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)) {
169 * If the buffer is stale then it was binval'ed
170 * since last read. This doesn't matter since the
171 * caller isn't allowed to use the data anyway.
173 else if (XFS_BUF_ISSTALE(bp))
174 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
176 ASSERT(bp->b_transp == tp);
179 ASSERT(atomic_read(&bip->bli_refcount) > 0);
181 trace_xfs_trans_get_buf_recur(bip);
186 * We always specify the XBF_DONT_BLOCK flag within a transaction
187 * so that get_buf does not try to push out a delayed write buffer
188 * which might cause another transaction to take place (if the
189 * buffer was delayed alloc). Such recursive transactions can
190 * easily deadlock with our current transaction as well as cause
191 * us to run out of stack space.
193 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
198 ASSERT(!bp->b_error);
200 _xfs_trans_bjoin(tp, bp, 1);
201 trace_xfs_trans_get_buf(bp->b_fspriv);
206 * Get and lock the superblock buffer of this file system for the
209 * We don't need to use incore_match() here, because the superblock
210 * buffer is a private buffer which we keep a pointer to in the
214 xfs_trans_getsb(xfs_trans_t *tp,
215 struct xfs_mount *mp,
219 xfs_buf_log_item_t *bip;
222 * Default to just trying to lock the superblock buffer
226 return (xfs_getsb(mp, flags));
230 * If the superblock buffer already has this transaction
231 * pointer in its b_fsprivate2 field, then we know we already
232 * have it locked. In this case we just increment the lock
233 * recursion count and return the buffer to the caller.
236 if (bp->b_transp == tp) {
239 ASSERT(atomic_read(&bip->bli_refcount) > 0);
241 trace_xfs_trans_getsb_recur(bip);
245 bp = xfs_getsb(mp, flags);
249 _xfs_trans_bjoin(tp, bp, 1);
250 trace_xfs_trans_getsb(bp->b_fspriv);
255 xfs_buftarg_t *xfs_error_target;
258 int xfs_error_mod = 33;
262 * Get and lock the buffer for the caller if it is not already
263 * locked within the given transaction. If it has not yet been
264 * read in, read it from disk. If it is already locked
265 * within the transaction and already read in, just increment its
266 * lock recursion count and return a pointer to it.
268 * If the transaction pointer is NULL, make this just a normal
275 xfs_buftarg_t *target,
282 xfs_buf_log_item_t *bip;
286 flags = XBF_LOCK | XBF_MAPPED;
289 * Default to a normal get_buf() call if the tp is NULL.
292 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
294 return (flags & XBF_TRYLOCK) ?
295 EAGAIN : XFS_ERROR(ENOMEM);
299 xfs_buf_ioerror_alert(bp, __func__);
305 if (xfs_error_target == target) {
306 if (((xfs_req_num++) % xfs_error_mod) == 0) {
308 xfs_debug(mp, "Returning error!");
309 return XFS_ERROR(EIO);
314 if (XFS_FORCED_SHUTDOWN(mp))
321 * If we find the buffer in the cache with this transaction
322 * pointer in its b_fsprivate2 field, then we know we already
323 * have it locked. If it is already read in we just increment
324 * the lock recursion count and return the buffer to the caller.
325 * If the buffer is not yet read in, then we read it in, increment
326 * the lock recursion count, and return it to the caller.
328 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
330 ASSERT(xfs_buf_islocked(bp));
331 ASSERT(bp->b_transp == tp);
332 ASSERT(bp->b_fspriv != NULL);
333 ASSERT(!bp->b_error);
334 if (!(XFS_BUF_ISDONE(bp))) {
335 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
336 ASSERT(!XFS_BUF_ISASYNC(bp));
338 xfsbdstrat(tp->t_mountp, bp);
339 error = xfs_buf_iowait(bp);
341 xfs_buf_ioerror_alert(bp, __func__);
344 * We can gracefully recover from most read
345 * errors. Ones we can't are those that happen
346 * after the transaction's already dirty.
348 if (tp->t_flags & XFS_TRANS_DIRTY)
349 xfs_force_shutdown(tp->t_mountp,
350 SHUTDOWN_META_IO_ERROR);
355 * We never locked this buf ourselves, so we shouldn't
356 * brelse it either. Just get out.
358 if (XFS_FORCED_SHUTDOWN(mp)) {
359 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
361 return XFS_ERROR(EIO);
368 ASSERT(atomic_read(&bip->bli_refcount) > 0);
369 trace_xfs_trans_read_buf_recur(bip);
375 * We always specify the XBF_DONT_BLOCK flag within a transaction
376 * so that get_buf does not try to push out a delayed write buffer
377 * which might cause another transaction to take place (if the
378 * buffer was delayed alloc). Such recursive transactions can
379 * easily deadlock with our current transaction as well as cause
380 * us to run out of stack space.
382 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
385 return (flags & XBF_TRYLOCK) ?
386 0 : XFS_ERROR(ENOMEM);
392 xfs_buf_ioerror_alert(bp, __func__);
393 if (tp->t_flags & XFS_TRANS_DIRTY)
394 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
399 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
400 if (xfs_error_target == target) {
401 if (((xfs_req_num++) % xfs_error_mod) == 0) {
402 xfs_force_shutdown(tp->t_mountp,
403 SHUTDOWN_META_IO_ERROR);
405 xfs_debug(mp, "Returning trans error!");
406 return XFS_ERROR(EIO);
411 if (XFS_FORCED_SHUTDOWN(mp))
414 _xfs_trans_bjoin(tp, bp, 1);
415 trace_xfs_trans_read_buf(bp->b_fspriv);
422 * the theory here is that buffer is good but we're
423 * bailing out because the filesystem is being forcibly
424 * shut down. So we should leave the b_flags alone since
425 * the buffer's not staled and just get out.
428 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
429 xfs_notice(mp, "about to pop assert, bp == 0x%p", bp);
431 ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) !=
432 (XBF_STALE|XBF_DELWRI));
434 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
437 return XFS_ERROR(EIO);
442 * Release the buffer bp which was previously acquired with one of the
443 * xfs_trans_... buffer allocation routines if the buffer has not
444 * been modified within this transaction. If the buffer is modified
445 * within this transaction, do decrement the recursion count but do
446 * not release the buffer even if the count goes to 0. If the buffer is not
447 * modified within the transaction, decrement the recursion count and
448 * release the buffer if the recursion count goes to 0.
450 * If the buffer is to be released and it was not modified before
451 * this transaction began, then free the buf_log_item associated with it.
453 * If the transaction pointer is NULL, make this just a normal
457 xfs_trans_brelse(xfs_trans_t *tp,
460 xfs_buf_log_item_t *bip;
463 * Default to a normal brelse() call if the tp is NULL.
466 ASSERT(bp->b_transp == NULL);
471 ASSERT(bp->b_transp == tp);
473 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
474 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
475 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
476 ASSERT(atomic_read(&bip->bli_refcount) > 0);
478 trace_xfs_trans_brelse(bip);
481 * If the release is just for a recursive lock,
482 * then decrement the count and return.
484 if (bip->bli_recur > 0) {
490 * If the buffer is dirty within this transaction, we can't
491 * release it until we commit.
493 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
497 * If the buffer has been invalidated, then we can't release
498 * it until the transaction commits to disk unless it is re-dirtied
499 * as part of this transaction. This prevents us from pulling
500 * the item from the AIL before we should.
502 if (bip->bli_flags & XFS_BLI_STALE)
505 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
508 * Free up the log item descriptor tracking the released item.
510 xfs_trans_del_item(&bip->bli_item);
513 * Clear the hold flag in the buf log item if it is set.
514 * We wouldn't want the next user of the buffer to
517 if (bip->bli_flags & XFS_BLI_HOLD) {
518 bip->bli_flags &= ~XFS_BLI_HOLD;
522 * Drop our reference to the buf log item.
524 atomic_dec(&bip->bli_refcount);
527 * If the buf item is not tracking data in the log, then
528 * we must free it before releasing the buffer back to the
529 * free pool. Before releasing the buffer to the free pool,
530 * clear the transaction pointer in b_fsprivate2 to dissolve
531 * its relation to this transaction.
533 if (!xfs_buf_item_dirty(bip)) {
535 ASSERT(bp->b_pincount == 0);
537 ASSERT(atomic_read(&bip->bli_refcount) == 0);
538 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
539 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
540 xfs_buf_item_relse(bp);
548 * Mark the buffer as not needing to be unlocked when the buf item's
549 * IOP_UNLOCK() routine is called. The buffer must already be locked
550 * and associated with the given transaction.
554 xfs_trans_bhold(xfs_trans_t *tp,
557 xfs_buf_log_item_t *bip = bp->b_fspriv;
559 ASSERT(bp->b_transp == tp);
561 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
562 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
563 ASSERT(atomic_read(&bip->bli_refcount) > 0);
565 bip->bli_flags |= XFS_BLI_HOLD;
566 trace_xfs_trans_bhold(bip);
570 * Cancel the previous buffer hold request made on this buffer
571 * for this transaction.
574 xfs_trans_bhold_release(xfs_trans_t *tp,
577 xfs_buf_log_item_t *bip = bp->b_fspriv;
579 ASSERT(bp->b_transp == tp);
581 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
582 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
583 ASSERT(atomic_read(&bip->bli_refcount) > 0);
584 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
586 bip->bli_flags &= ~XFS_BLI_HOLD;
587 trace_xfs_trans_bhold_release(bip);
591 * This is called to mark bytes first through last inclusive of the given
592 * buffer as needing to be logged when the transaction is committed.
593 * The buffer must already be associated with the given transaction.
595 * First and last are numbers relative to the beginning of this buffer,
596 * so the first byte in the buffer is numbered 0 regardless of the
600 xfs_trans_log_buf(xfs_trans_t *tp,
605 xfs_buf_log_item_t *bip = bp->b_fspriv;
607 ASSERT(bp->b_transp == tp);
609 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
610 ASSERT(bp->b_iodone == NULL ||
611 bp->b_iodone == xfs_buf_iodone_callbacks);
614 * Mark the buffer as needing to be written out eventually,
615 * and set its iodone function to remove the buffer's buf log
616 * item from the AIL and free it when the buffer is flushed
617 * to disk. See xfs_buf_attach_iodone() for more details
618 * on li_cb and xfs_buf_iodone_callbacks().
619 * If we end up aborting this transaction, we trap this buffer
620 * inside the b_bdstrat callback so that this won't get written to
625 ASSERT(atomic_read(&bip->bli_refcount) > 0);
626 bp->b_iodone = xfs_buf_iodone_callbacks;
627 bip->bli_item.li_cb = xfs_buf_iodone;
629 xfs_buf_delwri_queue(bp);
631 trace_xfs_trans_log_buf(bip);
634 * If we invalidated the buffer within this transaction, then
635 * cancel the invalidation now that we're dirtying the buffer
636 * again. There are no races with the code in xfs_buf_item_unpin(),
637 * because we have a reference to the buffer this entire time.
639 if (bip->bli_flags & XFS_BLI_STALE) {
640 bip->bli_flags &= ~XFS_BLI_STALE;
641 ASSERT(XFS_BUF_ISSTALE(bp));
643 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
646 tp->t_flags |= XFS_TRANS_DIRTY;
647 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
648 bip->bli_flags |= XFS_BLI_LOGGED;
649 xfs_buf_item_log(bip, first, last);
654 * This called to invalidate a buffer that is being used within
655 * a transaction. Typically this is because the blocks in the
656 * buffer are being freed, so we need to prevent it from being
657 * written out when we're done. Allowing it to be written again
658 * might overwrite data in the free blocks if they are reallocated
661 * We prevent the buffer from being written out by clearing the
662 * B_DELWRI flag. We can't always
663 * get rid of the buf log item at this point, though, because
664 * the buffer may still be pinned by another transaction. If that
665 * is the case, then we'll wait until the buffer is committed to
666 * disk for the last time (we can tell by the ref count) and
667 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
668 * will keep the buffer locked so that the buffer and buf log item
676 xfs_buf_log_item_t *bip = bp->b_fspriv;
678 ASSERT(bp->b_transp == tp);
680 ASSERT(atomic_read(&bip->bli_refcount) > 0);
682 trace_xfs_trans_binval(bip);
684 if (bip->bli_flags & XFS_BLI_STALE) {
686 * If the buffer is already invalidated, then
689 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
690 ASSERT(XFS_BUF_ISSTALE(bp));
691 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
692 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
693 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
694 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
695 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
700 * Clear the dirty bit in the buffer and set the STALE flag
701 * in the buf log item. The STALE flag will be used in
702 * xfs_buf_item_unpin() to determine if it should clean up
703 * when the last reference to the buf item is given up.
704 * We set the XFS_BLF_CANCEL flag in the buf log format structure
705 * and log the buf item. This will be used at recovery time
706 * to determine that copies of the buffer in the log before
707 * this should not be replayed.
708 * We mark the item descriptor and the transaction dirty so
709 * that we'll hold the buffer until after the commit.
711 * Since we're invalidating the buffer, we also clear the state
712 * about which parts of the buffer have been logged. We also
713 * clear the flag indicating that this is an inode buffer since
714 * the data in the buffer will no longer be valid.
716 * We set the stale bit in the buffer as well since we're getting
720 bip->bli_flags |= XFS_BLI_STALE;
721 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
722 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
723 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
724 memset((char *)(bip->bli_format.blf_data_map), 0,
725 (bip->bli_format.blf_map_size * sizeof(uint)));
726 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
727 tp->t_flags |= XFS_TRANS_DIRTY;
731 * This call is used to indicate that the buffer contains on-disk inodes which
732 * must be handled specially during recovery. They require special handling
733 * because only the di_next_unlinked from the inodes in the buffer should be
734 * recovered. The rest of the data in the buffer is logged via the inodes
737 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
738 * transferred to the buffer's log format structure so that we'll know what to
739 * do at recovery time.
746 xfs_buf_log_item_t *bip = bp->b_fspriv;
748 ASSERT(bp->b_transp == tp);
750 ASSERT(atomic_read(&bip->bli_refcount) > 0);
752 bip->bli_flags |= XFS_BLI_INODE_BUF;
756 * This call is used to indicate that the buffer is going to
757 * be staled and was an inode buffer. This means it gets
758 * special processing during unpin - where any inodes
759 * associated with the buffer should be removed from ail.
760 * There is also special processing during recovery,
761 * any replay of the inodes in the buffer needs to be
762 * prevented as the buffer may have been reused.
765 xfs_trans_stale_inode_buf(
769 xfs_buf_log_item_t *bip = bp->b_fspriv;
771 ASSERT(bp->b_transp == tp);
773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
775 bip->bli_flags |= XFS_BLI_STALE_INODE;
776 bip->bli_item.li_cb = xfs_buf_iodone;
780 * Mark the buffer as being one which contains newly allocated
781 * inodes. We need to make sure that even if this buffer is
782 * relogged as an 'inode buf' we still recover all of the inode
783 * images in the face of a crash. This works in coordination with
784 * xfs_buf_item_committed() to ensure that the buffer remains in the
785 * AIL at its original location even after it has been relogged.
789 xfs_trans_inode_alloc_buf(
793 xfs_buf_log_item_t *bip = bp->b_fspriv;
795 ASSERT(bp->b_transp == tp);
797 ASSERT(atomic_read(&bip->bli_refcount) > 0);
799 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
804 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
805 * dquots. However, unlike in inode buffer recovery, dquot buffers get
806 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
807 * The only thing that makes dquot buffers different from regular
808 * buffers is that we must not replay dquot bufs when recovering
809 * if a _corresponding_ quotaoff has happened. We also have to distinguish
810 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
811 * can be turned off independently.
820 xfs_buf_log_item_t *bip = bp->b_fspriv;
822 ASSERT(bp->b_transp == tp);
824 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
825 type == XFS_BLF_PDQUOT_BUF ||
826 type == XFS_BLF_GDQUOT_BUF);
827 ASSERT(atomic_read(&bip->bli_refcount) > 0);
829 bip->bli_format.blf_flags |= type;