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_daddr(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_is_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 for the given transaction.
173 struct xfs_trans *tp)
175 struct xfs_buf *bp = tp->t_mountp->m_sb_bp;
178 * Just increment the lock recursion count if the buffer is already
179 * attached to this transaction.
181 if (bp->b_transp == tp) {
182 struct xfs_buf_log_item *bip = bp->b_log_item;
185 ASSERT(atomic_read(&bip->bli_refcount) > 0);
188 trace_xfs_trans_getsb_recur(bip);
192 _xfs_trans_bjoin(tp, bp, 1);
194 trace_xfs_trans_getsb(bp->b_log_item);
201 * Get and lock the buffer for the caller if it is not already
202 * locked within the given transaction. If it has not yet been
203 * read in, read it from disk. If it is already locked
204 * within the transaction and already read in, just increment its
205 * lock recursion count and return a pointer to it.
207 * If the transaction pointer is NULL, make this just a normal
211 xfs_trans_read_buf_map(
212 struct xfs_mount *mp,
213 struct xfs_trans *tp,
214 struct xfs_buftarg *target,
215 struct xfs_buf_map *map,
217 xfs_buf_flags_t flags,
218 struct xfs_buf **bpp,
219 const struct xfs_buf_ops *ops)
221 struct xfs_buf *bp = NULL;
222 struct xfs_buf_log_item *bip;
227 * If we find the buffer in the cache with this transaction
228 * pointer in its b_fsprivate2 field, then we know we already
229 * have it locked. If it is already read in we just increment
230 * the lock recursion count and return the buffer to the caller.
231 * If the buffer is not yet read in, then we read it in, increment
232 * the lock recursion count, and return it to the caller.
235 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
237 ASSERT(xfs_buf_islocked(bp));
238 ASSERT(bp->b_transp == tp);
239 ASSERT(bp->b_log_item != NULL);
240 ASSERT(!bp->b_error);
241 ASSERT(bp->b_flags & XBF_DONE);
244 * We never locked this buf ourselves, so we shouldn't
245 * brelse it either. Just get out.
247 if (xfs_is_shutdown(mp)) {
248 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
253 * Check if the caller is trying to read a buffer that is
254 * already attached to the transaction yet has no buffer ops
255 * assigned. Ops are usually attached when the buffer is
256 * attached to the transaction, or by the read caller if
257 * special circumstances. That didn't happen, which is not
258 * how this is supposed to go.
260 * If the buffer passes verification we'll let this go, but if
261 * not we have to shut down. Let the transaction cleanup code
262 * release this buffer when it kills the tranaction.
264 ASSERT(bp->b_ops != NULL);
265 error = xfs_buf_reverify(bp, ops);
267 xfs_buf_ioerror_alert(bp, __return_address);
269 if (tp->t_flags & XFS_TRANS_DIRTY)
270 xfs_force_shutdown(tp->t_mountp,
271 SHUTDOWN_META_IO_ERROR);
273 /* bad CRC means corrupted metadata */
274 if (error == -EFSBADCRC)
275 error = -EFSCORRUPTED;
279 bip = bp->b_log_item;
282 ASSERT(atomic_read(&bip->bli_refcount) > 0);
283 trace_xfs_trans_read_buf_recur(bip);
284 ASSERT(bp->b_ops != NULL || ops == NULL);
289 error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
295 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
296 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
303 if (xfs_is_shutdown(mp)) {
305 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
310 _xfs_trans_bjoin(tp, bp, 1);
311 trace_xfs_trans_read_buf(bp->b_log_item);
313 ASSERT(bp->b_ops != NULL || ops == NULL);
319 /* Has this buffer been dirtied by anyone? */
321 xfs_trans_buf_is_dirty(
324 struct xfs_buf_log_item *bip = bp->b_log_item;
328 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
329 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
333 * Release a buffer previously joined to the transaction. If the buffer is
334 * modified within this transaction, decrement the recursion count but do not
335 * release the buffer even if the count goes to 0. If the buffer is not modified
336 * within the transaction, decrement the recursion count and release the buffer
337 * if the recursion count goes to 0.
339 * If the buffer is to be released and it was not already dirty before this
340 * transaction began, then also free the buf_log_item associated with it.
342 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
346 struct xfs_trans *tp,
349 struct xfs_buf_log_item *bip = bp->b_log_item;
351 ASSERT(bp->b_transp == tp);
358 trace_xfs_trans_brelse(bip);
359 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
360 ASSERT(atomic_read(&bip->bli_refcount) > 0);
363 * If the release is for a recursive lookup, then decrement the count
366 if (bip->bli_recur > 0) {
372 * If the buffer is invalidated or dirty in this transaction, we can't
373 * release it until we commit.
375 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
377 if (bip->bli_flags & XFS_BLI_STALE)
381 * Unlink the log item from the transaction and clear the hold flag, if
382 * set. We wouldn't want the next user of the buffer to get confused.
384 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
385 xfs_trans_del_item(&bip->bli_item);
386 bip->bli_flags &= ~XFS_BLI_HOLD;
388 /* drop the reference to the bli */
389 xfs_buf_item_put(bip);
396 * Mark the buffer as not needing to be unlocked when the buf item's
397 * iop_committing() routine is called. The buffer must already be locked
398 * and associated with the given transaction.
406 struct xfs_buf_log_item *bip = bp->b_log_item;
408 ASSERT(bp->b_transp == tp);
410 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
411 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
412 ASSERT(atomic_read(&bip->bli_refcount) > 0);
414 bip->bli_flags |= XFS_BLI_HOLD;
415 trace_xfs_trans_bhold(bip);
419 * Cancel the previous buffer hold request made on this buffer
420 * for this transaction.
423 xfs_trans_bhold_release(
427 struct xfs_buf_log_item *bip = bp->b_log_item;
429 ASSERT(bp->b_transp == tp);
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);
434 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
436 bip->bli_flags &= ~XFS_BLI_HOLD;
437 trace_xfs_trans_bhold_release(bip);
441 * Mark a buffer dirty in the transaction.
445 struct xfs_trans *tp,
448 struct xfs_buf_log_item *bip = bp->b_log_item;
450 ASSERT(bp->b_transp == tp);
454 * Mark the buffer as needing to be written out eventually,
455 * and set its iodone function to remove the buffer's buf log
456 * item from the AIL and free it when the buffer is flushed
459 bp->b_flags |= XBF_DONE;
461 ASSERT(atomic_read(&bip->bli_refcount) > 0);
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 bp->b_flags |= _XBF_INODES;
606 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
610 * This call is used to indicate that the buffer is going to
611 * be staled and was an inode buffer. This means it gets
612 * special processing during unpin - where any inodes
613 * associated with the buffer should be removed from ail.
614 * There is also special processing during recovery,
615 * any replay of the inodes in the buffer needs to be
616 * prevented as the buffer may have been reused.
619 xfs_trans_stale_inode_buf(
623 struct xfs_buf_log_item *bip = bp->b_log_item;
625 ASSERT(bp->b_transp == tp);
627 ASSERT(atomic_read(&bip->bli_refcount) > 0);
629 bip->bli_flags |= XFS_BLI_STALE_INODE;
630 bp->b_flags |= _XBF_INODES;
631 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
635 * Mark the buffer as being one which contains newly allocated
636 * inodes. We need to make sure that even if this buffer is
637 * relogged as an 'inode buf' we still recover all of the inode
638 * images in the face of a crash. This works in coordination with
639 * xfs_buf_item_committed() to ensure that the buffer remains in the
640 * AIL at its original location even after it has been relogged.
644 xfs_trans_inode_alloc_buf(
648 struct xfs_buf_log_item *bip = bp->b_log_item;
650 ASSERT(bp->b_transp == tp);
652 ASSERT(atomic_read(&bip->bli_refcount) > 0);
654 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
655 bp->b_flags |= _XBF_INODES;
656 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
660 * Mark the buffer as ordered for this transaction. This means that the contents
661 * of the buffer are not recorded in the transaction but it is tracked in the
662 * AIL as though it was. This allows us to record logical changes in
663 * transactions rather than the physical changes we make to the buffer without
664 * changing writeback ordering constraints of metadata buffers.
667 xfs_trans_ordered_buf(
668 struct xfs_trans *tp,
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 if (xfs_buf_item_dirty_format(bip))
680 bip->bli_flags |= XFS_BLI_ORDERED;
681 trace_xfs_buf_item_ordered(bip);
684 * We don't log a dirty range of an ordered buffer but it still needs
685 * to be marked dirty and that it has been logged.
687 xfs_trans_dirty_buf(tp, bp);
692 * Set the type of the buffer for log recovery so that it can correctly identify
693 * and hence attach the correct buffer ops to the buffer after replay.
696 xfs_trans_buf_set_type(
697 struct xfs_trans *tp,
701 struct xfs_buf_log_item *bip = bp->b_log_item;
706 ASSERT(bp->b_transp == tp);
708 ASSERT(atomic_read(&bip->bli_refcount) > 0);
710 xfs_blft_to_flags(&bip->__bli_format, type);
714 xfs_trans_buf_copy_type(
715 struct xfs_buf *dst_bp,
716 struct xfs_buf *src_bp)
718 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
719 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
722 type = xfs_blft_from_flags(&sbip->__bli_format);
723 xfs_blft_to_flags(&dbip->__bli_format, type);
727 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
728 * dquots. However, unlike in inode buffer recovery, dquot buffers get
729 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
730 * The only thing that makes dquot buffers different from regular
731 * buffers is that we must not replay dquot bufs when recovering
732 * if a _corresponding_ quotaoff has happened. We also have to distinguish
733 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
734 * can be turned off independently.
743 struct xfs_buf_log_item *bip = bp->b_log_item;
745 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
746 type == XFS_BLF_PDQUOT_BUF ||
747 type == XFS_BLF_GDQUOT_BUF);
749 bip->__bli_format.blf_flags |= type;
752 case XFS_BLF_UDQUOT_BUF:
753 type = XFS_BLFT_UDQUOT_BUF;
755 case XFS_BLF_PDQUOT_BUF:
756 type = XFS_BLFT_PDQUOT_BUF;
758 case XFS_BLF_GDQUOT_BUF:
759 type = XFS_BLFT_GDQUOT_BUF;
762 type = XFS_BLFT_UNKNOWN_BUF;
766 bp->b_flags |= _XBF_DQUOTS;
767 xfs_trans_buf_set_type(tp, bp, type);