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"
22 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_alloc_btree.h"
28 #include "xfs_ialloc_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_buf_item.h"
32 #include "xfs_trans_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
37 * Check to see if a buffer matching the given parameters is already
38 * a part of the given transaction.
40 STATIC
struct xfs_buf
*
41 xfs_trans_buf_item_match(
43 struct xfs_buftarg
*target
,
44 struct xfs_buf_map
*map
,
47 struct xfs_log_item_desc
*lidp
;
48 struct xfs_buf_log_item
*blip
;
52 for (i
= 0; i
< nmaps
; i
++)
55 list_for_each_entry(lidp
, &tp
->t_items
, lid_trans
) {
56 blip
= (struct xfs_buf_log_item
*)lidp
->lid_item
;
57 if (blip
->bli_item
.li_type
== XFS_LI_BUF
&&
58 blip
->bli_buf
->b_target
== target
&&
59 XFS_BUF_ADDR(blip
->bli_buf
) == map
[0].bm_bn
&&
60 blip
->bli_buf
->b_length
== len
) {
61 ASSERT(blip
->bli_buf
->b_map_count
== nmaps
);
70 * Add the locked buffer to the transaction.
72 * The buffer must be locked, and it cannot be associated with any
75 * If the buffer does not yet have a buf log item associated with it,
76 * then allocate one for it. Then add the buf item to the transaction.
84 struct xfs_buf_log_item
*bip
;
86 ASSERT(bp
->b_transp
== NULL
);
89 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
90 * it doesn't have one yet, then allocate one and initialize it.
91 * The checks to see if one is there are in xfs_buf_item_init().
93 xfs_buf_item_init(bp
, tp
->t_mountp
);
95 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
96 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
));
97 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
102 * Take a reference for this transaction on the buf item.
104 atomic_inc(&bip
->bli_refcount
);
107 * Get a log_item_desc to point at the new item.
109 xfs_trans_add_item(tp
, &bip
->bli_item
);
112 * Initialize b_fsprivate2 so we can find it with incore_match()
113 * in xfs_trans_get_buf() and friends above.
121 struct xfs_trans
*tp
,
124 _xfs_trans_bjoin(tp
, bp
, 0);
125 trace_xfs_trans_bjoin(bp
->b_fspriv
);
129 * Get and lock the buffer for the caller if it is not already
130 * locked within the given transaction. If it is already locked
131 * within the transaction, just increment its lock recursion count
132 * and return a pointer to it.
134 * If the transaction pointer is NULL, make this just a normal
138 xfs_trans_get_buf_map(
139 struct xfs_trans
*tp
,
140 struct xfs_buftarg
*target
,
141 struct xfs_buf_map
*map
,
143 xfs_buf_flags_t flags
)
146 xfs_buf_log_item_t
*bip
;
149 return xfs_buf_get_map(target
, map
, nmaps
, flags
);
152 * If we find the buffer in the cache with this transaction
153 * pointer in its b_fsprivate2 field, then we know we already
154 * have it locked. In this case we just increment the lock
155 * recursion count and return the buffer to the caller.
157 bp
= xfs_trans_buf_item_match(tp
, target
, map
, nmaps
);
159 ASSERT(xfs_buf_islocked(bp
));
160 if (XFS_FORCED_SHUTDOWN(tp
->t_mountp
)) {
165 ASSERT(bp
->b_transp
== tp
);
168 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
170 trace_xfs_trans_get_buf_recur(bip
);
174 bp
= xfs_buf_get_map(target
, map
, nmaps
, flags
);
179 ASSERT(!bp
->b_error
);
181 _xfs_trans_bjoin(tp
, bp
, 1);
182 trace_xfs_trans_get_buf(bp
->b_fspriv
);
187 * Get and lock the superblock buffer of this file system for the
190 * We don't need to use incore_match() here, because the superblock
191 * buffer is a private buffer which we keep a pointer to in the
195 xfs_trans_getsb(xfs_trans_t
*tp
,
196 struct xfs_mount
*mp
,
200 xfs_buf_log_item_t
*bip
;
203 * Default to just trying to lock the superblock buffer
207 return (xfs_getsb(mp
, flags
));
211 * If the superblock buffer already has this transaction
212 * pointer in its b_fsprivate2 field, then we know we already
213 * have it locked. In this case we just increment the lock
214 * recursion count and return the buffer to the caller.
217 if (bp
->b_transp
== tp
) {
220 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
222 trace_xfs_trans_getsb_recur(bip
);
226 bp
= xfs_getsb(mp
, flags
);
230 _xfs_trans_bjoin(tp
, bp
, 1);
231 trace_xfs_trans_getsb(bp
->b_fspriv
);
236 xfs_buftarg_t
*xfs_error_target
;
239 int xfs_error_mod
= 33;
243 * Get and lock the buffer for the caller if it is not already
244 * locked within the given transaction. If it has not yet been
245 * read in, read it from disk. If it is already locked
246 * within the transaction and already read in, just increment its
247 * lock recursion count and return a pointer to it.
249 * If the transaction pointer is NULL, make this just a normal
253 xfs_trans_read_buf_map(
254 struct xfs_mount
*mp
,
255 struct xfs_trans
*tp
,
256 struct xfs_buftarg
*target
,
257 struct xfs_buf_map
*map
,
259 xfs_buf_flags_t flags
,
260 struct xfs_buf
**bpp
)
263 xfs_buf_log_item_t
*bip
;
268 bp
= xfs_buf_read_map(target
, map
, nmaps
, flags
);
270 return (flags
& XBF_TRYLOCK
) ?
271 EAGAIN
: XFS_ERROR(ENOMEM
);
275 xfs_buf_ioerror_alert(bp
, __func__
);
283 if (xfs_error_target
== target
) {
284 if (((xfs_req_num
++) % xfs_error_mod
) == 0) {
286 xfs_debug(mp
, "Returning error!");
287 return XFS_ERROR(EIO
);
292 if (XFS_FORCED_SHUTDOWN(mp
))
299 * If we find the buffer in the cache with this transaction
300 * pointer in its b_fsprivate2 field, then we know we already
301 * have it locked. If it is already read in we just increment
302 * the lock recursion count and return the buffer to the caller.
303 * If the buffer is not yet read in, then we read it in, increment
304 * the lock recursion count, and return it to the caller.
306 bp
= xfs_trans_buf_item_match(tp
, target
, map
, nmaps
);
308 ASSERT(xfs_buf_islocked(bp
));
309 ASSERT(bp
->b_transp
== tp
);
310 ASSERT(bp
->b_fspriv
!= NULL
);
311 ASSERT(!bp
->b_error
);
312 if (!(XFS_BUF_ISDONE(bp
))) {
313 trace_xfs_trans_read_buf_io(bp
, _RET_IP_
);
314 ASSERT(!XFS_BUF_ISASYNC(bp
));
316 xfsbdstrat(tp
->t_mountp
, bp
);
317 error
= xfs_buf_iowait(bp
);
319 xfs_buf_ioerror_alert(bp
, __func__
);
322 * We can gracefully recover from most read
323 * errors. Ones we can't are those that happen
324 * after the transaction's already dirty.
326 if (tp
->t_flags
& XFS_TRANS_DIRTY
)
327 xfs_force_shutdown(tp
->t_mountp
,
328 SHUTDOWN_META_IO_ERROR
);
333 * We never locked this buf ourselves, so we shouldn't
334 * brelse it either. Just get out.
336 if (XFS_FORCED_SHUTDOWN(mp
)) {
337 trace_xfs_trans_read_buf_shut(bp
, _RET_IP_
);
339 return XFS_ERROR(EIO
);
346 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
347 trace_xfs_trans_read_buf_recur(bip
);
352 bp
= xfs_buf_read_map(target
, map
, nmaps
, flags
);
355 return (flags
& XBF_TRYLOCK
) ?
356 0 : XFS_ERROR(ENOMEM
);
362 xfs_buf_ioerror_alert(bp
, __func__
);
363 if (tp
->t_flags
& XFS_TRANS_DIRTY
)
364 xfs_force_shutdown(tp
->t_mountp
, SHUTDOWN_META_IO_ERROR
);
369 if (xfs_do_error
&& !(tp
->t_flags
& XFS_TRANS_DIRTY
)) {
370 if (xfs_error_target
== target
) {
371 if (((xfs_req_num
++) % xfs_error_mod
) == 0) {
372 xfs_force_shutdown(tp
->t_mountp
,
373 SHUTDOWN_META_IO_ERROR
);
375 xfs_debug(mp
, "Returning trans error!");
376 return XFS_ERROR(EIO
);
381 if (XFS_FORCED_SHUTDOWN(mp
))
384 _xfs_trans_bjoin(tp
, bp
, 1);
385 trace_xfs_trans_read_buf(bp
->b_fspriv
);
391 trace_xfs_trans_read_buf_shut(bp
, _RET_IP_
);
394 return XFS_ERROR(EIO
);
399 * Release the buffer bp which was previously acquired with one of the
400 * xfs_trans_... buffer allocation routines if the buffer has not
401 * been modified within this transaction. If the buffer is modified
402 * within this transaction, do decrement the recursion count but do
403 * not release the buffer even if the count goes to 0. If the buffer is not
404 * modified within the transaction, decrement the recursion count and
405 * release the buffer if the recursion count goes to 0.
407 * If the buffer is to be released and it was not modified before
408 * this transaction began, then free the buf_log_item associated with it.
410 * If the transaction pointer is NULL, make this just a normal
414 xfs_trans_brelse(xfs_trans_t
*tp
,
417 xfs_buf_log_item_t
*bip
;
420 * Default to a normal brelse() call if the tp is NULL.
423 ASSERT(bp
->b_transp
== NULL
);
428 ASSERT(bp
->b_transp
== tp
);
430 ASSERT(bip
->bli_item
.li_type
== XFS_LI_BUF
);
431 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
432 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
));
433 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
435 trace_xfs_trans_brelse(bip
);
438 * If the release is just for a recursive lock,
439 * then decrement the count and return.
441 if (bip
->bli_recur
> 0) {
447 * If the buffer is dirty within this transaction, we can't
448 * release it until we commit.
450 if (bip
->bli_item
.li_desc
->lid_flags
& XFS_LID_DIRTY
)
454 * If the buffer has been invalidated, then we can't release
455 * it until the transaction commits to disk unless it is re-dirtied
456 * as part of this transaction. This prevents us from pulling
457 * the item from the AIL before we should.
459 if (bip
->bli_flags
& XFS_BLI_STALE
)
462 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
465 * Free up the log item descriptor tracking the released item.
467 xfs_trans_del_item(&bip
->bli_item
);
470 * Clear the hold flag in the buf log item if it is set.
471 * We wouldn't want the next user of the buffer to
474 if (bip
->bli_flags
& XFS_BLI_HOLD
) {
475 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
479 * Drop our reference to the buf log item.
481 atomic_dec(&bip
->bli_refcount
);
484 * If the buf item is not tracking data in the log, then
485 * we must free it before releasing the buffer back to the
486 * free pool. Before releasing the buffer to the free pool,
487 * clear the transaction pointer in b_fsprivate2 to dissolve
488 * its relation to this transaction.
490 if (!xfs_buf_item_dirty(bip
)) {
492 ASSERT(bp->b_pincount == 0);
494 ASSERT(atomic_read(&bip
->bli_refcount
) == 0);
495 ASSERT(!(bip
->bli_item
.li_flags
& XFS_LI_IN_AIL
));
496 ASSERT(!(bip
->bli_flags
& XFS_BLI_INODE_ALLOC_BUF
));
497 xfs_buf_item_relse(bp
);
505 * Mark the buffer as not needing to be unlocked when the buf item's
506 * IOP_UNLOCK() routine is called. The buffer must already be locked
507 * and associated with the given transaction.
511 xfs_trans_bhold(xfs_trans_t
*tp
,
514 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
516 ASSERT(bp
->b_transp
== tp
);
518 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
519 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
));
520 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
522 bip
->bli_flags
|= XFS_BLI_HOLD
;
523 trace_xfs_trans_bhold(bip
);
527 * Cancel the previous buffer hold request made on this buffer
528 * for this transaction.
531 xfs_trans_bhold_release(xfs_trans_t
*tp
,
534 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
536 ASSERT(bp
->b_transp
== tp
);
538 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
539 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
));
540 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
541 ASSERT(bip
->bli_flags
& XFS_BLI_HOLD
);
543 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
544 trace_xfs_trans_bhold_release(bip
);
548 * This is called to mark bytes first through last inclusive of the given
549 * buffer as needing to be logged when the transaction is committed.
550 * The buffer must already be associated with the given transaction.
552 * First and last are numbers relative to the beginning of this buffer,
553 * so the first byte in the buffer is numbered 0 regardless of the
557 xfs_trans_log_buf(xfs_trans_t
*tp
,
562 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
564 ASSERT(bp
->b_transp
== tp
);
566 ASSERT(first
<= last
&& last
< BBTOB(bp
->b_length
));
567 ASSERT(bp
->b_iodone
== NULL
||
568 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
571 * Mark the buffer as needing to be written out eventually,
572 * and set its iodone function to remove the buffer's buf log
573 * item from the AIL and free it when the buffer is flushed
574 * to disk. See xfs_buf_attach_iodone() for more details
575 * on li_cb and xfs_buf_iodone_callbacks().
576 * If we end up aborting this transaction, we trap this buffer
577 * inside the b_bdstrat callback so that this won't get written to
582 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
583 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
584 bip
->bli_item
.li_cb
= xfs_buf_iodone
;
586 trace_xfs_trans_log_buf(bip
);
589 * If we invalidated the buffer within this transaction, then
590 * cancel the invalidation now that we're dirtying the buffer
591 * again. There are no races with the code in xfs_buf_item_unpin(),
592 * because we have a reference to the buffer this entire time.
594 if (bip
->bli_flags
& XFS_BLI_STALE
) {
595 bip
->bli_flags
&= ~XFS_BLI_STALE
;
596 ASSERT(XFS_BUF_ISSTALE(bp
));
598 bip
->bli_format
.blf_flags
&= ~XFS_BLF_CANCEL
;
601 tp
->t_flags
|= XFS_TRANS_DIRTY
;
602 bip
->bli_item
.li_desc
->lid_flags
|= XFS_LID_DIRTY
;
603 bip
->bli_flags
|= XFS_BLI_LOGGED
;
604 xfs_buf_item_log(bip
, first
, last
);
609 * Invalidate a buffer that is being used within a transaction.
611 * Typically this is because the blocks in the buffer are being freed, so we
612 * need to prevent it from being written out when we're done. Allowing it
613 * to be written again might overwrite data in the free blocks if they are
614 * reallocated to a file.
616 * We prevent the buffer from being written out by marking it stale. We can't
617 * get rid of the buf log item at this point because the buffer may still be
618 * pinned by another transaction. If that is the case, then we'll wait until
619 * the buffer is committed to disk for the last time (we can tell by the ref
620 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
621 * keep the buffer locked so that the buffer and buf log item are not reused.
623 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
624 * the buf item. This will be used at recovery time to determine that copies
625 * of the buffer in the log before this should not be replayed.
627 * We mark the item descriptor and the transaction dirty so that we'll hold
628 * the buffer until after the commit.
630 * Since we're invalidating the buffer, we also clear the state about which
631 * parts of the buffer have been logged. We also clear the flag indicating
632 * that this is an inode buffer since the data in the buffer will no longer
635 * We set the stale bit in the buffer as well since we're getting rid of it.
642 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
644 ASSERT(bp
->b_transp
== tp
);
646 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
648 trace_xfs_trans_binval(bip
);
650 if (bip
->bli_flags
& XFS_BLI_STALE
) {
652 * If the buffer is already invalidated, then
655 ASSERT(XFS_BUF_ISSTALE(bp
));
656 ASSERT(!(bip
->bli_flags
& (XFS_BLI_LOGGED
| XFS_BLI_DIRTY
)));
657 ASSERT(!(bip
->bli_format
.blf_flags
& XFS_BLF_INODE_BUF
));
658 ASSERT(bip
->bli_format
.blf_flags
& XFS_BLF_CANCEL
);
659 ASSERT(bip
->bli_item
.li_desc
->lid_flags
& XFS_LID_DIRTY
);
660 ASSERT(tp
->t_flags
& XFS_TRANS_DIRTY
);
666 bip
->bli_flags
|= XFS_BLI_STALE
;
667 bip
->bli_flags
&= ~(XFS_BLI_INODE_BUF
| XFS_BLI_LOGGED
| XFS_BLI_DIRTY
);
668 bip
->bli_format
.blf_flags
&= ~XFS_BLF_INODE_BUF
;
669 bip
->bli_format
.blf_flags
|= XFS_BLF_CANCEL
;
670 memset((char *)(bip
->bli_format
.blf_data_map
), 0,
671 (bip
->bli_format
.blf_map_size
* sizeof(uint
)));
672 bip
->bli_item
.li_desc
->lid_flags
|= XFS_LID_DIRTY
;
673 tp
->t_flags
|= XFS_TRANS_DIRTY
;
677 * This call is used to indicate that the buffer contains on-disk inodes which
678 * must be handled specially during recovery. They require special handling
679 * because only the di_next_unlinked from the inodes in the buffer should be
680 * recovered. The rest of the data in the buffer is logged via the inodes
683 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
684 * transferred to the buffer's log format structure so that we'll know what to
685 * do at recovery time.
692 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
694 ASSERT(bp
->b_transp
== tp
);
696 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
698 bip
->bli_flags
|= XFS_BLI_INODE_BUF
;
702 * This call is used to indicate that the buffer is going to
703 * be staled and was an inode buffer. This means it gets
704 * special processing during unpin - where any inodes
705 * associated with the buffer should be removed from ail.
706 * There is also special processing during recovery,
707 * any replay of the inodes in the buffer needs to be
708 * prevented as the buffer may have been reused.
711 xfs_trans_stale_inode_buf(
715 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
717 ASSERT(bp
->b_transp
== tp
);
719 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
721 bip
->bli_flags
|= XFS_BLI_STALE_INODE
;
722 bip
->bli_item
.li_cb
= xfs_buf_iodone
;
726 * Mark the buffer as being one which contains newly allocated
727 * inodes. We need to make sure that even if this buffer is
728 * relogged as an 'inode buf' we still recover all of the inode
729 * images in the face of a crash. This works in coordination with
730 * xfs_buf_item_committed() to ensure that the buffer remains in the
731 * AIL at its original location even after it has been relogged.
735 xfs_trans_inode_alloc_buf(
739 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
741 ASSERT(bp
->b_transp
== tp
);
743 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
745 bip
->bli_flags
|= XFS_BLI_INODE_ALLOC_BUF
;
750 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
751 * dquots. However, unlike in inode buffer recovery, dquot buffers get
752 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
753 * The only thing that makes dquot buffers different from regular
754 * buffers is that we must not replay dquot bufs when recovering
755 * if a _corresponding_ quotaoff has happened. We also have to distinguish
756 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
757 * can be turned off independently.
766 xfs_buf_log_item_t
*bip
= bp
->b_fspriv
;
768 ASSERT(bp
->b_transp
== tp
);
770 ASSERT(type
== XFS_BLF_UDQUOT_BUF
||
771 type
== XFS_BLF_PDQUOT_BUF
||
772 type
== XFS_BLF_GDQUOT_BUF
);
773 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
775 bip
->bli_format
.blf_flags
|= type
;