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_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trans_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
35 kmem_zone_t
*xfs_buf_item_zone
;
37 #ifdef XFS_TRANS_DEBUG
39 * This function uses an alternate strategy for tracking the bytes
40 * that the user requests to be logged. This can then be used
41 * in conjunction with the bli_orig array in the buf log item to
42 * catch bugs in our callers' code.
44 * We also double check the bits set in xfs_buf_item_log using a
45 * simple algorithm to check that every byte is accounted for.
48 xfs_buf_item_log_debug(
49 xfs_buf_log_item_t
*bip
,
62 ASSERT(bip
->bli_logged
!= NULL
);
64 nbytes
= last
- first
+ 1;
65 bfset(bip
->bli_logged
, first
, nbytes
);
66 for (x
= 0; x
< nbytes
; x
++) {
67 chunk_num
= byte
>> XFS_BLI_SHIFT
;
68 word_num
= chunk_num
>> BIT_TO_WORD_SHIFT
;
69 bit_num
= chunk_num
& (NBWORD
- 1);
70 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
71 bit_set
= *wordp
& (1 << bit_num
);
78 * This function is called when we flush something into a buffer without
79 * logging it. This happens for things like inodes which are logged
80 * separately from the buffer.
83 xfs_buf_item_flush_log_debug(
88 xfs_buf_log_item_t
*bip
;
91 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
92 if ((bip
== NULL
) || (bip
->bli_item
.li_type
!= XFS_LI_BUF
)) {
96 ASSERT(bip
->bli_logged
!= NULL
);
97 nbytes
= last
- first
+ 1;
98 bfset(bip
->bli_logged
, first
, nbytes
);
102 * This function is called to verify that our callers have logged
103 * all the bytes that they changed.
105 * It does this by comparing the original copy of the buffer stored in
106 * the buf log item's bli_orig array to the current copy of the buffer
107 * and ensuring that all bytes which mismatch are set in the bli_logged
108 * array of the buf log item.
111 xfs_buf_item_log_check(
112 xfs_buf_log_item_t
*bip
)
119 ASSERT(bip
->bli_orig
!= NULL
);
120 ASSERT(bip
->bli_logged
!= NULL
);
123 ASSERT(XFS_BUF_COUNT(bp
) > 0);
124 ASSERT(XFS_BUF_PTR(bp
) != NULL
);
125 orig
= bip
->bli_orig
;
126 buffer
= XFS_BUF_PTR(bp
);
127 for (x
= 0; x
< XFS_BUF_COUNT(bp
); x
++) {
128 if (orig
[x
] != buffer
[x
] && !btst(bip
->bli_logged
, x
))
130 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
135 #define xfs_buf_item_log_debug(x,y,z)
136 #define xfs_buf_item_log_check(x)
139 STATIC
void xfs_buf_error_relse(xfs_buf_t
*bp
);
140 STATIC
void xfs_buf_do_callbacks(xfs_buf_t
*bp
, xfs_log_item_t
*lip
);
143 * This returns the number of log iovecs needed to log the
144 * given buf log item.
146 * It calculates this as 1 iovec for the buf log format structure
147 * and 1 for each stretch of non-contiguous chunks to be logged.
148 * Contiguous chunks are logged in a single iovec.
150 * If the XFS_BLI_STALE flag has been set, then log nothing.
154 xfs_buf_log_item_t
*bip
)
161 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
162 if (bip
->bli_flags
& XFS_BLI_STALE
) {
164 * The buffer is stale, so all we need to log
165 * is the buf log format structure with the
168 trace_xfs_buf_item_size_stale(bip
);
169 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
);
174 ASSERT(bip
->bli_flags
& XFS_BLI_LOGGED
);
176 last_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
177 bip
->bli_format
.blf_map_size
, 0);
178 ASSERT(last_bit
!= -1);
180 while (last_bit
!= -1) {
182 * This takes the bit number to start looking from and
183 * returns the next set bit from there. It returns -1
184 * if there are no more bits set or the start bit is
185 * beyond the end of the bitmap.
187 next_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
188 bip
->bli_format
.blf_map_size
,
191 * If we run out of bits, leave the loop,
192 * else if we find a new set of bits bump the number of vecs,
193 * else keep scanning the current set of bits.
195 if (next_bit
== -1) {
197 } else if (next_bit
!= last_bit
+ 1) {
200 } else if (xfs_buf_offset(bp
, next_bit
* XFS_BLI_CHUNK
) !=
201 (xfs_buf_offset(bp
, last_bit
* XFS_BLI_CHUNK
) +
210 trace_xfs_buf_item_size(bip
);
215 * This is called to fill in the vector of log iovecs for the
216 * given log buf item. It fills the first entry with a buf log
217 * format structure, and the rest point to contiguous chunks
222 xfs_buf_log_item_t
*bip
,
223 xfs_log_iovec_t
*log_vector
)
227 xfs_log_iovec_t
*vecp
;
235 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
236 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
237 (bip
->bli_flags
& XFS_BLI_STALE
));
242 * The size of the base structure is the size of the
243 * declared structure plus the space for the extra words
244 * of the bitmap. We subtract one from the map size, because
245 * the first element of the bitmap is accounted for in the
246 * size of the base structure.
249 (uint
)(sizeof(xfs_buf_log_format_t
) +
250 ((bip
->bli_format
.blf_map_size
- 1) * sizeof(uint
)));
251 vecp
->i_addr
= (xfs_caddr_t
)&bip
->bli_format
;
252 vecp
->i_len
= base_size
;
253 vecp
->i_type
= XLOG_REG_TYPE_BFORMAT
;
257 if (bip
->bli_flags
& XFS_BLI_STALE
) {
259 * The buffer is stale, so all we need to log
260 * is the buf log format structure with the
263 trace_xfs_buf_item_format_stale(bip
);
264 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
);
265 bip
->bli_format
.blf_size
= nvecs
;
270 * Fill in an iovec for each set of contiguous chunks.
272 first_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
273 bip
->bli_format
.blf_map_size
, 0);
274 ASSERT(first_bit
!= -1);
275 last_bit
= first_bit
;
279 * This takes the bit number to start looking from and
280 * returns the next set bit from there. It returns -1
281 * if there are no more bits set or the start bit is
282 * beyond the end of the bitmap.
284 next_bit
= xfs_next_bit(bip
->bli_format
.blf_data_map
,
285 bip
->bli_format
.blf_map_size
,
288 * If we run out of bits fill in the last iovec and get
290 * Else if we start a new set of bits then fill in the
291 * iovec for the series we were looking at and start
292 * counting the bits in the new one.
293 * Else we're still in the same set of bits so just
294 * keep counting and scanning.
296 if (next_bit
== -1) {
297 buffer_offset
= first_bit
* XFS_BLI_CHUNK
;
298 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
299 vecp
->i_len
= nbits
* XFS_BLI_CHUNK
;
300 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
303 } else if (next_bit
!= last_bit
+ 1) {
304 buffer_offset
= first_bit
* XFS_BLI_CHUNK
;
305 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
306 vecp
->i_len
= nbits
* XFS_BLI_CHUNK
;
307 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
310 first_bit
= next_bit
;
313 } else if (xfs_buf_offset(bp
, next_bit
<< XFS_BLI_SHIFT
) !=
314 (xfs_buf_offset(bp
, last_bit
<< XFS_BLI_SHIFT
) +
316 buffer_offset
= first_bit
* XFS_BLI_CHUNK
;
317 vecp
->i_addr
= xfs_buf_offset(bp
, buffer_offset
);
318 vecp
->i_len
= nbits
* XFS_BLI_CHUNK
;
319 vecp
->i_type
= XLOG_REG_TYPE_BCHUNK
;
320 /* You would think we need to bump the nvecs here too, but we do not
321 * this number is used by recovery, and it gets confused by the boundary
326 first_bit
= next_bit
;
334 bip
->bli_format
.blf_size
= nvecs
;
337 * Check to make sure everything is consistent.
339 trace_xfs_buf_item_format(bip
);
340 xfs_buf_item_log_check(bip
);
344 * This is called to pin the buffer associated with the buf log item in memory
345 * so it cannot be written out. Simply call bpin() on the buffer to do this.
347 * We also always take a reference to the buffer log item here so that the bli
348 * is held while the item is pinned in memory. This means that we can
349 * unconditionally drop the reference count a transaction holds when the
350 * transaction is completed.
355 xfs_buf_log_item_t
*bip
)
360 ASSERT(XFS_BUF_ISBUSY(bp
));
361 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
362 ASSERT((bip
->bli_flags
& XFS_BLI_LOGGED
) ||
363 (bip
->bli_flags
& XFS_BLI_STALE
));
364 atomic_inc(&bip
->bli_refcount
);
365 trace_xfs_buf_item_pin(bip
);
371 * This is called to unpin the buffer associated with the buf log
372 * item which was previously pinned with a call to xfs_buf_item_pin().
373 * Just call bunpin() on the buffer to do this.
375 * Also drop the reference to the buf item for the current transaction.
376 * If the XFS_BLI_STALE flag is set and we are the last reference,
377 * then free up the buf log item and unlock the buffer.
381 xfs_buf_log_item_t
*bip
)
383 struct xfs_ail
*ailp
;
386 int stale
= bip
->bli_flags
& XFS_BLI_STALE
;
390 ASSERT(XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*) == bip
);
391 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
392 trace_xfs_buf_item_unpin(bip
);
394 freed
= atomic_dec_and_test(&bip
->bli_refcount
);
395 ailp
= bip
->bli_item
.li_ailp
;
397 if (freed
&& stale
) {
398 ASSERT(bip
->bli_flags
& XFS_BLI_STALE
);
399 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
400 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
401 ASSERT(XFS_BUF_ISSTALE(bp
));
402 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
);
403 trace_xfs_buf_item_unpin_stale(bip
);
406 * If we get called here because of an IO error, we may
407 * or may not have the item on the AIL. xfs_trans_ail_delete()
408 * will take care of that situation.
409 * xfs_trans_ail_delete() drops the AIL lock.
411 if (bip
->bli_flags
& XFS_BLI_STALE_INODE
) {
412 xfs_buf_do_callbacks(bp
, (xfs_log_item_t
*)bip
);
413 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
414 XFS_BUF_CLR_IODONE_FUNC(bp
);
416 spin_lock(&ailp
->xa_lock
);
417 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)bip
);
418 xfs_buf_item_relse(bp
);
419 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) == NULL
);
426 * this is called from uncommit in the forced-shutdown path.
427 * we need to check to see if the reference count on the log item
428 * is going to drop to zero. If so, unpin will free the log item
429 * so we need to free the item's descriptor (that points to the item)
430 * in the transaction.
433 xfs_buf_item_unpin_remove(
434 xfs_buf_log_item_t
*bip
,
437 /* will xfs_buf_item_unpin() call xfs_buf_item_relse()? */
438 if ((atomic_read(&bip
->bli_refcount
) == 1) &&
439 (bip
->bli_flags
& XFS_BLI_STALE
)) {
441 * yes -- We can safely do some work here and then call
442 * buf_item_unpin to do the rest because we are
443 * are holding the buffer locked so no one else will be
444 * able to bump up the refcount. We have to remove the
445 * log item from the transaction as we are about to release
446 * our reference to the buffer. If we don't, the unlock that
447 * occurs later in the xfs_trans_uncommit() will try to
448 * reference the buffer which we no longer have a hold on.
450 struct xfs_log_item_desc
*lidp
;
452 ASSERT(XFS_BUF_VALUSEMA(bip
->bli_buf
) <= 0);
453 trace_xfs_buf_item_unpin_stale(bip
);
455 lidp
= xfs_trans_find_item(tp
, (xfs_log_item_t
*)bip
);
456 xfs_trans_free_item(tp
, lidp
);
459 * Since the transaction no longer refers to the buffer, the
460 * buffer should no longer refer to the transaction.
462 XFS_BUF_SET_FSPRIVATE2(bip
->bli_buf
, NULL
);
464 xfs_buf_item_unpin(bip
);
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 xfs_buf_log_item_t
*bip
)
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 xfs_buf_log_item_t
*bip
)
526 /* Clear the buffer's association with this transaction. */
527 XFS_BUF_SET_FSPRIVATE2(bp
, NULL
);
530 * If this is a transaction abort, don't return early. Instead, allow
531 * the brelse to happen. Normally it would be done for stale
532 * (cancelled) buffers at unpin time, but we'll never go through the
533 * pin/unpin cycle if we abort inside commit.
535 aborted
= (bip
->bli_item
.li_flags
& XFS_LI_ABORTED
) != 0;
538 * Before possibly freeing the buf item, determine if we should
539 * release the buffer at the end of this routine.
541 hold
= bip
->bli_flags
& XFS_BLI_HOLD
;
543 /* Clear the per transaction state. */
544 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_HOLD
);
547 * If the buf item is marked stale, then don't do anything. We'll
548 * unlock the buffer and free the buf item when the buffer is unpinned
551 if (bip
->bli_flags
& XFS_BLI_STALE
) {
552 trace_xfs_buf_item_unlock_stale(bip
);
553 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
);
555 atomic_dec(&bip
->bli_refcount
);
560 trace_xfs_buf_item_unlock(bip
);
563 * If the buf item isn't tracking any data, free it, otherwise drop the
564 * reference we hold to it.
566 if (xfs_bitmap_empty(bip
->bli_format
.blf_data_map
,
567 bip
->bli_format
.blf_map_size
))
568 xfs_buf_item_relse(bp
);
570 atomic_dec(&bip
->bli_refcount
);
577 * This is called to find out where the oldest active copy of the
578 * buf log item in the on disk log resides now that the last log
579 * write of it completed at the given lsn.
580 * We always re-log all the dirty data in a buffer, so usually the
581 * latest copy in the on disk log is the only one that matters. For
582 * those cases we simply return the given lsn.
584 * The one exception to this is for buffers full of newly allocated
585 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
586 * flag set, indicating that only the di_next_unlinked fields from the
587 * inodes in the buffers will be replayed during recovery. If the
588 * original newly allocated inode images have not yet been flushed
589 * when the buffer is so relogged, then we need to make sure that we
590 * keep the old images in the 'active' portion of the log. We do this
591 * by returning the original lsn of that transaction here rather than
595 xfs_buf_item_committed(
596 xfs_buf_log_item_t
*bip
,
599 trace_xfs_buf_item_committed(bip
);
601 if ((bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
) &&
602 (bip
->bli_item
.li_lsn
!= 0)) {
603 return bip
->bli_item
.li_lsn
;
609 * The buffer is locked, but is not a delayed write buffer. This happens
610 * if we race with IO completion and hence we don't want to try to write it
611 * again. Just release the buffer.
615 xfs_buf_log_item_t
*bip
)
619 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
620 trace_xfs_buf_item_push(bip
);
623 ASSERT(!XFS_BUF_ISDELAYWRITE(bp
));
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 xfs_buf_log_item_t
*bip
)
639 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
640 trace_xfs_buf_item_pushbuf(bip
);
643 ASSERT(XFS_BUF_ISDELAYWRITE(bp
));
644 xfs_buf_delwri_promote(bp
);
650 xfs_buf_item_committing(xfs_buf_log_item_t
*bip
, xfs_lsn_t commit_lsn
)
655 * This is the ops vector shared by all buf log items.
657 static struct xfs_item_ops xfs_buf_item_ops
= {
658 .iop_size
= (uint(*)(xfs_log_item_t
*))xfs_buf_item_size
,
659 .iop_format
= (void(*)(xfs_log_item_t
*, xfs_log_iovec_t
*))
661 .iop_pin
= (void(*)(xfs_log_item_t
*))xfs_buf_item_pin
,
662 .iop_unpin
= (void(*)(xfs_log_item_t
*))xfs_buf_item_unpin
,
663 .iop_unpin_remove
= (void(*)(xfs_log_item_t
*, xfs_trans_t
*))
664 xfs_buf_item_unpin_remove
,
665 .iop_trylock
= (uint(*)(xfs_log_item_t
*))xfs_buf_item_trylock
,
666 .iop_unlock
= (void(*)(xfs_log_item_t
*))xfs_buf_item_unlock
,
667 .iop_committed
= (xfs_lsn_t(*)(xfs_log_item_t
*, xfs_lsn_t
))
668 xfs_buf_item_committed
,
669 .iop_push
= (void(*)(xfs_log_item_t
*))xfs_buf_item_push
,
670 .iop_pushbuf
= (void(*)(xfs_log_item_t
*))xfs_buf_item_pushbuf
,
671 .iop_committing
= (void(*)(xfs_log_item_t
*, xfs_lsn_t
))
672 xfs_buf_item_committing
677 * Allocate a new buf log item to go with the given buffer.
678 * Set the buffer's b_fsprivate field to point to the new
679 * buf log item. If there are other item's attached to the
680 * buffer (see xfs_buf_attach_iodone() below), then put the
681 * buf log item at the front.
689 xfs_buf_log_item_t
*bip
;
694 * Check to see if there is already a buf log item for
695 * this buffer. If there is, it is guaranteed to be
696 * the first. If we do already have one, there is
697 * nothing to do here so return.
699 if (bp
->b_mount
!= mp
)
701 XFS_BUF_SET_BDSTRAT_FUNC(bp
, xfs_bdstrat_cb
);
702 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
703 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
704 if (lip
->li_type
== XFS_LI_BUF
) {
710 * chunks is the number of XFS_BLI_CHUNK size pieces
711 * the buffer can be divided into. Make sure not to
712 * truncate any pieces. map_size is the size of the
713 * bitmap needed to describe the chunks of the buffer.
715 chunks
= (int)((XFS_BUF_COUNT(bp
) + (XFS_BLI_CHUNK
- 1)) >> XFS_BLI_SHIFT
);
716 map_size
= (int)((chunks
+ NBWORD
) >> BIT_TO_WORD_SHIFT
);
718 bip
= (xfs_buf_log_item_t
*)kmem_zone_zalloc(xfs_buf_item_zone
,
720 xfs_log_item_init(mp
, &bip
->bli_item
, XFS_LI_BUF
, &xfs_buf_item_ops
);
723 bip
->bli_format
.blf_type
= XFS_LI_BUF
;
724 bip
->bli_format
.blf_blkno
= (__int64_t
)XFS_BUF_ADDR(bp
);
725 bip
->bli_format
.blf_len
= (ushort
)BTOBB(XFS_BUF_COUNT(bp
));
726 bip
->bli_format
.blf_map_size
= map_size
;
728 #ifdef XFS_TRANS_DEBUG
730 * Allocate the arrays for tracking what needs to be logged
731 * and what our callers request to be logged. bli_orig
732 * holds a copy of the original, clean buffer for comparison
733 * against, and bli_logged keeps a 1 bit flag per byte in
734 * the buffer to indicate which bytes the callers have asked
737 bip
->bli_orig
= (char *)kmem_alloc(XFS_BUF_COUNT(bp
), KM_SLEEP
);
738 memcpy(bip
->bli_orig
, XFS_BUF_PTR(bp
), XFS_BUF_COUNT(bp
));
739 bip
->bli_logged
= (char *)kmem_zalloc(XFS_BUF_COUNT(bp
) / NBBY
, KM_SLEEP
);
743 * Put the buf item into the list of items attached to the
744 * buffer at the front.
746 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
747 bip
->bli_item
.li_bio_list
=
748 XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
750 XFS_BUF_SET_FSPRIVATE(bp
, bip
);
755 * Mark bytes first through last inclusive as dirty in the buf
760 xfs_buf_log_item_t
*bip
,
775 * Mark the item as having some dirty data for
776 * quick reference in xfs_buf_item_dirty.
778 bip
->bli_flags
|= XFS_BLI_DIRTY
;
781 * Convert byte offsets to bit numbers.
783 first_bit
= first
>> XFS_BLI_SHIFT
;
784 last_bit
= last
>> XFS_BLI_SHIFT
;
787 * Calculate the total number of bits to be set.
789 bits_to_set
= last_bit
- first_bit
+ 1;
792 * Get a pointer to the first word in the bitmap
795 word_num
= first_bit
>> BIT_TO_WORD_SHIFT
;
796 wordp
= &(bip
->bli_format
.blf_data_map
[word_num
]);
799 * Calculate the starting bit in the first word.
801 bit
= first_bit
& (uint
)(NBWORD
- 1);
804 * First set any bits in the first word of our range.
805 * If it starts at bit 0 of the word, it will be
806 * set below rather than here. That is what the variable
807 * bit tells us. The variable bits_set tracks the number
808 * of bits that have been set so far. End_bit is the number
809 * of the last bit to be set in this word plus one.
812 end_bit
= MIN(bit
+ bits_to_set
, (uint
)NBWORD
);
813 mask
= ((1 << (end_bit
- bit
)) - 1) << bit
;
816 bits_set
= end_bit
- bit
;
822 * Now set bits a whole word at a time that are between
823 * first_bit and last_bit.
825 while ((bits_to_set
- bits_set
) >= NBWORD
) {
826 *wordp
|= 0xffffffff;
832 * Finally, set any bits left to be set in one last partial word.
834 end_bit
= bits_to_set
- bits_set
;
836 mask
= (1 << end_bit
) - 1;
840 xfs_buf_item_log_debug(bip
, first
, last
);
845 * Return 1 if the buffer has some data that has been logged (at any
846 * point, not just the current transaction) and 0 if not.
850 xfs_buf_log_item_t
*bip
)
852 return (bip
->bli_flags
& XFS_BLI_DIRTY
);
857 xfs_buf_log_item_t
*bip
)
859 #ifdef XFS_TRANS_DEBUG
860 kmem_free(bip
->bli_orig
);
861 kmem_free(bip
->bli_logged
);
862 #endif /* XFS_TRANS_DEBUG */
864 kmem_zone_free(xfs_buf_item_zone
, bip
);
868 * This is called when the buf log item is no longer needed. It should
869 * free the buf log item associated with the given buffer and clear
870 * the buffer's pointer to the buf log item. If there are no more
871 * items in the list, clear the b_iodone field of the buffer (see
872 * xfs_buf_attach_iodone() below).
878 xfs_buf_log_item_t
*bip
;
880 trace_xfs_buf_item_relse(bp
, _RET_IP_
);
882 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
883 XFS_BUF_SET_FSPRIVATE(bp
, bip
->bli_item
.li_bio_list
);
884 if ((XFS_BUF_FSPRIVATE(bp
, void *) == NULL
) &&
885 (XFS_BUF_IODONE_FUNC(bp
) != NULL
)) {
886 XFS_BUF_CLR_IODONE_FUNC(bp
);
889 xfs_buf_item_free(bip
);
894 * Add the given log item with its callback to the list of callbacks
895 * to be called when the buffer's I/O completes. If it is not set
896 * already, set the buffer's b_iodone() routine to be
897 * xfs_buf_iodone_callbacks() and link the log item into the list of
898 * items rooted at b_fsprivate. Items are always added as the second
899 * entry in the list if there is a first, because the buf item code
900 * assumes that the buf log item is first.
903 xfs_buf_attach_iodone(
905 void (*cb
)(xfs_buf_t
*, xfs_log_item_t
*),
908 xfs_log_item_t
*head_lip
;
910 ASSERT(XFS_BUF_ISBUSY(bp
));
911 ASSERT(XFS_BUF_VALUSEMA(bp
) <= 0);
914 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
915 head_lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
916 lip
->li_bio_list
= head_lip
->li_bio_list
;
917 head_lip
->li_bio_list
= lip
;
919 XFS_BUF_SET_FSPRIVATE(bp
, lip
);
922 ASSERT((XFS_BUF_IODONE_FUNC(bp
) == xfs_buf_iodone_callbacks
) ||
923 (XFS_BUF_IODONE_FUNC(bp
) == NULL
));
924 XFS_BUF_SET_IODONE_FUNC(bp
, xfs_buf_iodone_callbacks
);
928 xfs_buf_do_callbacks(
932 xfs_log_item_t
*nlip
;
934 while (lip
!= NULL
) {
935 nlip
= lip
->li_bio_list
;
936 ASSERT(lip
->li_cb
!= NULL
);
938 * Clear the next pointer so we don't have any
939 * confusion if the item is added to another buf.
940 * Don't touch the log item after calling its
941 * callback, because it could have freed itself.
943 lip
->li_bio_list
= NULL
;
950 * This is the iodone() function for buffers which have had callbacks
951 * attached to them by xfs_buf_attach_iodone(). It should remove each
952 * log item from the buffer's list and call the callback of each in turn.
953 * When done, the buffer's fsprivate field is set to NULL and the buffer
954 * is unlocked with a call to iodone().
957 xfs_buf_iodone_callbacks(
961 static ulong lasttime
;
962 static xfs_buftarg_t
*lasttarg
;
965 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
966 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
968 if (XFS_BUF_GETERROR(bp
) != 0) {
970 * If we've already decided to shutdown the filesystem
971 * because of IO errors, there's no point in giving this
975 if (XFS_FORCED_SHUTDOWN(mp
)) {
976 ASSERT(XFS_BUF_TARGET(bp
) == mp
->m_ddev_targp
);
977 XFS_BUF_SUPER_STALE(bp
);
978 trace_xfs_buf_item_iodone(bp
, _RET_IP_
);
979 xfs_buf_do_callbacks(bp
, lip
);
980 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
981 XFS_BUF_CLR_IODONE_FUNC(bp
);
986 if ((XFS_BUF_TARGET(bp
) != lasttarg
) ||
987 (time_after(jiffies
, (lasttime
+ 5*HZ
)))) {
989 cmn_err(CE_ALERT
, "Device %s, XFS metadata write error"
990 " block 0x%llx in %s",
991 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp
)),
992 (__uint64_t
)XFS_BUF_ADDR(bp
), mp
->m_fsname
);
994 lasttarg
= XFS_BUF_TARGET(bp
);
996 if (XFS_BUF_ISASYNC(bp
)) {
998 * If the write was asynchronous then noone will be
999 * looking for the error. Clear the error state
1000 * and write the buffer out again delayed write.
1002 * XXXsup This is OK, so long as we catch these
1003 * before we start the umount; we don't want these
1004 * DELWRI metadata bufs to be hanging around.
1006 XFS_BUF_ERROR(bp
,0); /* errno of 0 unsets the flag */
1008 if (!(XFS_BUF_ISSTALE(bp
))) {
1009 XFS_BUF_DELAYWRITE(bp
);
1011 XFS_BUF_SET_START(bp
);
1013 ASSERT(XFS_BUF_IODONE_FUNC(bp
));
1014 trace_xfs_buf_item_iodone_async(bp
, _RET_IP_
);
1018 * If the write of the buffer was not asynchronous,
1019 * then we want to make sure to return the error
1020 * to the caller of bwrite(). Because of this we
1021 * cannot clear the B_ERROR state at this point.
1022 * Instead we install a callback function that
1023 * will be called when the buffer is released, and
1024 * that routine will clear the error state and
1025 * set the buffer to be written out again after
1028 /* We actually overwrite the existing b-relse
1029 function at times, but we're gonna be shutting down
1031 XFS_BUF_SET_BRELSE_FUNC(bp
,xfs_buf_error_relse
);
1033 XFS_BUF_FINISH_IOWAIT(bp
);
1038 xfs_buf_do_callbacks(bp
, lip
);
1039 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
1040 XFS_BUF_CLR_IODONE_FUNC(bp
);
1045 * This is a callback routine attached to a buffer which gets an error
1046 * when being written out synchronously.
1049 xfs_buf_error_relse(
1052 xfs_log_item_t
*lip
;
1055 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
1056 mp
= (xfs_mount_t
*)lip
->li_mountp
;
1057 ASSERT(XFS_BUF_TARGET(bp
) == mp
->m_ddev_targp
);
1061 XFS_BUF_UNDELAYWRITE(bp
);
1062 XFS_BUF_ERROR(bp
,0);
1064 trace_xfs_buf_error_relse(bp
, _RET_IP_
);
1066 if (! XFS_FORCED_SHUTDOWN(mp
))
1067 xfs_force_shutdown(mp
, SHUTDOWN_META_IO_ERROR
);
1069 * We have to unpin the pinned buffers so do the
1072 xfs_buf_do_callbacks(bp
, lip
);
1073 XFS_BUF_SET_FSPRIVATE(bp
, NULL
);
1074 XFS_BUF_CLR_IODONE_FUNC(bp
);
1075 XFS_BUF_SET_BRELSE_FUNC(bp
,NULL
);
1081 * This is the iodone() function for buffers which have been
1082 * logged. It is called when they are eventually flushed out.
1083 * It should remove the buf item from the AIL, and free the buf item.
1084 * It is called by xfs_buf_iodone_callbacks() above which will take
1085 * care of cleaning up the buffer itself.
1091 xfs_buf_log_item_t
*bip
)
1093 struct xfs_ail
*ailp
= bip
->bli_item
.li_ailp
;
1095 ASSERT(bip
->bli_buf
== bp
);
1100 * If we are forcibly shutting down, this may well be
1101 * off the AIL already. That's because we simulate the
1102 * log-committed callbacks to unpin these buffers. Or we may never
1103 * have put this item on AIL because of the transaction was
1104 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1106 * Either way, AIL is useless if we're forcing a shutdown.
1108 spin_lock(&ailp
->xa_lock
);
1109 xfs_trans_ail_delete(ailp
, (xfs_log_item_t
*)bip
);
1110 xfs_buf_item_free(bip
);