1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_inode.h"
14 #include "xfs_trans.h"
15 #include "xfs_buf_item.h"
16 #include "xfs_trans_priv.h"
17 #include "xfs_error.h"
18 #include "xfs_trace.h"
21 * Check to see if a buffer matching the given parameters is already
22 * a part of the given transaction.
24 STATIC
struct xfs_buf
*
25 xfs_trans_buf_item_match(
27 struct xfs_buftarg
*target
,
28 struct xfs_buf_map
*map
,
31 struct xfs_log_item
*lip
;
32 struct xfs_buf_log_item
*blip
;
36 for (i
= 0; i
< nmaps
; i
++)
39 list_for_each_entry(lip
, &tp
->t_items
, li_trans
) {
40 blip
= (struct xfs_buf_log_item
*)lip
;
41 if (blip
->bli_item
.li_type
== XFS_LI_BUF
&&
42 blip
->bli_buf
->b_target
== target
&&
43 XFS_BUF_ADDR(blip
->bli_buf
) == map
[0].bm_bn
&&
44 blip
->bli_buf
->b_length
== len
) {
45 ASSERT(blip
->bli_buf
->b_map_count
== nmaps
);
54 * Add the locked buffer to the transaction.
56 * The buffer must be locked, and it cannot be associated with any
59 * If the buffer does not yet have a buf log item associated with it,
60 * then allocate one for it. Then add the buf item to the transaction.
68 struct xfs_buf_log_item
*bip
;
70 ASSERT(bp
->b_transp
== NULL
);
73 * The xfs_buf_log_item pointer is stored in b_log_item. If
74 * it doesn't have one yet, then allocate one and initialize it.
75 * The checks to see if one is there are in xfs_buf_item_init().
77 xfs_buf_item_init(bp
, tp
->t_mountp
);
79 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
80 ASSERT(!(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
));
81 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
86 * Take a reference for this transaction on the buf item.
88 atomic_inc(&bip
->bli_refcount
);
91 * Attach the item to the transaction so we can find it in
92 * xfs_trans_get_buf() and friends.
94 xfs_trans_add_item(tp
, &bip
->bli_item
);
101 struct xfs_trans
*tp
,
104 _xfs_trans_bjoin(tp
, bp
, 0);
105 trace_xfs_trans_bjoin(bp
->b_log_item
);
109 * Get and lock the buffer for the caller if it is not already
110 * locked within the given transaction. If it is already locked
111 * within the transaction, just increment its lock recursion count
112 * and return a pointer to it.
114 * If the transaction pointer is NULL, make this just a normal
118 xfs_trans_get_buf_map(
119 struct xfs_trans
*tp
,
120 struct xfs_buftarg
*target
,
121 struct xfs_buf_map
*map
,
123 xfs_buf_flags_t flags
)
126 struct xfs_buf_log_item
*bip
;
129 return xfs_buf_get_map(target
, map
, nmaps
, flags
);
132 * If we find the buffer in the cache with this transaction
133 * pointer in its b_fsprivate2 field, then we know we already
134 * have it locked. In this case we just increment the lock
135 * recursion count and return the buffer to the caller.
137 bp
= xfs_trans_buf_item_match(tp
, target
, map
, nmaps
);
139 ASSERT(xfs_buf_islocked(bp
));
140 if (XFS_FORCED_SHUTDOWN(tp
->t_mountp
)) {
142 bp
->b_flags
|= XBF_DONE
;
145 ASSERT(bp
->b_transp
== tp
);
146 bip
= bp
->b_log_item
;
148 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
150 trace_xfs_trans_get_buf_recur(bip
);
154 bp
= xfs_buf_get_map(target
, map
, nmaps
, flags
);
159 ASSERT(!bp
->b_error
);
161 _xfs_trans_bjoin(tp
, bp
, 1);
162 trace_xfs_trans_get_buf(bp
->b_log_item
);
167 * Get and lock the superblock buffer of this file system for the
170 * We don't need to use incore_match() here, because the superblock
171 * buffer is a private buffer which we keep a pointer to in the
177 struct xfs_mount
*mp
,
181 struct xfs_buf_log_item
*bip
;
184 * Default to just trying to lock the superblock buffer
188 return xfs_getsb(mp
, flags
);
191 * If the superblock buffer already has this transaction
192 * pointer in its b_fsprivate2 field, then we know we already
193 * have it locked. In this case we just increment the lock
194 * recursion count and return the buffer to the caller.
197 if (bp
->b_transp
== tp
) {
198 bip
= bp
->b_log_item
;
200 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
202 trace_xfs_trans_getsb_recur(bip
);
206 bp
= xfs_getsb(mp
, flags
);
210 _xfs_trans_bjoin(tp
, bp
, 1);
211 trace_xfs_trans_getsb(bp
->b_log_item
);
216 * Get and lock the buffer for the caller if it is not already
217 * locked within the given transaction. If it has not yet been
218 * read in, read it from disk. If it is already locked
219 * within the transaction and already read in, just increment its
220 * lock recursion count and return a pointer to it.
222 * If the transaction pointer is NULL, make this just a normal
226 xfs_trans_read_buf_map(
227 struct xfs_mount
*mp
,
228 struct xfs_trans
*tp
,
229 struct xfs_buftarg
*target
,
230 struct xfs_buf_map
*map
,
232 xfs_buf_flags_t flags
,
233 struct xfs_buf
**bpp
,
234 const struct xfs_buf_ops
*ops
)
236 struct xfs_buf
*bp
= NULL
;
237 struct xfs_buf_log_item
*bip
;
242 * If we find the buffer in the cache with this transaction
243 * pointer in its b_fsprivate2 field, then we know we already
244 * have it locked. If it is already read in we just increment
245 * the lock recursion count and return the buffer to the caller.
246 * If the buffer is not yet read in, then we read it in, increment
247 * the lock recursion count, and return it to the caller.
250 bp
= xfs_trans_buf_item_match(tp
, target
, map
, nmaps
);
252 ASSERT(xfs_buf_islocked(bp
));
253 ASSERT(bp
->b_transp
== tp
);
254 ASSERT(bp
->b_log_item
!= NULL
);
255 ASSERT(!bp
->b_error
);
256 ASSERT(bp
->b_flags
& XBF_DONE
);
259 * We never locked this buf ourselves, so we shouldn't
260 * brelse it either. Just get out.
262 if (XFS_FORCED_SHUTDOWN(mp
)) {
263 trace_xfs_trans_read_buf_shut(bp
, _RET_IP_
);
268 * Check if the caller is trying to read a buffer that is
269 * already attached to the transaction yet has no buffer ops
270 * assigned. Ops are usually attached when the buffer is
271 * attached to the transaction, or by the read caller if
272 * special circumstances. That didn't happen, which is not
273 * how this is supposed to go.
275 * If the buffer passes verification we'll let this go, but if
276 * not we have to shut down. Let the transaction cleanup code
277 * release this buffer when it kills the tranaction.
279 ASSERT(bp
->b_ops
!= NULL
);
280 error
= xfs_buf_ensure_ops(bp
, ops
);
282 xfs_buf_ioerror_alert(bp
, __func__
);
284 if (tp
->t_flags
& XFS_TRANS_DIRTY
)
285 xfs_force_shutdown(tp
->t_mountp
,
286 SHUTDOWN_META_IO_ERROR
);
288 /* bad CRC means corrupted metadata */
289 if (error
== -EFSBADCRC
)
290 error
= -EFSCORRUPTED
;
294 bip
= bp
->b_log_item
;
297 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
298 trace_xfs_trans_read_buf_recur(bip
);
299 ASSERT(bp
->b_ops
!= NULL
|| ops
== NULL
);
304 bp
= xfs_buf_read_map(target
, map
, nmaps
, flags
, ops
);
306 if (!(flags
& XBF_TRYLOCK
))
308 return tp
? 0 : -EAGAIN
;
312 * If we've had a read error, then the contents of the buffer are
313 * invalid and should not be used. To ensure that a followup read tries
314 * to pull the buffer from disk again, we clear the XBF_DONE flag and
315 * mark the buffer stale. This ensures that anyone who has a current
316 * reference to the buffer will interpret it's contents correctly and
317 * future cache lookups will also treat it as an empty, uninitialised
322 if (!XFS_FORCED_SHUTDOWN(mp
))
323 xfs_buf_ioerror_alert(bp
, __func__
);
324 bp
->b_flags
&= ~XBF_DONE
;
327 if (tp
&& (tp
->t_flags
& XFS_TRANS_DIRTY
))
328 xfs_force_shutdown(tp
->t_mountp
, SHUTDOWN_META_IO_ERROR
);
331 /* bad CRC means corrupted metadata */
332 if (error
== -EFSBADCRC
)
333 error
= -EFSCORRUPTED
;
337 if (XFS_FORCED_SHUTDOWN(mp
)) {
339 trace_xfs_trans_read_buf_shut(bp
, _RET_IP_
);
344 _xfs_trans_bjoin(tp
, bp
, 1);
345 trace_xfs_trans_read_buf(bp
->b_log_item
);
347 ASSERT(bp
->b_ops
!= NULL
|| ops
== NULL
);
353 /* Has this buffer been dirtied by anyone? */
355 xfs_trans_buf_is_dirty(
358 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
362 ASSERT(bip
->bli_item
.li_type
== XFS_LI_BUF
);
363 return test_bit(XFS_LI_DIRTY
, &bip
->bli_item
.li_flags
);
367 * Release a buffer previously joined to the transaction. If the buffer is
368 * modified within this transaction, decrement the recursion count but do not
369 * release the buffer even if the count goes to 0. If the buffer is not modified
370 * within the transaction, decrement the recursion count and release the buffer
371 * if the recursion count goes to 0.
373 * If the buffer is to be released and it was not already dirty before this
374 * transaction began, then also free the buf_log_item associated with it.
376 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
380 struct xfs_trans
*tp
,
383 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
385 ASSERT(bp
->b_transp
== tp
);
392 trace_xfs_trans_brelse(bip
);
393 ASSERT(bip
->bli_item
.li_type
== XFS_LI_BUF
);
394 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
397 * If the release is for a recursive lookup, then decrement the count
400 if (bip
->bli_recur
> 0) {
406 * If the buffer is invalidated or dirty in this transaction, we can't
407 * release it until we commit.
409 if (test_bit(XFS_LI_DIRTY
, &bip
->bli_item
.li_flags
))
411 if (bip
->bli_flags
& XFS_BLI_STALE
)
415 * Unlink the log item from the transaction and clear the hold flag, if
416 * set. We wouldn't want the next user of the buffer to get confused.
418 ASSERT(!(bip
->bli_flags
& XFS_BLI_LOGGED
));
419 xfs_trans_del_item(&bip
->bli_item
);
420 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
422 /* drop the reference to the bli */
423 xfs_buf_item_put(bip
);
430 * Mark the buffer as not needing to be unlocked when the buf item's
431 * iop_unlock() routine is called. The buffer must already be locked
432 * and associated with the given transaction.
440 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
442 ASSERT(bp
->b_transp
== tp
);
444 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
445 ASSERT(!(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
));
446 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
448 bip
->bli_flags
|= XFS_BLI_HOLD
;
449 trace_xfs_trans_bhold(bip
);
453 * Cancel the previous buffer hold request made on this buffer
454 * for this transaction.
457 xfs_trans_bhold_release(
461 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
463 ASSERT(bp
->b_transp
== tp
);
465 ASSERT(!(bip
->bli_flags
& XFS_BLI_STALE
));
466 ASSERT(!(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
));
467 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
468 ASSERT(bip
->bli_flags
& XFS_BLI_HOLD
);
470 bip
->bli_flags
&= ~XFS_BLI_HOLD
;
471 trace_xfs_trans_bhold_release(bip
);
475 * Mark a buffer dirty in the transaction.
479 struct xfs_trans
*tp
,
482 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
484 ASSERT(bp
->b_transp
== tp
);
486 ASSERT(bp
->b_iodone
== NULL
||
487 bp
->b_iodone
== xfs_buf_iodone_callbacks
);
490 * Mark the buffer as needing to be written out eventually,
491 * and set its iodone function to remove the buffer's buf log
492 * item from the AIL and free it when the buffer is flushed
493 * to disk. See xfs_buf_attach_iodone() for more details
494 * on li_cb and xfs_buf_iodone_callbacks().
495 * If we end up aborting this transaction, we trap this buffer
496 * inside the b_bdstrat callback so that this won't get written to
499 bp
->b_flags
|= XBF_DONE
;
501 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
502 bp
->b_iodone
= xfs_buf_iodone_callbacks
;
503 bip
->bli_item
.li_cb
= xfs_buf_iodone
;
506 * If we invalidated the buffer within this transaction, then
507 * cancel the invalidation now that we're dirtying the buffer
508 * again. There are no races with the code in xfs_buf_item_unpin(),
509 * because we have a reference to the buffer this entire time.
511 if (bip
->bli_flags
& XFS_BLI_STALE
) {
512 bip
->bli_flags
&= ~XFS_BLI_STALE
;
513 ASSERT(bp
->b_flags
& XBF_STALE
);
514 bp
->b_flags
&= ~XBF_STALE
;
515 bip
->__bli_format
.blf_flags
&= ~XFS_BLF_CANCEL
;
517 bip
->bli_flags
|= XFS_BLI_DIRTY
| XFS_BLI_LOGGED
;
519 tp
->t_flags
|= XFS_TRANS_DIRTY
;
520 set_bit(XFS_LI_DIRTY
, &bip
->bli_item
.li_flags
);
524 * This is called to mark bytes first through last inclusive of the given
525 * buffer as needing to be logged when the transaction is committed.
526 * The buffer must already be associated with the given transaction.
528 * First and last are numbers relative to the beginning of this buffer,
529 * so the first byte in the buffer is numbered 0 regardless of the
534 struct xfs_trans
*tp
,
539 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
541 ASSERT(first
<= last
&& last
< BBTOB(bp
->b_length
));
542 ASSERT(!(bip
->bli_flags
& XFS_BLI_ORDERED
));
544 xfs_trans_dirty_buf(tp
, bp
);
546 trace_xfs_trans_log_buf(bip
);
547 xfs_buf_item_log(bip
, first
, last
);
552 * Invalidate a buffer that is being used within a transaction.
554 * Typically this is because the blocks in the buffer are being freed, so we
555 * need to prevent it from being written out when we're done. Allowing it
556 * to be written again might overwrite data in the free blocks if they are
557 * reallocated to a file.
559 * We prevent the buffer from being written out by marking it stale. We can't
560 * get rid of the buf log item at this point because the buffer may still be
561 * pinned by another transaction. If that is the case, then we'll wait until
562 * the buffer is committed to disk for the last time (we can tell by the ref
563 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
564 * keep the buffer locked so that the buffer and buf log item are not reused.
566 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
567 * the buf item. This will be used at recovery time to determine that copies
568 * of the buffer in the log before this should not be replayed.
570 * We mark the item descriptor and the transaction dirty so that we'll hold
571 * the buffer until after the commit.
573 * Since we're invalidating the buffer, we also clear the state about which
574 * parts of the buffer have been logged. We also clear the flag indicating
575 * that this is an inode buffer since the data in the buffer will no longer
578 * We set the stale bit in the buffer as well since we're getting rid of it.
585 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
588 ASSERT(bp
->b_transp
== tp
);
590 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
592 trace_xfs_trans_binval(bip
);
594 if (bip
->bli_flags
& XFS_BLI_STALE
) {
596 * If the buffer is already invalidated, then
599 ASSERT(bp
->b_flags
& XBF_STALE
);
600 ASSERT(!(bip
->bli_flags
& (XFS_BLI_LOGGED
| XFS_BLI_DIRTY
)));
601 ASSERT(!(bip
->__bli_format
.blf_flags
& XFS_BLF_INODE_BUF
));
602 ASSERT(!(bip
->__bli_format
.blf_flags
& XFS_BLFT_MASK
));
603 ASSERT(bip
->__bli_format
.blf_flags
& XFS_BLF_CANCEL
);
604 ASSERT(test_bit(XFS_LI_DIRTY
, &bip
->bli_item
.li_flags
));
605 ASSERT(tp
->t_flags
& XFS_TRANS_DIRTY
);
611 bip
->bli_flags
|= XFS_BLI_STALE
;
612 bip
->bli_flags
&= ~(XFS_BLI_INODE_BUF
| XFS_BLI_LOGGED
| XFS_BLI_DIRTY
);
613 bip
->__bli_format
.blf_flags
&= ~XFS_BLF_INODE_BUF
;
614 bip
->__bli_format
.blf_flags
|= XFS_BLF_CANCEL
;
615 bip
->__bli_format
.blf_flags
&= ~XFS_BLFT_MASK
;
616 for (i
= 0; i
< bip
->bli_format_count
; i
++) {
617 memset(bip
->bli_formats
[i
].blf_data_map
, 0,
618 (bip
->bli_formats
[i
].blf_map_size
* sizeof(uint
)));
620 set_bit(XFS_LI_DIRTY
, &bip
->bli_item
.li_flags
);
621 tp
->t_flags
|= XFS_TRANS_DIRTY
;
625 * This call is used to indicate that the buffer contains on-disk inodes which
626 * must be handled specially during recovery. They require special handling
627 * because only the di_next_unlinked from the inodes in the buffer should be
628 * recovered. The rest of the data in the buffer is logged via the inodes
631 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
632 * transferred to the buffer's log format structure so that we'll know what to
633 * do at recovery time.
640 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
642 ASSERT(bp
->b_transp
== tp
);
644 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
646 bip
->bli_flags
|= XFS_BLI_INODE_BUF
;
647 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DINO_BUF
);
651 * This call is used to indicate that the buffer is going to
652 * be staled and was an inode buffer. This means it gets
653 * special processing during unpin - where any inodes
654 * associated with the buffer should be removed from ail.
655 * There is also special processing during recovery,
656 * any replay of the inodes in the buffer needs to be
657 * prevented as the buffer may have been reused.
660 xfs_trans_stale_inode_buf(
664 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
666 ASSERT(bp
->b_transp
== tp
);
668 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
670 bip
->bli_flags
|= XFS_BLI_STALE_INODE
;
671 bip
->bli_item
.li_cb
= xfs_buf_iodone
;
672 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DINO_BUF
);
676 * Mark the buffer as being one which contains newly allocated
677 * inodes. We need to make sure that even if this buffer is
678 * relogged as an 'inode buf' we still recover all of the inode
679 * images in the face of a crash. This works in coordination with
680 * xfs_buf_item_committed() to ensure that the buffer remains in the
681 * AIL at its original location even after it has been relogged.
685 xfs_trans_inode_alloc_buf(
689 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
691 ASSERT(bp
->b_transp
== tp
);
693 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
695 bip
->bli_flags
|= XFS_BLI_INODE_ALLOC_BUF
;
696 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DINO_BUF
);
700 * Mark the buffer as ordered for this transaction. This means that the contents
701 * of the buffer are not recorded in the transaction but it is tracked in the
702 * AIL as though it was. This allows us to record logical changes in
703 * transactions rather than the physical changes we make to the buffer without
704 * changing writeback ordering constraints of metadata buffers.
707 xfs_trans_ordered_buf(
708 struct xfs_trans
*tp
,
711 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
713 ASSERT(bp
->b_transp
== tp
);
715 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
717 if (xfs_buf_item_dirty_format(bip
))
720 bip
->bli_flags
|= XFS_BLI_ORDERED
;
721 trace_xfs_buf_item_ordered(bip
);
724 * We don't log a dirty range of an ordered buffer but it still needs
725 * to be marked dirty and that it has been logged.
727 xfs_trans_dirty_buf(tp
, bp
);
732 * Set the type of the buffer for log recovery so that it can correctly identify
733 * and hence attach the correct buffer ops to the buffer after replay.
736 xfs_trans_buf_set_type(
737 struct xfs_trans
*tp
,
741 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
746 ASSERT(bp
->b_transp
== tp
);
748 ASSERT(atomic_read(&bip
->bli_refcount
) > 0);
750 xfs_blft_to_flags(&bip
->__bli_format
, type
);
754 xfs_trans_buf_copy_type(
755 struct xfs_buf
*dst_bp
,
756 struct xfs_buf
*src_bp
)
758 struct xfs_buf_log_item
*sbip
= src_bp
->b_log_item
;
759 struct xfs_buf_log_item
*dbip
= dst_bp
->b_log_item
;
762 type
= xfs_blft_from_flags(&sbip
->__bli_format
);
763 xfs_blft_to_flags(&dbip
->__bli_format
, type
);
767 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
768 * dquots. However, unlike in inode buffer recovery, dquot buffers get
769 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
770 * The only thing that makes dquot buffers different from regular
771 * buffers is that we must not replay dquot bufs when recovering
772 * if a _corresponding_ quotaoff has happened. We also have to distinguish
773 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
774 * can be turned off independently.
783 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
785 ASSERT(type
== XFS_BLF_UDQUOT_BUF
||
786 type
== XFS_BLF_PDQUOT_BUF
||
787 type
== XFS_BLF_GDQUOT_BUF
);
789 bip
->__bli_format
.blf_flags
|= type
;
792 case XFS_BLF_UDQUOT_BUF
:
793 type
= XFS_BLFT_UDQUOT_BUF
;
795 case XFS_BLF_PDQUOT_BUF
:
796 type
= XFS_BLFT_PDQUOT_BUF
;
798 case XFS_BLF_GDQUOT_BUF
:
799 type
= XFS_BLFT_GDQUOT_BUF
;
802 type
= XFS_BLFT_UNKNOWN_BUF
;
806 xfs_trans_buf_set_type(tp
, bp
, type
);