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_);
267 bip = bp->b_log_item;
270 ASSERT(atomic_read(&bip->bli_refcount) > 0);
271 trace_xfs_trans_read_buf_recur(bip);
276 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
278 if (!(flags & XBF_TRYLOCK))
280 return tp ? 0 : -EAGAIN;
284 * If we've had a read error, then the contents of the buffer are
285 * invalid and should not be used. To ensure that a followup read tries
286 * to pull the buffer from disk again, we clear the XBF_DONE flag and
287 * mark the buffer stale. This ensures that anyone who has a current
288 * reference to the buffer will interpret it's contents correctly and
289 * future cache lookups will also treat it as an empty, uninitialised
294 if (!XFS_FORCED_SHUTDOWN(mp))
295 xfs_buf_ioerror_alert(bp, __func__);
296 bp->b_flags &= ~XBF_DONE;
299 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
300 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
303 /* bad CRC means corrupted metadata */
304 if (error == -EFSBADCRC)
305 error = -EFSCORRUPTED;
309 if (XFS_FORCED_SHUTDOWN(mp)) {
311 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
316 _xfs_trans_bjoin(tp, bp, 1);
317 trace_xfs_trans_read_buf(bp->b_log_item);
325 * Release a buffer previously joined to the transaction. If the buffer is
326 * modified within this transaction, decrement the recursion count but do not
327 * release the buffer even if the count goes to 0. If the buffer is not modified
328 * within the transaction, decrement the recursion count and release the buffer
329 * if the recursion count goes to 0.
331 * If the buffer is to be released and it was not already dirty before this
332 * transaction began, then also free the buf_log_item associated with it.
334 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
338 struct xfs_trans *tp,
341 struct xfs_buf_log_item *bip = bp->b_log_item;
343 ASSERT(bp->b_transp == tp);
350 trace_xfs_trans_brelse(bip);
351 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
352 ASSERT(atomic_read(&bip->bli_refcount) > 0);
355 * If the release is for a recursive lookup, then decrement the count
358 if (bip->bli_recur > 0) {
364 * If the buffer is invalidated or dirty in this transaction, we can't
365 * release it until we commit.
367 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
369 if (bip->bli_flags & XFS_BLI_STALE)
373 * Unlink the log item from the transaction and clear the hold flag, if
374 * set. We wouldn't want the next user of the buffer to get confused.
376 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
377 xfs_trans_del_item(&bip->bli_item);
378 bip->bli_flags &= ~XFS_BLI_HOLD;
380 /* drop the reference to the bli */
381 xfs_buf_item_put(bip);
388 * Mark the buffer as not needing to be unlocked when the buf item's
389 * iop_unlock() routine is called. The buffer must already be locked
390 * and associated with the given transaction.
398 struct xfs_buf_log_item *bip = bp->b_log_item;
400 ASSERT(bp->b_transp == tp);
402 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
403 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
404 ASSERT(atomic_read(&bip->bli_refcount) > 0);
406 bip->bli_flags |= XFS_BLI_HOLD;
407 trace_xfs_trans_bhold(bip);
411 * Cancel the previous buffer hold request made on this buffer
412 * for this transaction.
415 xfs_trans_bhold_release(
419 struct xfs_buf_log_item *bip = bp->b_log_item;
421 ASSERT(bp->b_transp == tp);
423 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
424 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
425 ASSERT(atomic_read(&bip->bli_refcount) > 0);
426 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
428 bip->bli_flags &= ~XFS_BLI_HOLD;
429 trace_xfs_trans_bhold_release(bip);
433 * Mark a buffer dirty in the transaction.
437 struct xfs_trans *tp,
440 struct xfs_buf_log_item *bip = bp->b_log_item;
442 ASSERT(bp->b_transp == tp);
444 ASSERT(bp->b_iodone == NULL ||
445 bp->b_iodone == xfs_buf_iodone_callbacks);
448 * Mark the buffer as needing to be written out eventually,
449 * and set its iodone function to remove the buffer's buf log
450 * item from the AIL and free it when the buffer is flushed
451 * to disk. See xfs_buf_attach_iodone() for more details
452 * on li_cb and xfs_buf_iodone_callbacks().
453 * If we end up aborting this transaction, we trap this buffer
454 * inside the b_bdstrat callback so that this won't get written to
457 bp->b_flags |= XBF_DONE;
459 ASSERT(atomic_read(&bip->bli_refcount) > 0);
460 bp->b_iodone = xfs_buf_iodone_callbacks;
461 bip->bli_item.li_cb = xfs_buf_iodone;
464 * If we invalidated the buffer within this transaction, then
465 * cancel the invalidation now that we're dirtying the buffer
466 * again. There are no races with the code in xfs_buf_item_unpin(),
467 * because we have a reference to the buffer this entire time.
469 if (bip->bli_flags & XFS_BLI_STALE) {
470 bip->bli_flags &= ~XFS_BLI_STALE;
471 ASSERT(bp->b_flags & XBF_STALE);
472 bp->b_flags &= ~XBF_STALE;
473 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
475 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
477 tp->t_flags |= XFS_TRANS_DIRTY;
478 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
482 * This is called to mark bytes first through last inclusive of the given
483 * buffer as needing to be logged when the transaction is committed.
484 * The buffer must already be associated with the given transaction.
486 * First and last are numbers relative to the beginning of this buffer,
487 * so the first byte in the buffer is numbered 0 regardless of the
492 struct xfs_trans *tp,
497 struct xfs_buf_log_item *bip = bp->b_log_item;
499 ASSERT(first <= last && last < BBTOB(bp->b_length));
500 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
502 xfs_trans_dirty_buf(tp, bp);
504 trace_xfs_trans_log_buf(bip);
505 xfs_buf_item_log(bip, first, last);
510 * Invalidate a buffer that is being used within a transaction.
512 * Typically this is because the blocks in the buffer are being freed, so we
513 * need to prevent it from being written out when we're done. Allowing it
514 * to be written again might overwrite data in the free blocks if they are
515 * reallocated to a file.
517 * We prevent the buffer from being written out by marking it stale. We can't
518 * get rid of the buf log item at this point because the buffer may still be
519 * pinned by another transaction. If that is the case, then we'll wait until
520 * the buffer is committed to disk for the last time (we can tell by the ref
521 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
522 * keep the buffer locked so that the buffer and buf log item are not reused.
524 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
525 * the buf item. This will be used at recovery time to determine that copies
526 * of the buffer in the log before this should not be replayed.
528 * We mark the item descriptor and the transaction dirty so that we'll hold
529 * the buffer until after the commit.
531 * Since we're invalidating the buffer, we also clear the state about which
532 * parts of the buffer have been logged. We also clear the flag indicating
533 * that this is an inode buffer since the data in the buffer will no longer
536 * We set the stale bit in the buffer as well since we're getting rid of it.
543 struct xfs_buf_log_item *bip = bp->b_log_item;
546 ASSERT(bp->b_transp == tp);
548 ASSERT(atomic_read(&bip->bli_refcount) > 0);
550 trace_xfs_trans_binval(bip);
552 if (bip->bli_flags & XFS_BLI_STALE) {
554 * If the buffer is already invalidated, then
557 ASSERT(bp->b_flags & XBF_STALE);
558 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
559 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
560 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
561 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
562 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
563 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
569 bip->bli_flags |= XFS_BLI_STALE;
570 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
571 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
572 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
573 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
574 for (i = 0; i < bip->bli_format_count; i++) {
575 memset(bip->bli_formats[i].blf_data_map, 0,
576 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
578 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
579 tp->t_flags |= XFS_TRANS_DIRTY;
583 * This call is used to indicate that the buffer contains on-disk inodes which
584 * must be handled specially during recovery. They require special handling
585 * because only the di_next_unlinked from the inodes in the buffer should be
586 * recovered. The rest of the data in the buffer is logged via the inodes
589 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
590 * transferred to the buffer's log format structure so that we'll know what to
591 * do at recovery time.
598 struct xfs_buf_log_item *bip = bp->b_log_item;
600 ASSERT(bp->b_transp == tp);
602 ASSERT(atomic_read(&bip->bli_refcount) > 0);
604 bip->bli_flags |= XFS_BLI_INODE_BUF;
605 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
609 * This call is used to indicate that the buffer is going to
610 * be staled and was an inode buffer. This means it gets
611 * special processing during unpin - where any inodes
612 * associated with the buffer should be removed from ail.
613 * There is also special processing during recovery,
614 * any replay of the inodes in the buffer needs to be
615 * prevented as the buffer may have been reused.
618 xfs_trans_stale_inode_buf(
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_STALE_INODE;
629 bip->bli_item.li_cb = xfs_buf_iodone;
630 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
634 * Mark the buffer as being one which contains newly allocated
635 * inodes. We need to make sure that even if this buffer is
636 * relogged as an 'inode buf' we still recover all of the inode
637 * images in the face of a crash. This works in coordination with
638 * xfs_buf_item_committed() to ensure that the buffer remains in the
639 * AIL at its original location even after it has been relogged.
643 xfs_trans_inode_alloc_buf(
647 struct xfs_buf_log_item *bip = bp->b_log_item;
649 ASSERT(bp->b_transp == tp);
651 ASSERT(atomic_read(&bip->bli_refcount) > 0);
653 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
654 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
658 * Mark the buffer as ordered for this transaction. This means that the contents
659 * of the buffer are not recorded in the transaction but it is tracked in the
660 * AIL as though it was. This allows us to record logical changes in
661 * transactions rather than the physical changes we make to the buffer without
662 * changing writeback ordering constraints of metadata buffers.
665 xfs_trans_ordered_buf(
666 struct xfs_trans *tp,
669 struct xfs_buf_log_item *bip = bp->b_log_item;
671 ASSERT(bp->b_transp == tp);
673 ASSERT(atomic_read(&bip->bli_refcount) > 0);
675 if (xfs_buf_item_dirty_format(bip))
678 bip->bli_flags |= XFS_BLI_ORDERED;
679 trace_xfs_buf_item_ordered(bip);
682 * We don't log a dirty range of an ordered buffer but it still needs
683 * to be marked dirty and that it has been logged.
685 xfs_trans_dirty_buf(tp, bp);
690 * Set the type of the buffer for log recovery so that it can correctly identify
691 * and hence attach the correct buffer ops to the buffer after replay.
694 xfs_trans_buf_set_type(
695 struct xfs_trans *tp,
699 struct xfs_buf_log_item *bip = bp->b_log_item;
704 ASSERT(bp->b_transp == tp);
706 ASSERT(atomic_read(&bip->bli_refcount) > 0);
708 xfs_blft_to_flags(&bip->__bli_format, type);
712 xfs_trans_buf_copy_type(
713 struct xfs_buf *dst_bp,
714 struct xfs_buf *src_bp)
716 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
717 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
720 type = xfs_blft_from_flags(&sbip->__bli_format);
721 xfs_blft_to_flags(&dbip->__bli_format, type);
725 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
726 * dquots. However, unlike in inode buffer recovery, dquot buffers get
727 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
728 * The only thing that makes dquot buffers different from regular
729 * buffers is that we must not replay dquot bufs when recovering
730 * if a _corresponding_ quotaoff has happened. We also have to distinguish
731 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
732 * can be turned off independently.
741 struct xfs_buf_log_item *bip = bp->b_log_item;
743 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
744 type == XFS_BLF_PDQUOT_BUF ||
745 type == XFS_BLF_GDQUOT_BUF);
747 bip->__bli_format.blf_flags |= type;
750 case XFS_BLF_UDQUOT_BUF:
751 type = XFS_BLFT_UDQUOT_BUF;
753 case XFS_BLF_PDQUOT_BUF:
754 type = XFS_BLFT_PDQUOT_BUF;
756 case XFS_BLF_GDQUOT_BUF:
757 type = XFS_BLFT_GDQUOT_BUF;
760 type = XFS_BLFT_UNKNOWN_BUF;
764 xfs_trans_buf_set_type(tp, bp, type);