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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_buf_item.h"
30 #include "xfs_trans_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
35 * Check to see if a buffer matching the given parameters is already
36 * a part of the given transaction.
38 STATIC struct xfs_buf *
39 xfs_trans_buf_item_match(
41 struct xfs_buftarg *target,
42 struct xfs_buf_map *map,
45 struct xfs_log_item_desc *lidp;
46 struct xfs_buf_log_item *blip;
50 for (i = 0; i < nmaps; i++)
53 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
54 blip = (struct xfs_buf_log_item *)lidp->lid_item;
55 if (blip->bli_item.li_type == XFS_LI_BUF &&
56 blip->bli_buf->b_target == target &&
57 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
58 blip->bli_buf->b_length == len) {
59 ASSERT(blip->bli_buf->b_map_count == nmaps);
68 * Add the locked buffer to the transaction.
70 * The buffer must be locked, and it cannot be associated with any
73 * If the buffer does not yet have a buf log item associated with it,
74 * then allocate one for it. Then add the buf item to the transaction.
82 struct xfs_buf_log_item *bip;
84 ASSERT(bp->b_transp == NULL);
87 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
88 * it doesn't have one yet, then allocate one and initialize it.
89 * The checks to see if one is there are in xfs_buf_item_init().
91 xfs_buf_item_init(bp, tp->t_mountp);
93 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
94 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
95 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
100 * Take a reference for this transaction on the buf item.
102 atomic_inc(&bip->bli_refcount);
105 * Get a log_item_desc to point at the new item.
107 xfs_trans_add_item(tp, &bip->bli_item);
110 * Initialize b_fsprivate2 so we can find it with incore_match()
111 * in xfs_trans_get_buf() and friends above.
119 struct xfs_trans *tp,
122 _xfs_trans_bjoin(tp, bp, 0);
123 trace_xfs_trans_bjoin(bp->b_fspriv);
127 * Get and lock the buffer for the caller if it is not already
128 * locked within the given transaction. If it is already locked
129 * within the transaction, just increment its lock recursion count
130 * and return a pointer to it.
132 * If the transaction pointer is NULL, make this just a normal
136 xfs_trans_get_buf_map(
137 struct xfs_trans *tp,
138 struct xfs_buftarg *target,
139 struct xfs_buf_map *map,
141 xfs_buf_flags_t flags)
144 xfs_buf_log_item_t *bip;
147 return xfs_buf_get_map(target, map, nmaps, flags);
150 * If we find the buffer in the cache with this transaction
151 * pointer in its b_fsprivate2 field, then we know we already
152 * have it locked. In this case we just increment the lock
153 * recursion count and return the buffer to the caller.
155 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
157 ASSERT(xfs_buf_islocked(bp));
158 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
163 ASSERT(bp->b_transp == tp);
166 ASSERT(atomic_read(&bip->bli_refcount) > 0);
168 trace_xfs_trans_get_buf_recur(bip);
172 bp = xfs_buf_get_map(target, map, nmaps, flags);
177 ASSERT(!bp->b_error);
179 _xfs_trans_bjoin(tp, bp, 1);
180 trace_xfs_trans_get_buf(bp->b_fspriv);
185 * Get and lock the superblock buffer of this file system for the
188 * We don't need to use incore_match() here, because the superblock
189 * buffer is a private buffer which we keep a pointer to in the
193 xfs_trans_getsb(xfs_trans_t *tp,
194 struct xfs_mount *mp,
198 xfs_buf_log_item_t *bip;
201 * Default to just trying to lock the superblock buffer
205 return (xfs_getsb(mp, flags));
209 * If the superblock buffer already has this transaction
210 * pointer in its b_fsprivate2 field, then we know we already
211 * have it locked. In this case we just increment the lock
212 * recursion count and return the buffer to the caller.
215 if (bp->b_transp == tp) {
218 ASSERT(atomic_read(&bip->bli_refcount) > 0);
220 trace_xfs_trans_getsb_recur(bip);
224 bp = xfs_getsb(mp, flags);
228 _xfs_trans_bjoin(tp, bp, 1);
229 trace_xfs_trans_getsb(bp->b_fspriv);
234 xfs_buftarg_t *xfs_error_target;
237 int xfs_error_mod = 33;
241 * Get and lock the buffer for the caller if it is not already
242 * locked within the given transaction. If it has not yet been
243 * read in, read it from disk. If it is already locked
244 * within the transaction and already read in, just increment its
245 * lock recursion count and return a pointer to it.
247 * If the transaction pointer is NULL, make this just a normal
251 xfs_trans_read_buf_map(
252 struct xfs_mount *mp,
253 struct xfs_trans *tp,
254 struct xfs_buftarg *target,
255 struct xfs_buf_map *map,
257 xfs_buf_flags_t flags,
258 struct xfs_buf **bpp,
259 const struct xfs_buf_ops *ops)
262 xfs_buf_log_item_t *bip;
267 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
269 return (flags & XBF_TRYLOCK) ?
270 EAGAIN : XFS_ERROR(ENOMEM);
274 xfs_buf_ioerror_alert(bp, __func__);
282 if (xfs_error_target == target) {
283 if (((xfs_req_num++) % xfs_error_mod) == 0) {
285 xfs_debug(mp, "Returning error!");
286 return XFS_ERROR(EIO);
291 if (XFS_FORCED_SHUTDOWN(mp))
298 * If we find the buffer in the cache with this transaction
299 * pointer in its b_fsprivate2 field, then we know we already
300 * have it locked. If it is already read in we just increment
301 * the lock recursion count and return the buffer to the caller.
302 * If the buffer is not yet read in, then we read it in, increment
303 * the lock recursion count, and return it to the caller.
305 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
307 ASSERT(xfs_buf_islocked(bp));
308 ASSERT(bp->b_transp == tp);
309 ASSERT(bp->b_fspriv != NULL);
310 ASSERT(!bp->b_error);
311 if (!(XFS_BUF_ISDONE(bp))) {
312 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
313 ASSERT(!XFS_BUF_ISASYNC(bp));
314 ASSERT(bp->b_iodone == NULL);
317 xfsbdstrat(tp->t_mountp, bp);
318 error = xfs_buf_iowait(bp);
320 xfs_buf_ioerror_alert(bp, __func__);
323 * We can gracefully recover from most read
324 * errors. Ones we can't are those that happen
325 * after the transaction's already dirty.
327 if (tp->t_flags & XFS_TRANS_DIRTY)
328 xfs_force_shutdown(tp->t_mountp,
329 SHUTDOWN_META_IO_ERROR);
334 * We never locked this buf ourselves, so we shouldn't
335 * brelse it either. Just get out.
337 if (XFS_FORCED_SHUTDOWN(mp)) {
338 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
340 return XFS_ERROR(EIO);
347 ASSERT(atomic_read(&bip->bli_refcount) > 0);
348 trace_xfs_trans_read_buf_recur(bip);
353 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
356 return (flags & XBF_TRYLOCK) ?
357 0 : XFS_ERROR(ENOMEM);
363 xfs_buf_ioerror_alert(bp, __func__);
364 if (tp->t_flags & XFS_TRANS_DIRTY)
365 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
370 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
371 if (xfs_error_target == target) {
372 if (((xfs_req_num++) % xfs_error_mod) == 0) {
373 xfs_force_shutdown(tp->t_mountp,
374 SHUTDOWN_META_IO_ERROR);
376 xfs_debug(mp, "Returning trans error!");
377 return XFS_ERROR(EIO);
382 if (XFS_FORCED_SHUTDOWN(mp))
385 _xfs_trans_bjoin(tp, bp, 1);
386 trace_xfs_trans_read_buf(bp->b_fspriv);
392 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
395 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;
605 * If we have an ordered buffer we are not logging any dirty range but
606 * it still needs to be marked dirty and that it has been logged.
608 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
609 if (!(bip->bli_flags & XFS_BLI_ORDERED))
610 xfs_buf_item_log(bip, first, last);
615 * Invalidate a buffer that is being used within a transaction.
617 * Typically this is because the blocks in the buffer are being freed, so we
618 * need to prevent it from being written out when we're done. Allowing it
619 * to be written again might overwrite data in the free blocks if they are
620 * reallocated to a file.
622 * We prevent the buffer from being written out by marking it stale. We can't
623 * get rid of the buf log item at this point because the buffer may still be
624 * pinned by another transaction. If that is the case, then we'll wait until
625 * the buffer is committed to disk for the last time (we can tell by the ref
626 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
627 * keep the buffer locked so that the buffer and buf log item are not reused.
629 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
630 * the buf item. This will be used at recovery time to determine that copies
631 * of the buffer in the log before this should not be replayed.
633 * We mark the item descriptor and the transaction dirty so that we'll hold
634 * the buffer until after the commit.
636 * Since we're invalidating the buffer, we also clear the state about which
637 * parts of the buffer have been logged. We also clear the flag indicating
638 * that this is an inode buffer since the data in the buffer will no longer
641 * We set the stale bit in the buffer as well since we're getting rid of it.
648 xfs_buf_log_item_t *bip = bp->b_fspriv;
651 ASSERT(bp->b_transp == tp);
653 ASSERT(atomic_read(&bip->bli_refcount) > 0);
655 trace_xfs_trans_binval(bip);
657 if (bip->bli_flags & XFS_BLI_STALE) {
659 * If the buffer is already invalidated, then
662 ASSERT(XFS_BUF_ISSTALE(bp));
663 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
664 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
665 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
666 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
667 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
668 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
674 bip->bli_flags |= XFS_BLI_STALE;
675 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
676 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
677 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
678 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
679 for (i = 0; i < bip->bli_format_count; i++) {
680 memset(bip->bli_formats[i].blf_data_map, 0,
681 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
683 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
684 tp->t_flags |= XFS_TRANS_DIRTY;
688 * This call is used to indicate that the buffer contains on-disk inodes which
689 * must be handled specially during recovery. They require special handling
690 * because only the di_next_unlinked from the inodes in the buffer should be
691 * recovered. The rest of the data in the buffer is logged via the inodes
694 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
695 * transferred to the buffer's log format structure so that we'll know what to
696 * do at recovery time.
703 xfs_buf_log_item_t *bip = bp->b_fspriv;
705 ASSERT(bp->b_transp == tp);
707 ASSERT(atomic_read(&bip->bli_refcount) > 0);
709 bip->bli_flags |= XFS_BLI_INODE_BUF;
710 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
714 * This call is used to indicate that the buffer is going to
715 * be staled and was an inode buffer. This means it gets
716 * special processing during unpin - where any inodes
717 * associated with the buffer should be removed from ail.
718 * There is also special processing during recovery,
719 * any replay of the inodes in the buffer needs to be
720 * prevented as the buffer may have been reused.
723 xfs_trans_stale_inode_buf(
727 xfs_buf_log_item_t *bip = bp->b_fspriv;
729 ASSERT(bp->b_transp == tp);
731 ASSERT(atomic_read(&bip->bli_refcount) > 0);
733 bip->bli_flags |= XFS_BLI_STALE_INODE;
734 bip->bli_item.li_cb = xfs_buf_iodone;
735 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
739 * Mark the buffer as being one which contains newly allocated
740 * inodes. We need to make sure that even if this buffer is
741 * relogged as an 'inode buf' we still recover all of the inode
742 * images in the face of a crash. This works in coordination with
743 * xfs_buf_item_committed() to ensure that the buffer remains in the
744 * AIL at its original location even after it has been relogged.
748 xfs_trans_inode_alloc_buf(
752 xfs_buf_log_item_t *bip = bp->b_fspriv;
754 ASSERT(bp->b_transp == tp);
756 ASSERT(atomic_read(&bip->bli_refcount) > 0);
758 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
759 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
763 * Mark the buffer as ordered for this transaction. This means
764 * that the contents of the buffer are not recorded in the transaction
765 * but it is tracked in the AIL as though it was. This allows us
766 * to record logical changes in transactions rather than the physical
767 * changes we make to the buffer without changing writeback ordering
768 * constraints of metadata buffers.
771 xfs_trans_ordered_buf(
772 struct xfs_trans *tp,
775 struct xfs_buf_log_item *bip = bp->b_fspriv;
777 ASSERT(bp->b_transp == tp);
779 ASSERT(atomic_read(&bip->bli_refcount) > 0);
781 bip->bli_flags |= XFS_BLI_ORDERED;
782 trace_xfs_buf_item_ordered(bip);
786 * Set the type of the buffer for log recovery so that it can correctly identify
787 * and hence attach the correct buffer ops to the buffer after replay.
790 xfs_trans_buf_set_type(
791 struct xfs_trans *tp,
795 struct xfs_buf_log_item *bip = bp->b_fspriv;
800 ASSERT(bp->b_transp == tp);
802 ASSERT(atomic_read(&bip->bli_refcount) > 0);
804 xfs_blft_to_flags(&bip->__bli_format, type);
808 xfs_trans_buf_copy_type(
809 struct xfs_buf *dst_bp,
810 struct xfs_buf *src_bp)
812 struct xfs_buf_log_item *sbip = src_bp->b_fspriv;
813 struct xfs_buf_log_item *dbip = dst_bp->b_fspriv;
816 type = xfs_blft_from_flags(&sbip->__bli_format);
817 xfs_blft_to_flags(&dbip->__bli_format, type);
821 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
822 * dquots. However, unlike in inode buffer recovery, dquot buffers get
823 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
824 * The only thing that makes dquot buffers different from regular
825 * buffers is that we must not replay dquot bufs when recovering
826 * if a _corresponding_ quotaoff has happened. We also have to distinguish
827 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
828 * can be turned off independently.
837 struct xfs_buf_log_item *bip = bp->b_fspriv;
839 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
840 type == XFS_BLF_PDQUOT_BUF ||
841 type == XFS_BLF_GDQUOT_BUF);
843 bip->__bli_format.blf_flags |= type;
846 case XFS_BLF_UDQUOT_BUF:
847 type = XFS_BLFT_UDQUOT_BUF;
849 case XFS_BLF_PDQUOT_BUF:
850 type = XFS_BLFT_PDQUOT_BUF;
852 case XFS_BLF_GDQUOT_BUF:
853 type = XFS_BLFT_GDQUOT_BUF;
856 type = XFS_BLFT_UNKNOWN_BUF;
860 xfs_trans_buf_set_type(tp, bp, type);