes
= last
- first
+ 1;
70 bfset(bip
->bli_logged
, first
, nbytes
);
71 for (x
= 0; x
< nbytes
; x
++) {
72 chunk_num
= byte
>> XFS_BLF_SHIFT
;
73 word_num
= chunk_num
>> BIT_TO_WORD_SHIFT
;
74 bit_num
= chunk_num
& (NBWORD
- 1);
75 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
76 bit_set
= *wordp
& (1 << bit_num
);
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
88 xfs_buf_item_flush_log_debug(
93 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
96 if (bip
== NULL
|| (bip
->bli_item
.li_type
!= XFS_LI_BUF
))
99 ASSERT(bip
->bli_logged
!= NULL
);
100 nbytes
= last
- first
+ 1;
101 bfset(bip
->bli_logged
, first
, nbytes
);
105 * This function is called to verify that our callers have logged
106 * all the bytes that they changed.
108 * It does this by comparing the original copy of the buffer stored in
109 * the buf log item's bli_orig array to the current copy of the buffer
110 * and ensuring that all bytes which mismatch are set in the bli_logged
111 * array of the buf log item.
114 xfs_buf_item_log_check(
115 xfs_buf_log_item_t
*bip
)
122 ASSERT(bip
->bli_orig
!= NULL
);
123 ASSERT(bip
->bli_logged
!= NULL
);
126 ASSERT(bp
->b_length
> 0);
127 ASSERT(bp
->b_addr
!= NULL
);
128 orig
= bip
->bli_orig
;
130 for (x
= 0; x
< BBTOB(bp
->b_length
); x
++) {
131 if (orig
[x
] != buffer
[x
] && !btst(bip
->bli_logged
, x
)) {
132 xfs_emerg(bp
->b_mount
,
133 "%s: bip %x buffer %x orig %x index %d",
134 __func__
, bip
, bp
, orig
, x
);
140 #define xfs_buf_item_log_debug(x,y,z)
141 #define xfs_buf_item_log_check(x)
144 STATIC
void xfs_buf_do_callbacks(struct xfs_buf
*bp
);
147 * This returns the number of log iovecs needed to log the
148 * given buf log item.
150 * It calculates this as 1 iovec for the buf log format structure
151 * and 1 for each stretch of non-contiguous chunks to be logged.
152 * Contiguous chunks are logged in a single iovec.
154 * If the XFS_BLI_STALE flag has been set, then log nothing.
158 struct xfs_log_item
*lip
)
160 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
161 struct xfs_buf
*bp
= bip
->bli_buf
;
166 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
167 if (bip
->bli_flags
& XFS_BLI_STALE
) {
169 * The buffer is stale, so all we need to log
170 * is the buf log format structure with the
173 trace_xfs_buf_item_size_stale(bip
);
174 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
178 ASSERT(bip
->bli_flags
& XFS_BLI_LOGGED
);
180 last_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
181 bip
->bli_format
.blf_map_size
, 0);
182 ASSERT(last_bit
!= -1);
184 while (last_bit
!= -1) {
186 * This takes the bit number to start looking from and
187 * returns the next set bit from there. It returns -1
188 * if there are no more bits set or the start bit is
189 * beyond the end of the bitmap.
191 next_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
192 bip
->bli_format
.blf_map_size
,
195 * If we run out of bits, leave the loop,
196 * else if we find a new set of bits bump the number of vecs,
197 * else keep scanning the current set of bits.
199 if (next_bit
== -1) {
201 } else if (next_bit
!= last_bit
+ 1) {
204 } else if (xfs_buf_offset(bp
, next_bit
* XFS_BLF_CHUNK
) !=
205 (xfs_buf_offset(bp
, last_bit
* XFS_BLF_CHUNK
) +
214 trace_xfs_buf_item_size(bip
);
219 * This is called to fill in the vector of log iovecs for the
220 * given log buf item. It fills the first entry with a buf log
221 * format structure, and the rest point to contiguous chunks
226 struct xfs_log_item
*lip
,
227 struct xfs_log_iovec
*vecp
)
229 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
230 struct xfs_buf
*bp
= bip
->bli_buf
;
239 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
240 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
241 (bip
->bli_flags
& XFS_BLI_STALE
));
244 * The size of the base structure is the size of the
245 * declared structure plus the space for the extra words
246 * of the bitmap. We subtract one from the map size, because
247 * the first element of the bitmap is accounted for in the
248 * size of the base structure.
251 (uint
)(sizeof(xfs_buf_log_format_t
) +
252 ((bip
->bli_format
.blf_map_size
- 1) * sizeof(uint
)));
253 vecp
->i_addr
= &bip
->bli_format
;
254 vecp
->i_len
= base_size
;
255 vecp
->i_type
= XLOG_REG_TYPE_BFORMAT
;
260 * If it is an inode buffer, transfer the in-memory state to the
261 * format flags and clear the in-memory state. We do not transfer
262 * this state if the inode buffer allocation has not yet been committed
263 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
264 * correct replay of the inode allocation.
266 if (bip
->bli_flags
& XFS_BLI_INODE_BUF
) {
267 if (!((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
268 xfs_log_item_in_current_chkpt(lip
)))
269 bip
->bli_format
.blf_flags
|= XFS_BLF_INODE_BUF
;
270 bip
->bli_flags
&= ~XFS_BLI_INODE_BUF
;
273 if (bip
->bli_flags
& XFS_BLI_STALE
) {
275 * The buffer is stale, so all we need to log
276 * is the buf log format structure with the
279 trace_xfs_buf_item_format_stale(bip
);
280 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
281 bip
->bli_format
.blf_size
= nvecs
;
286 * Fill in an iovec for each set of contiguous chunks.
288 first_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
289 bip
->bli_format
.blf_map_size
, 0);
290 ASSERT(first_bit
!= -1);
291 last_bit
= first_bit
;
295 * This takes the bit number to start looking from and
296 * returns the next set bit from there. It returns -1
297 * if there are no more bits set or the start bit is
298 * beyond the end of the bitmap.
300 next_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
301 bip
->bli_format
.blf_map_size
,
304 * If we run out of bits fill in the last iovec and get
306 * Else if we start a new set of bits then fill in the
307 * iovec for the series we were looking at and start
308 * counting the bits in the new one.
309 * Else we're still in the same set of bits so just
310 * keep counting and scanning.
312 if (next_bit
== -1) {
313 buffer_offset
= first_bit
* XFS_BLF_CHUNK
;
314 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
315 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
316 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
319 } else if (next_bit
!= last_bit
+ 1) {
320 buffer_offset
= first_bit
* XFS_BLF_CHUNK
;
321 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
322 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
323 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
326 first_bit
= next_bit
;
329 } else if (xfs_buf_offset(bp
, next_bit
<< XFS_BLF_SHIFT
) !=
330 (xfs_buf_offset(bp
, last_bit
<< XFS_BLF_SHIFT
) +
332 buffer_offset
= first_bit
* XFS_BLF_CHUNK
;
333 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
334 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
335 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
336 /* You would think we need to bump the nvecs here too, but we do not
337 * this number is used by recovery, and it gets confused by the boundary
342 first_bit
= next_bit
;
350 bip
->bli_format
.blf_size
= nvecs
;
353 * Check to make sure everything is consistent.
355 trace_xfs_buf_item_format(bip
);
356 xfs_buf_item_log_check(bip
);
360 * This is called to pin the buffer associated with the buf log item in memory
361 * so it cannot be written out.
363 * We also always take a reference to the buffer log item here so that the bli
364 * is held while the item is pinned in memory. This means that we can
365 * unconditionally drop the reference count a transaction holds when the
366 * transaction is completed.
370 struct xfs_log_item
*lip
)
372 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
374 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
375 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
376 (bip
->bli_flags
& XFS_BLI_STALE
));
378 trace_xfs_buf_item_pin(bip
);
380 atomic_inc(&bip
->bli_refcount
);
381 atomic_inc(&bip
->bli_buf
->b_pin_count
);
385 * This is called to unpin the buffer associated with the buf log
386 * item which was previously pinned with a call to xfs_buf_item_pin().
388 * Also drop the reference to the buf item for the current transaction.
389 * If the XFS_BLI_STALE flag is set and we are the last reference,
390 * then free up the buf log item and unlock the buffer.
392 * If the remove flag is set we are called from uncommit in the
393 * forced-shutdown path. If that is true and the reference count on
394 * the log item is going to drop to zero we need to free the item's
395 * descriptor in the transaction.
399 struct xfs_log_item
*lip
,
402 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
403 xfs_buf_t
*bp
= bip
->bli_buf
;
404 struct xfs_ail
*ailp
= lip
->li_ailp
;
405 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
408 ASSERT(bp
->b_fspriv
== bip
);
409 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
411 trace_xfs_buf_item_unpin(bip
);
413 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
415 if (atomic_dec_and_test(&bp
->b_pin_count
))
416 wake_up_all(&bp
->b_waiters
);
418 if (freed
&& stale
) {
419 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
420 ASSERT(xfs_buf_islocked(bp
));
421 ASSERT(XFS_BUF_ISSTALE(bp
));
422 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
424 trace_xfs_buf_item_unpin_stale(bip
);
428 * If we are in a transaction context, we have to
429 * remove the log item from the transaction as we are
430 * about to release our reference to the buffer. If we
431 * don't, the unlock that occurs later in
432 * xfs_trans_uncommit() will try to reference the
433 * buffer which we no longer have a hold on.
436 xfs_trans_del_item(lip
);
439 * Since the transaction no longer refers to the buffer,
440 * the buffer should no longer refer to the transaction.
446 * If we get called here because of an IO error, we may
447 * or may not have the item on the AIL. xfs_trans_ail_delete()
448 * will take care of that situation.
449 * xfs_trans_ail_delete() drops the AIL lock.
451 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
452 xfs_buf_do_callbacks(bp
);
456 spin_lock(&ailp
->xa_lock
);
457 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_LOG_IO_ERROR
);
458 xfs_buf_item_relse(bp
);
459 ASSERT(bp
->b_fspriv
== NULL
);
462 } else if (freed
&& remove
) {
464 xfs_buf_ioerror(bp
, EIO
);
467 xfs_buf_ioend(bp
, 0);
473 struct xfs_log_item
*lip
,
474 struct list_head
*buffer_list
)
476 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
477 struct xfs_buf
*bp
= bip
->bli_buf
;
478 uint rval
= XFS_ITEM_SUCCESS
;
480 if (xfs_buf_ispinned(bp
))
481 return XFS_ITEM_PINNED
;
482 if (!xfs_buf_trylock(bp
))
483 return XFS_ITEM_LOCKED
;
485 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
487 trace_xfs_buf_item_push(bip
);
489 if (!xfs_buf_delwri_queue(bp
, buffer_list
))
490 rval
= XFS_ITEM_FLUSHING
;
496 * Release the buffer associated with the buf log item. If there is no dirty
497 * logged data associated with the buffer recorded in the buf log item, then
498 * free the buf log item and remove the reference to it in the buffer.
500 * This call ignores the recursion count. It is only called when the buffer
501 * should REALLY be unlocked, regardless of the recursion count.
503 * We unconditionally drop the transaction's reference to the log item. If the
504 * item was logged, then another reference was taken when it was pinned, so we
505 * can safely drop the transaction reference now. This also allows us to avoid
506 * potential races with the unpin code freeing the bli by not referencing the
507 * bli after we've dropped the reference count.
509 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
510 * if necessary but do not unlock the buffer. This is for support of
511 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
516 struct xfs_log_item
*lip
)
518 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
519 struct xfs_buf
*bp
= bip
->bli_buf
;
523 /* Clear the buffer's association with this transaction. */
527 * If this is a transaction abort, don't return early. Instead, allow
528 * the brelse to happen. Normally it would be done for stale
529 * (cancelled) buffers at unpin time, but we'll never go through the
530 * pin/unpin cycle if we abort inside commit.
532 aborted
= (lip
->li_flags
& XFS_LI_ABORTED
) != 0;
535 * Before possibly freeing the buf item, determine if we should
536 * release the buffer at the end of this routine.
538 hold
= bip
->bli_flags
& XFS_BLI_HOLD
;
540 /* Clear the per transaction state. */
541 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
);
544 * If the buf item is marked stale, then don't do anything. We'll
545 * unlock the buffer and free the buf item when the buffer is unpinned
548 if (bip
->bli_flags
& XFS_BLI_STALE
) {
549 trace_xfs_buf_item_unlock_stale(bip
);
550 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
552 atomic_dec(&bip
->bli_refcount
);
557 trace_xfs_buf_item_unlock(bip
);
560 * If the buf item isn't tracking any data, free it, otherwise drop the
561 * reference we hold to it.
563 if (xfs_bitmap_empty(bip
->bli_format
.blf_data_map
,
564 bip
->bli_format
.blf_map_size
))
565 xfs_buf_item_relse(bp
);
567 atomic_dec(&bip
->bli_refcount
);
574 * This is called to find out where the oldest active copy of the
575 * buf log item in the on disk log resides now that the last log
576 * write of it completed at the given lsn.
577 * We always re-log all the dirty data in a buffer, so usually the
578 * latest copy in the on disk log is the only one that matters. For
579 * those cases we simply return the given lsn.
581 * The one exception to this is for buffers full of newly allocated
582 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
583 * flag set, indicating that only the di_next_unlinked fields from the
584 * inodes in the buffers will be replayed during recovery. If the
585 * original newly allocated inode images have not yet been flushed
586 * when the buffer is so relogged, then we need to make sure that we
587 * keep the old images in the 'active' portion of the log. We do this
588 * by returning the original lsn of that transaction here rather than
592 xfs_buf_item_committed(
593 struct xfs_log_item
*lip
,
596 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
598 trace_xfs_buf_item_committed(bip
);
600 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
606 xfs_buf_item_committing(
607 struct xfs_log_item
*lip
,
608 xfs_lsn_t commit_lsn
)
613 * This is the ops vector shared by all buf log items.
615 static const struct xfs_item_ops xfs_buf_item_ops
= {
616 .iop_size
= xfs_buf_item_size
,
617 .iop_format
= xfs_buf_item_format
,
618 .iop_pin
= xfs_buf_item_pin
,
619 .iop_unpin
= xfs_buf_item_unpin
,
620 .iop_unlock
= xfs_buf_item_unlock
,
621 .iop_committed
= xfs_buf_item_committed
,
622 .iop_push
= xfs_buf_item_push
,
623 .iop_committing
= xfs_buf_item_committing
628 * Allocate a new buf log item to go with the given buffer.
629 * Set the buffer's b_fsprivate field to point to the new
630 * buf log item. If there are other item's attached to the
631 * buffer (see xfs_buf_attach_iodone() below), then put the
632 * buf log item at the front.
639 xfs_log_item_t
*lip
= bp
->b_fspriv
;
640 xfs_buf_log_item_t
*bip
;
645 * Check to see if there is already a buf log item for
646 * this buffer. If there is, it is guaranteed to be
647 * the first. If we do already have one, there is
648 * nothing to do here so return.
650 ASSERT(bp
->b_target
->bt_mount
== mp
);
651 if (lip
!= NULL
&& lip
->li_type
== XFS_LI_BUF
)
655 * chunks is the number of XFS_BLF_CHUNK size pieces
656 * the buffer can be divided into. Make sure not to
657 * truncate any pieces. map_size is the size of the
658 * bitmap needed to describe the chunks of the buffer.
660 chunks
= (int)((BBTOB(bp
->b_length
) + (XFS_BLF_CHUNK
- 1)) >>
662 map_size
= (int)((chunks
+ NBWORD
) >> BIT_TO_WORD_SHIFT
);
664 bip
= (xfs_buf_log_item_t
*)kmem_zone_zalloc(xfs_buf_item_zone
,
666 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
669 bip
->bli_format
.blf_type
= XFS_LI_BUF
;
670 bip
->bli_format
.blf_blkno
= (__int64_t
)XFS_BUF_ADDR(bp
);
671 bip
->bli_format
.blf_len
= (ushort
)bp
->b_length
;
672 bip
->bli_format
.blf_map_size
= map_size
;
674 #ifdef XFS_TRANS_DEBUG
676 * Allocate the arrays for tracking what needs to be logged
677 * and what our callers request to be logged. bli_orig
678 * holds a copy of the original, clean buffer for comparison
679 * against, and bli_logged keeps a 1 bit flag per byte in
680 * the buffer to indicate which bytes the callers have asked
683 bip
->bli_orig
= kmem_alloc(BBTOB(bp
->b_length
), KM_SLEEP
);
684 memcpy(bip
->bli_orig
, bp
->b_addr
, BBTOB(bp
->b_length
));
685 bip
->bli_logged
= kmem_zalloc(BBTOB(bp
->b_length
) / NBBY
, KM_SLEEP
);
689 * Put the buf item into the list of items attached to the
690 * buffer at the front.
693 bip
->bli_item
.li_bio_list
= bp
->b_fspriv
;
699 * Mark bytes first through last inclusive as dirty in the buf
704 xfs_buf_log_item_t
*bip
,
719 * Mark the item as having some dirty data for
720 * quick reference in xfs_buf_item_dirty.
722 bip
->bli_flags
|= XFS_BLI_DIRTY
;
725 * Convert byte offsets to bit numbers.
727 first_bit
= first
>> XFS_BLF_SHIFT
;
728 last_bit
= last
>> XFS_BLF_SHIFT
;
731 * Calculate the total number of bits to be set.
733 bits_to_set
= last_bit
- first_bit
+ 1;
736 * Get a pointer to the first word in the bitmap
739 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
740 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
743 * Calculate the starting bit in the first word.
745 bit
= first_bit
& (uint
)(NBWORD
- 1);
748 * First set any bits in the first word of our range.
749 * If it starts at bit 0 of the word, it will be
750 * set below rather than here. That is what the variable
751 * bit tells us. The variable bits_set tracks the number
752 * of bits that have been set so far. End_bit is the number
753 * of the last bit to be set in this word plus one.
756 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
757 mask
= ((1 << (end_bit
- bit
)) - 1) << bit
;
760 bits_set
= end_bit
- bit
;
766 * Now set bits a whole word at a time that are between
767 * first_bit and last_bit.
769 while ((bits_to_set
- bits_set
) >= NBWORD
) {
770 *wordp
|= 0xffffffff;
776 * Finally, set any bits left to be set in one last partial word.
778 end_bit
= bits_to_set
- bits_set
;
780 mask
= (1 << end_bit
) - 1;
784 xfs_buf_item_log_debug(bip
, first
, last
);
789 * Return 1 if the buffer has some data that has been logged (at any
790 * point, not just the current transaction) and 0 if not.
794 xfs_buf_log_item_t
*bip
)
796 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
801 xfs_buf_log_item_t
*bip
)
803 #ifdef XFS_TRANS_DEBUG
804 kmem_free(bip
->bli_orig
);
805 kmem_free(bip
->bli_logged
);
806 #endif /* XFS_TRANS_DEBUG */
808 kmem_zone_free(xfs_buf_item_zone
, bip
);
812 * This is called when the buf log item is no longer needed. It should
813 * free the buf log item associated with the given buffer and clear
814 * the buffer's pointer to the buf log item. If there are no more
815 * items in the list, clear the b_iodone field of the buffer (see
816 * xfs_buf_attach_iodone() below).
822 xfs_buf_log_item_t
*bip
;
824 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
827 bp
->b_fspriv
= bip
->bli_item
.li_bio_list
;
828 if (bp
->b_fspriv
== NULL
)
832 xfs_buf_item_free(bip
);
837 * Add the given log item with its callback to the list of callbacks
838 * to be called when the buffer's I/O completes. If it is not set
839 * already, set the buffer's b_iodone() routine to be
840 * xfs_buf_iodone_callbacks() and link the log item into the list of
841 * items rooted at b_fsprivate. Items are always added as the second
842 * entry in the list if there is a first, because the buf item code
843 * assumes that the buf log item is first.
846 xfs_buf_attach_iodone(
848 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
851 xfs_log_item_t
*head_lip
;
853 ASSERT(xfs_buf_islocked(bp
));
856 head_lip
= bp
->b_fspriv
;
858 lip
->li_bio_list
= head_lip
->li_bio_list
;
859 head_lip
->li_bio_list
= lip
;
864 ASSERT(bp
->b_iodone
== NULL
||
865 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
866 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
870 * We can have many callbacks on a buffer. Running the callbacks individually
871 * can cause a lot of contention on the AIL lock, so we allow for a single
872 * callback to be able to scan the remaining lip->li_bio_list for other items
873 * of the same type and callback to be processed in the first call.
875 * As a result, the loop walking the callback list below will also modify the
876 * list. it removes the first item from the list and then runs the callback.
877 * The loop then restarts from the new head of the list. This allows the
878 * callback to scan and modify the list attached to the buffer and we don't
879 * have to care about maintaining a next item pointer.
882 xfs_buf_do_callbacks(
885 struct xfs_log_item
*lip
;
887 while ((lip
= bp
->b_fspriv
) != NULL
) {
888 bp
->b_fspriv
= lip
->li_bio_list
;
889 ASSERT(lip
->li_cb
!= NULL
);
891 * Clear the next pointer so we don't have any
892 * confusion if the item is added to another buf.
893 * Don't touch the log item after calling its
894 * callback, because it could have freed itself.
896 lip
->li_bio_list
= NULL
;
902 * This is the iodone() function for buffers which have had callbacks
903 * attached to them by xfs_buf_attach_iodone(). It should remove each
904 * log item from the buffer's list and call the callback of each in turn.
905 * When done, the buffer's fsprivate field is set to NULL and the buffer
906 * is unlocked with a call to iodone().
909 xfs_buf_iodone_callbacks(
912 struct xfs_log_item
*lip
= bp
->b_fspriv
;
913 struct xfs_mount
*mp
= lip
->li_mountp
;
914 static ulong lasttime
;
915 static xfs_buftarg_t
*lasttarg
;
917 if (likely(!xfs_buf_geterror(bp
)))
921 * If we've already decided to shutdown the filesystem because of
922 * I/O errors, there's no point in giving this a retry.
924 if (XFS_FORCED_SHUTDOWN(mp
)) {
927 trace_xfs_buf_item_iodone(bp
, _RET_IP_
);
931 if (bp
->b_target
!= lasttarg
||
932 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
934 xfs_buf_ioerror_alert(bp
, __func__
);
936 lasttarg
= bp
->b_target
;
939 * If the write was asynchronous then no one will be looking for the
940 * error. Clear the error state and write the buffer out again.
942 * XXX: This helps against transient write errors, but we need to find
943 * a way to shut the filesystem down if the writes keep failing.
945 * In practice we'll shut the filesystem down soon as non-transient
946 * erorrs tend to affect the whole device and a failing log write
947 * will make us give up. But we really ought to do better here.
949 if (XFS_BUF_ISASYNC(bp
)) {
950 ASSERT(bp
->b_iodone
!= NULL
);
952 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
954 xfs_buf_ioerror(bp
, 0); /* errno of 0 unsets the flag */
956 if (!XFS_BUF_ISSTALE(bp
)) {
957 bp
->b_flags
|= XBF_WRITE
| XBF_ASYNC
| XBF_DONE
;
967 * If the write of the buffer was synchronous, we want to make
968 * sure to return the error to the caller of xfs_bwrite().
973 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
976 xfs_buf_do_callbacks(bp
);
979 xfs_buf_ioend(bp
, 0);
983 * This is the iodone() function for buffers which have been
984 * logged. It is called when they are eventually flushed out.
985 * It should remove the buf item from the AIL, and free the buf item.
986 * It is called by xfs_buf_iodone_callbacks() above which will take
987 * care of cleaning up the buffer itself.
992 struct xfs_log_item
*lip
)
994 struct xfs_ail
*ailp
= lip
->li_ailp
;
996 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1001 * If we are forcibly shutting down, this may well be
1002 * off the AIL already. That's because we simulate the
1003 * log-committed callbacks to unpin these buffers. Or we may never
1004 * have put this item on AIL because of the transaction was
1005 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1007 * Either way, AIL is useless if we're forcing a shutdown.
1009 spin_lock(&ailp
->xa_lock
);
1010 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1011 xfs_buf_item_free(BUF_ITEM(lip
));