2 * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
28 * For further information regarding this notice, see:
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
38 #include "xfs_trans.h"
39 #include "xfs_buf_item.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_trans_priv.h"
46 #include "xfs_error.h"
50 STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *,
52 STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *,
57 * Get and lock the buffer for the caller if it is not already
58 * locked within the given transaction. If it is already locked
59 * within the transaction, just increment its lock recursion count
60 * and return a pointer to it.
62 * Use the fast path function xfs_trans_buf_item_match() or the buffer
63 * cache routine incore_match() to find the buffer
64 * if it is already owned by this transaction.
66 * If we don't already own the buffer, use get_buf() to get it.
67 * If it doesn't yet have an associated xfs_buf_log_item structure,
68 * then allocate one and add the item to this transaction.
70 * If the transaction pointer is NULL, make this just a normal
74 xfs_trans_get_buf(xfs_trans_t *tp,
75 xfs_buftarg_t *target_dev,
81 xfs_buf_log_item_t *bip;
84 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
87 * Default to a normal get_buf() call if the tp is NULL.
90 bp = xfs_buf_get_flags(target_dev, blkno, len,
96 * If we find the buffer in the cache with this transaction
97 * pointer in its b_fsprivate2 field, then we know we already
98 * have it locked. In this case we just increment the lock
99 * recursion count and return the buffer to the caller.
101 if (tp->t_items.lic_next == NULL) {
102 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
104 bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
107 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
108 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
109 xfs_buftrace("TRANS GET RECUR SHUT", bp);
110 XFS_BUF_SUPER_STALE(bp);
113 * If the buffer is stale then it was binval'ed
114 * since last read. This doesn't matter since the
115 * caller isn't allowed to use the data anyway.
117 else if (XFS_BUF_ISSTALE(bp)) {
118 xfs_buftrace("TRANS GET RECUR STALE", bp);
119 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
121 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
122 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
124 ASSERT(atomic_read(&bip->bli_refcount) > 0);
126 xfs_buftrace("TRANS GET RECUR", bp);
127 xfs_buf_item_trace("GET RECUR", bip);
132 * We always specify the BUF_BUSY flag within a transaction so
133 * that get_buf does not try to push out a delayed write buffer
134 * which might cause another transaction to take place (if the
135 * buffer was delayed alloc). Such recursive transactions can
136 * easily deadlock with our current transaction as well as cause
137 * us to run out of stack space.
139 bp = xfs_buf_get_flags(target_dev, blkno, len, flags | BUF_BUSY);
144 ASSERT(!XFS_BUF_GETERROR(bp));
147 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
148 * it doesn't have one yet, then allocate one and initialize it.
149 * The checks to see if one is there are in xfs_buf_item_init().
151 xfs_buf_item_init(bp, tp->t_mountp);
154 * Set the recursion count for the buffer within this transaction
157 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
158 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
159 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
160 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
164 * Take a reference for this transaction on the buf item.
166 atomic_inc(&bip->bli_refcount);
169 * Get a log_item_desc to point at the new item.
171 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
174 * Initialize b_fsprivate2 so we can find it with incore_match()
177 XFS_BUF_SET_FSPRIVATE2(bp, tp);
179 xfs_buftrace("TRANS GET", bp);
180 xfs_buf_item_trace("GET", bip);
185 * Get and lock the superblock buffer of this file system for the
188 * We don't need to use incore_match() here, because the superblock
189 * buffer is a private buffer which we keep a pointer to in the
193 xfs_trans_getsb(xfs_trans_t *tp,
194 struct xfs_mount *mp,
198 xfs_buf_log_item_t *bip;
201 * Default to just trying to lock the superblock buffer
205 return (xfs_getsb(mp, flags));
209 * If the superblock buffer already has this transaction
210 * pointer in its b_fsprivate2 field, then we know we already
211 * have it locked. In this case we just increment the lock
212 * recursion count and return the buffer to the caller.
215 if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) {
216 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
218 ASSERT(atomic_read(&bip->bli_refcount) > 0);
220 xfs_buf_item_trace("GETSB RECUR", bip);
224 bp = xfs_getsb(mp, flags);
230 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
231 * it doesn't have one yet, then allocate one and initialize it.
232 * The checks to see if one is there are in xfs_buf_item_init().
234 xfs_buf_item_init(bp, mp);
237 * Set the recursion count for the buffer within this transaction
240 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
241 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
242 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
243 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
247 * Take a reference for this transaction on the buf item.
249 atomic_inc(&bip->bli_refcount);
252 * Get a log_item_desc to point at the new item.
254 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
257 * Initialize b_fsprivate2 so we can find it with incore_match()
260 XFS_BUF_SET_FSPRIVATE2(bp, tp);
262 xfs_buf_item_trace("GETSB", bip);
267 xfs_buftarg_t *xfs_error_target;
270 int xfs_error_mod = 33;
274 * Get and lock the buffer for the caller if it is not already
275 * locked within the given transaction. If it has not yet been
276 * read in, read it from disk. If it is already locked
277 * within the transaction and already read in, just increment its
278 * lock recursion count and return a pointer to it.
280 * Use the fast path function xfs_trans_buf_item_match() or the buffer
281 * cache routine incore_match() to find the buffer
282 * if it is already owned by this transaction.
284 * If we don't already own the buffer, use read_buf() to get it.
285 * If it doesn't yet have an associated xfs_buf_log_item structure,
286 * then allocate one and add the item to this transaction.
288 * If the transaction pointer is NULL, make this just a normal
295 xfs_buftarg_t *target,
302 xfs_buf_log_item_t *bip;
306 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
309 * Default to a normal get_buf() call if the tp is NULL.
312 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
314 return XFS_ERROR(ENOMEM);
316 if ((bp != NULL) && (XFS_BUF_GETERROR(bp) != 0)) {
317 xfs_ioerror_alert("xfs_trans_read_buf", mp,
319 error = XFS_BUF_GETERROR(bp);
324 if (xfs_do_error && (bp != NULL)) {
325 if (xfs_error_target == target) {
326 if (((xfs_req_num++) % xfs_error_mod) == 0) {
328 printk("Returning error!\n");
329 return XFS_ERROR(EIO);
334 if (XFS_FORCED_SHUTDOWN(mp))
341 * If we find the buffer in the cache with this transaction
342 * pointer in its b_fsprivate2 field, then we know we already
343 * have it locked. If it is already read in we just increment
344 * the lock recursion count and return the buffer to the caller.
345 * If the buffer is not yet read in, then we read it in, increment
346 * the lock recursion count, and return it to the caller.
348 if (tp->t_items.lic_next == NULL) {
349 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
351 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
354 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
355 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
356 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
357 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
358 if (!(XFS_BUF_ISDONE(bp))) {
359 xfs_buftrace("READ_BUF_INCORE !DONE", bp);
360 ASSERT(!XFS_BUF_ISASYNC(bp));
362 xfsbdstrat(tp->t_mountp, bp);
364 if (XFS_BUF_GETERROR(bp) != 0) {
365 xfs_ioerror_alert("xfs_trans_read_buf", mp,
367 error = XFS_BUF_GETERROR(bp);
370 * We can gracefully recover from most
371 * read errors. Ones we can't are those
372 * that happen after the transaction's
375 if (tp->t_flags & XFS_TRANS_DIRTY)
376 xfs_force_shutdown(tp->t_mountp,
377 XFS_METADATA_IO_ERROR);
382 * We never locked this buf ourselves, so we shouldn't
383 * brelse it either. Just get out.
385 if (XFS_FORCED_SHUTDOWN(mp)) {
386 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp);
388 return XFS_ERROR(EIO);
392 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
395 ASSERT(atomic_read(&bip->bli_refcount) > 0);
396 xfs_buf_item_trace("READ RECUR", bip);
402 * We always specify the BUF_BUSY flag within a transaction so
403 * that get_buf does not try to push out a delayed write buffer
404 * which might cause another transaction to take place (if the
405 * buffer was delayed alloc). Such recursive transactions can
406 * easily deadlock with our current transaction as well as cause
407 * us to run out of stack space.
409 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
414 if (XFS_BUF_GETERROR(bp) != 0) {
415 XFS_BUF_SUPER_STALE(bp);
416 xfs_buftrace("READ ERROR", bp);
417 error = XFS_BUF_GETERROR(bp);
419 xfs_ioerror_alert("xfs_trans_read_buf", mp,
421 if (tp->t_flags & XFS_TRANS_DIRTY)
422 xfs_force_shutdown(tp->t_mountp, XFS_METADATA_IO_ERROR);
427 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
428 if (xfs_error_target == target) {
429 if (((xfs_req_num++) % xfs_error_mod) == 0) {
430 xfs_force_shutdown(tp->t_mountp,
431 XFS_METADATA_IO_ERROR);
433 printk("Returning error in trans!\n");
434 return XFS_ERROR(EIO);
439 if (XFS_FORCED_SHUTDOWN(mp))
443 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
444 * it doesn't have one yet, then allocate one and initialize it.
445 * The checks to see if one is there are in xfs_buf_item_init().
447 xfs_buf_item_init(bp, tp->t_mountp);
450 * Set the recursion count for the buffer within this transaction
453 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
454 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
455 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
456 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
460 * Take a reference for this transaction on the buf item.
462 atomic_inc(&bip->bli_refcount);
465 * Get a log_item_desc to point at the new item.
467 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
470 * Initialize b_fsprivate2 so we can find it with incore_match()
473 XFS_BUF_SET_FSPRIVATE2(bp, tp);
475 xfs_buftrace("TRANS READ", bp);
476 xfs_buf_item_trace("READ", bip);
482 * the theory here is that buffer is good but we're
483 * bailing out because the filesystem is being forcibly
484 * shut down. So we should leave the b_flags alone since
485 * the buffer's not staled and just get out.
488 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
489 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
491 ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) !=
492 (XFS_B_STALE|XFS_B_DELWRI));
494 xfs_buftrace("READ_BUF XFSSHUTDN", bp);
497 return XFS_ERROR(EIO);
502 * Release the buffer bp which was previously acquired with one of the
503 * xfs_trans_... buffer allocation routines if the buffer has not
504 * been modified within this transaction. If the buffer is modified
505 * within this transaction, do decrement the recursion count but do
506 * not release the buffer even if the count goes to 0. If the buffer is not
507 * modified within the transaction, decrement the recursion count and
508 * release the buffer if the recursion count goes to 0.
510 * If the buffer is to be released and it was not modified before
511 * this transaction began, then free the buf_log_item associated with it.
513 * If the transaction pointer is NULL, make this just a normal
517 xfs_trans_brelse(xfs_trans_t *tp,
520 xfs_buf_log_item_t *bip;
522 xfs_log_item_desc_t *lidp;
525 * Default to a normal brelse() call if the tp is NULL.
528 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
530 * If there's a buf log item attached to the buffer,
531 * then let the AIL know that the buffer is being
534 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
535 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
536 if (lip->li_type == XFS_LI_BUF) {
537 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
538 xfs_trans_unlocked_item(
539 bip->bli_item.li_mountp,
547 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
548 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
549 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
550 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
551 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
552 ASSERT(atomic_read(&bip->bli_refcount) > 0);
555 * Find the item descriptor pointing to this buffer's
556 * log item. It must be there.
558 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
559 ASSERT(lidp != NULL);
562 * If the release is just for a recursive lock,
563 * then decrement the count and return.
565 if (bip->bli_recur > 0) {
567 xfs_buf_item_trace("RELSE RECUR", bip);
572 * If the buffer is dirty within this transaction, we can't
573 * release it until we commit.
575 if (lidp->lid_flags & XFS_LID_DIRTY) {
576 xfs_buf_item_trace("RELSE DIRTY", bip);
581 * If the buffer has been invalidated, then we can't release
582 * it until the transaction commits to disk unless it is re-dirtied
583 * as part of this transaction. This prevents us from pulling
584 * the item from the AIL before we should.
586 if (bip->bli_flags & XFS_BLI_STALE) {
587 xfs_buf_item_trace("RELSE STALE", bip);
591 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
592 xfs_buf_item_trace("RELSE", bip);
595 * Free up the log item descriptor tracking the released item.
597 xfs_trans_free_item(tp, lidp);
600 * Clear the hold flag in the buf log item if it is set.
601 * We wouldn't want the next user of the buffer to
604 if (bip->bli_flags & XFS_BLI_HOLD) {
605 bip->bli_flags &= ~XFS_BLI_HOLD;
609 * Drop our reference to the buf log item.
611 atomic_dec(&bip->bli_refcount);
614 * If the buf item is not tracking data in the log, then
615 * we must free it before releasing the buffer back to the
616 * free pool. Before releasing the buffer to the free pool,
617 * clear the transaction pointer in b_fsprivate2 to dissolve
618 * its relation to this transaction.
620 if (!xfs_buf_item_dirty(bip)) {
622 ASSERT(bp->b_pincount == 0);
624 ASSERT(atomic_read(&bip->bli_refcount) == 0);
625 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
626 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
627 xfs_buf_item_relse(bp);
630 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
633 * If we've still got a buf log item on the buffer, then
634 * tell the AIL that the buffer is being unlocked.
637 xfs_trans_unlocked_item(bip->bli_item.li_mountp,
638 (xfs_log_item_t*)bip);
646 * Add the locked buffer to the transaction.
647 * The buffer must be locked, and it cannot be associated with any
650 * If the buffer does not yet have a buf log item associated with it,
651 * then allocate one for it. Then add the buf item to the transaction.
654 xfs_trans_bjoin(xfs_trans_t *tp,
657 xfs_buf_log_item_t *bip;
659 ASSERT(XFS_BUF_ISBUSY(bp));
660 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
663 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
664 * it doesn't have one yet, then allocate one and initialize it.
665 * The checks to see if one is there are in xfs_buf_item_init().
667 xfs_buf_item_init(bp, tp->t_mountp);
668 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
669 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
670 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
671 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
674 * Take a reference for this transaction on the buf item.
676 atomic_inc(&bip->bli_refcount);
679 * Get a log_item_desc to point at the new item.
681 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
684 * Initialize b_fsprivate2 so we can find it with incore_match()
685 * in xfs_trans_get_buf() and friends above.
687 XFS_BUF_SET_FSPRIVATE2(bp, tp);
689 xfs_buf_item_trace("BJOIN", bip);
693 * Mark the buffer as not needing to be unlocked when the buf item's
694 * IOP_UNLOCK() routine is called. The buffer must already be locked
695 * and associated with the given transaction.
699 xfs_trans_bhold(xfs_trans_t *tp,
702 xfs_buf_log_item_t *bip;
704 ASSERT(XFS_BUF_ISBUSY(bp));
705 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
706 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
708 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
709 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
710 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
711 ASSERT(atomic_read(&bip->bli_refcount) > 0);
712 bip->bli_flags |= XFS_BLI_HOLD;
713 xfs_buf_item_trace("BHOLD", bip);
717 * This is called to mark bytes first through last inclusive of the given
718 * buffer as needing to be logged when the transaction is committed.
719 * The buffer must already be associated with the given transaction.
721 * First and last are numbers relative to the beginning of this buffer,
722 * so the first byte in the buffer is numbered 0 regardless of the
726 xfs_trans_log_buf(xfs_trans_t *tp,
731 xfs_buf_log_item_t *bip;
732 xfs_log_item_desc_t *lidp;
734 ASSERT(XFS_BUF_ISBUSY(bp));
735 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
736 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
737 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
738 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
739 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
742 * Mark the buffer as needing to be written out eventually,
743 * and set its iodone function to remove the buffer's buf log
744 * item from the AIL and free it when the buffer is flushed
745 * to disk. See xfs_buf_attach_iodone() for more details
746 * on li_cb and xfs_buf_iodone_callbacks().
747 * If we end up aborting this transaction, we trap this buffer
748 * inside the b_bdstrat callback so that this won't get written to
751 XFS_BUF_DELAYWRITE(bp);
754 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
755 ASSERT(atomic_read(&bip->bli_refcount) > 0);
756 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
757 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
760 * If we invalidated the buffer within this transaction, then
761 * cancel the invalidation now that we're dirtying the buffer
762 * again. There are no races with the code in xfs_buf_item_unpin(),
763 * because we have a reference to the buffer this entire time.
765 if (bip->bli_flags & XFS_BLI_STALE) {
766 xfs_buf_item_trace("BLOG UNSTALE", bip);
767 bip->bli_flags &= ~XFS_BLI_STALE;
768 ASSERT(XFS_BUF_ISSTALE(bp));
770 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
773 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
774 ASSERT(lidp != NULL);
776 tp->t_flags |= XFS_TRANS_DIRTY;
777 lidp->lid_flags |= XFS_LID_DIRTY;
778 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
779 bip->bli_flags |= XFS_BLI_LOGGED;
780 xfs_buf_item_log(bip, first, last);
781 xfs_buf_item_trace("BLOG", bip);
786 * This called to invalidate a buffer that is being used within
787 * a transaction. Typically this is because the blocks in the
788 * buffer are being freed, so we need to prevent it from being
789 * written out when we're done. Allowing it to be written again
790 * might overwrite data in the free blocks if they are reallocated
793 * We prevent the buffer from being written out by clearing the
794 * B_DELWRI flag. We can't always
795 * get rid of the buf log item at this point, though, because
796 * the buffer may still be pinned by another transaction. If that
797 * is the case, then we'll wait until the buffer is committed to
798 * disk for the last time (we can tell by the ref count) and
799 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
800 * will keep the buffer locked so that the buffer and buf log item
808 xfs_log_item_desc_t *lidp;
809 xfs_buf_log_item_t *bip;
811 ASSERT(XFS_BUF_ISBUSY(bp));
812 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
813 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
815 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
816 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
817 ASSERT(lidp != NULL);
818 ASSERT(atomic_read(&bip->bli_refcount) > 0);
820 if (bip->bli_flags & XFS_BLI_STALE) {
822 * If the buffer is already invalidated, then
825 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
826 ASSERT(XFS_BUF_ISSTALE(bp));
827 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
828 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
829 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
830 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
831 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
832 xfs_buftrace("XFS_BINVAL RECUR", bp);
833 xfs_buf_item_trace("BINVAL RECUR", bip);
838 * Clear the dirty bit in the buffer and set the STALE flag
839 * in the buf log item. The STALE flag will be used in
840 * xfs_buf_item_unpin() to determine if it should clean up
841 * when the last reference to the buf item is given up.
842 * We set the XFS_BLI_CANCEL flag in the buf log format structure
843 * and log the buf item. This will be used at recovery time
844 * to determine that copies of the buffer in the log before
845 * this should not be replayed.
846 * We mark the item descriptor and the transaction dirty so
847 * that we'll hold the buffer until after the commit.
849 * Since we're invalidating the buffer, we also clear the state
850 * about which parts of the buffer have been logged. We also
851 * clear the flag indicating that this is an inode buffer since
852 * the data in the buffer will no longer be valid.
854 * We set the stale bit in the buffer as well since we're getting
857 XFS_BUF_UNDELAYWRITE(bp);
859 bip->bli_flags |= XFS_BLI_STALE;
860 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
861 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
862 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
863 memset((char *)(bip->bli_format.blf_data_map), 0,
864 (bip->bli_format.blf_map_size * sizeof(uint)));
865 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
866 tp->t_flags |= XFS_TRANS_DIRTY;
867 xfs_buftrace("XFS_BINVAL", bp);
868 xfs_buf_item_trace("BINVAL", bip);
872 * This call is used to indicate that the buffer contains on-disk
873 * inodes which must be handled specially during recovery. They
874 * require special handling because only the di_next_unlinked from
875 * the inodes in the buffer should be recovered. The rest of the
876 * data in the buffer is logged via the inodes themselves.
878 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
879 * format structure so that we'll know what to do at recovery time.
887 xfs_buf_log_item_t *bip;
889 ASSERT(XFS_BUF_ISBUSY(bp));
890 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
891 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
893 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
894 ASSERT(atomic_read(&bip->bli_refcount) > 0);
896 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
900 * This call is used to indicate that the buffer is going to
901 * be staled and was an inode buffer. This means it gets
902 * special processing during unpin - where any inodes
903 * associated with the buffer should be removed from ail.
904 * There is also special processing during recovery,
905 * any replay of the inodes in the buffer needs to be
906 * prevented as the buffer may have been reused.
909 xfs_trans_stale_inode_buf(
913 xfs_buf_log_item_t *bip;
915 ASSERT(XFS_BUF_ISBUSY(bp));
916 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
917 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
919 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
920 ASSERT(atomic_read(&bip->bli_refcount) > 0);
922 bip->bli_flags |= XFS_BLI_STALE_INODE;
923 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
930 * Mark the buffer as being one which contains newly allocated
931 * inodes. We need to make sure that even if this buffer is
932 * relogged as an 'inode buf' we still recover all of the inode
933 * images in the face of a crash. This works in coordination with
934 * xfs_buf_item_committed() to ensure that the buffer remains in the
935 * AIL at its original location even after it has been relogged.
939 xfs_trans_inode_alloc_buf(
943 xfs_buf_log_item_t *bip;
945 ASSERT(XFS_BUF_ISBUSY(bp));
946 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
947 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
949 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
950 ASSERT(atomic_read(&bip->bli_refcount) > 0);
952 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
957 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
958 * dquots. However, unlike in inode buffer recovery, dquot buffers get
959 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
960 * The only thing that makes dquot buffers different from regular
961 * buffers is that we must not replay dquot bufs when recovering
962 * if a _corresponding_ quotaoff has happened. We also have to distinguish
963 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
964 * can be turned off independently.
973 xfs_buf_log_item_t *bip;
975 ASSERT(XFS_BUF_ISBUSY(bp));
976 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
977 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
978 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
979 type == XFS_BLI_GDQUOT_BUF);
981 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
982 ASSERT(atomic_read(&bip->bli_refcount) > 0);
984 bip->bli_format.blf_flags |= type;
988 * Check to see if a buffer matching the given parameters is already
989 * a part of the given transaction. Only check the first, embedded
990 * chunk, since we don't want to spend all day scanning large transactions.
993 xfs_trans_buf_item_match(
995 xfs_buftarg_t *target,
999 xfs_log_item_chunk_t *licp;
1000 xfs_log_item_desc_t *lidp;
1001 xfs_buf_log_item_t *blip;
1007 licp = &tp->t_items;
1008 if (!XFS_LIC_ARE_ALL_FREE(licp)) {
1009 for (i = 0; i < licp->lic_unused; i++) {
1011 * Skip unoccupied slots.
1013 if (XFS_LIC_ISFREE(licp, i)) {
1017 lidp = XFS_LIC_SLOT(licp, i);
1018 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1019 if (blip->bli_item.li_type != XFS_LI_BUF) {
1024 if ((XFS_BUF_TARGET(bp) == target) &&
1025 (XFS_BUF_ADDR(bp) == blkno) &&
1026 (XFS_BUF_COUNT(bp) == len)) {
1028 * We found it. Break out and
1029 * return the pointer to the buffer.
1041 * Check to see if a buffer matching the given parameters is already
1042 * a part of the given transaction. Check all the chunks, we
1043 * want to be thorough.
1046 xfs_trans_buf_item_match_all(
1048 xfs_buftarg_t *target,
1052 xfs_log_item_chunk_t *licp;
1053 xfs_log_item_desc_t *lidp;
1054 xfs_buf_log_item_t *blip;
1060 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
1061 if (XFS_LIC_ARE_ALL_FREE(licp)) {
1062 ASSERT(licp == &tp->t_items);
1063 ASSERT(licp->lic_next == NULL);
1066 for (i = 0; i < licp->lic_unused; i++) {
1068 * Skip unoccupied slots.
1070 if (XFS_LIC_ISFREE(licp, i)) {
1074 lidp = XFS_LIC_SLOT(licp, i);
1075 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1076 if (blip->bli_item.li_type != XFS_LI_BUF) {
1081 if ((XFS_BUF_TARGET(bp) == target) &&
1082 (XFS_BUF_ADDR(bp) == blkno) &&
1083 (XFS_BUF_COUNT(bp) == len)) {
1085 * We found it. Break out and
1086 * return the pointer to the buffer.