2 * Copyright (c) 2000-2002,2005 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"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_trans_priv.h"
39 #include "xfs_error.h"
43 STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *,
45 STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *,
50 * Get and lock the buffer for the caller if it is not already
51 * locked within the given transaction. If it is already locked
52 * within the transaction, just increment its lock recursion count
53 * and return a pointer to it.
55 * Use the fast path function xfs_trans_buf_item_match() or the buffer
56 * cache routine incore_match() to find the buffer
57 * if it is already owned by this transaction.
59 * If we don't already own the buffer, use get_buf() to get it.
60 * If it doesn't yet have an associated xfs_buf_log_item structure,
61 * then allocate one and add the item to this transaction.
63 * If the transaction pointer is NULL, make this just a normal
67 xfs_trans_get_buf(xfs_trans_t *tp,
68 xfs_buftarg_t *target_dev,
74 xfs_buf_log_item_t *bip;
77 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
80 * Default to a normal get_buf() call if the tp is NULL.
83 bp = xfs_buf_get_flags(target_dev, blkno, len,
89 * If we find the buffer in the cache with this transaction
90 * pointer in its b_fsprivate2 field, then we know we already
91 * have it locked. In this case we just increment the lock
92 * recursion count and return the buffer to the caller.
94 if (tp->t_items.lic_next == NULL) {
95 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
97 bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
100 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
101 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
102 xfs_buftrace("TRANS GET RECUR SHUT", bp);
103 XFS_BUF_SUPER_STALE(bp);
106 * If the buffer is stale then it was binval'ed
107 * since last read. This doesn't matter since the
108 * caller isn't allowed to use the data anyway.
110 else if (XFS_BUF_ISSTALE(bp)) {
111 xfs_buftrace("TRANS GET RECUR STALE", bp);
112 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
114 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
115 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
117 ASSERT(atomic_read(&bip->bli_refcount) > 0);
119 xfs_buftrace("TRANS GET RECUR", bp);
120 xfs_buf_item_trace("GET RECUR", bip);
125 * We always specify the BUF_BUSY flag within a transaction so
126 * that get_buf does not try to push out a delayed write buffer
127 * which might cause another transaction to take place (if the
128 * buffer was delayed alloc). Such recursive transactions can
129 * easily deadlock with our current transaction as well as cause
130 * us to run out of stack space.
132 bp = xfs_buf_get_flags(target_dev, blkno, len, flags | BUF_BUSY);
137 ASSERT(!XFS_BUF_GETERROR(bp));
140 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
141 * it doesn't have one yet, then allocate one and initialize it.
142 * The checks to see if one is there are in xfs_buf_item_init().
144 xfs_buf_item_init(bp, tp->t_mountp);
147 * Set the recursion count for the buffer within this transaction
150 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
151 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
152 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
153 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
157 * Take a reference for this transaction on the buf item.
159 atomic_inc(&bip->bli_refcount);
162 * Get a log_item_desc to point at the new item.
164 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
167 * Initialize b_fsprivate2 so we can find it with incore_match()
170 XFS_BUF_SET_FSPRIVATE2(bp, tp);
172 xfs_buftrace("TRANS GET", bp);
173 xfs_buf_item_trace("GET", bip);
178 * Get and lock the superblock buffer of this file system for the
181 * We don't need to use incore_match() here, because the superblock
182 * buffer is a private buffer which we keep a pointer to in the
186 xfs_trans_getsb(xfs_trans_t *tp,
187 struct xfs_mount *mp,
191 xfs_buf_log_item_t *bip;
194 * Default to just trying to lock the superblock buffer
198 return (xfs_getsb(mp, flags));
202 * If the superblock buffer already has this transaction
203 * pointer in its b_fsprivate2 field, then we know we already
204 * have it locked. In this case we just increment the lock
205 * recursion count and return the buffer to the caller.
208 if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) {
209 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
211 ASSERT(atomic_read(&bip->bli_refcount) > 0);
213 xfs_buf_item_trace("GETSB RECUR", bip);
217 bp = xfs_getsb(mp, flags);
223 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
224 * it doesn't have one yet, then allocate one and initialize it.
225 * The checks to see if one is there are in xfs_buf_item_init().
227 xfs_buf_item_init(bp, mp);
230 * Set the recursion count for the buffer within this transaction
233 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
234 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
235 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
236 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
240 * Take a reference for this transaction on the buf item.
242 atomic_inc(&bip->bli_refcount);
245 * Get a log_item_desc to point at the new item.
247 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
250 * Initialize b_fsprivate2 so we can find it with incore_match()
253 XFS_BUF_SET_FSPRIVATE2(bp, tp);
255 xfs_buf_item_trace("GETSB", bip);
260 xfs_buftarg_t *xfs_error_target;
263 int xfs_error_mod = 33;
267 * Get and lock the buffer for the caller if it is not already
268 * locked within the given transaction. If it has not yet been
269 * read in, read it from disk. If it is already locked
270 * within the transaction and already read in, just increment its
271 * lock recursion count and return a pointer to it.
273 * Use the fast path function xfs_trans_buf_item_match() or the buffer
274 * cache routine incore_match() to find the buffer
275 * if it is already owned by this transaction.
277 * If we don't already own the buffer, use read_buf() to get it.
278 * If it doesn't yet have an associated xfs_buf_log_item structure,
279 * then allocate one and add the item to this transaction.
281 * If the transaction pointer is NULL, make this just a normal
288 xfs_buftarg_t *target,
295 xfs_buf_log_item_t *bip;
299 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
302 * Default to a normal get_buf() call if the tp is NULL.
305 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
307 return (flags & XFS_BUF_TRYLOCK) ?
308 EAGAIN : XFS_ERROR(ENOMEM);
310 if ((bp != NULL) && (XFS_BUF_GETERROR(bp) != 0)) {
311 xfs_ioerror_alert("xfs_trans_read_buf", mp,
313 error = XFS_BUF_GETERROR(bp);
318 if (xfs_do_error && (bp != NULL)) {
319 if (xfs_error_target == target) {
320 if (((xfs_req_num++) % xfs_error_mod) == 0) {
322 cmn_err(CE_DEBUG, "Returning error!\n");
323 return XFS_ERROR(EIO);
328 if (XFS_FORCED_SHUTDOWN(mp))
335 * If we find the buffer in the cache with this transaction
336 * pointer in its b_fsprivate2 field, then we know we already
337 * have it locked. If it is already read in we just increment
338 * the lock recursion count and return the buffer to the caller.
339 * If the buffer is not yet read in, then we read it in, increment
340 * the lock recursion count, and return it to the caller.
342 if (tp->t_items.lic_next == NULL) {
343 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
345 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
348 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
349 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
350 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
351 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
352 if (!(XFS_BUF_ISDONE(bp))) {
353 xfs_buftrace("READ_BUF_INCORE !DONE", bp);
354 ASSERT(!XFS_BUF_ISASYNC(bp));
356 xfsbdstrat(tp->t_mountp, bp);
357 error = xfs_iowait(bp);
359 xfs_ioerror_alert("xfs_trans_read_buf", mp,
363 * We can gracefully recover from most read
364 * errors. Ones we can't are those that happen
365 * after the transaction's already dirty.
367 if (tp->t_flags & XFS_TRANS_DIRTY)
368 xfs_force_shutdown(tp->t_mountp,
369 SHUTDOWN_META_IO_ERROR);
374 * We never locked this buf ourselves, so we shouldn't
375 * brelse it either. Just get out.
377 if (XFS_FORCED_SHUTDOWN(mp)) {
378 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp);
380 return XFS_ERROR(EIO);
384 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
387 ASSERT(atomic_read(&bip->bli_refcount) > 0);
388 xfs_buf_item_trace("READ RECUR", bip);
394 * We always specify the BUF_BUSY flag within a transaction so
395 * that get_buf does not try to push out a delayed write buffer
396 * which might cause another transaction to take place (if the
397 * buffer was delayed alloc). Such recursive transactions can
398 * easily deadlock with our current transaction as well as cause
399 * us to run out of stack space.
401 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
406 if (XFS_BUF_GETERROR(bp) != 0) {
407 XFS_BUF_SUPER_STALE(bp);
408 xfs_buftrace("READ ERROR", bp);
409 error = XFS_BUF_GETERROR(bp);
411 xfs_ioerror_alert("xfs_trans_read_buf", mp,
413 if (tp->t_flags & XFS_TRANS_DIRTY)
414 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
419 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
420 if (xfs_error_target == target) {
421 if (((xfs_req_num++) % xfs_error_mod) == 0) {
422 xfs_force_shutdown(tp->t_mountp,
423 SHUTDOWN_META_IO_ERROR);
425 cmn_err(CE_DEBUG, "Returning trans error!\n");
426 return XFS_ERROR(EIO);
431 if (XFS_FORCED_SHUTDOWN(mp))
435 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
436 * it doesn't have one yet, then allocate one and initialize it.
437 * The checks to see if one is there are in xfs_buf_item_init().
439 xfs_buf_item_init(bp, tp->t_mountp);
442 * Set the recursion count for the buffer within this transaction
445 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
446 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
447 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
448 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
452 * Take a reference for this transaction on the buf item.
454 atomic_inc(&bip->bli_refcount);
457 * Get a log_item_desc to point at the new item.
459 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
462 * Initialize b_fsprivate2 so we can find it with incore_match()
465 XFS_BUF_SET_FSPRIVATE2(bp, tp);
467 xfs_buftrace("TRANS READ", bp);
468 xfs_buf_item_trace("READ", bip);
474 * the theory here is that buffer is good but we're
475 * bailing out because the filesystem is being forcibly
476 * shut down. So we should leave the b_flags alone since
477 * the buffer's not staled and just get out.
480 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
481 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
483 ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) !=
484 (XFS_B_STALE|XFS_B_DELWRI));
486 xfs_buftrace("READ_BUF XFSSHUTDN", bp);
489 return XFS_ERROR(EIO);
494 * Release the buffer bp which was previously acquired with one of the
495 * xfs_trans_... buffer allocation routines if the buffer has not
496 * been modified within this transaction. If the buffer is modified
497 * within this transaction, do decrement the recursion count but do
498 * not release the buffer even if the count goes to 0. If the buffer is not
499 * modified within the transaction, decrement the recursion count and
500 * release the buffer if the recursion count goes to 0.
502 * If the buffer is to be released and it was not modified before
503 * this transaction began, then free the buf_log_item associated with it.
505 * If the transaction pointer is NULL, make this just a normal
509 xfs_trans_brelse(xfs_trans_t *tp,
512 xfs_buf_log_item_t *bip;
514 xfs_log_item_desc_t *lidp;
517 * Default to a normal brelse() call if the tp is NULL.
520 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
522 * If there's a buf log item attached to the buffer,
523 * then let the AIL know that the buffer is being
526 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
527 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
528 if (lip->li_type == XFS_LI_BUF) {
529 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
530 xfs_trans_unlocked_item(
531 bip->bli_item.li_mountp,
539 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
540 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
541 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
542 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
543 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
544 ASSERT(atomic_read(&bip->bli_refcount) > 0);
547 * Find the item descriptor pointing to this buffer's
548 * log item. It must be there.
550 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
551 ASSERT(lidp != NULL);
554 * If the release is just for a recursive lock,
555 * then decrement the count and return.
557 if (bip->bli_recur > 0) {
559 xfs_buf_item_trace("RELSE RECUR", bip);
564 * If the buffer is dirty within this transaction, we can't
565 * release it until we commit.
567 if (lidp->lid_flags & XFS_LID_DIRTY) {
568 xfs_buf_item_trace("RELSE DIRTY", bip);
573 * If the buffer has been invalidated, then we can't release
574 * it until the transaction commits to disk unless it is re-dirtied
575 * as part of this transaction. This prevents us from pulling
576 * the item from the AIL before we should.
578 if (bip->bli_flags & XFS_BLI_STALE) {
579 xfs_buf_item_trace("RELSE STALE", bip);
583 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
584 xfs_buf_item_trace("RELSE", bip);
587 * Free up the log item descriptor tracking the released item.
589 xfs_trans_free_item(tp, lidp);
592 * Clear the hold flag in the buf log item if it is set.
593 * We wouldn't want the next user of the buffer to
596 if (bip->bli_flags & XFS_BLI_HOLD) {
597 bip->bli_flags &= ~XFS_BLI_HOLD;
601 * Drop our reference to the buf log item.
603 atomic_dec(&bip->bli_refcount);
606 * If the buf item is not tracking data in the log, then
607 * we must free it before releasing the buffer back to the
608 * free pool. Before releasing the buffer to the free pool,
609 * clear the transaction pointer in b_fsprivate2 to dissolve
610 * its relation to this transaction.
612 if (!xfs_buf_item_dirty(bip)) {
614 ASSERT(bp->b_pincount == 0);
616 ASSERT(atomic_read(&bip->bli_refcount) == 0);
617 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
618 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
619 xfs_buf_item_relse(bp);
622 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
625 * If we've still got a buf log item on the buffer, then
626 * tell the AIL that the buffer is being unlocked.
629 xfs_trans_unlocked_item(bip->bli_item.li_mountp,
630 (xfs_log_item_t*)bip);
638 * Add the locked buffer to the transaction.
639 * The buffer must be locked, and it cannot be associated with any
642 * If the buffer does not yet have a buf log item associated with it,
643 * then allocate one for it. Then add the buf item to the transaction.
646 xfs_trans_bjoin(xfs_trans_t *tp,
649 xfs_buf_log_item_t *bip;
651 ASSERT(XFS_BUF_ISBUSY(bp));
652 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
655 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
656 * it doesn't have one yet, then allocate one and initialize it.
657 * The checks to see if one is there are in xfs_buf_item_init().
659 xfs_buf_item_init(bp, tp->t_mountp);
660 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
661 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
662 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
663 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
666 * Take a reference for this transaction on the buf item.
668 atomic_inc(&bip->bli_refcount);
671 * Get a log_item_desc to point at the new item.
673 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
676 * Initialize b_fsprivate2 so we can find it with incore_match()
677 * in xfs_trans_get_buf() and friends above.
679 XFS_BUF_SET_FSPRIVATE2(bp, tp);
681 xfs_buf_item_trace("BJOIN", bip);
685 * Mark the buffer as not needing to be unlocked when the buf item's
686 * IOP_UNLOCK() routine is called. The buffer must already be locked
687 * and associated with the given transaction.
691 xfs_trans_bhold(xfs_trans_t *tp,
694 xfs_buf_log_item_t *bip;
696 ASSERT(XFS_BUF_ISBUSY(bp));
697 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
698 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
700 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
701 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
702 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
703 ASSERT(atomic_read(&bip->bli_refcount) > 0);
704 bip->bli_flags |= XFS_BLI_HOLD;
705 xfs_buf_item_trace("BHOLD", bip);
709 * Cancel the previous buffer hold request made on this buffer
710 * for this transaction.
713 xfs_trans_bhold_release(xfs_trans_t *tp,
716 xfs_buf_log_item_t *bip;
718 ASSERT(XFS_BUF_ISBUSY(bp));
719 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
720 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
722 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
723 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
724 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
725 ASSERT(atomic_read(&bip->bli_refcount) > 0);
726 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
727 bip->bli_flags &= ~XFS_BLI_HOLD;
728 xfs_buf_item_trace("BHOLD RELEASE", bip);
732 * This is called to mark bytes first through last inclusive of the given
733 * buffer as needing to be logged when the transaction is committed.
734 * The buffer must already be associated with the given transaction.
736 * First and last are numbers relative to the beginning of this buffer,
737 * so the first byte in the buffer is numbered 0 regardless of the
741 xfs_trans_log_buf(xfs_trans_t *tp,
746 xfs_buf_log_item_t *bip;
747 xfs_log_item_desc_t *lidp;
749 ASSERT(XFS_BUF_ISBUSY(bp));
750 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
751 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
752 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
753 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
754 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
757 * Mark the buffer as needing to be written out eventually,
758 * and set its iodone function to remove the buffer's buf log
759 * item from the AIL and free it when the buffer is flushed
760 * to disk. See xfs_buf_attach_iodone() for more details
761 * on li_cb and xfs_buf_iodone_callbacks().
762 * If we end up aborting this transaction, we trap this buffer
763 * inside the b_bdstrat callback so that this won't get written to
766 XFS_BUF_DELAYWRITE(bp);
769 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
770 ASSERT(atomic_read(&bip->bli_refcount) > 0);
771 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
772 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
775 * If we invalidated the buffer within this transaction, then
776 * cancel the invalidation now that we're dirtying the buffer
777 * again. There are no races with the code in xfs_buf_item_unpin(),
778 * because we have a reference to the buffer this entire time.
780 if (bip->bli_flags & XFS_BLI_STALE) {
781 xfs_buf_item_trace("BLOG UNSTALE", bip);
782 bip->bli_flags &= ~XFS_BLI_STALE;
783 ASSERT(XFS_BUF_ISSTALE(bp));
785 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
788 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
789 ASSERT(lidp != NULL);
791 tp->t_flags |= XFS_TRANS_DIRTY;
792 lidp->lid_flags |= XFS_LID_DIRTY;
793 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
794 bip->bli_flags |= XFS_BLI_LOGGED;
795 xfs_buf_item_log(bip, first, last);
796 xfs_buf_item_trace("BLOG", bip);
801 * This called to invalidate a buffer that is being used within
802 * a transaction. Typically this is because the blocks in the
803 * buffer are being freed, so we need to prevent it from being
804 * written out when we're done. Allowing it to be written again
805 * might overwrite data in the free blocks if they are reallocated
808 * We prevent the buffer from being written out by clearing the
809 * B_DELWRI flag. We can't always
810 * get rid of the buf log item at this point, though, because
811 * the buffer may still be pinned by another transaction. If that
812 * is the case, then we'll wait until the buffer is committed to
813 * disk for the last time (we can tell by the ref count) and
814 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
815 * will keep the buffer locked so that the buffer and buf log item
823 xfs_log_item_desc_t *lidp;
824 xfs_buf_log_item_t *bip;
826 ASSERT(XFS_BUF_ISBUSY(bp));
827 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
828 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
830 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
831 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
832 ASSERT(lidp != NULL);
833 ASSERT(atomic_read(&bip->bli_refcount) > 0);
835 if (bip->bli_flags & XFS_BLI_STALE) {
837 * If the buffer is already invalidated, then
840 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
841 ASSERT(XFS_BUF_ISSTALE(bp));
842 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
843 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
844 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
845 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
846 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
847 xfs_buftrace("XFS_BINVAL RECUR", bp);
848 xfs_buf_item_trace("BINVAL RECUR", bip);
853 * Clear the dirty bit in the buffer and set the STALE flag
854 * in the buf log item. The STALE flag will be used in
855 * xfs_buf_item_unpin() to determine if it should clean up
856 * when the last reference to the buf item is given up.
857 * We set the XFS_BLI_CANCEL flag in the buf log format structure
858 * and log the buf item. This will be used at recovery time
859 * to determine that copies of the buffer in the log before
860 * this should not be replayed.
861 * We mark the item descriptor and the transaction dirty so
862 * that we'll hold the buffer until after the commit.
864 * Since we're invalidating the buffer, we also clear the state
865 * about which parts of the buffer have been logged. We also
866 * clear the flag indicating that this is an inode buffer since
867 * the data in the buffer will no longer be valid.
869 * We set the stale bit in the buffer as well since we're getting
872 XFS_BUF_UNDELAYWRITE(bp);
874 bip->bli_flags |= XFS_BLI_STALE;
875 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
876 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
877 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
878 memset((char *)(bip->bli_format.blf_data_map), 0,
879 (bip->bli_format.blf_map_size * sizeof(uint)));
880 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
881 tp->t_flags |= XFS_TRANS_DIRTY;
882 xfs_buftrace("XFS_BINVAL", bp);
883 xfs_buf_item_trace("BINVAL", bip);
887 * This call is used to indicate that the buffer contains on-disk
888 * inodes which must be handled specially during recovery. They
889 * require special handling because only the di_next_unlinked from
890 * the inodes in the buffer should be recovered. The rest of the
891 * data in the buffer is logged via the inodes themselves.
893 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
894 * format structure so that we'll know what to do at recovery time.
902 xfs_buf_log_item_t *bip;
904 ASSERT(XFS_BUF_ISBUSY(bp));
905 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
906 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
908 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
909 ASSERT(atomic_read(&bip->bli_refcount) > 0);
911 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
915 * This call is used to indicate that the buffer is going to
916 * be staled and was an inode buffer. This means it gets
917 * special processing during unpin - where any inodes
918 * associated with the buffer should be removed from ail.
919 * There is also special processing during recovery,
920 * any replay of the inodes in the buffer needs to be
921 * prevented as the buffer may have been reused.
924 xfs_trans_stale_inode_buf(
928 xfs_buf_log_item_t *bip;
930 ASSERT(XFS_BUF_ISBUSY(bp));
931 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
932 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
934 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
935 ASSERT(atomic_read(&bip->bli_refcount) > 0);
937 bip->bli_flags |= XFS_BLI_STALE_INODE;
938 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
945 * Mark the buffer as being one which contains newly allocated
946 * inodes. We need to make sure that even if this buffer is
947 * relogged as an 'inode buf' we still recover all of the inode
948 * images in the face of a crash. This works in coordination with
949 * xfs_buf_item_committed() to ensure that the buffer remains in the
950 * AIL at its original location even after it has been relogged.
954 xfs_trans_inode_alloc_buf(
958 xfs_buf_log_item_t *bip;
960 ASSERT(XFS_BUF_ISBUSY(bp));
961 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
962 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
964 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
965 ASSERT(atomic_read(&bip->bli_refcount) > 0);
967 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
972 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
973 * dquots. However, unlike in inode buffer recovery, dquot buffers get
974 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
975 * The only thing that makes dquot buffers different from regular
976 * buffers is that we must not replay dquot bufs when recovering
977 * if a _corresponding_ quotaoff has happened. We also have to distinguish
978 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
979 * can be turned off independently.
988 xfs_buf_log_item_t *bip;
990 ASSERT(XFS_BUF_ISBUSY(bp));
991 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
992 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
993 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
994 type == XFS_BLI_PDQUOT_BUF ||
995 type == XFS_BLI_GDQUOT_BUF);
997 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
998 ASSERT(atomic_read(&bip->bli_refcount) > 0);
1000 bip->bli_format.blf_flags |= type;
1004 * Check to see if a buffer matching the given parameters is already
1005 * a part of the given transaction. Only check the first, embedded
1006 * chunk, since we don't want to spend all day scanning large transactions.
1009 xfs_trans_buf_item_match(
1011 xfs_buftarg_t *target,
1015 xfs_log_item_chunk_t *licp;
1016 xfs_log_item_desc_t *lidp;
1017 xfs_buf_log_item_t *blip;
1023 licp = &tp->t_items;
1024 if (!xfs_lic_are_all_free(licp)) {
1025 for (i = 0; i < licp->lic_unused; i++) {
1027 * Skip unoccupied slots.
1029 if (xfs_lic_isfree(licp, i)) {
1033 lidp = xfs_lic_slot(licp, i);
1034 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1035 if (blip->bli_item.li_type != XFS_LI_BUF) {
1040 if ((XFS_BUF_TARGET(bp) == target) &&
1041 (XFS_BUF_ADDR(bp) == blkno) &&
1042 (XFS_BUF_COUNT(bp) == len)) {
1044 * We found it. Break out and
1045 * return the pointer to the buffer.
1057 * Check to see if a buffer matching the given parameters is already
1058 * a part of the given transaction. Check all the chunks, we
1059 * want to be thorough.
1062 xfs_trans_buf_item_match_all(
1064 xfs_buftarg_t *target,
1068 xfs_log_item_chunk_t *licp;
1069 xfs_log_item_desc_t *lidp;
1070 xfs_buf_log_item_t *blip;
1076 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
1077 if (xfs_lic_are_all_free(licp)) {
1078 ASSERT(licp == &tp->t_items);
1079 ASSERT(licp->lic_next == NULL);
1082 for (i = 0; i < licp->lic_unused; i++) {
1084 * Skip unoccupied slots.
1086 if (xfs_lic_isfree(licp, i)) {
1090 lidp = xfs_lic_slot(licp, i);
1091 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1092 if (blip->bli_item.li_type != XFS_LI_BUF) {
1097 if ((XFS_BUF_TARGET(bp) == target) &&
1098 (XFS_BUF_ADDR(bp) == blkno) &&
1099 (XFS_BUF_COUNT(bp) == len)) {
1101 * We found it. Break out and
1102 * return the pointer to the buffer.