]>
git.proxmox.com Git - mirror_ubuntu-kernels.git/blob - fs/xfs/xfs_trans_buf.c
4e5c010f50406940a5b4758e300530a6922bae63
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
);
358 if (XFS_BUF_GETERROR(bp
) != 0) {
359 xfs_ioerror_alert("xfs_trans_read_buf", mp
,
361 error
= XFS_BUF_GETERROR(bp
);
364 * We can gracefully recover from most
365 * read errors. Ones we can't are those
366 * that happen after the transaction's
369 if (tp
->t_flags
& XFS_TRANS_DIRTY
)
370 xfs_force_shutdown(tp
->t_mountp
,
371 SHUTDOWN_META_IO_ERROR
);
376 * We never locked this buf ourselves, so we shouldn't
377 * brelse it either. Just get out.
379 if (XFS_FORCED_SHUTDOWN(mp
)) {
380 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp
);
382 return XFS_ERROR(EIO
);
386 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
389 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
390 xfs_buf_item_trace("READ RECUR", bip
);
396 * We always specify the BUF_BUSY flag within a transaction so
397 * that get_buf does not try to push out a delayed write buffer
398 * which might cause another transaction to take place (if the
399 * buffer was delayed alloc). Such recursive transactions can
400 * easily deadlock with our current transaction as well as cause
401 * us to run out of stack space.
403 bp
= xfs_buf_read_flags(target
, blkno
, len
, flags
| BUF_BUSY
);
408 if (XFS_BUF_GETERROR(bp
) != 0) {
409 XFS_BUF_SUPER_STALE(bp
);
410 xfs_buftrace("READ ERROR", bp
);
411 error
= XFS_BUF_GETERROR(bp
);
413 xfs_ioerror_alert("xfs_trans_read_buf", mp
,
415 if (tp
->t_flags
& XFS_TRANS_DIRTY
)
416 xfs_force_shutdown(tp
->t_mountp
, SHUTDOWN_META_IO_ERROR
);
421 if (xfs_do_error
&& !(tp
->t_flags
& XFS_TRANS_DIRTY
)) {
422 if (xfs_error_target
== target
) {
423 if (((xfs_req_num
++) % xfs_error_mod
) == 0) {
424 xfs_force_shutdown(tp
->t_mountp
,
425 SHUTDOWN_META_IO_ERROR
);
427 cmn_err(CE_DEBUG
, "Returning trans error!\n");
428 return XFS_ERROR(EIO
);
433 if (XFS_FORCED_SHUTDOWN(mp
))
437 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
438 * it doesn't have one yet, then allocate one and initialize it.
439 * The checks to see if one is there are in xfs_buf_item_init().
441 xfs_buf_item_init(bp
, tp
->t_mountp
);
444 * Set the recursion count for the buffer within this transaction
447 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
448 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
449 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
));
450 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
454 * Take a reference for this transaction on the buf item.
456 atomic_inc(&bip
->bli_refcount
);
459 * Get a log_item_desc to point at the new item.
461 (void) xfs_trans_add_item(tp
, (xfs_log_item_t
*)bip
);
464 * Initialize b_fsprivate2 so we can find it with incore_match()
467 XFS_BUF_SET_FSPRIVATE2(bp
, tp
);
469 xfs_buftrace("TRANS READ", bp
);
470 xfs_buf_item_trace("READ", bip
);
476 * the theory here is that buffer is good but we're
477 * bailing out because the filesystem is being forcibly
478 * shut down. So we should leave the b_flags alone since
479 * the buffer's not staled and just get out.
482 if (XFS_BUF_ISSTALE(bp
) && XFS_BUF_ISDELAYWRITE(bp
))
483 cmn_err(CE_NOTE
, "about to pop assert, bp == 0x%p", bp
);
485 ASSERT((XFS_BUF_BFLAGS(bp
) & (XFS_B_STALE
|XFS_B_DELWRI
)) !=
486 (XFS_B_STALE
|XFS_B_DELWRI
));
488 xfs_buftrace("READ_BUF XFSSHUTDN", bp
);
491 return XFS_ERROR(EIO
);
496 * Release the buffer bp which was previously acquired with one of the
497 * xfs_trans_... buffer allocation routines if the buffer has not
498 * been modified within this transaction. If the buffer is modified
499 * within this transaction, do decrement the recursion count but do
500 * not release the buffer even if the count goes to 0. If the buffer is not
501 * modified within the transaction, decrement the recursion count and
502 * release the buffer if the recursion count goes to 0.
504 * If the buffer is to be released and it was not modified before
505 * this transaction began, then free the buf_log_item associated with it.
507 * If the transaction pointer is NULL, make this just a normal
511 xfs_trans_brelse(xfs_trans_t
*tp
,
514 xfs_buf_log_item_t
*bip
;
516 xfs_log_item_desc_t
*lidp
;
519 * Default to a normal brelse() call if the tp is NULL.
522 ASSERT(XFS_BUF_FSPRIVATE2(bp
, void *) == NULL
);
524 * If there's a buf log item attached to the buffer,
525 * then let the AIL know that the buffer is being
528 if (XFS_BUF_FSPRIVATE(bp
, void *) != NULL
) {
529 lip
= XFS_BUF_FSPRIVATE(bp
, xfs_log_item_t
*);
530 if (lip
->li_type
== XFS_LI_BUF
) {
531 bip
= XFS_BUF_FSPRIVATE(bp
,xfs_buf_log_item_t
*);
532 xfs_trans_unlocked_item(
533 bip
->bli_item
.li_mountp
,
541 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
542 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
543 ASSERT(bip
->bli_item
.li_type
== XFS_LI_BUF
);
544 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
545 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
));
546 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
549 * Find the item descriptor pointing to this buffer's
550 * log item. It must be there.
552 lidp
= xfs_trans_find_item(tp
, (xfs_log_item_t
*)bip
);
553 ASSERT(lidp
!= NULL
);
556 * If the release is just for a recursive lock,
557 * then decrement the count and return.
559 if (bip
->bli_recur
> 0) {
561 xfs_buf_item_trace("RELSE RECUR", bip
);
566 * If the buffer is dirty within this transaction, we can't
567 * release it until we commit.
569 if (lidp
->lid_flags
& XFS_LID_DIRTY
) {
570 xfs_buf_item_trace("RELSE DIRTY", bip
);
575 * If the buffer has been invalidated, then we can't release
576 * it until the transaction commits to disk unless it is re-dirtied
577 * as part of this transaction. This prevents us from pulling
578 * the item from the AIL before we should.
580 if (bip
->bli_flags
& XFS_BLI_STALE
) {
581 xfs_buf_item_trace("RELSE STALE", bip
);
585 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
586 xfs_buf_item_trace("RELSE", bip
);
589 * Free up the log item descriptor tracking the released item.
591 xfs_trans_free_item(tp
, lidp
);
594 * Clear the hold flag in the buf log item if it is set.
595 * We wouldn't want the next user of the buffer to
598 if (bip
->bli_flags
& XFS_BLI_HOLD
) {
599 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
603 * Drop our reference to the buf log item.
605 atomic_dec(&bip
->bli_refcount
);
608 * If the buf item is not tracking data in the log, then
609 * we must free it before releasing the buffer back to the
610 * free pool. Before releasing the buffer to the free pool,
611 * clear the transaction pointer in b_fsprivate2 to dissolve
612 * its relation to this transaction.
614 if (!xfs_buf_item_dirty(bip
)) {
616 ASSERT(bp->b_pincount == 0);
618 ASSERT(atomic_read(&bip
->bli_refcount
) == 0);
619 ASSERT(!(bip
->bli_item
.li_flags
& XFS_LI_IN_AIL
));
620 ASSERT(!(bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
));
621 xfs_buf_item_relse(bp
);
624 XFS_BUF_SET_FSPRIVATE2(bp
, NULL
);
627 * If we've still got a buf log item on the buffer, then
628 * tell the AIL that the buffer is being unlocked.
631 xfs_trans_unlocked_item(bip
->bli_item
.li_mountp
,
632 (xfs_log_item_t
*)bip
);
640 * Add the locked buffer to the transaction.
641 * The buffer must be locked, and it cannot be associated with any
644 * If the buffer does not yet have a buf log item associated with it,
645 * then allocate one for it. Then add the buf item to the transaction.
648 xfs_trans_bjoin(xfs_trans_t
*tp
,
651 xfs_buf_log_item_t
*bip
;
653 ASSERT(XFS_BUF_ISBUSY(bp
));
654 ASSERT(XFS_BUF_FSPRIVATE2(bp
, void *) == NULL
);
657 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
658 * it doesn't have one yet, then allocate one and initialize it.
659 * The checks to see if one is there are in xfs_buf_item_init().
661 xfs_buf_item_init(bp
, tp
->t_mountp
);
662 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
663 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
664 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
));
665 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
668 * Take a reference for this transaction on the buf item.
670 atomic_inc(&bip
->bli_refcount
);
673 * Get a log_item_desc to point at the new item.
675 (void) xfs_trans_add_item(tp
, (xfs_log_item_t
*)bip
);
678 * Initialize b_fsprivate2 so we can find it with incore_match()
679 * in xfs_trans_get_buf() and friends above.
681 XFS_BUF_SET_FSPRIVATE2(bp
, tp
);
683 xfs_buf_item_trace("BJOIN", bip
);
687 * Mark the buffer as not needing to be unlocked when the buf item's
688 * IOP_UNLOCK() routine is called. The buffer must already be locked
689 * and associated with the given transaction.
693 xfs_trans_bhold(xfs_trans_t
*tp
,
696 xfs_buf_log_item_t
*bip
;
698 ASSERT(XFS_BUF_ISBUSY(bp
));
699 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
700 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
702 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
703 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
704 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
));
705 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
706 bip
->bli_flags
|= XFS_BLI_HOLD
;
707 xfs_buf_item_trace("BHOLD", bip
);
711 * Cancel the previous buffer hold request made on this buffer
712 * for this transaction.
715 xfs_trans_bhold_release(xfs_trans_t
*tp
,
718 xfs_buf_log_item_t
*bip
;
720 ASSERT(XFS_BUF_ISBUSY(bp
));
721 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
722 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
724 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
725 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
726 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
));
727 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
728 ASSERT(bip
->bli_flags
& XFS_BLI_HOLD
);
729 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
730 xfs_buf_item_trace("BHOLD RELEASE", bip
);
734 * This is called to mark bytes first through last inclusive of the given
735 * buffer as needing to be logged when the transaction is committed.
736 * The buffer must already be associated with the given transaction.
738 * First and last are numbers relative to the beginning of this buffer,
739 * so the first byte in the buffer is numbered 0 regardless of the
743 xfs_trans_log_buf(xfs_trans_t
*tp
,
748 xfs_buf_log_item_t
*bip
;
749 xfs_log_item_desc_t
*lidp
;
751 ASSERT(XFS_BUF_ISBUSY(bp
));
752 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
753 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
754 ASSERT((first
<= last
) && (last
< XFS_BUF_COUNT(bp
)));
755 ASSERT((XFS_BUF_IODONE_FUNC(bp
) == NULL
) ||
756 (XFS_BUF_IODONE_FUNC(bp
) == xfs_buf_iodone_callbacks
));
759 * Mark the buffer as needing to be written out eventually,
760 * and set its iodone function to remove the buffer's buf log
761 * item from the AIL and free it when the buffer is flushed
762 * to disk. See xfs_buf_attach_iodone() for more details
763 * on li_cb and xfs_buf_iodone_callbacks().
764 * If we end up aborting this transaction, we trap this buffer
765 * inside the b_bdstrat callback so that this won't get written to
768 XFS_BUF_DELAYWRITE(bp
);
771 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
772 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
773 XFS_BUF_SET_IODONE_FUNC(bp
, xfs_buf_iodone_callbacks
);
774 bip
->bli_item
.li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*))xfs_buf_iodone
;
777 * If we invalidated the buffer within this transaction, then
778 * cancel the invalidation now that we're dirtying the buffer
779 * again. There are no races with the code in xfs_buf_item_unpin(),
780 * because we have a reference to the buffer this entire time.
782 if (bip
->bli_flags
& XFS_BLI_STALE
) {
783 xfs_buf_item_trace("BLOG UNSTALE", bip
);
784 bip
->bli_flags
&= ~XFS_BLI_STALE
;
785 ASSERT(XFS_BUF_ISSTALE(bp
));
787 bip
->bli_format
.blf_flags
&= ~XFS_BLI_CANCEL
;
790 lidp
= xfs_trans_find_item(tp
, (xfs_log_item_t
*)bip
);
791 ASSERT(lidp
!= NULL
);
793 tp
->t_flags
|= XFS_TRANS_DIRTY
;
794 lidp
->lid_flags
|= XFS_LID_DIRTY
;
795 lidp
->lid_flags
&= ~XFS_LID_BUF_STALE
;
796 bip
->bli_flags
|= XFS_BLI_LOGGED
;
797 xfs_buf_item_log(bip
, first
, last
);
798 xfs_buf_item_trace("BLOG", bip
);
803 * This called to invalidate a buffer that is being used within
804 * a transaction. Typically this is because the blocks in the
805 * buffer are being freed, so we need to prevent it from being
806 * written out when we're done. Allowing it to be written again
807 * might overwrite data in the free blocks if they are reallocated
810 * We prevent the buffer from being written out by clearing the
811 * B_DELWRI flag. We can't always
812 * get rid of the buf log item at this point, though, because
813 * the buffer may still be pinned by another transaction. If that
814 * is the case, then we'll wait until the buffer is committed to
815 * disk for the last time (we can tell by the ref count) and
816 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
817 * will keep the buffer locked so that the buffer and buf log item
825 xfs_log_item_desc_t
*lidp
;
826 xfs_buf_log_item_t
*bip
;
828 ASSERT(XFS_BUF_ISBUSY(bp
));
829 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
830 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
832 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
833 lidp
= xfs_trans_find_item(tp
, (xfs_log_item_t
*)bip
);
834 ASSERT(lidp
!= NULL
);
835 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
837 if (bip
->bli_flags
& XFS_BLI_STALE
) {
839 * If the buffer is already invalidated, then
842 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp
)));
843 ASSERT(XFS_BUF_ISSTALE(bp
));
844 ASSERT(!(bip
->bli_flags
& (XFS_BLI_LOGGED
| XFS_BLI_DIRTY
)));
845 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLI_INODE_BUF
));
846 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLI_CANCEL
);
847 ASSERT(lidp
->lid_flags
& XFS_LID_DIRTY
);
848 ASSERT(tp
->t_flags
& XFS_TRANS_DIRTY
);
849 xfs_buftrace("XFS_BINVAL RECUR", bp
);
850 xfs_buf_item_trace("BINVAL RECUR", bip
);
855 * Clear the dirty bit in the buffer and set the STALE flag
856 * in the buf log item. The STALE flag will be used in
857 * xfs_buf_item_unpin() to determine if it should clean up
858 * when the last reference to the buf item is given up.
859 * We set the XFS_BLI_CANCEL flag in the buf log format structure
860 * and log the buf item. This will be used at recovery time
861 * to determine that copies of the buffer in the log before
862 * this should not be replayed.
863 * We mark the item descriptor and the transaction dirty so
864 * that we'll hold the buffer until after the commit.
866 * Since we're invalidating the buffer, we also clear the state
867 * about which parts of the buffer have been logged. We also
868 * clear the flag indicating that this is an inode buffer since
869 * the data in the buffer will no longer be valid.
871 * We set the stale bit in the buffer as well since we're getting
874 XFS_BUF_UNDELAYWRITE(bp
);
876 bip
->bli_flags
|= XFS_BLI_STALE
;
877 bip
->bli_flags
&= ~(XFS_BLI_LOGGED
| XFS_BLI_DIRTY
);
878 bip
->bli_format
.blf_flags
&= ~XFS_BLI_INODE_BUF
;
879 bip
->bli_format
.blf_flags
|= XFS_BLI_CANCEL
;
880 memset((char *)(bip
->bli_format
.blf_data_map
), 0,
881 (bip
->bli_format
.blf_map_size
* sizeof(uint
)));
882 lidp
->lid_flags
|= XFS_LID_DIRTY
|XFS_LID_BUF_STALE
;
883 tp
->t_flags
|= XFS_TRANS_DIRTY
;
884 xfs_buftrace("XFS_BINVAL", bp
);
885 xfs_buf_item_trace("BINVAL", bip
);
889 * This call is used to indicate that the buffer contains on-disk
890 * inodes which must be handled specially during recovery. They
891 * require special handling because only the di_next_unlinked from
892 * the inodes in the buffer should be recovered. The rest of the
893 * data in the buffer is logged via the inodes themselves.
895 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
896 * format structure so that we'll know what to do at recovery time.
904 xfs_buf_log_item_t
*bip
;
906 ASSERT(XFS_BUF_ISBUSY(bp
));
907 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
908 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
910 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
911 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
913 bip
->bli_format
.blf_flags
|= XFS_BLI_INODE_BUF
;
917 * This call is used to indicate that the buffer is going to
918 * be staled and was an inode buffer. This means it gets
919 * special processing during unpin - where any inodes
920 * associated with the buffer should be removed from ail.
921 * There is also special processing during recovery,
922 * any replay of the inodes in the buffer needs to be
923 * prevented as the buffer may have been reused.
926 xfs_trans_stale_inode_buf(
930 xfs_buf_log_item_t
*bip
;
932 ASSERT(XFS_BUF_ISBUSY(bp
));
933 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
934 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
936 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
937 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
939 bip
->bli_flags
|= XFS_BLI_STALE_INODE
;
940 bip
->bli_item
.li_cb
= (void(*)(xfs_buf_t
*,xfs_log_item_t
*))
947 * Mark the buffer as being one which contains newly allocated
948 * inodes. We need to make sure that even if this buffer is
949 * relogged as an 'inode buf' we still recover all of the inode
950 * images in the face of a crash. This works in coordination with
951 * xfs_buf_item_committed() to ensure that the buffer remains in the
952 * AIL at its original location even after it has been relogged.
956 xfs_trans_inode_alloc_buf(
960 xfs_buf_log_item_t
*bip
;
962 ASSERT(XFS_BUF_ISBUSY(bp
));
963 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
964 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
966 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
967 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
969 bip
->bli_flags
|= XFS_BLI_INODE_ALLOC_BUF
;
974 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
975 * dquots. However, unlike in inode buffer recovery, dquot buffers get
976 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
977 * The only thing that makes dquot buffers different from regular
978 * buffers is that we must not replay dquot bufs when recovering
979 * if a _corresponding_ quotaoff has happened. We also have to distinguish
980 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
981 * can be turned off independently.
990 xfs_buf_log_item_t
*bip
;
992 ASSERT(XFS_BUF_ISBUSY(bp
));
993 ASSERT(XFS_BUF_FSPRIVATE2(bp
, xfs_trans_t
*) == tp
);
994 ASSERT(XFS_BUF_FSPRIVATE(bp
, void *) != NULL
);
995 ASSERT(type
== XFS_BLI_UDQUOT_BUF
||
996 type
== XFS_BLI_PDQUOT_BUF
||
997 type
== XFS_BLI_GDQUOT_BUF
);
999 bip
= XFS_BUF_FSPRIVATE(bp
, xfs_buf_log_item_t
*);
1000 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
1002 bip
->bli_format
.blf_flags
|= type
;
1006 * Check to see if a buffer matching the given parameters is already
1007 * a part of the given transaction. Only check the first, embedded
1008 * chunk, since we don't want to spend all day scanning large transactions.
1011 xfs_trans_buf_item_match(
1013 xfs_buftarg_t
*target
,
1017 xfs_log_item_chunk_t
*licp
;
1018 xfs_log_item_desc_t
*lidp
;
1019 xfs_buf_log_item_t
*blip
;
1025 licp
= &tp
->t_items
;
1026 if (!XFS_LIC_ARE_ALL_FREE(licp
)) {
1027 for (i
= 0; i
< licp
->lic_unused
; i
++) {
1029 * Skip unoccupied slots.
1031 if (XFS_LIC_ISFREE(licp
, i
)) {
1035 lidp
= XFS_LIC_SLOT(licp
, i
);
1036 blip
= (xfs_buf_log_item_t
*)lidp
->lid_item
;
1037 if (blip
->bli_item
.li_type
!= XFS_LI_BUF
) {
1042 if ((XFS_BUF_TARGET(bp
) == target
) &&
1043 (XFS_BUF_ADDR(bp
) == blkno
) &&
1044 (XFS_BUF_COUNT(bp
) == len
)) {
1046 * We found it. Break out and
1047 * return the pointer to the buffer.
1059 * Check to see if a buffer matching the given parameters is already
1060 * a part of the given transaction. Check all the chunks, we
1061 * want to be thorough.
1064 xfs_trans_buf_item_match_all(
1066 xfs_buftarg_t
*target
,
1070 xfs_log_item_chunk_t
*licp
;
1071 xfs_log_item_desc_t
*lidp
;
1072 xfs_buf_log_item_t
*blip
;
1078 for (licp
= &tp
->t_items
; licp
!= NULL
; licp
= licp
->lic_next
) {
1079 if (XFS_LIC_ARE_ALL_FREE(licp
)) {
1080 ASSERT(licp
== &tp
->t_items
);
1081 ASSERT(licp
->lic_next
== NULL
);
1084 for (i
= 0; i
< licp
->lic_unused
; i
++) {
1086 * Skip unoccupied slots.
1088 if (XFS_LIC_ISFREE(licp
, i
)) {
1092 lidp
= XFS_LIC_SLOT(licp
, i
);
1093 blip
= (xfs_buf_log_item_t
*)lidp
->lid_item
;
1094 if (blip
->bli_item
.li_type
!= XFS_LI_BUF
) {
1099 if ((XFS_BUF_TARGET(bp
) == target
) &&
1100 (XFS_BUF_ADDR(bp
) == blkno
) &&
1101 (XFS_BUF_COUNT(bp
) == len
)) {
1103 * We found it. Break out and
1104 * return the pointer to the buffer.