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"
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_BLI_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 xfs_buf_item_trace("SIZE STALE", bip);
168 ASSERT(bip->bli_format.blf_flags & XFS_BLI_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_BLI_CHUNK) !=
200 (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) +
209 xfs_buf_item_trace("SIZE NORM", 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 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BFORMAT);
256 if (bip->bli_flags & XFS_BLI_STALE) {
258 * The buffer is stale, so all we need to log
259 * is the buf log format structure with the
262 xfs_buf_item_trace("FORMAT STALE", bip);
263 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
264 bip->bli_format.blf_size = nvecs;
269 * Fill in an iovec for each set of contiguous chunks.
271 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
272 bip->bli_format.blf_map_size, 0);
273 ASSERT(first_bit != -1);
274 last_bit = first_bit;
278 * This takes the bit number to start looking from and
279 * returns the next set bit from there. It returns -1
280 * if there are no more bits set or the start bit is
281 * beyond the end of the bitmap.
283 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
284 bip->bli_format.blf_map_size,
287 * If we run out of bits fill in the last iovec and get
289 * Else if we start a new set of bits then fill in the
290 * iovec for the series we were looking at and start
291 * counting the bits in the new one.
292 * Else we're still in the same set of bits so just
293 * keep counting and scanning.
295 if (next_bit == -1) {
296 buffer_offset = first_bit * XFS_BLI_CHUNK;
297 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
298 vecp->i_len = nbits * XFS_BLI_CHUNK;
299 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
302 } else if (next_bit != last_bit + 1) {
303 buffer_offset = first_bit * XFS_BLI_CHUNK;
304 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
305 vecp->i_len = nbits * XFS_BLI_CHUNK;
306 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
309 first_bit = next_bit;
312 } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) !=
313 (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) +
315 buffer_offset = first_bit * XFS_BLI_CHUNK;
316 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
317 vecp->i_len = nbits * XFS_BLI_CHUNK;
318 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
319 /* You would think we need to bump the nvecs here too, but we do not
320 * this number is used by recovery, and it gets confused by the boundary
325 first_bit = next_bit;
333 bip->bli_format.blf_size = nvecs;
336 * Check to make sure everything is consistent.
338 xfs_buf_item_trace("FORMAT NORM", bip);
339 xfs_buf_item_log_check(bip);
343 * This is called to pin the buffer associated with the buf log
344 * item in memory so it cannot be written out. Simply call bpin()
345 * on the buffer to do this.
349 xfs_buf_log_item_t *bip)
354 ASSERT(XFS_BUF_ISBUSY(bp));
355 ASSERT(atomic_read(&bip->bli_refcount) > 0);
356 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
357 (bip->bli_flags & XFS_BLI_STALE));
358 xfs_buf_item_trace("PIN", bip);
359 xfs_buftrace("XFS_PIN", bp);
365 * This is called to unpin the buffer associated with the buf log
366 * item which was previously pinned with a call to xfs_buf_item_pin().
367 * Just call bunpin() on the buffer to do this.
369 * Also drop the reference to the buf item for the current transaction.
370 * If the XFS_BLI_STALE flag is set and we are the last reference,
371 * then free up the buf log item and unlock the buffer.
375 xfs_buf_log_item_t *bip,
384 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
385 ASSERT(atomic_read(&bip->bli_refcount) > 0);
386 xfs_buf_item_trace("UNPIN", bip);
387 xfs_buftrace("XFS_UNPIN", bp);
389 freed = atomic_dec_and_test(&bip->bli_refcount);
390 mp = bip->bli_item.li_mountp;
392 if (freed && stale) {
393 ASSERT(bip->bli_flags & XFS_BLI_STALE);
394 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
395 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
396 ASSERT(XFS_BUF_ISSTALE(bp));
397 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
398 xfs_buf_item_trace("UNPIN STALE", bip);
399 xfs_buftrace("XFS_UNPIN STALE", bp);
401 * If we get called here because of an IO error, we may
402 * or may not have the item on the AIL. xfs_trans_delete_ail()
403 * will take care of that situation.
404 * xfs_trans_delete_ail() drops the AIL lock.
406 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
407 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
408 XFS_BUF_SET_FSPRIVATE(bp, NULL);
409 XFS_BUF_CLR_IODONE_FUNC(bp);
411 spin_lock(&mp->m_ail_lock);
412 xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip);
413 xfs_buf_item_relse(bp);
414 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
421 * this is called from uncommit in the forced-shutdown path.
422 * we need to check to see if the reference count on the log item
423 * is going to drop to zero. If so, unpin will free the log item
424 * so we need to free the item's descriptor (that points to the item)
425 * in the transaction.
428 xfs_buf_item_unpin_remove(
429 xfs_buf_log_item_t *bip,
433 xfs_log_item_desc_t *lidp;
438 * will xfs_buf_item_unpin() call xfs_buf_item_relse()?
440 if ((atomic_read(&bip->bli_refcount) == 1) &&
441 (bip->bli_flags & XFS_BLI_STALE)) {
442 ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0);
443 xfs_buf_item_trace("UNPIN REMOVE", bip);
444 xfs_buftrace("XFS_UNPIN_REMOVE", bp);
446 * yes -- clear the xaction descriptor in-use flag
447 * and free the chunk if required. We can safely
448 * do some work here and then call buf_item_unpin
449 * to do the rest because if the if is true, then
450 * we are holding the buffer locked so no one else
451 * will be able to bump up the refcount.
453 lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip);
454 stale = lidp->lid_flags & XFS_LID_BUF_STALE;
455 xfs_trans_free_item(tp, lidp);
457 * Since the transaction no longer refers to the buffer,
458 * the buffer should no longer refer to the transaction.
460 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
463 xfs_buf_item_unpin(bip, stale);
469 * This is called to attempt to lock the buffer associated with this
470 * buf log item. Don't sleep on the buffer lock. If we can't get
471 * the lock right away, return 0. If we can get the lock, pull the
472 * buffer from the free list, mark it busy, and return 1.
475 xfs_buf_item_trylock(
476 xfs_buf_log_item_t *bip)
482 if (XFS_BUF_ISPINNED(bp)) {
483 return XFS_ITEM_PINNED;
486 if (!XFS_BUF_CPSEMA(bp)) {
487 return XFS_ITEM_LOCKED;
491 * Remove the buffer from the free list. Only do this
492 * if it's on the free list. Private buffers like the
493 * superblock buffer are not.
497 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
498 xfs_buf_item_trace("TRYLOCK SUCCESS", bip);
499 return XFS_ITEM_SUCCESS;
503 * Release the buffer associated with the buf log item.
504 * If there is no dirty logged data associated with the
505 * buffer recorded in the buf log item, then free the
506 * buf log item and remove the reference to it in the
509 * This call ignores the recursion count. It is only called
510 * when the buffer should REALLY be unlocked, regardless
511 * of the recursion count.
513 * If the XFS_BLI_HOLD flag is set in the buf log item, then
514 * free the log item if necessary but do not unlock the buffer.
515 * This is for support of xfs_trans_bhold(). Make sure the
516 * XFS_BLI_HOLD field is cleared if we don't free the item.
520 xfs_buf_log_item_t *bip)
527 xfs_buftrace("XFS_UNLOCK", bp);
530 * Clear the buffer's association with this transaction.
532 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
535 * If this is a transaction abort, don't return early.
536 * Instead, allow the brelse to happen.
537 * Normally it would be done for stale (cancelled) buffers
538 * at unpin time, but we'll never go through the pin/unpin
539 * cycle if we abort inside commit.
541 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
544 * If the buf item is marked stale, then don't do anything.
545 * We'll unlock the buffer and free the buf item when the
546 * buffer is unpinned for the last time.
548 if (bip->bli_flags & XFS_BLI_STALE) {
549 bip->bli_flags &= ~XFS_BLI_LOGGED;
550 xfs_buf_item_trace("UNLOCK STALE", bip);
551 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
557 * Drop the transaction's reference to the log item if
558 * it was not logged as part of the transaction. Otherwise
559 * we'll drop the reference in xfs_buf_item_unpin() when
560 * the transaction is really through with the buffer.
562 if (!(bip->bli_flags & XFS_BLI_LOGGED)) {
563 atomic_dec(&bip->bli_refcount);
566 * Clear the logged flag since this is per
569 bip->bli_flags &= ~XFS_BLI_LOGGED;
573 * Before possibly freeing the buf item, determine if we should
574 * release the buffer at the end of this routine.
576 hold = bip->bli_flags & XFS_BLI_HOLD;
577 xfs_buf_item_trace("UNLOCK", bip);
580 * If the buf item isn't tracking any data, free it.
581 * Otherwise, if XFS_BLI_HOLD is set clear it.
583 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
584 bip->bli_format.blf_map_size)) {
585 xfs_buf_item_relse(bp);
587 bip->bli_flags &= ~XFS_BLI_HOLD;
591 * Release the buffer if XFS_BLI_HOLD was not set.
599 * This is called to find out where the oldest active copy of the
600 * buf log item in the on disk log resides now that the last log
601 * write of it completed at the given lsn.
602 * We always re-log all the dirty data in a buffer, so usually the
603 * latest copy in the on disk log is the only one that matters. For
604 * those cases we simply return the given lsn.
606 * The one exception to this is for buffers full of newly allocated
607 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
608 * flag set, indicating that only the di_next_unlinked fields from the
609 * inodes in the buffers will be replayed during recovery. If the
610 * original newly allocated inode images have not yet been flushed
611 * when the buffer is so relogged, then we need to make sure that we
612 * keep the old images in the 'active' portion of the log. We do this
613 * by returning the original lsn of that transaction here rather than
617 xfs_buf_item_committed(
618 xfs_buf_log_item_t *bip,
621 xfs_buf_item_trace("COMMITTED", bip);
622 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
623 (bip->bli_item.li_lsn != 0)) {
624 return bip->bli_item.li_lsn;
630 * This is called to asynchronously write the buffer associated with this
631 * buf log item out to disk. The buffer will already have been locked by
632 * a successful call to xfs_buf_item_trylock(). If the buffer still has
633 * B_DELWRI set, then get it going out to disk with a call to bawrite().
634 * If not, then just release the buffer.
638 xfs_buf_log_item_t *bip)
642 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
643 xfs_buf_item_trace("PUSH", bip);
647 if (XFS_BUF_ISDELAYWRITE(bp)) {
649 error = xfs_bawrite(bip->bli_item.li_mountp, bp);
651 xfs_fs_cmn_err(CE_WARN, bip->bli_item.li_mountp,
652 "xfs_buf_item_push: pushbuf error %d on bip %p, bp %p",
661 xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
666 * This is the ops vector shared by all buf log items.
668 static struct xfs_item_ops xfs_buf_item_ops = {
669 .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
670 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
672 .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
673 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
674 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
675 xfs_buf_item_unpin_remove,
676 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
677 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
678 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
679 xfs_buf_item_committed,
680 .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push,
682 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
683 xfs_buf_item_committing
688 * Allocate a new buf log item to go with the given buffer.
689 * Set the buffer's b_fsprivate field to point to the new
690 * buf log item. If there are other item's attached to the
691 * buffer (see xfs_buf_attach_iodone() below), then put the
692 * buf log item at the front.
700 xfs_buf_log_item_t *bip;
705 * Check to see if there is already a buf log item for
706 * this buffer. If there is, it is guaranteed to be
707 * the first. If we do already have one, there is
708 * nothing to do here so return.
710 if (XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *) != mp)
711 XFS_BUF_SET_FSPRIVATE3(bp, mp);
712 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
713 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
714 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
715 if (lip->li_type == XFS_LI_BUF) {
721 * chunks is the number of XFS_BLI_CHUNK size pieces
722 * the buffer can be divided into. Make sure not to
723 * truncate any pieces. map_size is the size of the
724 * bitmap needed to describe the chunks of the buffer.
726 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT);
727 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
729 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
731 bip->bli_item.li_type = XFS_LI_BUF;
732 bip->bli_item.li_ops = &xfs_buf_item_ops;
733 bip->bli_item.li_mountp = mp;
735 bip->bli_format.blf_type = XFS_LI_BUF;
736 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
737 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
738 bip->bli_format.blf_map_size = map_size;
740 bip->bli_trace = ktrace_alloc(XFS_BLI_TRACE_SIZE, KM_SLEEP);
743 #ifdef XFS_TRANS_DEBUG
745 * Allocate the arrays for tracking what needs to be logged
746 * and what our callers request to be logged. bli_orig
747 * holds a copy of the original, clean buffer for comparison
748 * against, and bli_logged keeps a 1 bit flag per byte in
749 * the buffer to indicate which bytes the callers have asked
752 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
753 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
754 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
758 * Put the buf item into the list of items attached to the
759 * buffer at the front.
761 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
762 bip->bli_item.li_bio_list =
763 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
765 XFS_BUF_SET_FSPRIVATE(bp, bip);
770 * Mark bytes first through last inclusive as dirty in the buf
775 xfs_buf_log_item_t *bip,
790 * Mark the item as having some dirty data for
791 * quick reference in xfs_buf_item_dirty.
793 bip->bli_flags |= XFS_BLI_DIRTY;
796 * Convert byte offsets to bit numbers.
798 first_bit = first >> XFS_BLI_SHIFT;
799 last_bit = last >> XFS_BLI_SHIFT;
802 * Calculate the total number of bits to be set.
804 bits_to_set = last_bit - first_bit + 1;
807 * Get a pointer to the first word in the bitmap
810 word_num = first_bit >> BIT_TO_WORD_SHIFT;
811 wordp = &(bip->bli_format.blf_data_map[word_num]);
814 * Calculate the starting bit in the first word.
816 bit = first_bit & (uint)(NBWORD - 1);
819 * First set any bits in the first word of our range.
820 * If it starts at bit 0 of the word, it will be
821 * set below rather than here. That is what the variable
822 * bit tells us. The variable bits_set tracks the number
823 * of bits that have been set so far. End_bit is the number
824 * of the last bit to be set in this word plus one.
827 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
828 mask = ((1 << (end_bit - bit)) - 1) << bit;
831 bits_set = end_bit - bit;
837 * Now set bits a whole word at a time that are between
838 * first_bit and last_bit.
840 while ((bits_to_set - bits_set) >= NBWORD) {
841 *wordp |= 0xffffffff;
847 * Finally, set any bits left to be set in one last partial word.
849 end_bit = bits_to_set - bits_set;
851 mask = (1 << end_bit) - 1;
855 xfs_buf_item_log_debug(bip, first, last);
860 * Return 1 if the buffer has some data that has been logged (at any
861 * point, not just the current transaction) and 0 if not.
865 xfs_buf_log_item_t *bip)
867 return (bip->bli_flags & XFS_BLI_DIRTY);
871 * This is called when the buf log item is no longer needed. It should
872 * free the buf log item associated with the given buffer and clear
873 * the buffer's pointer to the buf log item. If there are no more
874 * items in the list, clear the b_iodone field of the buffer (see
875 * xfs_buf_attach_iodone() below).
881 xfs_buf_log_item_t *bip;
883 xfs_buftrace("XFS_RELSE", bp);
884 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
885 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
886 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
887 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
888 XFS_BUF_CLR_IODONE_FUNC(bp);
891 #ifdef XFS_TRANS_DEBUG
892 kmem_free(bip->bli_orig, XFS_BUF_COUNT(bp));
893 bip->bli_orig = NULL;
894 kmem_free(bip->bli_logged, XFS_BUF_COUNT(bp) / NBBY);
895 bip->bli_logged = NULL;
896 #endif /* XFS_TRANS_DEBUG */
899 ktrace_free(bip->bli_trace);
901 kmem_zone_free(xfs_buf_item_zone, bip);
906 * Add the given log item with its callback to the list of callbacks
907 * to be called when the buffer's I/O completes. If it is not set
908 * already, set the buffer's b_iodone() routine to be
909 * xfs_buf_iodone_callbacks() and link the log item into the list of
910 * items rooted at b_fsprivate. Items are always added as the second
911 * entry in the list if there is a first, because the buf item code
912 * assumes that the buf log item is first.
915 xfs_buf_attach_iodone(
917 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
920 xfs_log_item_t *head_lip;
922 ASSERT(XFS_BUF_ISBUSY(bp));
923 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
926 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
927 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
928 lip->li_bio_list = head_lip->li_bio_list;
929 head_lip->li_bio_list = lip;
931 XFS_BUF_SET_FSPRIVATE(bp, lip);
934 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
935 (XFS_BUF_IODONE_FUNC(bp) == NULL));
936 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
940 xfs_buf_do_callbacks(
944 xfs_log_item_t *nlip;
946 while (lip != NULL) {
947 nlip = lip->li_bio_list;
948 ASSERT(lip->li_cb != NULL);
950 * Clear the next pointer so we don't have any
951 * confusion if the item is added to another buf.
952 * Don't touch the log item after calling its
953 * callback, because it could have freed itself.
955 lip->li_bio_list = NULL;
962 * This is the iodone() function for buffers which have had callbacks
963 * attached to them by xfs_buf_attach_iodone(). It should remove each
964 * log item from the buffer's list and call the callback of each in turn.
965 * When done, the buffer's fsprivate field is set to NULL and the buffer
966 * is unlocked with a call to iodone().
969 xfs_buf_iodone_callbacks(
973 static ulong lasttime;
974 static xfs_buftarg_t *lasttarg;
977 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
978 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
980 if (XFS_BUF_GETERROR(bp) != 0) {
982 * If we've already decided to shutdown the filesystem
983 * because of IO errors, there's no point in giving this
987 if (XFS_FORCED_SHUTDOWN(mp)) {
988 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
989 XFS_BUF_SUPER_STALE(bp);
990 xfs_buftrace("BUF_IODONE_CB", bp);
991 xfs_buf_do_callbacks(bp, lip);
992 XFS_BUF_SET_FSPRIVATE(bp, NULL);
993 XFS_BUF_CLR_IODONE_FUNC(bp);
996 * XFS_SHUT flag gets set when we go thru the
997 * entire buffer cache and deliberately start
998 * throwing away delayed write buffers.
999 * Since there's no biowait done on those,
1000 * we should just brelse them.
1002 if (XFS_BUF_ISSHUT(bp)) {
1012 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
1013 (time_after(jiffies, (lasttime + 5*HZ)))) {
1015 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
1016 " block 0x%llx in %s",
1017 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
1018 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
1020 lasttarg = XFS_BUF_TARGET(bp);
1022 if (XFS_BUF_ISASYNC(bp)) {
1024 * If the write was asynchronous then noone will be
1025 * looking for the error. Clear the error state
1026 * and write the buffer out again delayed write.
1028 * XXXsup This is OK, so long as we catch these
1029 * before we start the umount; we don't want these
1030 * DELWRI metadata bufs to be hanging around.
1032 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1034 if (!(XFS_BUF_ISSTALE(bp))) {
1035 XFS_BUF_DELAYWRITE(bp);
1037 XFS_BUF_SET_START(bp);
1039 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1040 xfs_buftrace("BUF_IODONE ASYNC", bp);
1044 * If the write of the buffer was not asynchronous,
1045 * then we want to make sure to return the error
1046 * to the caller of bwrite(). Because of this we
1047 * cannot clear the B_ERROR state at this point.
1048 * Instead we install a callback function that
1049 * will be called when the buffer is released, and
1050 * that routine will clear the error state and
1051 * set the buffer to be written out again after
1054 /* We actually overwrite the existing b-relse
1055 function at times, but we're gonna be shutting down
1057 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1059 XFS_BUF_V_IODONESEMA(bp);
1063 #ifdef XFSERRORDEBUG
1064 xfs_buftrace("XFS BUFCB NOERR", bp);
1066 xfs_buf_do_callbacks(bp, lip);
1067 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1068 XFS_BUF_CLR_IODONE_FUNC(bp);
1073 * This is a callback routine attached to a buffer which gets an error
1074 * when being written out synchronously.
1077 xfs_buf_error_relse(
1080 xfs_log_item_t *lip;
1083 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1084 mp = (xfs_mount_t *)lip->li_mountp;
1085 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1089 XFS_BUF_UNDELAYWRITE(bp);
1090 XFS_BUF_ERROR(bp,0);
1091 xfs_buftrace("BUF_ERROR_RELSE", bp);
1092 if (! XFS_FORCED_SHUTDOWN(mp))
1093 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1095 * We have to unpin the pinned buffers so do the
1098 xfs_buf_do_callbacks(bp, lip);
1099 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1100 XFS_BUF_CLR_IODONE_FUNC(bp);
1101 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1107 * This is the iodone() function for buffers which have been
1108 * logged. It is called when they are eventually flushed out.
1109 * It should remove the buf item from the AIL, and free the buf item.
1110 * It is called by xfs_buf_iodone_callbacks() above which will take
1111 * care of cleaning up the buffer itself.
1117 xfs_buf_log_item_t *bip)
1119 struct xfs_mount *mp;
1121 ASSERT(bip->bli_buf == bp);
1123 mp = bip->bli_item.li_mountp;
1126 * If we are forcibly shutting down, this may well be
1127 * off the AIL already. That's because we simulate the
1128 * log-committed callbacks to unpin these buffers. Or we may never
1129 * have put this item on AIL because of the transaction was
1130 * aborted forcibly. xfs_trans_delete_ail() takes care of these.
1132 * Either way, AIL is useless if we're forcing a shutdown.
1134 spin_lock(&mp->m_ail_lock);
1136 * xfs_trans_delete_ail() drops the AIL lock.
1138 xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip);
1140 #ifdef XFS_TRANS_DEBUG
1141 kmem_free(bip->bli_orig, XFS_BUF_COUNT(bp));
1142 bip->bli_orig = NULL;
1143 kmem_free(bip->bli_logged, XFS_BUF_COUNT(bp) / NBBY);
1144 bip->bli_logged = NULL;
1145 #endif /* XFS_TRANS_DEBUG */
1147 #ifdef XFS_BLI_TRACE
1148 ktrace_free(bip->bli_trace);
1150 kmem_zone_free(xfs_buf_item_zone, bip);
1153 #if defined(XFS_BLI_TRACE)
1157 xfs_buf_log_item_t *bip)
1160 ASSERT(bip->bli_trace != NULL);
1163 ktrace_enter(bip->bli_trace,
1165 (void *)bip->bli_buf,
1166 (void *)((unsigned long)bip->bli_flags),
1167 (void *)((unsigned long)bip->bli_recur),
1168 (void *)((unsigned long)atomic_read(&bip->bli_refcount)),
1169 (void *)((unsigned long)
1170 (0xFFFFFFFF & XFS_BUF_ADDR(bp) >> 32)),
1171 (void *)((unsigned long)(0xFFFFFFFF & XFS_BUF_ADDR(bp))),
1172 (void *)((unsigned long)XFS_BUF_COUNT(bp)),
1173 (void *)((unsigned long)XFS_BUF_BFLAGS(bp)),
1174 XFS_BUF_FSPRIVATE(bp, void *),
1175 XFS_BUF_FSPRIVATE2(bp, void *),
1176 (void *)(unsigned long)XFS_BUF_ISPINNED(bp),
1177 (void *)XFS_BUF_IODONE_FUNC(bp),
1178 (void *)((unsigned long)(XFS_BUF_VALUSEMA(bp))),
1179 (void *)bip->bli_item.li_desc,
1180 (void *)((unsigned long)bip->bli_item.li_flags));
1182 #endif /* XFS_BLI_TRACE */