2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_buf_item.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
34 kmem_zone_t
*xfs_buf_item_zone
;
36 static inline struct xfs_buf_log_item
*BUF_ITEM(struct xfs_log_item
*lip
)
38 return container_of(lip
, struct xfs_buf_log_item
, bli_item
);
42 #ifdef XFS_TRANS_DEBUG
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
53 xfs_buf_item_log_debug(
54 xfs_buf_log_item_t
*bip
,
67 ASSERT(bip
->bli_logged
!= NULL
);
69 nbytes
= last
- first
+ 1;
70 bfset(bip
->bli_logged
, first
, nbytes
);
71 for (x
= 0; x
< nbytes
; x
++) {
72 chunk_num
= byte
>> XFS_BLF_SHIFT
;
73 word_num
= chunk_num
>> BIT_TO_WORD_SHIFT
;
74 bit_num
= chunk_num
& (NBWORD
- 1);
75 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
76 bit_set
= *wordp
& (1 << bit_num
);
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
88 xfs_buf_item_flush_log_debug(
93 xfs_buf_log_item_t
*bip
;
96 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
97 if ((bip
== NULL
) || (bip
->bli_item
.li_type
!= XFS_LI_BUF
)) {
101 ASSERT(bip
->bli_logged
!= NULL
);
102 nbytes
= last
- first
+ 1;
103 bfset(bip
->bli_logged
, first
, nbytes
);
107 * This function is called to verify that our callers have logged
108 * all the bytes that they changed.
110 * It does this by comparing the original copy of the buffer stored in
111 * the buf log item's bli_orig array to the current copy of the buffer
112 * and ensuring that all bytes which mismatch are set in the bli_logged
113 * array of the buf log item.
116 xfs_buf_item_log_check(
117 xfs_buf_log_item_t
*bip
)
124 ASSERT(bip
->bli_orig
!= NULL
);
125 ASSERT(bip
->bli_logged
!= NULL
);
128 ASSERT(XFS_BUF_COUNT(bp
) > 0);
129 ASSERT(XFS_BUF_PTR(bp
) != NULL
);
130 orig
= bip
->bli_orig
;
131 buffer
= XFS_BUF_PTR(bp
);
132 for (x
= 0; x
< XFS_BUF_COUNT(bp
); x
++) {
133 if (orig
[x
] != buffer
[x
] && !btst(bip
->bli_logged
, x
))
135 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
140 #define xfs_buf_item_log_debug(x,y,z)
141 #define xfs_buf_item_log_check(x)
144 STATIC
void xfs_buf_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(XFS_BUF_ISBUSY(bip
->bli_buf
));
375 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
376 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
377 (bip
->bli_flags
& XFS_BLI_STALE
));
379 trace_xfs_buf_item_pin(bip
);
381 atomic_inc(&bip
->bli_refcount
);
382 atomic_inc(&bip
->bli_buf
->b_pin_count
);
386 * This is called to unpin the buffer associated with the buf log
387 * item which was previously pinned with a call to xfs_buf_item_pin().
389 * Also drop the reference to the buf item for the current transaction.
390 * If the XFS_BLI_STALE flag is set and we are the last reference,
391 * then free up the buf log item and unlock the buffer.
393 * If the remove flag is set we are called from uncommit in the
394 * forced-shutdown path. If that is true and the reference count on
395 * the log item is going to drop to zero we need to free the item's
396 * descriptor in the transaction.
400 struct xfs_log_item
*lip
,
403 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
404 xfs_buf_t
*bp
= bip
->bli_buf
;
405 struct xfs_ail
*ailp
= lip
->li_ailp
;
406 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
409 ASSERT(XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*) == bip
);
410 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
412 trace_xfs_buf_item_unpin(bip
);
414 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
416 if (atomic_dec_and_test(&bp
->b_pin_count
))
417 wake_up_all(&bp
->b_waiters
);
419 if (freed
&& stale
) {
420 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
421 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
422 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
423 ASSERT(XFS_BUF_ISSTALE(bp
));
424 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
426 trace_xfs_buf_item_unpin_stale(bip
);
430 * If we are in a transaction context, we have to
431 * remove the log item from the transaction as we are
432 * about to release our reference to the buffer. If we
433 * don't, the unlock that occurs later in
434 * xfs_trans_uncommit() will try to reference the
435 * buffer which we no longer have a hold on.
438 xfs_trans_del_item(lip
);
441 * Since the transaction no longer refers to the buffer,
442 * the buffer should no longer refer to the transaction.
444 XFS_BUF_SET_FSPRIVATE2(bp
, NULL
);
448 * If we get called here because of an IO error, we may
449 * or may not have the item on the AIL. xfs_trans_ail_delete()
450 * will take care of that situation.
451 * xfs_trans_ail_delete() drops the AIL lock.
453 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
454 xfs_buf_do_callbacks(bp
);
455 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
456 XFS_BUF_CLR_IODONE_FUNC(bp
);
458 spin_lock(&ailp
->xa_lock
);
459 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)bip
);
460 xfs_buf_item_relse(bp
);
461 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
);
468 * This is called to attempt to lock the buffer associated with this
469 * buf log item. Don't sleep on the buffer lock. If we can't get
470 * the lock right away, return 0. If we can get the lock, take a
471 * reference to the buffer. If this is a delayed write buffer that
472 * needs AIL help to be written back, invoke the pushbuf routine
473 * rather than the normal success path.
476 xfs_buf_item_trylock(
477 struct xfs_log_item
*lip
)
479 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
480 struct xfs_buf
*bp
= bip
->bli_buf
;
482 if (XFS_BUF_ISPINNED(bp
))
483 return XFS_ITEM_PINNED
;
484 if (!XFS_BUF_CPSEMA(bp
))
485 return XFS_ITEM_LOCKED
;
487 /* take a reference to the buffer. */
490 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
491 trace_xfs_buf_item_trylock(bip
);
492 if (XFS_BUF_ISDELAYWRITE(bp
))
493 return XFS_ITEM_PUSHBUF
;
494 return XFS_ITEM_SUCCESS
;
498 * Release the buffer associated with the buf log item. If there is no dirty
499 * logged data associated with the buffer recorded in the buf log item, then
500 * free the buf log item and remove the reference to it in the buffer.
502 * This call ignores the recursion count. It is only called when the buffer
503 * should REALLY be unlocked, regardless of the recursion count.
505 * We unconditionally drop the transaction's reference to the log item. If the
506 * item was logged, then another reference was taken when it was pinned, so we
507 * can safely drop the transaction reference now. This also allows us to avoid
508 * potential races with the unpin code freeing the bli by not referencing the
509 * bli after we've dropped the reference count.
511 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
512 * if necessary but do not unlock the buffer. This is for support of
513 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
518 struct xfs_log_item
*lip
)
520 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
521 struct xfs_buf
*bp
= bip
->bli_buf
;
525 /* Clear the buffer's association with this transaction. */
526 XFS_BUF_SET_FSPRIVATE2(bp
, NULL
);
529 * If this is a transaction abort, don't return early. Instead, allow
530 * the brelse to happen. Normally it would be done for stale
531 * (cancelled) buffers at unpin time, but we'll never go through the
532 * pin/unpin cycle if we abort inside commit.
534 aborted
= (lip
->li_flags
& XFS_LI_ABORTED
) != 0;
537 * Before possibly freeing the buf item, determine if we should
538 * release the buffer at the end of this routine.
540 hold
= bip
->bli_flags
& XFS_BLI_HOLD
;
542 /* Clear the per transaction state. */
543 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
);
546 * If the buf item is marked stale, then don't do anything. We'll
547 * unlock the buffer and free the buf item when the buffer is unpinned
550 if (bip
->bli_flags
& XFS_BLI_STALE
) {
551 trace_xfs_buf_item_unlock_stale(bip
);
552 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
554 atomic_dec(&bip
->bli_refcount
);
559 trace_xfs_buf_item_unlock(bip
);
562 * If the buf item isn't tracking any data, free it, otherwise drop the
563 * reference we hold to it.
565 if (xfs_bitmap_empty(bip
->bli_format
.blf_data_map
,
566 bip
->bli_format
.blf_map_size
))
567 xfs_buf_item_relse(bp
);
569 atomic_dec(&bip
->bli_refcount
);
576 * This is called to find out where the oldest active copy of the
577 * buf log item in the on disk log resides now that the last log
578 * write of it completed at the given lsn.
579 * We always re-log all the dirty data in a buffer, so usually the
580 * latest copy in the on disk log is the only one that matters. For
581 * those cases we simply return the given lsn.
583 * The one exception to this is for buffers full of newly allocated
584 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
585 * flag set, indicating that only the di_next_unlinked fields from the
586 * inodes in the buffers will be replayed during recovery. If the
587 * original newly allocated inode images have not yet been flushed
588 * when the buffer is so relogged, then we need to make sure that we
589 * keep the old images in the 'active' portion of the log. We do this
590 * by returning the original lsn of that transaction here rather than
594 xfs_buf_item_committed(
595 struct xfs_log_item
*lip
,
598 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
600 trace_xfs_buf_item_committed(bip
);
602 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
608 * The buffer is locked, but is not a delayed write buffer. This happens
609 * if we race with IO completion and hence we don't want to try to write it
610 * again. Just release the buffer.
614 struct xfs_log_item
*lip
)
616 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
617 struct xfs_buf
*bp
= bip
->bli_buf
;
619 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
620 ASSERT(!XFS_BUF_ISDELAYWRITE(bp
));
622 trace_xfs_buf_item_push(bip
);
628 * The buffer is locked and is a delayed write buffer. Promote the buffer
629 * in the delayed write queue as the caller knows that they must invoke
630 * the xfsbufd to get this buffer written. We have to unlock the buffer
631 * to allow the xfsbufd to write it, too.
634 xfs_buf_item_pushbuf(
635 struct xfs_log_item
*lip
)
637 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
638 struct xfs_buf
*bp
= bip
->bli_buf
;
640 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
641 ASSERT(XFS_BUF_ISDELAYWRITE(bp
));
643 trace_xfs_buf_item_pushbuf(bip
);
645 xfs_buf_delwri_promote(bp
);
650 xfs_buf_item_committing(
651 struct xfs_log_item
*lip
,
652 xfs_lsn_t commit_lsn
)
657 * This is the ops vector shared by all buf log items.
659 static struct xfs_item_ops xfs_buf_item_ops
= {
660 .iop_size
= xfs_buf_item_size
,
661 .iop_format
= xfs_buf_item_format
,
662 .iop_pin
= xfs_buf_item_pin
,
663 .iop_unpin
= xfs_buf_item_unpin
,
664 .iop_trylock
= xfs_buf_item_trylock
,
665 .iop_unlock
= xfs_buf_item_unlock
,
666 .iop_committed
= xfs_buf_item_committed
,
667 .iop_push
= xfs_buf_item_push
,
668 .iop_pushbuf
= xfs_buf_item_pushbuf
,
669 .iop_committing
= xfs_buf_item_committing
674 * Allocate a new buf log item to go with the given buffer.
675 * Set the buffer's b_fsprivate field to point to the new
676 * buf log item. If there are other item's attached to the
677 * buffer (see xfs_buf_attach_iodone() below), then put the
678 * buf log item at the front.
686 xfs_buf_log_item_t
*bip
;
691 * Check to see if there is already a buf log item for
692 * this buffer. If there is, it is guaranteed to be
693 * the first. If we do already have one, there is
694 * nothing to do here so return.
696 ASSERT(bp
->b_target
->bt_mount
== mp
);
697 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
698 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
699 if (lip
->li_type
== XFS_LI_BUF
) {
705 * chunks is the number of XFS_BLF_CHUNK size pieces
706 * the buffer can be divided into. Make sure not to
707 * truncate any pieces. map_size is the size of the
708 * bitmap needed to describe the chunks of the buffer.
710 chunks
= (int)((XFS_BUF_COUNT(bp
) + (XFS_BLF_CHUNK
- 1)) >> XFS_BLF_SHIFT
);
711 map_size
= (int)((chunks
+ NBWORD
) >> BIT_TO_WORD_SHIFT
);
713 bip
= (xfs_buf_log_item_t
*)kmem_zone_zalloc(xfs_buf_item_zone
,
715 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
718 bip
->bli_format
.blf_type
= XFS_LI_BUF
;
719 bip
->bli_format
.blf_blkno
= (__int64_t
)XFS_BUF_ADDR(bp
);
720 bip
->bli_format
.blf_len
= (ushort
)BTOBB(XFS_BUF_COUNT(bp
));
721 bip
->bli_format
.blf_map_size
= map_size
;
723 #ifdef XFS_TRANS_DEBUG
725 * Allocate the arrays for tracking what needs to be logged
726 * and what our callers request to be logged. bli_orig
727 * holds a copy of the original, clean buffer for comparison
728 * against, and bli_logged keeps a 1 bit flag per byte in
729 * the buffer to indicate which bytes the callers have asked
732 bip
->bli_orig
= (char *)kmem_alloc(XFS_BUF_COUNT(bp
), KM_SLEEP
);
733 memcpy(bip
->bli_orig
, XFS_BUF_PTR(bp
), XFS_BUF_COUNT(bp
));
734 bip
->bli_logged
= (char *)kmem_zalloc(XFS_BUF_COUNT(bp
) / NBBY
, KM_SLEEP
);
738 * Put the buf item into the list of items attached to the
739 * buffer at the front.
741 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
742 bip
->bli_item
.li_bio_list
=
743 XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
745 XFS_BUF_SET_FSPRIVATE(bp
, bip
);
750 * Mark bytes first through last inclusive as dirty in the buf
755 xfs_buf_log_item_t
*bip
,
770 * Mark the item as having some dirty data for
771 * quick reference in xfs_buf_item_dirty.
773 bip
->bli_flags
|= XFS_BLI_DIRTY
;
776 * Convert byte offsets to bit numbers.
778 first_bit
= first
>> XFS_BLF_SHIFT
;
779 last_bit
= last
>> XFS_BLF_SHIFT
;
782 * Calculate the total number of bits to be set.
784 bits_to_set
= last_bit
- first_bit
+ 1;
787 * Get a pointer to the first word in the bitmap
790 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
791 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
794 * Calculate the starting bit in the first word.
796 bit
= first_bit
& (uint
)(NBWORD
- 1);
799 * First set any bits in the first word of our range.
800 * If it starts at bit 0 of the word, it will be
801 * set below rather than here. That is what the variable
802 * bit tells us. The variable bits_set tracks the number
803 * of bits that have been set so far. End_bit is the number
804 * of the last bit to be set in this word plus one.
807 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
808 mask
= ((1 << (end_bit
- bit
)) - 1) << bit
;
811 bits_set
= end_bit
- bit
;
817 * Now set bits a whole word at a time that are between
818 * first_bit and last_bit.
820 while ((bits_to_set
- bits_set
) >= NBWORD
) {
821 *wordp
|= 0xffffffff;
827 * Finally, set any bits left to be set in one last partial word.
829 end_bit
= bits_to_set
- bits_set
;
831 mask
= (1 << end_bit
) - 1;
835 xfs_buf_item_log_debug(bip
, first
, last
);
840 * Return 1 if the buffer has some data that has been logged (at any
841 * point, not just the current transaction) and 0 if not.
845 xfs_buf_log_item_t
*bip
)
847 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
852 xfs_buf_log_item_t
*bip
)
854 #ifdef XFS_TRANS_DEBUG
855 kmem_free(bip
->bli_orig
);
856 kmem_free(bip
->bli_logged
);
857 #endif /* XFS_TRANS_DEBUG */
859 kmem_zone_free(xfs_buf_item_zone
, bip
);
863 * This is called when the buf log item is no longer needed. It should
864 * free the buf log item associated with the given buffer and clear
865 * the buffer's pointer to the buf log item. If there are no more
866 * items in the list, clear the b_iodone field of the buffer (see
867 * xfs_buf_attach_iodone() below).
873 xfs_buf_log_item_t
*bip
;
875 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
877 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
878 XFS_BUF_SET_FSPRIVATE(bp
, bip
->bli_item
.li_bio_list
);
879 if ((XFS_BUF_FSPRIVATE(bp
, void *) == NULL
) &&
880 (XFS_BUF_IODONE_FUNC(bp
) != NULL
)) {
881 XFS_BUF_CLR_IODONE_FUNC(bp
);
884 xfs_buf_item_free(bip
);
889 * Add the given log item with its callback to the list of callbacks
890 * to be called when the buffer's I/O completes. If it is not set
891 * already, set the buffer's b_iodone() routine to be
892 * xfs_buf_iodone_callbacks() and link the log item into the list of
893 * items rooted at b_fsprivate. Items are always added as the second
894 * entry in the list if there is a first, because the buf item code
895 * assumes that the buf log item is first.
898 xfs_buf_attach_iodone(
900 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
903 xfs_log_item_t
*head_lip
;
905 ASSERT(XFS_BUF_ISBUSY(bp
));
906 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
909 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
910 head_lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
911 lip
->li_bio_list
= head_lip
->li_bio_list
;
912 head_lip
->li_bio_list
= lip
;
914 XFS_BUF_SET_FSPRIVATE(bp
, lip
);
917 ASSERT((XFS_BUF_IODONE_FUNC(bp
) == xfs_buf_iodone_callbacks
) ||
918 (XFS_BUF_IODONE_FUNC(bp
) == NULL
));
919 XFS_BUF_SET_IODONE_FUNC(bp
, xfs_buf_iodone_callbacks
);
923 * We can have many callbacks on a buffer. Running the callbacks individually
924 * can cause a lot of contention on the AIL lock, so we allow for a single
925 * callback to be able to scan the remaining lip->li_bio_list for other items
926 * of the same type and callback to be processed in the first call.
928 * As a result, the loop walking the callback list below will also modify the
929 * list. it removes the first item from the list and then runs the callback.
930 * The loop then restarts from the new head of the list. This allows the
931 * callback to scan and modify the list attached to the buffer and we don't
932 * have to care about maintaining a next item pointer.
935 xfs_buf_do_callbacks(
938 struct xfs_log_item
*lip
;
940 while ((lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*)) != NULL
) {
941 XFS_BUF_SET_FSPRIVATE(bp
, lip
->li_bio_list
);
942 ASSERT(lip
->li_cb
!= NULL
);
944 * Clear the next pointer so we don't have any
945 * confusion if the item is added to another buf.
946 * Don't touch the log item after calling its
947 * callback, because it could have freed itself.
949 lip
->li_bio_list
= NULL
;
955 * This is the iodone() function for buffers which have had callbacks
956 * attached to them by xfs_buf_attach_iodone(). It should remove each
957 * log item from the buffer's list and call the callback of each in turn.
958 * When done, the buffer's fsprivate field is set to NULL and the buffer
959 * is unlocked with a call to iodone().
962 xfs_buf_iodone_callbacks(
965 struct xfs_log_item
*lip
= bp
->b_fspriv
;
966 struct xfs_mount
*mp
= lip
->li_mountp
;
967 static ulong lasttime
;
968 static xfs_buftarg_t
*lasttarg
;
970 if (likely(!XFS_BUF_GETERROR(bp
)))
974 * If we've already decided to shutdown the filesystem because of
975 * I/O errors, there's no point in giving this a retry.
977 if (XFS_FORCED_SHUTDOWN(mp
)) {
978 XFS_BUF_SUPER_STALE(bp
);
979 trace_xfs_buf_item_iodone(bp
, _RET_IP_
);
983 if (XFS_BUF_TARGET(bp
) != lasttarg
||
984 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
986 cmn_err(CE_ALERT
, "Device %s, XFS metadata write error"
987 " block 0x%llx in %s",
988 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp
)),
989 (__uint64_t
)XFS_BUF_ADDR(bp
), mp
->m_fsname
);
991 lasttarg
= XFS_BUF_TARGET(bp
);
994 * If the write was asynchronous then noone will be looking for the
995 * error. Clear the error state and write the buffer out again.
997 * During sync or umount we'll write all pending buffers again
998 * synchronous, which will catch these errors if they keep hanging
1001 if (XFS_BUF_ISASYNC(bp
)) {
1002 XFS_BUF_ERROR(bp
, 0); /* errno of 0 unsets the flag */
1004 if (!XFS_BUF_ISSTALE(bp
)) {
1005 XFS_BUF_DELAYWRITE(bp
);
1007 XFS_BUF_SET_START(bp
);
1009 ASSERT(XFS_BUF_IODONE_FUNC(bp
));
1010 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1016 * If the write of the buffer was synchronous, we want to make
1017 * sure to return the error to the caller of xfs_bwrite().
1021 XFS_BUF_UNDELAYWRITE(bp
);
1023 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1024 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1027 xfs_buf_do_callbacks(bp
);
1028 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
1029 XFS_BUF_CLR_IODONE_FUNC(bp
);
1030 xfs_buf_ioend(bp
, 0);
1034 * This is the iodone() function for buffers which have been
1035 * logged. It is called when they are eventually flushed out.
1036 * It should remove the buf item from the AIL, and free the buf item.
1037 * It is called by xfs_buf_iodone_callbacks() above which will take
1038 * care of cleaning up the buffer itself.
1043 struct xfs_log_item
*lip
)
1045 struct xfs_ail
*ailp
= lip
->li_ailp
;
1047 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1052 * If we are forcibly shutting down, this may well be
1053 * off the AIL already. That's because we simulate the
1054 * log-committed callbacks to unpin these buffers. Or we may never
1055 * have put this item on AIL because of the transaction was
1056 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1058 * Either way, AIL is useless if we're forcing a shutdown.
1060 spin_lock(&ailp
->xa_lock
);
1061 xfs_trans_ail_delete(ailp
, lip
);
1062 xfs_buf_item_free(BUF_ITEM(lip
));