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"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_buf_item.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_error.h"
30 #include "xfs_trace.h"
33 kmem_zone_t
*xfs_buf_item_zone
;
35 static inline struct xfs_buf_log_item
*BUF_ITEM(struct xfs_log_item
*lip
)
37 return container_of(lip
, struct xfs_buf_log_item
, bli_item
);
41 #ifdef XFS_TRANS_DEBUG
43 * This function uses an alternate strategy for tracking the bytes
44 * that the user requests to be logged. This can then be used
45 * in conjunction with the bli_orig array in the buf log item to
46 * catch bugs in our callers' code.
48 * We also double check the bits set in xfs_buf_item_log using a
49 * simple algorithm to check that every byte is accounted for.
52 xfs_buf_item_log_debug(
53 xfs_buf_log_item_t
*bip
,
66 ASSERT(bip
->bli_logged
!= NULL
);
68 nbytes
= last
- first
+ 1;
69 bfset(bip
->bli_logged
, first
, nbytes
);
70 for (x
= 0; x
< nbytes
; x
++) {
71 chunk_num
= byte
>> XFS_BLF_SHIFT
;
72 word_num
= chunk_num
>> BIT_TO_WORD_SHIFT
;
73 bit_num
= chunk_num
& (NBWORD
- 1);
74 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
75 bit_set
= *wordp
& (1 << bit_num
);
82 * This function is called when we flush something into a buffer without
83 * logging it. This happens for things like inodes which are logged
84 * separately from the buffer.
87 xfs_buf_item_flush_log_debug(
92 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
95 if (bip
== NULL
|| (bip
->bli_item
.li_type
!= XFS_LI_BUF
))
98 ASSERT(bip
->bli_logged
!= NULL
);
99 nbytes
= last
- first
+ 1;
100 bfset(bip
->bli_logged
, first
, nbytes
);
104 * This function is called to verify that our callers have logged
105 * all the bytes that they changed.
107 * It does this by comparing the original copy of the buffer stored in
108 * the buf log item's bli_orig array to the current copy of the buffer
109 * and ensuring that all bytes which mismatch are set in the bli_logged
110 * array of the buf log item.
113 xfs_buf_item_log_check(
114 xfs_buf_log_item_t
*bip
)
121 ASSERT(bip
->bli_orig
!= NULL
);
122 ASSERT(bip
->bli_logged
!= NULL
);
125 ASSERT(bp
->b_length
> 0);
126 ASSERT(bp
->b_addr
!= NULL
);
127 orig
= bip
->bli_orig
;
129 for (x
= 0; x
< BBTOB(bp
->b_length
); x
++) {
130 if (orig
[x
] != buffer
[x
] && !btst(bip
->bli_logged
, x
)) {
131 xfs_emerg(bp
->b_mount
,
132 "%s: bip %x buffer %x orig %x index %d",
133 __func__
, bip
, bp
, orig
, x
);
139 #define xfs_buf_item_log_debug(x,y,z)
140 #define xfs_buf_item_log_check(x)
143 STATIC
void xfs_buf_do_callbacks(struct xfs_buf
*bp
);
146 * This returns the number of log iovecs needed to log the
147 * given buf log item.
149 * It calculates this as 1 iovec for the buf log format structure
150 * and 1 for each stretch of non-contiguous chunks to be logged.
151 * Contiguous chunks are logged in a single iovec.
153 * If the XFS_BLI_STALE flag has been set, then log nothing.
156 xfs_buf_item_size_segment(
157 struct xfs_buf_log_item
*bip
,
158 struct xfs_buf_log_format
*blfp
)
160 struct xfs_buf
*bp
= bip
->bli_buf
;
165 last_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
170 * initial count for a dirty buffer is 2 vectors - the format structure
171 * and the first dirty region.
175 while (last_bit
!= -1) {
177 * This takes the bit number to start looking from and
178 * returns the next set bit from there. It returns -1
179 * if there are no more bits set or the start bit is
180 * beyond the end of the bitmap.
182 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
185 * If we run out of bits, leave the loop,
186 * else if we find a new set of bits bump the number of vecs,
187 * else keep scanning the current set of bits.
189 if (next_bit
== -1) {
191 } else if (next_bit
!= last_bit
+ 1) {
194 } else if (xfs_buf_offset(bp
, next_bit
* XFS_BLF_CHUNK
) !=
195 (xfs_buf_offset(bp
, last_bit
* XFS_BLF_CHUNK
) +
208 * This returns the number of log iovecs needed to log the given buf log item.
210 * It calculates this as 1 iovec for the buf log format structure and 1 for each
211 * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged
214 * Discontiguous buffers need a format structure per region that that is being
215 * logged. This makes the changes in the buffer appear to log recovery as though
216 * they came from separate buffers, just like would occur if multiple buffers
217 * were used instead of a single discontiguous buffer. This enables
218 * discontiguous buffers to be in-memory constructs, completely transparent to
219 * what ends up on disk.
221 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
226 struct xfs_log_item
*lip
)
228 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
232 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
233 if (bip
->bli_flags
& XFS_BLI_STALE
) {
235 * The buffer is stale, so all we need to log
236 * is the buf log format structure with the
239 trace_xfs_buf_item_size_stale(bip
);
240 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
241 return bip
->bli_format_count
;
244 ASSERT(bip
->bli_flags
& XFS_BLI_LOGGED
);
247 * the vector count is based on the number of buffer vectors we have
248 * dirty bits in. This will only be greater than one when we have a
249 * compound buffer with more than one segment dirty. Hence for compound
250 * buffers we need to track which segment the dirty bits correspond to,
251 * and when we move from one segment to the next increment the vector
252 * count for the extra buf log format structure that will need to be
256 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
257 nvecs
+= xfs_buf_item_size_segment(bip
, &bip
->bli_formats
[i
]);
260 trace_xfs_buf_item_size(bip
);
264 static struct xfs_log_iovec
*
265 xfs_buf_item_format_segment(
266 struct xfs_buf_log_item
*bip
,
267 struct xfs_log_iovec
*vecp
,
269 struct xfs_buf_log_format
*blfp
)
271 struct xfs_buf
*bp
= bip
->bli_buf
;
280 /* copy the flags across from the base format item */
281 blfp
->blf_flags
= bip
->bli_format
.blf_flags
;
284 * Base size is the actual size of the ondisk structure - it reflects
285 * the actual size of the dirty bitmap rather than the size of the in
288 base_size
= offsetof(struct xfs_buf_log_format
, blf_data_map
) +
289 (blfp
->blf_map_size
* sizeof(blfp
->blf_data_map
[0]));
291 vecp
->i_len
= base_size
;
292 vecp
->i_type
= XLOG_REG_TYPE_BFORMAT
;
296 if (bip
->bli_flags
& XFS_BLI_STALE
) {
298 * The buffer is stale, so all we need to log
299 * is the buf log format structure with the
302 trace_xfs_buf_item_format_stale(bip
);
303 ASSERT(blfp
->blf_flags
& XFS_BLF_CANCEL
);
304 blfp
->blf_size
= nvecs
;
309 * Fill in an iovec for each set of contiguous chunks.
311 first_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
, 0);
312 ASSERT(first_bit
!= -1);
313 last_bit
= first_bit
;
317 * This takes the bit number to start looking from and
318 * returns the next set bit from there. It returns -1
319 * if there are no more bits set or the start bit is
320 * beyond the end of the bitmap.
322 next_bit
= xfs_next_bit(blfp
->blf_data_map
, blfp
->blf_map_size
,
325 * If we run out of bits fill in the last iovec and get
327 * Else if we start a new set of bits then fill in the
328 * iovec for the series we were looking at and start
329 * counting the bits in the new one.
330 * Else we're still in the same set of bits so just
331 * keep counting and scanning.
333 if (next_bit
== -1) {
334 buffer_offset
= offset
+ first_bit
* XFS_BLF_CHUNK
;
335 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
336 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
337 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
340 } else if (next_bit
!= last_bit
+ 1) {
341 buffer_offset
= offset
+ first_bit
* XFS_BLF_CHUNK
;
342 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
343 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
344 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
347 first_bit
= next_bit
;
350 } else if (xfs_buf_offset(bp
, offset
+
351 (next_bit
<< XFS_BLF_SHIFT
)) !=
352 (xfs_buf_offset(bp
, offset
+
353 (last_bit
<< XFS_BLF_SHIFT
)) +
355 buffer_offset
= offset
+ first_bit
* XFS_BLF_CHUNK
;
356 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
357 vecp
->i_len
= nbits
* XFS_BLF_CHUNK
;
358 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
360 * You would think we need to bump the nvecs here too, but we do not
361 * this number is used by recovery, and it gets confused by the boundary
366 first_bit
= next_bit
;
374 bip
->bli_format
.blf_size
= nvecs
;
379 * This is called to fill in the vector of log iovecs for the
380 * given log buf item. It fills the first entry with a buf log
381 * format structure, and the rest point to contiguous chunks
386 struct xfs_log_item
*lip
,
387 struct xfs_log_iovec
*vecp
)
389 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
390 struct xfs_buf
*bp
= bip
->bli_buf
;
394 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
395 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
396 (bip
->bli_flags
& XFS_BLI_STALE
));
399 * If it is an inode buffer, transfer the in-memory state to the
400 * format flags and clear the in-memory state. We do not transfer
401 * this state if the inode buffer allocation has not yet been committed
402 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
403 * correct replay of the inode allocation.
405 if (bip
->bli_flags
& XFS_BLI_INODE_BUF
) {
406 if (!((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
407 xfs_log_item_in_current_chkpt(lip
)))
408 bip
->bli_format
.blf_flags
|= XFS_BLF_INODE_BUF
;
409 bip
->bli_flags
&= ~XFS_BLI_INODE_BUF
;
412 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
413 vecp
= xfs_buf_item_format_segment(bip
, vecp
, offset
,
414 &bip
->bli_formats
[i
]);
415 offset
+= bp
->b_maps
[i
].bm_len
;
419 * Check to make sure everything is consistent.
421 trace_xfs_buf_item_format(bip
);
422 xfs_buf_item_log_check(bip
);
426 * This is called to pin the buffer associated with the buf log item in memory
427 * so it cannot be written out.
429 * We also always take a reference to the buffer log item here so that the bli
430 * is held while the item is pinned in memory. This means that we can
431 * unconditionally drop the reference count a transaction holds when the
432 * transaction is completed.
436 struct xfs_log_item
*lip
)
438 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
440 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
441 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
442 (bip
->bli_flags
& XFS_BLI_STALE
));
444 trace_xfs_buf_item_pin(bip
);
446 atomic_inc(&bip
->bli_refcount
);
447 atomic_inc(&bip
->bli_buf
->b_pin_count
);
451 * This is called to unpin the buffer associated with the buf log
452 * item which was previously pinned with a call to xfs_buf_item_pin().
454 * Also drop the reference to the buf item for the current transaction.
455 * If the XFS_BLI_STALE flag is set and we are the last reference,
456 * then free up the buf log item and unlock the buffer.
458 * If the remove flag is set we are called from uncommit in the
459 * forced-shutdown path. If that is true and the reference count on
460 * the log item is going to drop to zero we need to free the item's
461 * descriptor in the transaction.
465 struct xfs_log_item
*lip
,
468 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
469 xfs_buf_t
*bp
= bip
->bli_buf
;
470 struct xfs_ail
*ailp
= lip
->li_ailp
;
471 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
474 ASSERT(bp
->b_fspriv
== bip
);
475 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
477 trace_xfs_buf_item_unpin(bip
);
479 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
481 if (atomic_dec_and_test(&bp
->b_pin_count
))
482 wake_up_all(&bp
->b_waiters
);
484 if (freed
&& stale
) {
485 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
486 ASSERT(xfs_buf_islocked(bp
));
487 ASSERT(XFS_BUF_ISSTALE(bp
));
488 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
490 trace_xfs_buf_item_unpin_stale(bip
);
494 * If we are in a transaction context, we have to
495 * remove the log item from the transaction as we are
496 * about to release our reference to the buffer. If we
497 * don't, the unlock that occurs later in
498 * xfs_trans_uncommit() will try to reference the
499 * buffer which we no longer have a hold on.
502 xfs_trans_del_item(lip
);
505 * Since the transaction no longer refers to the buffer,
506 * the buffer should no longer refer to the transaction.
512 * If we get called here because of an IO error, we may
513 * or may not have the item on the AIL. xfs_trans_ail_delete()
514 * will take care of that situation.
515 * xfs_trans_ail_delete() drops the AIL lock.
517 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
518 xfs_buf_do_callbacks(bp
);
522 spin_lock(&ailp
->xa_lock
);
523 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_LOG_IO_ERROR
);
524 xfs_buf_item_relse(bp
);
525 ASSERT(bp
->b_fspriv
== NULL
);
528 } else if (freed
&& remove
) {
530 xfs_buf_ioerror(bp
, EIO
);
533 xfs_buf_ioend(bp
, 0);
539 struct xfs_log_item
*lip
,
540 struct list_head
*buffer_list
)
542 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
543 struct xfs_buf
*bp
= bip
->bli_buf
;
544 uint rval
= XFS_ITEM_SUCCESS
;
546 if (xfs_buf_ispinned(bp
))
547 return XFS_ITEM_PINNED
;
548 if (!xfs_buf_trylock(bp
))
549 return XFS_ITEM_LOCKED
;
551 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
553 trace_xfs_buf_item_push(bip
);
555 if (!xfs_buf_delwri_queue(bp
, buffer_list
))
556 rval
= XFS_ITEM_FLUSHING
;
562 * Release the buffer associated with the buf log item. If there is no dirty
563 * logged data associated with the buffer recorded in the buf log item, then
564 * free the buf log item and remove the reference to it in the buffer.
566 * This call ignores the recursion count. It is only called when the buffer
567 * should REALLY be unlocked, regardless of the recursion count.
569 * We unconditionally drop the transaction's reference to the log item. If the
570 * item was logged, then another reference was taken when it was pinned, so we
571 * can safely drop the transaction reference now. This also allows us to avoid
572 * potential races with the unpin code freeing the bli by not referencing the
573 * bli after we've dropped the reference count.
575 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
576 * if necessary but do not unlock the buffer. This is for support of
577 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
582 struct xfs_log_item
*lip
)
584 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
585 struct xfs_buf
*bp
= bip
->bli_buf
;
589 /* Clear the buffer's association with this transaction. */
593 * If this is a transaction abort, don't return early. Instead, allow
594 * the brelse to happen. Normally it would be done for stale
595 * (cancelled) buffers at unpin time, but we'll never go through the
596 * pin/unpin cycle if we abort inside commit.
598 aborted
= (lip
->li_flags
& XFS_LI_ABORTED
) != 0;
601 * Before possibly freeing the buf item, determine if we should
602 * release the buffer at the end of this routine.
604 hold
= bip
->bli_flags
& XFS_BLI_HOLD
;
606 /* Clear the per transaction state. */
607 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
);
610 * If the buf item is marked stale, then don't do anything. We'll
611 * unlock the buffer and free the buf item when the buffer is unpinned
614 if (bip
->bli_flags
& XFS_BLI_STALE
) {
615 trace_xfs_buf_item_unlock_stale(bip
);
616 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
618 atomic_dec(&bip
->bli_refcount
);
623 trace_xfs_buf_item_unlock(bip
);
626 * If the buf item isn't tracking any data, free it, otherwise drop the
627 * reference we hold to it.
629 if (xfs_bitmap_empty(bip
->bli_format
.blf_data_map
,
630 bip
->bli_format
.blf_map_size
))
631 xfs_buf_item_relse(bp
);
633 atomic_dec(&bip
->bli_refcount
);
640 * This is called to find out where the oldest active copy of the
641 * buf log item in the on disk log resides now that the last log
642 * write of it completed at the given lsn.
643 * We always re-log all the dirty data in a buffer, so usually the
644 * latest copy in the on disk log is the only one that matters. For
645 * those cases we simply return the given lsn.
647 * The one exception to this is for buffers full of newly allocated
648 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
649 * flag set, indicating that only the di_next_unlinked fields from the
650 * inodes in the buffers will be replayed during recovery. If the
651 * original newly allocated inode images have not yet been flushed
652 * when the buffer is so relogged, then we need to make sure that we
653 * keep the old images in the 'active' portion of the log. We do this
654 * by returning the original lsn of that transaction here rather than
658 xfs_buf_item_committed(
659 struct xfs_log_item
*lip
,
662 struct xfs_buf_log_item
*bip
= BUF_ITEM(lip
);
664 trace_xfs_buf_item_committed(bip
);
666 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) && lip
->li_lsn
!= 0)
672 xfs_buf_item_committing(
673 struct xfs_log_item
*lip
,
674 xfs_lsn_t commit_lsn
)
679 * This is the ops vector shared by all buf log items.
681 static const struct xfs_item_ops xfs_buf_item_ops
= {
682 .iop_size
= xfs_buf_item_size
,
683 .iop_format
= xfs_buf_item_format
,
684 .iop_pin
= xfs_buf_item_pin
,
685 .iop_unpin
= xfs_buf_item_unpin
,
686 .iop_unlock
= xfs_buf_item_unlock
,
687 .iop_committed
= xfs_buf_item_committed
,
688 .iop_push
= xfs_buf_item_push
,
689 .iop_committing
= xfs_buf_item_committing
693 xfs_buf_item_get_format(
694 struct xfs_buf_log_item
*bip
,
697 ASSERT(bip
->bli_formats
== NULL
);
698 bip
->bli_format_count
= count
;
701 bip
->bli_formats
= &bip
->bli_format
;
705 bip
->bli_formats
= kmem_zalloc(count
* sizeof(struct xfs_buf_log_format
),
707 if (!bip
->bli_formats
)
713 xfs_buf_item_free_format(
714 struct xfs_buf_log_item
*bip
)
716 if (bip
->bli_formats
!= &bip
->bli_format
) {
717 kmem_free(bip
->bli_formats
);
718 bip
->bli_formats
= NULL
;
723 * Allocate a new buf log item to go with the given buffer.
724 * Set the buffer's b_fsprivate field to point to the new
725 * buf log item. If there are other item's attached to the
726 * buffer (see xfs_buf_attach_iodone() below), then put the
727 * buf log item at the front.
734 xfs_log_item_t
*lip
= bp
->b_fspriv
;
735 xfs_buf_log_item_t
*bip
;
742 * Check to see if there is already a buf log item for
743 * this buffer. If there is, it is guaranteed to be
744 * the first. If we do already have one, there is
745 * nothing to do here so return.
747 ASSERT(bp
->b_target
->bt_mount
== mp
);
748 if (lip
!= NULL
&& lip
->li_type
== XFS_LI_BUF
)
751 bip
= kmem_zone_zalloc(xfs_buf_item_zone
, KM_SLEEP
);
752 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
757 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
758 * can be divided into. Make sure not to truncate any pieces.
759 * map_size is the size of the bitmap needed to describe the
760 * chunks of the buffer.
762 * Discontiguous buffer support follows the layout of the underlying
763 * buffer. This makes the implementation as simple as possible.
765 error
= xfs_buf_item_get_format(bip
, bp
->b_map_count
);
768 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
769 chunks
= DIV_ROUND_UP(BBTOB(bp
->b_maps
[i
].bm_len
),
771 map_size
= DIV_ROUND_UP(chunks
, NBWORD
);
773 bip
->bli_formats
[i
].blf_type
= XFS_LI_BUF
;
774 bip
->bli_formats
[i
].blf_blkno
= bp
->b_maps
[i
].bm_bn
;
775 bip
->bli_formats
[i
].blf_len
= bp
->b_maps
[i
].bm_len
;
776 bip
->bli_formats
[i
].blf_map_size
= map_size
;
779 #ifdef XFS_TRANS_DEBUG
781 * Allocate the arrays for tracking what needs to be logged
782 * and what our callers request to be logged. bli_orig
783 * holds a copy of the original, clean buffer for comparison
784 * against, and bli_logged keeps a 1 bit flag per byte in
785 * the buffer to indicate which bytes the callers have asked
788 bip
->bli_orig
= kmem_alloc(BBTOB(bp
->b_length
), KM_SLEEP
);
789 memcpy(bip
->bli_orig
, bp
->b_addr
, BBTOB(bp
->b_length
));
790 bip
->bli_logged
= kmem_zalloc(BBTOB(bp
->b_length
) / NBBY
, KM_SLEEP
);
794 * Put the buf item into the list of items attached to the
795 * buffer at the front.
798 bip
->bli_item
.li_bio_list
= bp
->b_fspriv
;
804 * Mark bytes first through last inclusive as dirty in the buf
808 xfs_buf_item_log_segment(
809 struct xfs_buf_log_item
*bip
,
825 * Convert byte offsets to bit numbers.
827 first_bit
= first
>> XFS_BLF_SHIFT
;
828 last_bit
= last
>> XFS_BLF_SHIFT
;
831 * Calculate the total number of bits to be set.
833 bits_to_set
= last_bit
- first_bit
+ 1;
836 * Get a pointer to the first word in the bitmap
839 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
840 wordp
= &map
[word_num
];
843 * Calculate the starting bit in the first word.
845 bit
= first_bit
& (uint
)(NBWORD
- 1);
848 * First set any bits in the first word of our range.
849 * If it starts at bit 0 of the word, it will be
850 * set below rather than here. That is what the variable
851 * bit tells us. The variable bits_set tracks the number
852 * of bits that have been set so far. End_bit is the number
853 * of the last bit to be set in this word plus one.
856 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
857 mask
= ((1 << (end_bit
- bit
)) - 1) << bit
;
860 bits_set
= end_bit
- bit
;
866 * Now set bits a whole word at a time that are between
867 * first_bit and last_bit.
869 while ((bits_to_set
- bits_set
) >= NBWORD
) {
870 *wordp
|= 0xffffffff;
876 * Finally, set any bits left to be set in one last partial word.
878 end_bit
= bits_to_set
- bits_set
;
880 mask
= (1 << end_bit
) - 1;
884 xfs_buf_item_log_debug(bip
, first
, last
);
888 * Mark bytes first through last inclusive as dirty in the buf
893 xfs_buf_log_item_t
*bip
,
900 struct xfs_buf
*bp
= bip
->bli_buf
;
903 * Mark the item as having some dirty data for
904 * quick reference in xfs_buf_item_dirty.
906 bip
->bli_flags
|= XFS_BLI_DIRTY
;
909 * walk each buffer segment and mark them dirty appropriately.
912 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
915 end
= start
+ BBTOB(bp
->b_maps
[i
].bm_len
);
917 start
+= BBTOB(bp
->b_maps
[i
].bm_len
);
925 xfs_buf_item_log_segment(bip
, first
, end
,
926 &bip
->bli_formats
[i
].blf_data_map
[0]);
928 start
+= bp
->b_maps
[i
].bm_len
;
934 * Return 1 if the buffer has some data that has been logged (at any
935 * point, not just the current transaction) and 0 if not.
939 xfs_buf_log_item_t
*bip
)
941 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
946 xfs_buf_log_item_t
*bip
)
948 #ifdef XFS_TRANS_DEBUG
949 kmem_free(bip
->bli_orig
);
950 kmem_free(bip
->bli_logged
);
951 #endif /* XFS_TRANS_DEBUG */
953 xfs_buf_item_free_format(bip
);
954 kmem_zone_free(xfs_buf_item_zone
, bip
);
958 * This is called when the buf log item is no longer needed. It should
959 * free the buf log item associated with the given buffer and clear
960 * the buffer's pointer to the buf log item. If there are no more
961 * items in the list, clear the b_iodone field of the buffer (see
962 * xfs_buf_attach_iodone() below).
968 xfs_buf_log_item_t
*bip
;
970 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
973 bp
->b_fspriv
= bip
->bli_item
.li_bio_list
;
974 if (bp
->b_fspriv
== NULL
)
978 xfs_buf_item_free(bip
);
983 * Add the given log item with its callback to the list of callbacks
984 * to be called when the buffer's I/O completes. If it is not set
985 * already, set the buffer's b_iodone() routine to be
986 * xfs_buf_iodone_callbacks() and link the log item into the list of
987 * items rooted at b_fsprivate. Items are always added as the second
988 * entry in the list if there is a first, because the buf item code
989 * assumes that the buf log item is first.
992 xfs_buf_attach_iodone(
994 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
997 xfs_log_item_t
*head_lip
;
999 ASSERT(xfs_buf_islocked(bp
));
1002 head_lip
= bp
->b_fspriv
;
1004 lip
->li_bio_list
= head_lip
->li_bio_list
;
1005 head_lip
->li_bio_list
= lip
;
1010 ASSERT(bp
->b_iodone
== NULL
||
1011 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
1012 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
1016 * We can have many callbacks on a buffer. Running the callbacks individually
1017 * can cause a lot of contention on the AIL lock, so we allow for a single
1018 * callback to be able to scan the remaining lip->li_bio_list for other items
1019 * of the same type and callback to be processed in the first call.
1021 * As a result, the loop walking the callback list below will also modify the
1022 * list. it removes the first item from the list and then runs the callback.
1023 * The loop then restarts from the new head of the list. This allows the
1024 * callback to scan and modify the list attached to the buffer and we don't
1025 * have to care about maintaining a next item pointer.
1028 xfs_buf_do_callbacks(
1031 struct xfs_log_item
*lip
;
1033 while ((lip
= bp
->b_fspriv
) != NULL
) {
1034 bp
->b_fspriv
= lip
->li_bio_list
;
1035 ASSERT(lip
->li_cb
!= NULL
);
1037 * Clear the next pointer so we don't have any
1038 * confusion if the item is added to another buf.
1039 * Don't touch the log item after calling its
1040 * callback, because it could have freed itself.
1042 lip
->li_bio_list
= NULL
;
1043 lip
->li_cb(bp
, lip
);
1048 * This is the iodone() function for buffers which have had callbacks
1049 * attached to them by xfs_buf_attach_iodone(). It should remove each
1050 * log item from the buffer's list and call the callback of each in turn.
1051 * When done, the buffer's fsprivate field is set to NULL and the buffer
1052 * is unlocked with a call to iodone().
1055 xfs_buf_iodone_callbacks(
1058 struct xfs_log_item
*lip
= bp
->b_fspriv
;
1059 struct xfs_mount
*mp
= lip
->li_mountp
;
1060 static ulong lasttime
;
1061 static xfs_buftarg_t
*lasttarg
;
1063 if (likely(!xfs_buf_geterror(bp
)))
1067 * If we've already decided to shutdown the filesystem because of
1068 * I/O errors, there's no point in giving this a retry.
1070 if (XFS_FORCED_SHUTDOWN(mp
)) {
1073 trace_xfs_buf_item_iodone(bp
, _RET_IP_
);
1077 if (bp
->b_target
!= lasttarg
||
1078 time_after(jiffies
, (lasttime
+ 5*HZ
))) {
1080 xfs_buf_ioerror_alert(bp
, __func__
);
1082 lasttarg
= bp
->b_target
;
1085 * If the write was asynchronous then no one will be looking for the
1086 * error. Clear the error state and write the buffer out again.
1088 * XXX: This helps against transient write errors, but we need to find
1089 * a way to shut the filesystem down if the writes keep failing.
1091 * In practice we'll shut the filesystem down soon as non-transient
1092 * erorrs tend to affect the whole device and a failing log write
1093 * will make us give up. But we really ought to do better here.
1095 if (XFS_BUF_ISASYNC(bp
)) {
1096 ASSERT(bp
->b_iodone
!= NULL
);
1098 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1100 xfs_buf_ioerror(bp
, 0); /* errno of 0 unsets the flag */
1102 if (!XFS_BUF_ISSTALE(bp
)) {
1103 bp
->b_flags
|= XBF_WRITE
| XBF_ASYNC
| XBF_DONE
;
1104 xfs_buf_iorequest(bp
);
1113 * If the write of the buffer was synchronous, we want to make
1114 * sure to return the error to the caller of xfs_bwrite().
1119 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1122 xfs_buf_do_callbacks(bp
);
1123 bp
->b_fspriv
= NULL
;
1124 bp
->b_iodone
= NULL
;
1125 xfs_buf_ioend(bp
, 0);
1129 * This is the iodone() function for buffers which have been
1130 * logged. It is called when they are eventually flushed out.
1131 * It should remove the buf item from the AIL, and free the buf item.
1132 * It is called by xfs_buf_iodone_callbacks() above which will take
1133 * care of cleaning up the buffer itself.
1138 struct xfs_log_item
*lip
)
1140 struct xfs_ail
*ailp
= lip
->li_ailp
;
1142 ASSERT(BUF_ITEM(lip
)->bli_buf
== bp
);
1147 * If we are forcibly shutting down, this may well be
1148 * off the AIL already. That's because we simulate the
1149 * log-committed callbacks to unpin these buffers. Or we may never
1150 * have put this item on AIL because of the transaction was
1151 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1153 * Either way, AIL is useless if we're forcing a shutdown.
1155 spin_lock(&ailp
->xa_lock
);
1156 xfs_trans_ail_delete(ailp
, lip
, SHUTDOWN_CORRUPT_INCORE
);
1157 xfs_buf_item_free(BUF_ITEM(lip
));