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 #ifdef XFS_TRANS_DEBUG
38 * This function uses an alternate strategy for tracking the bytes
39 * that the user requests to be logged. This can then be used
40 * in conjunction with the bli_orig array in the buf log item to
41 * catch bugs in our callers' code.
43 * We also double check the bits set in xfs_buf_item_log using a
44 * simple algorithm to check that every byte is accounted for.
47 xfs_buf_item_log_debug(
48 xfs_buf_log_item_t *bip,
61 ASSERT(bip->bli_logged != NULL);
63 nbytes = last - first + 1;
64 bfset(bip->bli_logged, first, nbytes);
65 for (x = 0; x < nbytes; x++) {
66 chunk_num = byte >> XFS_BLF_SHIFT;
67 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
68 bit_num = chunk_num & (NBWORD - 1);
69 wordp = &(bip->bli_format.blf_data_map[word_num]);
70 bit_set = *wordp & (1 << bit_num);
77 * This function is called when we flush something into a buffer without
78 * logging it. This happens for things like inodes which are logged
79 * separately from the buffer.
82 xfs_buf_item_flush_log_debug(
87 xfs_buf_log_item_t *bip;
90 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
91 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
95 ASSERT(bip->bli_logged != NULL);
96 nbytes = last - first + 1;
97 bfset(bip->bli_logged, first, nbytes);
101 * This function is called to verify that our callers have logged
102 * all the bytes that they changed.
104 * It does this by comparing the original copy of the buffer stored in
105 * the buf log item's bli_orig array to the current copy of the buffer
106 * and ensuring that all bytes which mismatch are set in the bli_logged
107 * array of the buf log item.
110 xfs_buf_item_log_check(
111 xfs_buf_log_item_t *bip)
118 ASSERT(bip->bli_orig != NULL);
119 ASSERT(bip->bli_logged != NULL);
122 ASSERT(XFS_BUF_COUNT(bp) > 0);
123 ASSERT(XFS_BUF_PTR(bp) != NULL);
124 orig = bip->bli_orig;
125 buffer = XFS_BUF_PTR(bp);
126 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
127 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
129 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
134 #define xfs_buf_item_log_debug(x,y,z)
135 #define xfs_buf_item_log_check(x)
138 STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
139 STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
142 * This returns the number of log iovecs needed to log the
143 * given buf log item.
145 * It calculates this as 1 iovec for the buf log format structure
146 * and 1 for each stretch of non-contiguous chunks to be logged.
147 * Contiguous chunks are logged in a single iovec.
149 * If the XFS_BLI_STALE flag has been set, then log nothing.
153 xfs_buf_log_item_t *bip)
160 ASSERT(atomic_read(&bip->bli_refcount) > 0);
161 if (bip->bli_flags & XFS_BLI_STALE) {
163 * The buffer is stale, so all we need to log
164 * is the buf log format structure with the
167 trace_xfs_buf_item_size_stale(bip);
168 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
173 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
175 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
176 bip->bli_format.blf_map_size, 0);
177 ASSERT(last_bit != -1);
179 while (last_bit != -1) {
181 * This takes the bit number to start looking from and
182 * returns the next set bit from there. It returns -1
183 * if there are no more bits set or the start bit is
184 * beyond the end of the bitmap.
186 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
187 bip->bli_format.blf_map_size,
190 * If we run out of bits, leave the loop,
191 * else if we find a new set of bits bump the number of vecs,
192 * else keep scanning the current set of bits.
194 if (next_bit == -1) {
196 } else if (next_bit != last_bit + 1) {
199 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
200 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
209 trace_xfs_buf_item_size(bip);
214 * This is called to fill in the vector of log iovecs for the
215 * given log buf item. It fills the first entry with a buf log
216 * format structure, and the rest point to contiguous chunks
221 xfs_buf_log_item_t *bip,
222 xfs_log_iovec_t *log_vector)
226 xfs_log_iovec_t *vecp;
234 ASSERT(atomic_read(&bip->bli_refcount) > 0);
235 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
236 (bip->bli_flags & XFS_BLI_STALE));
241 * The size of the base structure is the size of the
242 * declared structure plus the space for the extra words
243 * of the bitmap. We subtract one from the map size, because
244 * the first element of the bitmap is accounted for in the
245 * size of the base structure.
248 (uint)(sizeof(xfs_buf_log_format_t) +
249 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
250 vecp->i_addr = (xfs_caddr_t)&bip->bli_format;
251 vecp->i_len = base_size;
252 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
257 * If it is an inode buffer, transfer the in-memory state to the
258 * format flags and clear the in-memory state. We do not transfer
259 * this state if the inode buffer allocation has not yet been committed
260 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
261 * correct replay of the inode allocation.
263 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
264 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
265 xfs_log_item_in_current_chkpt(&bip->bli_item)))
266 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
267 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
270 if (bip->bli_flags & XFS_BLI_STALE) {
272 * The buffer is stale, so all we need to log
273 * is the buf log format structure with the
276 trace_xfs_buf_item_format_stale(bip);
277 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
278 bip->bli_format.blf_size = nvecs;
283 * Fill in an iovec for each set of contiguous chunks.
285 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
286 bip->bli_format.blf_map_size, 0);
287 ASSERT(first_bit != -1);
288 last_bit = first_bit;
292 * This takes the bit number to start looking from and
293 * returns the next set bit from there. It returns -1
294 * if there are no more bits set or the start bit is
295 * beyond the end of the bitmap.
297 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
298 bip->bli_format.blf_map_size,
301 * If we run out of bits fill in the last iovec and get
303 * Else if we start a new set of bits then fill in the
304 * iovec for the series we were looking at and start
305 * counting the bits in the new one.
306 * Else we're still in the same set of bits so just
307 * keep counting and scanning.
309 if (next_bit == -1) {
310 buffer_offset = first_bit * XFS_BLF_CHUNK;
311 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
312 vecp->i_len = nbits * XFS_BLF_CHUNK;
313 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
316 } else if (next_bit != last_bit + 1) {
317 buffer_offset = first_bit * XFS_BLF_CHUNK;
318 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
319 vecp->i_len = nbits * XFS_BLF_CHUNK;
320 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
323 first_bit = next_bit;
326 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
327 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
329 buffer_offset = first_bit * XFS_BLF_CHUNK;
330 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
331 vecp->i_len = nbits * XFS_BLF_CHUNK;
332 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
333 /* You would think we need to bump the nvecs here too, but we do not
334 * this number is used by recovery, and it gets confused by the boundary
339 first_bit = next_bit;
347 bip->bli_format.blf_size = nvecs;
350 * Check to make sure everything is consistent.
352 trace_xfs_buf_item_format(bip);
353 xfs_buf_item_log_check(bip);
357 * This is called to pin the buffer associated with the buf log item in memory
358 * so it cannot be written out. Simply call bpin() on the buffer to do this.
360 * We also always take a reference to the buffer log item here so that the bli
361 * is held while the item is pinned in memory. This means that we can
362 * unconditionally drop the reference count a transaction holds when the
363 * transaction is completed.
368 xfs_buf_log_item_t *bip)
373 ASSERT(XFS_BUF_ISBUSY(bp));
374 ASSERT(atomic_read(&bip->bli_refcount) > 0);
375 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
376 (bip->bli_flags & XFS_BLI_STALE));
377 atomic_inc(&bip->bli_refcount);
378 trace_xfs_buf_item_pin(bip);
384 * This is called to unpin the buffer associated with the buf log
385 * item which was previously pinned with a call to xfs_buf_item_pin().
386 * Just call bunpin() on the buffer to do this.
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.
394 xfs_buf_log_item_t *bip)
396 struct xfs_ail *ailp;
399 int stale = bip->bli_flags & XFS_BLI_STALE;
403 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
404 ASSERT(atomic_read(&bip->bli_refcount) > 0);
405 trace_xfs_buf_item_unpin(bip);
407 freed = atomic_dec_and_test(&bip->bli_refcount);
408 ailp = bip->bli_item.li_ailp;
410 if (freed && stale) {
411 ASSERT(bip->bli_flags & XFS_BLI_STALE);
412 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
413 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
414 ASSERT(XFS_BUF_ISSTALE(bp));
415 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
416 trace_xfs_buf_item_unpin_stale(bip);
419 * If we get called here because of an IO error, we may
420 * or may not have the item on the AIL. xfs_trans_ail_delete()
421 * will take care of that situation.
422 * xfs_trans_ail_delete() drops the AIL lock.
424 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
425 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
426 XFS_BUF_SET_FSPRIVATE(bp, NULL);
427 XFS_BUF_CLR_IODONE_FUNC(bp);
429 spin_lock(&ailp->xa_lock);
430 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
431 xfs_buf_item_relse(bp);
432 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
439 * this is called from uncommit in the forced-shutdown path.
440 * we need to check to see if the reference count on the log item
441 * is going to drop to zero. If so, unpin will free the log item
442 * so we need to free the item's descriptor (that points to the item)
443 * in the transaction.
446 xfs_buf_item_unpin_remove(
447 xfs_buf_log_item_t *bip,
450 /* will xfs_buf_item_unpin() call xfs_buf_item_relse()? */
451 if ((atomic_read(&bip->bli_refcount) == 1) &&
452 (bip->bli_flags & XFS_BLI_STALE)) {
454 * yes -- We can safely do some work here and then call
455 * buf_item_unpin to do the rest because we are
456 * are holding the buffer locked so no one else will be
457 * able to bump up the refcount. We have to remove the
458 * log item from the transaction as we are about to release
459 * our reference to the buffer. If we don't, the unlock that
460 * occurs later in the xfs_trans_uncommit() will try to
461 * reference the buffer which we no longer have a hold on.
463 ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0);
464 trace_xfs_buf_item_unpin_stale(bip);
466 xfs_trans_del_item(&bip->bli_item);
469 * Since the transaction no longer refers to the buffer, the
470 * buffer should no longer refer to the transaction.
472 XFS_BUF_SET_FSPRIVATE2(bip->bli_buf, NULL);
474 xfs_buf_item_unpin(bip);
478 * This is called to attempt to lock the buffer associated with this
479 * buf log item. Don't sleep on the buffer lock. If we can't get
480 * the lock right away, return 0. If we can get the lock, take a
481 * reference to the buffer. If this is a delayed write buffer that
482 * needs AIL help to be written back, invoke the pushbuf routine
483 * rather than the normal success path.
486 xfs_buf_item_trylock(
487 xfs_buf_log_item_t *bip)
492 if (XFS_BUF_ISPINNED(bp))
493 return XFS_ITEM_PINNED;
494 if (!XFS_BUF_CPSEMA(bp))
495 return XFS_ITEM_LOCKED;
497 /* take a reference to the buffer. */
500 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
501 trace_xfs_buf_item_trylock(bip);
502 if (XFS_BUF_ISDELAYWRITE(bp))
503 return XFS_ITEM_PUSHBUF;
504 return XFS_ITEM_SUCCESS;
508 * Release the buffer associated with the buf log item. If there is no dirty
509 * logged data associated with the buffer recorded in the buf log item, then
510 * free the buf log item and remove the reference to it in the buffer.
512 * This call ignores the recursion count. It is only called when the buffer
513 * should REALLY be unlocked, regardless of the recursion count.
515 * We unconditionally drop the transaction's reference to the log item. If the
516 * item was logged, then another reference was taken when it was pinned, so we
517 * can safely drop the transaction reference now. This also allows us to avoid
518 * potential races with the unpin code freeing the bli by not referencing the
519 * bli after we've dropped the reference count.
521 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
522 * if necessary but do not unlock the buffer. This is for support of
523 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
528 xfs_buf_log_item_t *bip)
536 /* Clear the buffer's association with this transaction. */
537 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
540 * If this is a transaction abort, don't return early. Instead, allow
541 * the brelse to happen. Normally it would be done for stale
542 * (cancelled) buffers at unpin time, but we'll never go through the
543 * pin/unpin cycle if we abort inside commit.
545 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
548 * Before possibly freeing the buf item, determine if we should
549 * release the buffer at the end of this routine.
551 hold = bip->bli_flags & XFS_BLI_HOLD;
553 /* Clear the per transaction state. */
554 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
557 * If the buf item is marked stale, then don't do anything. We'll
558 * unlock the buffer and free the buf item when the buffer is unpinned
561 if (bip->bli_flags & XFS_BLI_STALE) {
562 trace_xfs_buf_item_unlock_stale(bip);
563 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
565 atomic_dec(&bip->bli_refcount);
570 trace_xfs_buf_item_unlock(bip);
573 * If the buf item isn't tracking any data, free it, otherwise drop the
574 * reference we hold to it.
576 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
577 bip->bli_format.blf_map_size))
578 xfs_buf_item_relse(bp);
580 atomic_dec(&bip->bli_refcount);
587 * This is called to find out where the oldest active copy of the
588 * buf log item in the on disk log resides now that the last log
589 * write of it completed at the given lsn.
590 * We always re-log all the dirty data in a buffer, so usually the
591 * latest copy in the on disk log is the only one that matters. For
592 * those cases we simply return the given lsn.
594 * The one exception to this is for buffers full of newly allocated
595 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
596 * flag set, indicating that only the di_next_unlinked fields from the
597 * inodes in the buffers will be replayed during recovery. If the
598 * original newly allocated inode images have not yet been flushed
599 * when the buffer is so relogged, then we need to make sure that we
600 * keep the old images in the 'active' portion of the log. We do this
601 * by returning the original lsn of that transaction here rather than
605 xfs_buf_item_committed(
606 xfs_buf_log_item_t *bip,
609 trace_xfs_buf_item_committed(bip);
611 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
612 (bip->bli_item.li_lsn != 0)) {
613 return bip->bli_item.li_lsn;
619 * The buffer is locked, but is not a delayed write buffer. This happens
620 * if we race with IO completion and hence we don't want to try to write it
621 * again. Just release the buffer.
625 xfs_buf_log_item_t *bip)
629 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
630 trace_xfs_buf_item_push(bip);
633 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
638 * The buffer is locked and is a delayed write buffer. Promote the buffer
639 * in the delayed write queue as the caller knows that they must invoke
640 * the xfsbufd to get this buffer written. We have to unlock the buffer
641 * to allow the xfsbufd to write it, too.
644 xfs_buf_item_pushbuf(
645 xfs_buf_log_item_t *bip)
649 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
650 trace_xfs_buf_item_pushbuf(bip);
653 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
654 xfs_buf_delwri_promote(bp);
660 xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
665 * This is the ops vector shared by all buf log items.
667 static struct xfs_item_ops xfs_buf_item_ops = {
668 .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
669 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
671 .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
672 .iop_unpin = (void(*)(xfs_log_item_t*))xfs_buf_item_unpin,
673 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
674 xfs_buf_item_unpin_remove,
675 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
676 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
677 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
678 xfs_buf_item_committed,
679 .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push,
680 .iop_pushbuf = (void(*)(xfs_log_item_t*))xfs_buf_item_pushbuf,
681 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
682 xfs_buf_item_committing
687 * Allocate a new buf log item to go with the given buffer.
688 * Set the buffer's b_fsprivate field to point to the new
689 * buf log item. If there are other item's attached to the
690 * buffer (see xfs_buf_attach_iodone() below), then put the
691 * buf log item at the front.
699 xfs_buf_log_item_t *bip;
704 * Check to see if there is already a buf log item for
705 * this buffer. If there is, it is guaranteed to be
706 * the first. If we do already have one, there is
707 * nothing to do here so return.
709 if (bp->b_mount != mp)
711 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
712 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
713 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
714 if (lip->li_type == XFS_LI_BUF) {
720 * chunks is the number of XFS_BLF_CHUNK size pieces
721 * the buffer can be divided into. Make sure not to
722 * truncate any pieces. map_size is the size of the
723 * bitmap needed to describe the chunks of the buffer.
725 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
726 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
728 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
730 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
733 bip->bli_format.blf_type = XFS_LI_BUF;
734 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
735 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
736 bip->bli_format.blf_map_size = map_size;
738 #ifdef XFS_TRANS_DEBUG
740 * Allocate the arrays for tracking what needs to be logged
741 * and what our callers request to be logged. bli_orig
742 * holds a copy of the original, clean buffer for comparison
743 * against, and bli_logged keeps a 1 bit flag per byte in
744 * the buffer to indicate which bytes the callers have asked
747 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
748 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
749 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
753 * Put the buf item into the list of items attached to the
754 * buffer at the front.
756 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
757 bip->bli_item.li_bio_list =
758 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
760 XFS_BUF_SET_FSPRIVATE(bp, bip);
765 * Mark bytes first through last inclusive as dirty in the buf
770 xfs_buf_log_item_t *bip,
785 * Mark the item as having some dirty data for
786 * quick reference in xfs_buf_item_dirty.
788 bip->bli_flags |= XFS_BLI_DIRTY;
791 * Convert byte offsets to bit numbers.
793 first_bit = first >> XFS_BLF_SHIFT;
794 last_bit = last >> XFS_BLF_SHIFT;
797 * Calculate the total number of bits to be set.
799 bits_to_set = last_bit - first_bit + 1;
802 * Get a pointer to the first word in the bitmap
805 word_num = first_bit >> BIT_TO_WORD_SHIFT;
806 wordp = &(bip->bli_format.blf_data_map[word_num]);
809 * Calculate the starting bit in the first word.
811 bit = first_bit & (uint)(NBWORD - 1);
814 * First set any bits in the first word of our range.
815 * If it starts at bit 0 of the word, it will be
816 * set below rather than here. That is what the variable
817 * bit tells us. The variable bits_set tracks the number
818 * of bits that have been set so far. End_bit is the number
819 * of the last bit to be set in this word plus one.
822 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
823 mask = ((1 << (end_bit - bit)) - 1) << bit;
826 bits_set = end_bit - bit;
832 * Now set bits a whole word at a time that are between
833 * first_bit and last_bit.
835 while ((bits_to_set - bits_set) >= NBWORD) {
836 *wordp |= 0xffffffff;
842 * Finally, set any bits left to be set in one last partial word.
844 end_bit = bits_to_set - bits_set;
846 mask = (1 << end_bit) - 1;
850 xfs_buf_item_log_debug(bip, first, last);
855 * Return 1 if the buffer has some data that has been logged (at any
856 * point, not just the current transaction) and 0 if not.
860 xfs_buf_log_item_t *bip)
862 return (bip->bli_flags & XFS_BLI_DIRTY);
867 xfs_buf_log_item_t *bip)
869 #ifdef XFS_TRANS_DEBUG
870 kmem_free(bip->bli_orig);
871 kmem_free(bip->bli_logged);
872 #endif /* XFS_TRANS_DEBUG */
874 kmem_zone_free(xfs_buf_item_zone, bip);
878 * This is called when the buf log item is no longer needed. It should
879 * free the buf log item associated with the given buffer and clear
880 * the buffer's pointer to the buf log item. If there are no more
881 * items in the list, clear the b_iodone field of the buffer (see
882 * xfs_buf_attach_iodone() below).
888 xfs_buf_log_item_t *bip;
890 trace_xfs_buf_item_relse(bp, _RET_IP_);
892 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
893 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
894 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
895 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
896 XFS_BUF_CLR_IODONE_FUNC(bp);
899 xfs_buf_item_free(bip);
904 * Add the given log item with its callback to the list of callbacks
905 * to be called when the buffer's I/O completes. If it is not set
906 * already, set the buffer's b_iodone() routine to be
907 * xfs_buf_iodone_callbacks() and link the log item into the list of
908 * items rooted at b_fsprivate. Items are always added as the second
909 * entry in the list if there is a first, because the buf item code
910 * assumes that the buf log item is first.
913 xfs_buf_attach_iodone(
915 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
918 xfs_log_item_t *head_lip;
920 ASSERT(XFS_BUF_ISBUSY(bp));
921 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
924 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
925 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
926 lip->li_bio_list = head_lip->li_bio_list;
927 head_lip->li_bio_list = lip;
929 XFS_BUF_SET_FSPRIVATE(bp, lip);
932 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
933 (XFS_BUF_IODONE_FUNC(bp) == NULL));
934 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
938 xfs_buf_do_callbacks(
942 xfs_log_item_t *nlip;
944 while (lip != NULL) {
945 nlip = lip->li_bio_list;
946 ASSERT(lip->li_cb != NULL);
948 * Clear the next pointer so we don't have any
949 * confusion if the item is added to another buf.
950 * Don't touch the log item after calling its
951 * callback, because it could have freed itself.
953 lip->li_bio_list = NULL;
960 * This is the iodone() function for buffers which have had callbacks
961 * attached to them by xfs_buf_attach_iodone(). It should remove each
962 * log item from the buffer's list and call the callback of each in turn.
963 * When done, the buffer's fsprivate field is set to NULL and the buffer
964 * is unlocked with a call to iodone().
967 xfs_buf_iodone_callbacks(
971 static ulong lasttime;
972 static xfs_buftarg_t *lasttarg;
975 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
976 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
978 if (XFS_BUF_GETERROR(bp) != 0) {
980 * If we've already decided to shutdown the filesystem
981 * because of IO errors, there's no point in giving this
985 if (XFS_FORCED_SHUTDOWN(mp)) {
986 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
987 XFS_BUF_SUPER_STALE(bp);
988 trace_xfs_buf_item_iodone(bp, _RET_IP_);
989 xfs_buf_do_callbacks(bp, lip);
990 XFS_BUF_SET_FSPRIVATE(bp, NULL);
991 XFS_BUF_CLR_IODONE_FUNC(bp);
996 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
997 (time_after(jiffies, (lasttime + 5*HZ)))) {
999 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
1000 " block 0x%llx in %s",
1001 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
1002 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
1004 lasttarg = XFS_BUF_TARGET(bp);
1006 if (XFS_BUF_ISASYNC(bp)) {
1008 * If the write was asynchronous then noone will be
1009 * looking for the error. Clear the error state
1010 * and write the buffer out again delayed write.
1012 * XXXsup This is OK, so long as we catch these
1013 * before we start the umount; we don't want these
1014 * DELWRI metadata bufs to be hanging around.
1016 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1018 if (!(XFS_BUF_ISSTALE(bp))) {
1019 XFS_BUF_DELAYWRITE(bp);
1021 XFS_BUF_SET_START(bp);
1023 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1024 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1028 * If the write of the buffer was not asynchronous,
1029 * then we want to make sure to return the error
1030 * to the caller of bwrite(). Because of this we
1031 * cannot clear the B_ERROR state at this point.
1032 * Instead we install a callback function that
1033 * will be called when the buffer is released, and
1034 * that routine will clear the error state and
1035 * set the buffer to be written out again after
1038 /* We actually overwrite the existing b-relse
1039 function at times, but we're gonna be shutting down
1041 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1043 XFS_BUF_FINISH_IOWAIT(bp);
1048 xfs_buf_do_callbacks(bp, lip);
1049 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1050 XFS_BUF_CLR_IODONE_FUNC(bp);
1055 * This is a callback routine attached to a buffer which gets an error
1056 * when being written out synchronously.
1059 xfs_buf_error_relse(
1062 xfs_log_item_t *lip;
1065 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1066 mp = (xfs_mount_t *)lip->li_mountp;
1067 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1071 XFS_BUF_UNDELAYWRITE(bp);
1072 XFS_BUF_ERROR(bp,0);
1074 trace_xfs_buf_error_relse(bp, _RET_IP_);
1076 if (! XFS_FORCED_SHUTDOWN(mp))
1077 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1079 * We have to unpin the pinned buffers so do the
1082 xfs_buf_do_callbacks(bp, lip);
1083 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1084 XFS_BUF_CLR_IODONE_FUNC(bp);
1085 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1091 * This is the iodone() function for buffers which have been
1092 * logged. It is called when they are eventually flushed out.
1093 * It should remove the buf item from the AIL, and free the buf item.
1094 * It is called by xfs_buf_iodone_callbacks() above which will take
1095 * care of cleaning up the buffer itself.
1101 xfs_buf_log_item_t *bip)
1103 struct xfs_ail *ailp = bip->bli_item.li_ailp;
1105 ASSERT(bip->bli_buf == bp);
1110 * If we are forcibly shutting down, this may well be
1111 * off the AIL already. That's because we simulate the
1112 * log-committed callbacks to unpin these buffers. Or we may never
1113 * have put this item on AIL because of the transaction was
1114 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1116 * Either way, AIL is useless if we're forcing a shutdown.
1118 spin_lock(&ailp->xa_lock);
1119 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
1120 xfs_buf_item_free(bip);