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 BBTOB(blip->bli_buf->b_length) == 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)) {
168 ASSERT(bp->b_transp == tp);
171 ASSERT(atomic_read(&bip->bli_refcount) > 0);
173 trace_xfs_trans_get_buf_recur(bip);
178 * We always specify the XBF_DONT_BLOCK flag within a transaction
179 * so that get_buf does not try to push out a delayed write buffer
180 * which might cause another transaction to take place (if the
181 * buffer was delayed alloc). Such recursive transactions can
182 * easily deadlock with our current transaction as well as cause
183 * us to run out of stack space.
185 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
190 ASSERT(!bp->b_error);
192 _xfs_trans_bjoin(tp, bp, 1);
193 trace_xfs_trans_get_buf(bp->b_fspriv);
198 * Get and lock the superblock buffer of this file system for the
201 * We don't need to use incore_match() here, because the superblock
202 * buffer is a private buffer which we keep a pointer to in the
206 xfs_trans_getsb(xfs_trans_t *tp,
207 struct xfs_mount *mp,
211 xfs_buf_log_item_t *bip;
214 * Default to just trying to lock the superblock buffer
218 return (xfs_getsb(mp, flags));
222 * If the superblock buffer already has this transaction
223 * pointer in its b_fsprivate2 field, then we know we already
224 * have it locked. In this case we just increment the lock
225 * recursion count and return the buffer to the caller.
228 if (bp->b_transp == tp) {
231 ASSERT(atomic_read(&bip->bli_refcount) > 0);
233 trace_xfs_trans_getsb_recur(bip);
237 bp = xfs_getsb(mp, flags);
241 _xfs_trans_bjoin(tp, bp, 1);
242 trace_xfs_trans_getsb(bp->b_fspriv);
247 xfs_buftarg_t *xfs_error_target;
250 int xfs_error_mod = 33;
254 * Get and lock the buffer for the caller if it is not already
255 * locked within the given transaction. If it has not yet been
256 * read in, read it from disk. If it is already locked
257 * within the transaction and already read in, just increment its
258 * lock recursion count and return a pointer to it.
260 * If the transaction pointer is NULL, make this just a normal
267 xfs_buftarg_t *target,
274 xfs_buf_log_item_t *bip;
278 flags = XBF_LOCK | XBF_MAPPED;
281 * Default to a normal get_buf() call if the tp is NULL.
284 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
286 return (flags & XBF_TRYLOCK) ?
287 EAGAIN : XFS_ERROR(ENOMEM);
291 xfs_buf_ioerror_alert(bp, __func__);
297 if (xfs_error_target == target) {
298 if (((xfs_req_num++) % xfs_error_mod) == 0) {
300 xfs_debug(mp, "Returning error!");
301 return XFS_ERROR(EIO);
306 if (XFS_FORCED_SHUTDOWN(mp))
313 * If we find the buffer in the cache with this transaction
314 * pointer in its b_fsprivate2 field, then we know we already
315 * have it locked. If it is already read in we just increment
316 * the lock recursion count and return the buffer to the caller.
317 * If the buffer is not yet read in, then we read it in, increment
318 * the lock recursion count, and return it to the caller.
320 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
322 ASSERT(xfs_buf_islocked(bp));
323 ASSERT(bp->b_transp == tp);
324 ASSERT(bp->b_fspriv != NULL);
325 ASSERT(!bp->b_error);
326 if (!(XFS_BUF_ISDONE(bp))) {
327 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
328 ASSERT(!XFS_BUF_ISASYNC(bp));
330 xfsbdstrat(tp->t_mountp, bp);
331 error = xfs_buf_iowait(bp);
333 xfs_buf_ioerror_alert(bp, __func__);
336 * We can gracefully recover from most read
337 * errors. Ones we can't are those that happen
338 * after the transaction's already dirty.
340 if (tp->t_flags & XFS_TRANS_DIRTY)
341 xfs_force_shutdown(tp->t_mountp,
342 SHUTDOWN_META_IO_ERROR);
347 * We never locked this buf ourselves, so we shouldn't
348 * brelse it either. Just get out.
350 if (XFS_FORCED_SHUTDOWN(mp)) {
351 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
353 return XFS_ERROR(EIO);
360 ASSERT(atomic_read(&bip->bli_refcount) > 0);
361 trace_xfs_trans_read_buf_recur(bip);
367 * We always specify the XBF_DONT_BLOCK flag within a transaction
368 * so that get_buf does not try to push out a delayed write buffer
369 * which might cause another transaction to take place (if the
370 * buffer was delayed alloc). Such recursive transactions can
371 * easily deadlock with our current transaction as well as cause
372 * us to run out of stack space.
374 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
377 return (flags & XBF_TRYLOCK) ?
378 0 : XFS_ERROR(ENOMEM);
384 xfs_buf_ioerror_alert(bp, __func__);
385 if (tp->t_flags & XFS_TRANS_DIRTY)
386 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
391 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
392 if (xfs_error_target == target) {
393 if (((xfs_req_num++) % xfs_error_mod) == 0) {
394 xfs_force_shutdown(tp->t_mountp,
395 SHUTDOWN_META_IO_ERROR);
397 xfs_debug(mp, "Returning trans error!");
398 return XFS_ERROR(EIO);
403 if (XFS_FORCED_SHUTDOWN(mp))
406 _xfs_trans_bjoin(tp, bp, 1);
407 trace_xfs_trans_read_buf(bp->b_fspriv);
413 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
416 return XFS_ERROR(EIO);
421 * Release the buffer bp which was previously acquired with one of the
422 * xfs_trans_... buffer allocation routines if the buffer has not
423 * been modified within this transaction. If the buffer is modified
424 * within this transaction, do decrement the recursion count but do
425 * not release the buffer even if the count goes to 0. If the buffer is not
426 * modified within the transaction, decrement the recursion count and
427 * release the buffer if the recursion count goes to 0.
429 * If the buffer is to be released and it was not modified before
430 * this transaction began, then free the buf_log_item associated with it.
432 * If the transaction pointer is NULL, make this just a normal
436 xfs_trans_brelse(xfs_trans_t *tp,
439 xfs_buf_log_item_t *bip;
442 * Default to a normal brelse() call if the tp is NULL.
445 ASSERT(bp->b_transp == NULL);
450 ASSERT(bp->b_transp == tp);
452 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
453 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
454 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
455 ASSERT(atomic_read(&bip->bli_refcount) > 0);
457 trace_xfs_trans_brelse(bip);
460 * If the release is just for a recursive lock,
461 * then decrement the count and return.
463 if (bip->bli_recur > 0) {
469 * If the buffer is dirty within this transaction, we can't
470 * release it until we commit.
472 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
476 * If the buffer has been invalidated, then we can't release
477 * it until the transaction commits to disk unless it is re-dirtied
478 * as part of this transaction. This prevents us from pulling
479 * the item from the AIL before we should.
481 if (bip->bli_flags & XFS_BLI_STALE)
484 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
487 * Free up the log item descriptor tracking the released item.
489 xfs_trans_del_item(&bip->bli_item);
492 * Clear the hold flag in the buf log item if it is set.
493 * We wouldn't want the next user of the buffer to
496 if (bip->bli_flags & XFS_BLI_HOLD) {
497 bip->bli_flags &= ~XFS_BLI_HOLD;
501 * Drop our reference to the buf log item.
503 atomic_dec(&bip->bli_refcount);
506 * If the buf item is not tracking data in the log, then
507 * we must free it before releasing the buffer back to the
508 * free pool. Before releasing the buffer to the free pool,
509 * clear the transaction pointer in b_fsprivate2 to dissolve
510 * its relation to this transaction.
512 if (!xfs_buf_item_dirty(bip)) {
514 ASSERT(bp->b_pincount == 0);
516 ASSERT(atomic_read(&bip->bli_refcount) == 0);
517 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
518 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
519 xfs_buf_item_relse(bp);
527 * Mark the buffer as not needing to be unlocked when the buf item's
528 * IOP_UNLOCK() routine is called. The buffer must already be locked
529 * and associated with the given transaction.
533 xfs_trans_bhold(xfs_trans_t *tp,
536 xfs_buf_log_item_t *bip = bp->b_fspriv;
538 ASSERT(bp->b_transp == tp);
540 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
541 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
542 ASSERT(atomic_read(&bip->bli_refcount) > 0);
544 bip->bli_flags |= XFS_BLI_HOLD;
545 trace_xfs_trans_bhold(bip);
549 * Cancel the previous buffer hold request made on this buffer
550 * for this transaction.
553 xfs_trans_bhold_release(xfs_trans_t *tp,
556 xfs_buf_log_item_t *bip = bp->b_fspriv;
558 ASSERT(bp->b_transp == tp);
560 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
561 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
562 ASSERT(atomic_read(&bip->bli_refcount) > 0);
563 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
565 bip->bli_flags &= ~XFS_BLI_HOLD;
566 trace_xfs_trans_bhold_release(bip);
570 * This is called to mark bytes first through last inclusive of the given
571 * buffer as needing to be logged when the transaction is committed.
572 * The buffer must already be associated with the given transaction.
574 * First and last are numbers relative to the beginning of this buffer,
575 * so the first byte in the buffer is numbered 0 regardless of the
579 xfs_trans_log_buf(xfs_trans_t *tp,
584 xfs_buf_log_item_t *bip = bp->b_fspriv;
586 ASSERT(bp->b_transp == tp);
588 ASSERT(first <= last && last < BBTOB(bp->b_length));
589 ASSERT(bp->b_iodone == NULL ||
590 bp->b_iodone == xfs_buf_iodone_callbacks);
593 * Mark the buffer as needing to be written out eventually,
594 * and set its iodone function to remove the buffer's buf log
595 * item from the AIL and free it when the buffer is flushed
596 * to disk. See xfs_buf_attach_iodone() for more details
597 * on li_cb and xfs_buf_iodone_callbacks().
598 * If we end up aborting this transaction, we trap this buffer
599 * inside the b_bdstrat callback so that this won't get written to
604 ASSERT(atomic_read(&bip->bli_refcount) > 0);
605 bp->b_iodone = xfs_buf_iodone_callbacks;
606 bip->bli_item.li_cb = xfs_buf_iodone;
608 trace_xfs_trans_log_buf(bip);
611 * If we invalidated the buffer within this transaction, then
612 * cancel the invalidation now that we're dirtying the buffer
613 * again. There are no races with the code in xfs_buf_item_unpin(),
614 * because we have a reference to the buffer this entire time.
616 if (bip->bli_flags & XFS_BLI_STALE) {
617 bip->bli_flags &= ~XFS_BLI_STALE;
618 ASSERT(XFS_BUF_ISSTALE(bp));
620 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
623 tp->t_flags |= XFS_TRANS_DIRTY;
624 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
625 bip->bli_flags |= XFS_BLI_LOGGED;
626 xfs_buf_item_log(bip, first, last);
631 * Invalidate a buffer that is being used within a transaction.
633 * Typically this is because the blocks in the buffer are being freed, so we
634 * need to prevent it from being written out when we're done. Allowing it
635 * to be written again might overwrite data in the free blocks if they are
636 * reallocated to a file.
638 * We prevent the buffer from being written out by marking it stale. We can't
639 * get rid of the buf log item at this point because the buffer may still be
640 * pinned by another transaction. If that is the case, then we'll wait until
641 * the buffer is committed to disk for the last time (we can tell by the ref
642 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
643 * keep the buffer locked so that the buffer and buf log item are not reused.
645 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
646 * the buf item. This will be used at recovery time to determine that copies
647 * of the buffer in the log before this should not be replayed.
649 * We mark the item descriptor and the transaction dirty so that we'll hold
650 * the buffer until after the commit.
652 * Since we're invalidating the buffer, we also clear the state about which
653 * parts of the buffer have been logged. We also clear the flag indicating
654 * that this is an inode buffer since the data in the buffer will no longer
657 * We set the stale bit in the buffer as well since we're getting rid of it.
664 xfs_buf_log_item_t *bip = bp->b_fspriv;
666 ASSERT(bp->b_transp == tp);
668 ASSERT(atomic_read(&bip->bli_refcount) > 0);
670 trace_xfs_trans_binval(bip);
672 if (bip->bli_flags & XFS_BLI_STALE) {
674 * If the buffer is already invalidated, then
677 ASSERT(XFS_BUF_ISSTALE(bp));
678 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
679 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
680 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
681 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
682 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
688 bip->bli_flags |= XFS_BLI_STALE;
689 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
690 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
691 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
692 memset((char *)(bip->bli_format.blf_data_map), 0,
693 (bip->bli_format.blf_map_size * sizeof(uint)));
694 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
695 tp->t_flags |= XFS_TRANS_DIRTY;
699 * This call is used to indicate that the buffer contains on-disk inodes which
700 * must be handled specially during recovery. They require special handling
701 * because only the di_next_unlinked from the inodes in the buffer should be
702 * recovered. The rest of the data in the buffer is logged via the inodes
705 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
706 * transferred to the buffer's log format structure so that we'll know what to
707 * do at recovery time.
714 xfs_buf_log_item_t *bip = bp->b_fspriv;
716 ASSERT(bp->b_transp == tp);
718 ASSERT(atomic_read(&bip->bli_refcount) > 0);
720 bip->bli_flags |= XFS_BLI_INODE_BUF;
724 * This call is used to indicate that the buffer is going to
725 * be staled and was an inode buffer. This means it gets
726 * special processing during unpin - where any inodes
727 * associated with the buffer should be removed from ail.
728 * There is also special processing during recovery,
729 * any replay of the inodes in the buffer needs to be
730 * prevented as the buffer may have been reused.
733 xfs_trans_stale_inode_buf(
737 xfs_buf_log_item_t *bip = bp->b_fspriv;
739 ASSERT(bp->b_transp == tp);
741 ASSERT(atomic_read(&bip->bli_refcount) > 0);
743 bip->bli_flags |= XFS_BLI_STALE_INODE;
744 bip->bli_item.li_cb = xfs_buf_iodone;
748 * Mark the buffer as being one which contains newly allocated
749 * inodes. We need to make sure that even if this buffer is
750 * relogged as an 'inode buf' we still recover all of the inode
751 * images in the face of a crash. This works in coordination with
752 * xfs_buf_item_committed() to ensure that the buffer remains in the
753 * AIL at its original location even after it has been relogged.
757 xfs_trans_inode_alloc_buf(
761 xfs_buf_log_item_t *bip = bp->b_fspriv;
763 ASSERT(bp->b_transp == tp);
765 ASSERT(atomic_read(&bip->bli_refcount) > 0);
767 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
772 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
773 * dquots. However, unlike in inode buffer recovery, dquot buffers get
774 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
775 * The only thing that makes dquot buffers different from regular
776 * buffers is that we must not replay dquot bufs when recovering
777 * if a _corresponding_ quotaoff has happened. We also have to distinguish
778 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
779 * can be turned off independently.
788 xfs_buf_log_item_t *bip = bp->b_fspriv;
790 ASSERT(bp->b_transp == tp);
792 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
793 type == XFS_BLF_PDQUOT_BUF ||
794 type == XFS_BLF_GDQUOT_BUF);
795 ASSERT(atomic_read(&bip->bli_refcount) > 0);
797 bip->bli_format.blf_flags |= type;