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_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_trans_priv.h"
35 #include "xfs_error.h"
37 #include "xfs_trace.h"
40 * Check to see if a buffer matching the given parameters is already
41 * a part of the given transaction.
43 STATIC struct xfs_buf *
44 xfs_trans_buf_item_match(
46 struct xfs_buftarg *target,
50 struct xfs_log_item_desc *lidp;
51 struct xfs_buf_log_item *blip;
54 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
55 blip = (struct xfs_buf_log_item *)lidp->lid_item;
56 if (blip->bli_item.li_type == XFS_LI_BUF &&
57 blip->bli_buf->b_target == target &&
58 XFS_BUF_ADDR(blip->bli_buf) == blkno &&
59 XFS_BUF_COUNT(blip->bli_buf) == len)
67 * Add the locked buffer to the transaction.
69 * The buffer must be locked, and it cannot be associated with any
72 * If the buffer does not yet have a buf log item associated with it,
73 * then allocate one for it. Then add the buf item to the transaction.
81 struct xfs_buf_log_item *bip;
83 ASSERT(bp->b_transp == NULL);
86 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
87 * it doesn't have one yet, then allocate one and initialize it.
88 * The checks to see if one is there are in xfs_buf_item_init().
90 xfs_buf_item_init(bp, tp->t_mountp);
92 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
93 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
94 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
99 * Take a reference for this transaction on the buf item.
101 atomic_inc(&bip->bli_refcount);
104 * Get a log_item_desc to point at the new item.
106 xfs_trans_add_item(tp, &bip->bli_item);
109 * Initialize b_fsprivate2 so we can find it with incore_match()
110 * in xfs_trans_get_buf() and friends above.
118 struct xfs_trans *tp,
121 _xfs_trans_bjoin(tp, bp, 0);
122 trace_xfs_trans_bjoin(bp->b_fspriv);
126 * Get and lock the buffer for the caller if it is not already
127 * locked within the given transaction. If it is already locked
128 * within the transaction, just increment its lock recursion count
129 * and return a pointer to it.
131 * If the transaction pointer is NULL, make this just a normal
135 xfs_trans_get_buf(xfs_trans_t *tp,
136 xfs_buftarg_t *target_dev,
142 xfs_buf_log_item_t *bip;
145 flags = XBF_LOCK | XBF_MAPPED;
148 * Default to a normal get_buf() call if the tp is NULL.
151 return xfs_buf_get(target_dev, blkno, len,
152 flags | XBF_DONT_BLOCK);
155 * If we find the buffer in the cache with this transaction
156 * pointer in its b_fsprivate2 field, then we know we already
157 * have it locked. In this case we just increment the lock
158 * recursion count and return the buffer to the caller.
160 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
162 ASSERT(xfs_buf_islocked(bp));
163 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
165 xfs_buf_delwri_dequeue(bp);
170 * If the buffer is stale then it was binval'ed
171 * since last read. This doesn't matter since the
172 * caller isn't allowed to use the data anyway.
174 else if (XFS_BUF_ISSTALE(bp))
175 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
177 ASSERT(bp->b_transp == tp);
180 ASSERT(atomic_read(&bip->bli_refcount) > 0);
182 trace_xfs_trans_get_buf_recur(bip);
187 * We always specify the XBF_DONT_BLOCK flag within a transaction
188 * so that get_buf does not try to push out a delayed write buffer
189 * which might cause another transaction to take place (if the
190 * buffer was delayed alloc). Such recursive transactions can
191 * easily deadlock with our current transaction as well as cause
192 * us to run out of stack space.
194 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
199 ASSERT(!bp->b_error);
201 _xfs_trans_bjoin(tp, bp, 1);
202 trace_xfs_trans_get_buf(bp->b_fspriv);
207 * Get and lock the superblock buffer of this file system for the
210 * We don't need to use incore_match() here, because the superblock
211 * buffer is a private buffer which we keep a pointer to in the
215 xfs_trans_getsb(xfs_trans_t *tp,
216 struct xfs_mount *mp,
220 xfs_buf_log_item_t *bip;
223 * Default to just trying to lock the superblock buffer
227 return (xfs_getsb(mp, flags));
231 * If the superblock buffer already has this transaction
232 * pointer in its b_fsprivate2 field, then we know we already
233 * have it locked. In this case we just increment the lock
234 * recursion count and return the buffer to the caller.
237 if (bp->b_transp == tp) {
240 ASSERT(atomic_read(&bip->bli_refcount) > 0);
242 trace_xfs_trans_getsb_recur(bip);
246 bp = xfs_getsb(mp, flags);
250 _xfs_trans_bjoin(tp, bp, 1);
251 trace_xfs_trans_getsb(bp->b_fspriv);
256 xfs_buftarg_t *xfs_error_target;
259 int xfs_error_mod = 33;
263 * Get and lock the buffer for the caller if it is not already
264 * locked within the given transaction. If it has not yet been
265 * read in, read it from disk. If it is already locked
266 * within the transaction and already read in, just increment its
267 * lock recursion count and return a pointer to it.
269 * If the transaction pointer is NULL, make this just a normal
276 xfs_buftarg_t *target,
283 xfs_buf_log_item_t *bip;
287 flags = XBF_LOCK | XBF_MAPPED;
290 * Default to a normal get_buf() call if the tp is NULL.
293 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
295 return (flags & XBF_TRYLOCK) ?
296 EAGAIN : XFS_ERROR(ENOMEM);
300 xfs_ioerror_alert("xfs_trans_read_buf", mp,
307 if (xfs_error_target == target) {
308 if (((xfs_req_num++) % xfs_error_mod) == 0) {
310 xfs_debug(mp, "Returning error!");
311 return XFS_ERROR(EIO);
316 if (XFS_FORCED_SHUTDOWN(mp))
323 * If we find the buffer in the cache with this transaction
324 * pointer in its b_fsprivate2 field, then we know we already
325 * have it locked. If it is already read in we just increment
326 * the lock recursion count and return the buffer to the caller.
327 * If the buffer is not yet read in, then we read it in, increment
328 * the lock recursion count, and return it to the caller.
330 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
332 ASSERT(xfs_buf_islocked(bp));
333 ASSERT(bp->b_transp == tp);
334 ASSERT(bp->b_fspriv != NULL);
335 ASSERT(!bp->b_error);
336 if (!(XFS_BUF_ISDONE(bp))) {
337 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
338 ASSERT(!XFS_BUF_ISASYNC(bp));
340 xfsbdstrat(tp->t_mountp, bp);
341 error = xfs_buf_iowait(bp);
343 xfs_ioerror_alert("xfs_trans_read_buf", mp,
347 * We can gracefully recover from most read
348 * errors. Ones we can't are those that happen
349 * after the transaction's already dirty.
351 if (tp->t_flags & XFS_TRANS_DIRTY)
352 xfs_force_shutdown(tp->t_mountp,
353 SHUTDOWN_META_IO_ERROR);
358 * We never locked this buf ourselves, so we shouldn't
359 * brelse it either. Just get out.
361 if (XFS_FORCED_SHUTDOWN(mp)) {
362 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
364 return XFS_ERROR(EIO);
371 ASSERT(atomic_read(&bip->bli_refcount) > 0);
372 trace_xfs_trans_read_buf_recur(bip);
378 * We always specify the XBF_DONT_BLOCK flag within a transaction
379 * so that get_buf does not try to push out a delayed write buffer
380 * which might cause another transaction to take place (if the
381 * buffer was delayed alloc). Such recursive transactions can
382 * easily deadlock with our current transaction as well as cause
383 * us to run out of stack space.
385 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
388 return (flags & XBF_TRYLOCK) ?
389 0 : XFS_ERROR(ENOMEM);
394 xfs_buf_delwri_dequeue(bp);
396 xfs_ioerror_alert("xfs_trans_read_buf", mp,
398 if (tp->t_flags & XFS_TRANS_DIRTY)
399 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
404 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
405 if (xfs_error_target == target) {
406 if (((xfs_req_num++) % xfs_error_mod) == 0) {
407 xfs_force_shutdown(tp->t_mountp,
408 SHUTDOWN_META_IO_ERROR);
410 xfs_debug(mp, "Returning trans error!");
411 return XFS_ERROR(EIO);
416 if (XFS_FORCED_SHUTDOWN(mp))
419 _xfs_trans_bjoin(tp, bp, 1);
420 trace_xfs_trans_read_buf(bp->b_fspriv);
427 * the theory here is that buffer is good but we're
428 * bailing out because the filesystem is being forcibly
429 * shut down. So we should leave the b_flags alone since
430 * the buffer's not staled and just get out.
433 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
434 xfs_notice(mp, "about to pop assert, bp == 0x%p", bp);
436 ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) !=
437 (XBF_STALE|XBF_DELWRI));
439 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
442 return XFS_ERROR(EIO);
447 * Release the buffer bp which was previously acquired with one of the
448 * xfs_trans_... buffer allocation routines if the buffer has not
449 * been modified within this transaction. If the buffer is modified
450 * within this transaction, do decrement the recursion count but do
451 * not release the buffer even if the count goes to 0. If the buffer is not
452 * modified within the transaction, decrement the recursion count and
453 * release the buffer if the recursion count goes to 0.
455 * If the buffer is to be released and it was not modified before
456 * this transaction began, then free the buf_log_item associated with it.
458 * If the transaction pointer is NULL, make this just a normal
462 xfs_trans_brelse(xfs_trans_t *tp,
465 xfs_buf_log_item_t *bip;
468 * Default to a normal brelse() call if the tp is NULL.
471 struct xfs_log_item *lip = bp->b_fspriv;
473 ASSERT(bp->b_transp == NULL);
476 * If there's a buf log item attached to the buffer,
477 * then let the AIL know that the buffer is being
480 if (lip != NULL && lip->li_type == XFS_LI_BUF) {
482 xfs_trans_unlocked_item(bip->bli_item.li_ailp, lip);
488 ASSERT(bp->b_transp == tp);
490 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
491 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
492 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
493 ASSERT(atomic_read(&bip->bli_refcount) > 0);
495 trace_xfs_trans_brelse(bip);
498 * If the release is just for a recursive lock,
499 * then decrement the count and return.
501 if (bip->bli_recur > 0) {
507 * If the buffer is dirty within this transaction, we can't
508 * release it until we commit.
510 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
514 * If the buffer has been invalidated, then we can't release
515 * it until the transaction commits to disk unless it is re-dirtied
516 * as part of this transaction. This prevents us from pulling
517 * the item from the AIL before we should.
519 if (bip->bli_flags & XFS_BLI_STALE)
522 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
525 * Free up the log item descriptor tracking the released item.
527 xfs_trans_del_item(&bip->bli_item);
530 * Clear the hold flag in the buf log item if it is set.
531 * We wouldn't want the next user of the buffer to
534 if (bip->bli_flags & XFS_BLI_HOLD) {
535 bip->bli_flags &= ~XFS_BLI_HOLD;
539 * Drop our reference to the buf log item.
541 atomic_dec(&bip->bli_refcount);
544 * If the buf item is not tracking data in the log, then
545 * we must free it before releasing the buffer back to the
546 * free pool. Before releasing the buffer to the free pool,
547 * clear the transaction pointer in b_fsprivate2 to dissolve
548 * its relation to this transaction.
550 if (!xfs_buf_item_dirty(bip)) {
552 ASSERT(bp->b_pincount == 0);
554 ASSERT(atomic_read(&bip->bli_refcount) == 0);
555 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
556 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
557 xfs_buf_item_relse(bp);
563 * If we've still got a buf log item on the buffer, then
564 * tell the AIL that the buffer is being unlocked.
567 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
568 (xfs_log_item_t*)bip);
576 * Mark the buffer as not needing to be unlocked when the buf item's
577 * IOP_UNLOCK() routine is called. The buffer must already be locked
578 * and associated with the given transaction.
582 xfs_trans_bhold(xfs_trans_t *tp,
585 xfs_buf_log_item_t *bip = bp->b_fspriv;
587 ASSERT(bp->b_transp == tp);
589 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
590 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
591 ASSERT(atomic_read(&bip->bli_refcount) > 0);
593 bip->bli_flags |= XFS_BLI_HOLD;
594 trace_xfs_trans_bhold(bip);
598 * Cancel the previous buffer hold request made on this buffer
599 * for this transaction.
602 xfs_trans_bhold_release(xfs_trans_t *tp,
605 xfs_buf_log_item_t *bip = bp->b_fspriv;
607 ASSERT(bp->b_transp == tp);
609 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
610 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
611 ASSERT(atomic_read(&bip->bli_refcount) > 0);
612 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
614 bip->bli_flags &= ~XFS_BLI_HOLD;
615 trace_xfs_trans_bhold_release(bip);
619 * This is called to mark bytes first through last inclusive of the given
620 * buffer as needing to be logged when the transaction is committed.
621 * The buffer must already be associated with the given transaction.
623 * First and last are numbers relative to the beginning of this buffer,
624 * so the first byte in the buffer is numbered 0 regardless of the
628 xfs_trans_log_buf(xfs_trans_t *tp,
633 xfs_buf_log_item_t *bip = bp->b_fspriv;
635 ASSERT(bp->b_transp == tp);
637 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
638 ASSERT(bp->b_iodone == NULL ||
639 bp->b_iodone == xfs_buf_iodone_callbacks);
642 * Mark the buffer as needing to be written out eventually,
643 * and set its iodone function to remove the buffer's buf log
644 * item from the AIL and free it when the buffer is flushed
645 * to disk. See xfs_buf_attach_iodone() for more details
646 * on li_cb and xfs_buf_iodone_callbacks().
647 * If we end up aborting this transaction, we trap this buffer
648 * inside the b_bdstrat callback so that this won't get written to
653 ASSERT(atomic_read(&bip->bli_refcount) > 0);
654 bp->b_iodone = xfs_buf_iodone_callbacks;
655 bip->bli_item.li_cb = xfs_buf_iodone;
657 xfs_buf_delwri_queue(bp);
659 trace_xfs_trans_log_buf(bip);
662 * If we invalidated the buffer within this transaction, then
663 * cancel the invalidation now that we're dirtying the buffer
664 * again. There are no races with the code in xfs_buf_item_unpin(),
665 * because we have a reference to the buffer this entire time.
667 if (bip->bli_flags & XFS_BLI_STALE) {
668 bip->bli_flags &= ~XFS_BLI_STALE;
669 ASSERT(XFS_BUF_ISSTALE(bp));
671 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
674 tp->t_flags |= XFS_TRANS_DIRTY;
675 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
676 bip->bli_flags |= XFS_BLI_LOGGED;
677 xfs_buf_item_log(bip, first, last);
682 * This called to invalidate a buffer that is being used within
683 * a transaction. Typically this is because the blocks in the
684 * buffer are being freed, so we need to prevent it from being
685 * written out when we're done. Allowing it to be written again
686 * might overwrite data in the free blocks if they are reallocated
689 * We prevent the buffer from being written out by clearing the
690 * B_DELWRI flag. We can't always
691 * get rid of the buf log item at this point, though, because
692 * the buffer may still be pinned by another transaction. If that
693 * is the case, then we'll wait until the buffer is committed to
694 * disk for the last time (we can tell by the ref count) and
695 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
696 * will keep the buffer locked so that the buffer and buf log item
704 xfs_buf_log_item_t *bip = bp->b_fspriv;
706 ASSERT(bp->b_transp == tp);
708 ASSERT(atomic_read(&bip->bli_refcount) > 0);
710 trace_xfs_trans_binval(bip);
712 if (bip->bli_flags & XFS_BLI_STALE) {
714 * If the buffer is already invalidated, then
717 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
718 ASSERT(XFS_BUF_ISSTALE(bp));
719 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
720 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
721 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
722 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
723 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
728 * Clear the dirty bit in the buffer and set the STALE flag
729 * in the buf log item. The STALE flag will be used in
730 * xfs_buf_item_unpin() to determine if it should clean up
731 * when the last reference to the buf item is given up.
732 * We set the XFS_BLF_CANCEL flag in the buf log format structure
733 * and log the buf item. This will be used at recovery time
734 * to determine that copies of the buffer in the log before
735 * this should not be replayed.
736 * We mark the item descriptor and the transaction dirty so
737 * that we'll hold the buffer until after the commit.
739 * Since we're invalidating the buffer, we also clear the state
740 * about which parts of the buffer have been logged. We also
741 * clear the flag indicating that this is an inode buffer since
742 * the data in the buffer will no longer be valid.
744 * We set the stale bit in the buffer as well since we're getting
747 xfs_buf_delwri_dequeue(bp);
749 bip->bli_flags |= XFS_BLI_STALE;
750 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
751 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
752 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
753 memset((char *)(bip->bli_format.blf_data_map), 0,
754 (bip->bli_format.blf_map_size * sizeof(uint)));
755 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
756 tp->t_flags |= XFS_TRANS_DIRTY;
760 * This call is used to indicate that the buffer contains on-disk inodes which
761 * must be handled specially during recovery. They require special handling
762 * because only the di_next_unlinked from the inodes in the buffer should be
763 * recovered. The rest of the data in the buffer is logged via the inodes
766 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
767 * transferred to the buffer's log format structure so that we'll know what to
768 * do at recovery time.
775 xfs_buf_log_item_t *bip = bp->b_fspriv;
777 ASSERT(bp->b_transp == tp);
779 ASSERT(atomic_read(&bip->bli_refcount) > 0);
781 bip->bli_flags |= XFS_BLI_INODE_BUF;
785 * This call is used to indicate that the buffer is going to
786 * be staled and was an inode buffer. This means it gets
787 * special processing during unpin - where any inodes
788 * associated with the buffer should be removed from ail.
789 * There is also special processing during recovery,
790 * any replay of the inodes in the buffer needs to be
791 * prevented as the buffer may have been reused.
794 xfs_trans_stale_inode_buf(
798 xfs_buf_log_item_t *bip = bp->b_fspriv;
800 ASSERT(bp->b_transp == tp);
802 ASSERT(atomic_read(&bip->bli_refcount) > 0);
804 bip->bli_flags |= XFS_BLI_STALE_INODE;
805 bip->bli_item.li_cb = xfs_buf_iodone;
809 * Mark the buffer as being one which contains newly allocated
810 * inodes. We need to make sure that even if this buffer is
811 * relogged as an 'inode buf' we still recover all of the inode
812 * images in the face of a crash. This works in coordination with
813 * xfs_buf_item_committed() to ensure that the buffer remains in the
814 * AIL at its original location even after it has been relogged.
818 xfs_trans_inode_alloc_buf(
822 xfs_buf_log_item_t *bip = bp->b_fspriv;
824 ASSERT(bp->b_transp == tp);
826 ASSERT(atomic_read(&bip->bli_refcount) > 0);
828 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
833 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
834 * dquots. However, unlike in inode buffer recovery, dquot buffers get
835 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
836 * The only thing that makes dquot buffers different from regular
837 * buffers is that we must not replay dquot bufs when recovering
838 * if a _corresponding_ quotaoff has happened. We also have to distinguish
839 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
840 * can be turned off independently.
849 xfs_buf_log_item_t *bip = bp->b_fspriv;
851 ASSERT(bp->b_transp == tp);
853 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
854 type == XFS_BLF_PDQUOT_BUF ||
855 type == XFS_BLF_GDQUOT_BUF);
856 ASSERT(atomic_read(&bip->bli_refcount) > 0);
858 bip->bli_format.blf_flags |= type;