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_trans.h"
14 #include "xfs_buf_item.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_trace.h"
19 * Check to see if a buffer matching the given parameters is already
20 * a part of the given transaction.
22 STATIC struct xfs_buf *
23 xfs_trans_buf_item_match(
25 struct xfs_buftarg *target,
26 struct xfs_buf_map *map,
29 struct xfs_log_item *lip;
30 struct xfs_buf_log_item *blip;
34 for (i = 0; i < nmaps; i++)
37 list_for_each_entry(lip, &tp->t_items, li_trans) {
38 blip = (struct xfs_buf_log_item *)lip;
39 if (blip->bli_item.li_type == XFS_LI_BUF &&
40 blip->bli_buf->b_target == target &&
41 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
42 blip->bli_buf->b_length == len) {
43 ASSERT(blip->bli_buf->b_map_count == nmaps);
52 * Add the locked buffer to the transaction.
54 * The buffer must be locked, and it cannot be associated with any
57 * If the buffer does not yet have a buf log item associated with it,
58 * then allocate one for it. Then add the buf item to the transaction.
66 struct xfs_buf_log_item *bip;
68 ASSERT(bp->b_transp == NULL);
71 * The xfs_buf_log_item pointer is stored in b_log_item. If
72 * it doesn't have one yet, then allocate one and initialize it.
73 * The checks to see if one is there are in xfs_buf_item_init().
75 xfs_buf_item_init(bp, tp->t_mountp);
77 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
78 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
79 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
84 * Take a reference for this transaction on the buf item.
86 atomic_inc(&bip->bli_refcount);
89 * Attach the item to the transaction so we can find it in
90 * xfs_trans_get_buf() and friends.
92 xfs_trans_add_item(tp, &bip->bli_item);
102 _xfs_trans_bjoin(tp, bp, 0);
103 trace_xfs_trans_bjoin(bp->b_log_item);
107 * Get and lock the buffer for the caller if it is not already
108 * locked within the given transaction. If it is already locked
109 * within the transaction, just increment its lock recursion count
110 * and return a pointer to it.
112 * If the transaction pointer is NULL, make this just a normal
116 xfs_trans_get_buf_map(
117 struct xfs_trans *tp,
118 struct xfs_buftarg *target,
119 struct xfs_buf_map *map,
121 xfs_buf_flags_t flags,
122 struct xfs_buf **bpp)
125 struct xfs_buf_log_item *bip;
130 return xfs_buf_get_map(target, map, nmaps, flags, bpp);
133 * If we find the buffer in the cache with this transaction
134 * pointer in its b_fsprivate2 field, then we know we already
135 * have it locked. In this case we just increment the lock
136 * recursion count and return the buffer to the caller.
138 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
140 ASSERT(xfs_buf_islocked(bp));
141 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
143 bp->b_flags |= XBF_DONE;
146 ASSERT(bp->b_transp == tp);
147 bip = bp->b_log_item;
149 ASSERT(atomic_read(&bip->bli_refcount) > 0);
151 trace_xfs_trans_get_buf_recur(bip);
156 error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
160 ASSERT(!bp->b_error);
162 _xfs_trans_bjoin(tp, bp, 1);
163 trace_xfs_trans_get_buf(bp->b_log_item);
169 * Get and lock the superblock buffer of this file system for the
172 * We don't need to use incore_match() here, because the superblock
173 * buffer is a private buffer which we keep a pointer to in the
179 struct xfs_mount *mp)
182 struct xfs_buf_log_item *bip;
185 * Default to just trying to lock the superblock buffer
189 return xfs_getsb(mp);
192 * If the superblock buffer already has this transaction
193 * pointer in its b_fsprivate2 field, then we know we already
194 * have it locked. In this case we just increment the lock
195 * recursion count and return the buffer to the caller.
198 if (bp->b_transp == tp) {
199 bip = bp->b_log_item;
201 ASSERT(atomic_read(&bip->bli_refcount) > 0);
203 trace_xfs_trans_getsb_recur(bip);
211 _xfs_trans_bjoin(tp, bp, 1);
212 trace_xfs_trans_getsb(bp->b_log_item);
217 * Get and lock the buffer for the caller if it is not already
218 * locked within the given transaction. If it has not yet been
219 * read in, read it from disk. If it is already locked
220 * within the transaction and already read in, just increment its
221 * lock recursion count and return a pointer to it.
223 * If the transaction pointer is NULL, make this just a normal
227 xfs_trans_read_buf_map(
228 struct xfs_mount *mp,
229 struct xfs_trans *tp,
230 struct xfs_buftarg *target,
231 struct xfs_buf_map *map,
233 xfs_buf_flags_t flags,
234 struct xfs_buf **bpp,
235 const struct xfs_buf_ops *ops)
237 struct xfs_buf *bp = NULL;
238 struct xfs_buf_log_item *bip;
243 * If we find the buffer in the cache with this transaction
244 * pointer in its b_fsprivate2 field, then we know we already
245 * have it locked. If it is already read in we just increment
246 * the lock recursion count and return the buffer to the caller.
247 * If the buffer is not yet read in, then we read it in, increment
248 * the lock recursion count, and return it to the caller.
251 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
253 ASSERT(xfs_buf_islocked(bp));
254 ASSERT(bp->b_transp == tp);
255 ASSERT(bp->b_log_item != NULL);
256 ASSERT(!bp->b_error);
257 ASSERT(bp->b_flags & XBF_DONE);
260 * We never locked this buf ourselves, so we shouldn't
261 * brelse it either. Just get out.
263 if (XFS_FORCED_SHUTDOWN(mp)) {
264 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
269 * Check if the caller is trying to read a buffer that is
270 * already attached to the transaction yet has no buffer ops
271 * assigned. Ops are usually attached when the buffer is
272 * attached to the transaction, or by the read caller if
273 * special circumstances. That didn't happen, which is not
274 * how this is supposed to go.
276 * If the buffer passes verification we'll let this go, but if
277 * not we have to shut down. Let the transaction cleanup code
278 * release this buffer when it kills the tranaction.
280 ASSERT(bp->b_ops != NULL);
281 error = xfs_buf_reverify(bp, ops);
283 xfs_buf_ioerror_alert(bp, __return_address);
285 if (tp->t_flags & XFS_TRANS_DIRTY)
286 xfs_force_shutdown(tp->t_mountp,
287 SHUTDOWN_META_IO_ERROR);
289 /* bad CRC means corrupted metadata */
290 if (error == -EFSBADCRC)
291 error = -EFSCORRUPTED;
295 bip = bp->b_log_item;
298 ASSERT(atomic_read(&bip->bli_refcount) > 0);
299 trace_xfs_trans_read_buf_recur(bip);
300 ASSERT(bp->b_ops != NULL || ops == NULL);
305 error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
311 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
312 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
319 if (XFS_FORCED_SHUTDOWN(mp)) {
321 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
326 _xfs_trans_bjoin(tp, bp, 1);
327 trace_xfs_trans_read_buf(bp->b_log_item);
329 ASSERT(bp->b_ops != NULL || ops == NULL);
335 /* Has this buffer been dirtied by anyone? */
337 xfs_trans_buf_is_dirty(
340 struct xfs_buf_log_item *bip = bp->b_log_item;
344 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
345 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
349 * Release a buffer previously joined to the transaction. If the buffer is
350 * modified within this transaction, decrement the recursion count but do not
351 * release the buffer even if the count goes to 0. If the buffer is not modified
352 * within the transaction, decrement the recursion count and release the buffer
353 * if the recursion count goes to 0.
355 * If the buffer is to be released and it was not already dirty before this
356 * transaction began, then also free the buf_log_item associated with it.
358 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
362 struct xfs_trans *tp,
365 struct xfs_buf_log_item *bip = bp->b_log_item;
367 ASSERT(bp->b_transp == tp);
374 trace_xfs_trans_brelse(bip);
375 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
376 ASSERT(atomic_read(&bip->bli_refcount) > 0);
379 * If the release is for a recursive lookup, then decrement the count
382 if (bip->bli_recur > 0) {
388 * If the buffer is invalidated or dirty in this transaction, we can't
389 * release it until we commit.
391 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
393 if (bip->bli_flags & XFS_BLI_STALE)
397 * Unlink the log item from the transaction and clear the hold flag, if
398 * set. We wouldn't want the next user of the buffer to get confused.
400 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
401 xfs_trans_del_item(&bip->bli_item);
402 bip->bli_flags &= ~XFS_BLI_HOLD;
404 /* drop the reference to the bli */
405 xfs_buf_item_put(bip);
412 * Mark the buffer as not needing to be unlocked when the buf item's
413 * iop_committing() routine is called. The buffer must already be locked
414 * and associated with the given transaction.
422 struct xfs_buf_log_item *bip = bp->b_log_item;
424 ASSERT(bp->b_transp == tp);
426 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
427 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
428 ASSERT(atomic_read(&bip->bli_refcount) > 0);
430 bip->bli_flags |= XFS_BLI_HOLD;
431 trace_xfs_trans_bhold(bip);
435 * Cancel the previous buffer hold request made on this buffer
436 * for this transaction.
439 xfs_trans_bhold_release(
443 struct xfs_buf_log_item *bip = bp->b_log_item;
445 ASSERT(bp->b_transp == tp);
447 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
448 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
449 ASSERT(atomic_read(&bip->bli_refcount) > 0);
450 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
452 bip->bli_flags &= ~XFS_BLI_HOLD;
453 trace_xfs_trans_bhold_release(bip);
457 * Mark a buffer dirty in the transaction.
461 struct xfs_trans *tp,
464 struct xfs_buf_log_item *bip = bp->b_log_item;
466 ASSERT(bp->b_transp == tp);
468 ASSERT(bp->b_iodone == NULL ||
469 bp->b_iodone == xfs_buf_iodone_callbacks);
472 * Mark the buffer as needing to be written out eventually,
473 * and set its iodone function to remove the buffer's buf log
474 * item from the AIL and free it when the buffer is flushed
475 * to disk. See xfs_buf_attach_iodone() for more details
476 * on li_cb and xfs_buf_iodone_callbacks().
477 * If we end up aborting this transaction, we trap this buffer
478 * inside the b_bdstrat callback so that this won't get written to
481 bp->b_flags |= XBF_DONE;
483 ASSERT(atomic_read(&bip->bli_refcount) > 0);
484 bp->b_iodone = xfs_buf_iodone_callbacks;
485 bip->bli_item.li_cb = xfs_buf_iodone;
488 * If we invalidated the buffer within this transaction, then
489 * cancel the invalidation now that we're dirtying the buffer
490 * again. There are no races with the code in xfs_buf_item_unpin(),
491 * because we have a reference to the buffer this entire time.
493 if (bip->bli_flags & XFS_BLI_STALE) {
494 bip->bli_flags &= ~XFS_BLI_STALE;
495 ASSERT(bp->b_flags & XBF_STALE);
496 bp->b_flags &= ~XBF_STALE;
497 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
499 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
501 tp->t_flags |= XFS_TRANS_DIRTY;
502 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
506 * This is called to mark bytes first through last inclusive of the given
507 * buffer as needing to be logged when the transaction is committed.
508 * The buffer must already be associated with the given transaction.
510 * First and last are numbers relative to the beginning of this buffer,
511 * so the first byte in the buffer is numbered 0 regardless of the
516 struct xfs_trans *tp,
521 struct xfs_buf_log_item *bip = bp->b_log_item;
523 ASSERT(first <= last && last < BBTOB(bp->b_length));
524 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
526 xfs_trans_dirty_buf(tp, bp);
528 trace_xfs_trans_log_buf(bip);
529 xfs_buf_item_log(bip, first, last);
534 * Invalidate a buffer that is being used within a transaction.
536 * Typically this is because the blocks in the buffer are being freed, so we
537 * need to prevent it from being written out when we're done. Allowing it
538 * to be written again might overwrite data in the free blocks if they are
539 * reallocated to a file.
541 * We prevent the buffer from being written out by marking it stale. We can't
542 * get rid of the buf log item at this point because the buffer may still be
543 * pinned by another transaction. If that is the case, then we'll wait until
544 * the buffer is committed to disk for the last time (we can tell by the ref
545 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
546 * keep the buffer locked so that the buffer and buf log item are not reused.
548 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
549 * the buf item. This will be used at recovery time to determine that copies
550 * of the buffer in the log before this should not be replayed.
552 * We mark the item descriptor and the transaction dirty so that we'll hold
553 * the buffer until after the commit.
555 * Since we're invalidating the buffer, we also clear the state about which
556 * parts of the buffer have been logged. We also clear the flag indicating
557 * that this is an inode buffer since the data in the buffer will no longer
560 * We set the stale bit in the buffer as well since we're getting rid of it.
567 struct xfs_buf_log_item *bip = bp->b_log_item;
570 ASSERT(bp->b_transp == tp);
572 ASSERT(atomic_read(&bip->bli_refcount) > 0);
574 trace_xfs_trans_binval(bip);
576 if (bip->bli_flags & XFS_BLI_STALE) {
578 * If the buffer is already invalidated, then
581 ASSERT(bp->b_flags & XBF_STALE);
582 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
583 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
584 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
585 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
586 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
587 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
593 bip->bli_flags |= XFS_BLI_STALE;
594 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
595 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
596 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
597 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
598 for (i = 0; i < bip->bli_format_count; i++) {
599 memset(bip->bli_formats[i].blf_data_map, 0,
600 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
602 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
603 tp->t_flags |= XFS_TRANS_DIRTY;
607 * This call is used to indicate that the buffer contains on-disk inodes which
608 * must be handled specially during recovery. They require special handling
609 * because only the di_next_unlinked from the inodes in the buffer should be
610 * recovered. The rest of the data in the buffer is logged via the inodes
613 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
614 * transferred to the buffer's log format structure so that we'll know what to
615 * do at recovery time.
622 struct xfs_buf_log_item *bip = bp->b_log_item;
624 ASSERT(bp->b_transp == tp);
626 ASSERT(atomic_read(&bip->bli_refcount) > 0);
628 bip->bli_flags |= XFS_BLI_INODE_BUF;
629 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
633 * This call is used to indicate that the buffer is going to
634 * be staled and was an inode buffer. This means it gets
635 * special processing during unpin - where any inodes
636 * associated with the buffer should be removed from ail.
637 * There is also special processing during recovery,
638 * any replay of the inodes in the buffer needs to be
639 * prevented as the buffer may have been reused.
642 xfs_trans_stale_inode_buf(
646 struct xfs_buf_log_item *bip = bp->b_log_item;
648 ASSERT(bp->b_transp == tp);
650 ASSERT(atomic_read(&bip->bli_refcount) > 0);
652 bip->bli_flags |= XFS_BLI_STALE_INODE;
653 bip->bli_item.li_cb = xfs_buf_iodone;
654 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
658 * Mark the buffer as being one which contains newly allocated
659 * inodes. We need to make sure that even if this buffer is
660 * relogged as an 'inode buf' we still recover all of the inode
661 * images in the face of a crash. This works in coordination with
662 * xfs_buf_item_committed() to ensure that the buffer remains in the
663 * AIL at its original location even after it has been relogged.
667 xfs_trans_inode_alloc_buf(
671 struct xfs_buf_log_item *bip = bp->b_log_item;
673 ASSERT(bp->b_transp == tp);
675 ASSERT(atomic_read(&bip->bli_refcount) > 0);
677 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
678 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
682 * Mark the buffer as ordered for this transaction. This means that the contents
683 * of the buffer are not recorded in the transaction but it is tracked in the
684 * AIL as though it was. This allows us to record logical changes in
685 * transactions rather than the physical changes we make to the buffer without
686 * changing writeback ordering constraints of metadata buffers.
689 xfs_trans_ordered_buf(
690 struct xfs_trans *tp,
693 struct xfs_buf_log_item *bip = bp->b_log_item;
695 ASSERT(bp->b_transp == tp);
697 ASSERT(atomic_read(&bip->bli_refcount) > 0);
699 if (xfs_buf_item_dirty_format(bip))
702 bip->bli_flags |= XFS_BLI_ORDERED;
703 trace_xfs_buf_item_ordered(bip);
706 * We don't log a dirty range of an ordered buffer but it still needs
707 * to be marked dirty and that it has been logged.
709 xfs_trans_dirty_buf(tp, bp);
714 * Set the type of the buffer for log recovery so that it can correctly identify
715 * and hence attach the correct buffer ops to the buffer after replay.
718 xfs_trans_buf_set_type(
719 struct xfs_trans *tp,
723 struct xfs_buf_log_item *bip = bp->b_log_item;
728 ASSERT(bp->b_transp == tp);
730 ASSERT(atomic_read(&bip->bli_refcount) > 0);
732 xfs_blft_to_flags(&bip->__bli_format, type);
736 xfs_trans_buf_copy_type(
737 struct xfs_buf *dst_bp,
738 struct xfs_buf *src_bp)
740 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
741 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
744 type = xfs_blft_from_flags(&sbip->__bli_format);
745 xfs_blft_to_flags(&dbip->__bli_format, type);
749 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
750 * dquots. However, unlike in inode buffer recovery, dquot buffers get
751 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
752 * The only thing that makes dquot buffers different from regular
753 * buffers is that we must not replay dquot bufs when recovering
754 * if a _corresponding_ quotaoff has happened. We also have to distinguish
755 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
756 * can be turned off independently.
765 struct xfs_buf_log_item *bip = bp->b_log_item;
767 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
768 type == XFS_BLF_PDQUOT_BUF ||
769 type == XFS_BLF_GDQUOT_BUF);
771 bip->__bli_format.blf_flags |= type;
774 case XFS_BLF_UDQUOT_BUF:
775 type = XFS_BLFT_UDQUOT_BUF;
777 case XFS_BLF_PDQUOT_BUF:
778 type = XFS_BLFT_PDQUOT_BUF;
780 case XFS_BLF_GDQUOT_BUF:
781 type = XFS_BLFT_GDQUOT_BUF;
784 type = XFS_BLFT_UNKNOWN_BUF;
788 xfs_trans_buf_set_type(tp, bp, type);