2 * linux/fs/ext3/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
25 #include <linux/highuid.h>
26 #include <linux/quotaops.h>
27 #include <linux/writeback.h>
28 #include <linux/mpage.h>
29 #include <linux/namei.h>
34 static int ext3_writepage_trans_blocks(struct inode *inode);
35 static int ext3_block_truncate_page(struct inode *inode, loff_t from);
38 * Test whether an inode is a fast symlink.
40 static int ext3_inode_is_fast_symlink(struct inode *inode)
42 int ea_blocks = EXT3_I(inode)->i_file_acl ?
43 (inode->i_sb->s_blocksize >> 9) : 0;
45 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
49 * The ext3 forget function must perform a revoke if we are freeing data
50 * which has been journaled. Metadata (eg. indirect blocks) must be
51 * revoked in all cases.
53 * "bh" may be NULL: a metadata block may have been freed from memory
54 * but there may still be a record of it in the journal, and that record
55 * still needs to be revoked.
57 int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
58 struct buffer_head *bh, ext3_fsblk_t blocknr)
64 trace_ext3_forget(inode, is_metadata, blocknr);
65 BUFFER_TRACE(bh, "enter");
67 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
69 bh, is_metadata, inode->i_mode,
70 test_opt(inode->i_sb, DATA_FLAGS));
72 /* Never use the revoke function if we are doing full data
73 * journaling: there is no need to, and a V1 superblock won't
74 * support it. Otherwise, only skip the revoke on un-journaled
77 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
78 (!is_metadata && !ext3_should_journal_data(inode))) {
80 BUFFER_TRACE(bh, "call journal_forget");
81 return ext3_journal_forget(handle, bh);
87 * data!=journal && (is_metadata || should_journal_data(inode))
89 BUFFER_TRACE(bh, "call ext3_journal_revoke");
90 err = ext3_journal_revoke(handle, blocknr, bh);
92 ext3_abort(inode->i_sb, __func__,
93 "error %d when attempting revoke", err);
94 BUFFER_TRACE(bh, "exit");
99 * Work out how many blocks we need to proceed with the next chunk of a
100 * truncate transaction.
102 static unsigned long blocks_for_truncate(struct inode *inode)
104 unsigned long needed;
106 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
108 /* Give ourselves just enough room to cope with inodes in which
109 * i_blocks is corrupt: we've seen disk corruptions in the past
110 * which resulted in random data in an inode which looked enough
111 * like a regular file for ext3 to try to delete it. Things
112 * will go a bit crazy if that happens, but at least we should
113 * try not to panic the whole kernel. */
117 /* But we need to bound the transaction so we don't overflow the
119 if (needed > EXT3_MAX_TRANS_DATA)
120 needed = EXT3_MAX_TRANS_DATA;
122 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
126 * Truncate transactions can be complex and absolutely huge. So we need to
127 * be able to restart the transaction at a conventient checkpoint to make
128 * sure we don't overflow the journal.
130 * start_transaction gets us a new handle for a truncate transaction,
131 * and extend_transaction tries to extend the existing one a bit. If
132 * extend fails, we need to propagate the failure up and restart the
133 * transaction in the top-level truncate loop. --sct
135 static handle_t *start_transaction(struct inode *inode)
139 result = ext3_journal_start(inode, blocks_for_truncate(inode));
143 ext3_std_error(inode->i_sb, PTR_ERR(result));
148 * Try to extend this transaction for the purposes of truncation.
150 * Returns 0 if we managed to create more room. If we can't create more
151 * room, and the transaction must be restarted we return 1.
153 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
155 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
157 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
163 * Restart the transaction associated with *handle. This does a commit,
164 * so before we call here everything must be consistently dirtied against
167 static int truncate_restart_transaction(handle_t *handle, struct inode *inode)
171 jbd_debug(2, "restarting handle %p\n", handle);
173 * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle
174 * At this moment, get_block can be called only for blocks inside
175 * i_size since page cache has been already dropped and writes are
176 * blocked by i_mutex. So we can safely drop the truncate_mutex.
178 mutex_unlock(&EXT3_I(inode)->truncate_mutex);
179 ret = ext3_journal_restart(handle, blocks_for_truncate(inode));
180 mutex_lock(&EXT3_I(inode)->truncate_mutex);
185 * Called at inode eviction from icache
187 void ext3_evict_inode (struct inode *inode)
189 struct ext3_inode_info *ei = EXT3_I(inode);
190 struct ext3_block_alloc_info *rsv;
194 trace_ext3_evict_inode(inode);
195 if (!inode->i_nlink && !is_bad_inode(inode)) {
196 dquot_initialize(inode);
201 * When journalling data dirty buffers are tracked only in the journal.
202 * So although mm thinks everything is clean and ready for reaping the
203 * inode might still have some pages to write in the running
204 * transaction or waiting to be checkpointed. Thus calling
205 * journal_invalidatepage() (via truncate_inode_pages()) to discard
206 * these buffers can cause data loss. Also even if we did not discard
207 * these buffers, we would have no way to find them after the inode
208 * is reaped and thus user could see stale data if he tries to read
209 * them before the transaction is checkpointed. So be careful and
210 * force everything to disk here... We use ei->i_datasync_tid to
211 * store the newest transaction containing inode's data.
213 * Note that directories do not have this problem because they don't
216 * The s_journal check handles the case when ext3_get_journal() fails
217 * and puts the journal inode.
219 if (inode->i_nlink && ext3_should_journal_data(inode) &&
220 EXT3_SB(inode->i_sb)->s_journal &&
221 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
222 inode->i_ino != EXT3_JOURNAL_INO) {
223 tid_t commit_tid = atomic_read(&ei->i_datasync_tid);
224 journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
226 log_start_commit(journal, commit_tid);
227 log_wait_commit(journal, commit_tid);
228 filemap_write_and_wait(&inode->i_data);
230 truncate_inode_pages(&inode->i_data, 0);
232 ext3_discard_reservation(inode);
233 rsv = ei->i_block_alloc_info;
234 ei->i_block_alloc_info = NULL;
241 handle = start_transaction(inode);
242 if (IS_ERR(handle)) {
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext3_orphan_del(NULL, inode);
256 ext3_truncate(inode);
258 * Kill off the orphan record created when the inode lost the last
259 * link. Note that ext3_orphan_del() has to be able to cope with the
260 * deletion of a non-existent orphan - ext3_truncate() could
261 * have removed the record.
263 ext3_orphan_del(handle, inode);
264 ei->i_dtime = get_seconds();
267 * One subtle ordering requirement: if anything has gone wrong
268 * (transaction abort, IO errors, whatever), then we can still
269 * do these next steps (the fs will already have been marked as
270 * having errors), but we can't free the inode if the mark_dirty
273 if (ext3_mark_inode_dirty(handle, inode)) {
274 /* If that failed, just dquot_drop() and be done with that */
278 ext3_xattr_delete_inode(handle, inode);
279 dquot_free_inode(inode);
282 ext3_free_inode(handle, inode);
284 ext3_journal_stop(handle);
294 struct buffer_head *bh;
297 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
299 p->key = *(p->p = v);
303 static int verify_chain(Indirect *from, Indirect *to)
305 while (from <= to && from->key == *from->p)
311 * ext3_block_to_path - parse the block number into array of offsets
312 * @inode: inode in question (we are only interested in its superblock)
313 * @i_block: block number to be parsed
314 * @offsets: array to store the offsets in
315 * @boundary: set this non-zero if the referred-to block is likely to be
316 * followed (on disk) by an indirect block.
318 * To store the locations of file's data ext3 uses a data structure common
319 * for UNIX filesystems - tree of pointers anchored in the inode, with
320 * data blocks at leaves and indirect blocks in intermediate nodes.
321 * This function translates the block number into path in that tree -
322 * return value is the path length and @offsets[n] is the offset of
323 * pointer to (n+1)th node in the nth one. If @block is out of range
324 * (negative or too large) warning is printed and zero returned.
326 * Note: function doesn't find node addresses, so no IO is needed. All
327 * we need to know is the capacity of indirect blocks (taken from the
332 * Portability note: the last comparison (check that we fit into triple
333 * indirect block) is spelled differently, because otherwise on an
334 * architecture with 32-bit longs and 8Kb pages we might get into trouble
335 * if our filesystem had 8Kb blocks. We might use long long, but that would
336 * kill us on x86. Oh, well, at least the sign propagation does not matter -
337 * i_block would have to be negative in the very beginning, so we would not
341 static int ext3_block_to_path(struct inode *inode,
342 long i_block, int offsets[4], int *boundary)
344 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
345 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
346 const long direct_blocks = EXT3_NDIR_BLOCKS,
347 indirect_blocks = ptrs,
348 double_blocks = (1 << (ptrs_bits * 2));
353 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
354 } else if (i_block < direct_blocks) {
355 offsets[n++] = i_block;
356 final = direct_blocks;
357 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
358 offsets[n++] = EXT3_IND_BLOCK;
359 offsets[n++] = i_block;
361 } else if ((i_block -= indirect_blocks) < double_blocks) {
362 offsets[n++] = EXT3_DIND_BLOCK;
363 offsets[n++] = i_block >> ptrs_bits;
364 offsets[n++] = i_block & (ptrs - 1);
366 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
367 offsets[n++] = EXT3_TIND_BLOCK;
368 offsets[n++] = i_block >> (ptrs_bits * 2);
369 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
370 offsets[n++] = i_block & (ptrs - 1);
373 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
376 *boundary = final - 1 - (i_block & (ptrs - 1));
381 * ext3_get_branch - read the chain of indirect blocks leading to data
382 * @inode: inode in question
383 * @depth: depth of the chain (1 - direct pointer, etc.)
384 * @offsets: offsets of pointers in inode/indirect blocks
385 * @chain: place to store the result
386 * @err: here we store the error value
388 * Function fills the array of triples <key, p, bh> and returns %NULL
389 * if everything went OK or the pointer to the last filled triple
390 * (incomplete one) otherwise. Upon the return chain[i].key contains
391 * the number of (i+1)-th block in the chain (as it is stored in memory,
392 * i.e. little-endian 32-bit), chain[i].p contains the address of that
393 * number (it points into struct inode for i==0 and into the bh->b_data
394 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
395 * block for i>0 and NULL for i==0. In other words, it holds the block
396 * numbers of the chain, addresses they were taken from (and where we can
397 * verify that chain did not change) and buffer_heads hosting these
400 * Function stops when it stumbles upon zero pointer (absent block)
401 * (pointer to last triple returned, *@err == 0)
402 * or when it gets an IO error reading an indirect block
403 * (ditto, *@err == -EIO)
404 * or when it notices that chain had been changed while it was reading
405 * (ditto, *@err == -EAGAIN)
406 * or when it reads all @depth-1 indirect blocks successfully and finds
407 * the whole chain, all way to the data (returns %NULL, *err == 0).
409 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
410 Indirect chain[4], int *err)
412 struct super_block *sb = inode->i_sb;
414 struct buffer_head *bh;
417 /* i_data is not going away, no lock needed */
418 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
422 bh = sb_bread(sb, le32_to_cpu(p->key));
425 /* Reader: pointers */
426 if (!verify_chain(chain, p))
428 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
446 * ext3_find_near - find a place for allocation with sufficient locality
448 * @ind: descriptor of indirect block.
450 * This function returns the preferred place for block allocation.
451 * It is used when heuristic for sequential allocation fails.
453 * + if there is a block to the left of our position - allocate near it.
454 * + if pointer will live in indirect block - allocate near that block.
455 * + if pointer will live in inode - allocate in the same
458 * In the latter case we colour the starting block by the callers PID to
459 * prevent it from clashing with concurrent allocations for a different inode
460 * in the same block group. The PID is used here so that functionally related
461 * files will be close-by on-disk.
463 * Caller must make sure that @ind is valid and will stay that way.
465 static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
467 struct ext3_inode_info *ei = EXT3_I(inode);
468 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
470 ext3_fsblk_t bg_start;
471 ext3_grpblk_t colour;
473 /* Try to find previous block */
474 for (p = ind->p - 1; p >= start; p--) {
476 return le32_to_cpu(*p);
479 /* No such thing, so let's try location of indirect block */
481 return ind->bh->b_blocknr;
484 * It is going to be referred to from the inode itself? OK, just put it
485 * into the same cylinder group then.
487 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
488 colour = (current->pid % 16) *
489 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
490 return bg_start + colour;
494 * ext3_find_goal - find a preferred place for allocation.
496 * @block: block we want
497 * @partial: pointer to the last triple within a chain
499 * Normally this function find the preferred place for block allocation,
503 static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
506 struct ext3_block_alloc_info *block_i;
508 block_i = EXT3_I(inode)->i_block_alloc_info;
511 * try the heuristic for sequential allocation,
512 * failing that at least try to get decent locality.
514 if (block_i && (block == block_i->last_alloc_logical_block + 1)
515 && (block_i->last_alloc_physical_block != 0)) {
516 return block_i->last_alloc_physical_block + 1;
519 return ext3_find_near(inode, partial);
523 * ext3_blks_to_allocate - Look up the block map and count the number
524 * of direct blocks need to be allocated for the given branch.
526 * @branch: chain of indirect blocks
527 * @k: number of blocks need for indirect blocks
528 * @blks: number of data blocks to be mapped.
529 * @blocks_to_boundary: the offset in the indirect block
531 * return the total number of blocks to be allocate, including the
532 * direct and indirect blocks.
534 static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
535 int blocks_to_boundary)
537 unsigned long count = 0;
540 * Simple case, [t,d]Indirect block(s) has not allocated yet
541 * then it's clear blocks on that path have not allocated
544 /* right now we don't handle cross boundary allocation */
545 if (blks < blocks_to_boundary + 1)
548 count += blocks_to_boundary + 1;
553 while (count < blks && count <= blocks_to_boundary &&
554 le32_to_cpu(*(branch[0].p + count)) == 0) {
561 * ext3_alloc_blocks - multiple allocate blocks needed for a branch
562 * @handle: handle for this transaction
564 * @goal: preferred place for allocation
565 * @indirect_blks: the number of blocks need to allocate for indirect
567 * @blks: number of blocks need to allocated for direct blocks
568 * @new_blocks: on return it will store the new block numbers for
569 * the indirect blocks(if needed) and the first direct block,
570 * @err: here we store the error value
572 * return the number of direct blocks allocated
574 static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
575 ext3_fsblk_t goal, int indirect_blks, int blks,
576 ext3_fsblk_t new_blocks[4], int *err)
579 unsigned long count = 0;
581 ext3_fsblk_t current_block = 0;
585 * Here we try to allocate the requested multiple blocks at once,
586 * on a best-effort basis.
587 * To build a branch, we should allocate blocks for
588 * the indirect blocks(if not allocated yet), and at least
589 * the first direct block of this branch. That's the
590 * minimum number of blocks need to allocate(required)
592 target = blks + indirect_blks;
596 /* allocating blocks for indirect blocks and direct blocks */
597 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
602 /* allocate blocks for indirect blocks */
603 while (index < indirect_blks && count) {
604 new_blocks[index++] = current_block++;
612 /* save the new block number for the first direct block */
613 new_blocks[index] = current_block;
615 /* total number of blocks allocated for direct blocks */
620 for (i = 0; i <index; i++)
621 ext3_free_blocks(handle, inode, new_blocks[i], 1);
626 * ext3_alloc_branch - allocate and set up a chain of blocks.
627 * @handle: handle for this transaction
629 * @indirect_blks: number of allocated indirect blocks
630 * @blks: number of allocated direct blocks
631 * @goal: preferred place for allocation
632 * @offsets: offsets (in the blocks) to store the pointers to next.
633 * @branch: place to store the chain in.
635 * This function allocates blocks, zeroes out all but the last one,
636 * links them into chain and (if we are synchronous) writes them to disk.
637 * In other words, it prepares a branch that can be spliced onto the
638 * inode. It stores the information about that chain in the branch[], in
639 * the same format as ext3_get_branch() would do. We are calling it after
640 * we had read the existing part of chain and partial points to the last
641 * triple of that (one with zero ->key). Upon the exit we have the same
642 * picture as after the successful ext3_get_block(), except that in one
643 * place chain is disconnected - *branch->p is still zero (we did not
644 * set the last link), but branch->key contains the number that should
645 * be placed into *branch->p to fill that gap.
647 * If allocation fails we free all blocks we've allocated (and forget
648 * their buffer_heads) and return the error value the from failed
649 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
650 * as described above and return 0.
652 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
653 int indirect_blks, int *blks, ext3_fsblk_t goal,
654 int *offsets, Indirect *branch)
656 int blocksize = inode->i_sb->s_blocksize;
659 struct buffer_head *bh;
661 ext3_fsblk_t new_blocks[4];
662 ext3_fsblk_t current_block;
664 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
665 *blks, new_blocks, &err);
669 branch[0].key = cpu_to_le32(new_blocks[0]);
671 * metadata blocks and data blocks are allocated.
673 for (n = 1; n <= indirect_blks; n++) {
675 * Get buffer_head for parent block, zero it out
676 * and set the pointer to new one, then send
679 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
686 BUFFER_TRACE(bh, "call get_create_access");
687 err = ext3_journal_get_create_access(handle, bh);
694 memset(bh->b_data, 0, blocksize);
695 branch[n].p = (__le32 *) bh->b_data + offsets[n];
696 branch[n].key = cpu_to_le32(new_blocks[n]);
697 *branch[n].p = branch[n].key;
698 if ( n == indirect_blks) {
699 current_block = new_blocks[n];
701 * End of chain, update the last new metablock of
702 * the chain to point to the new allocated
703 * data blocks numbers
705 for (i=1; i < num; i++)
706 *(branch[n].p + i) = cpu_to_le32(++current_block);
708 BUFFER_TRACE(bh, "marking uptodate");
709 set_buffer_uptodate(bh);
712 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
713 err = ext3_journal_dirty_metadata(handle, bh);
720 /* Allocation failed, free what we already allocated */
721 for (i = 1; i <= n ; i++) {
722 BUFFER_TRACE(branch[i].bh, "call journal_forget");
723 ext3_journal_forget(handle, branch[i].bh);
725 for (i = 0; i < indirect_blks; i++)
726 ext3_free_blocks(handle, inode, new_blocks[i], 1);
728 ext3_free_blocks(handle, inode, new_blocks[i], num);
734 * ext3_splice_branch - splice the allocated branch onto inode.
735 * @handle: handle for this transaction
737 * @block: (logical) number of block we are adding
738 * @where: location of missing link
739 * @num: number of indirect blocks we are adding
740 * @blks: number of direct blocks we are adding
742 * This function fills the missing link and does all housekeeping needed in
743 * inode (->i_blocks, etc.). In case of success we end up with the full
744 * chain to new block and return 0.
746 static int ext3_splice_branch(handle_t *handle, struct inode *inode,
747 long block, Indirect *where, int num, int blks)
751 struct ext3_block_alloc_info *block_i;
752 ext3_fsblk_t current_block;
753 struct ext3_inode_info *ei = EXT3_I(inode);
756 block_i = ei->i_block_alloc_info;
758 * If we're splicing into a [td]indirect block (as opposed to the
759 * inode) then we need to get write access to the [td]indirect block
763 BUFFER_TRACE(where->bh, "get_write_access");
764 err = ext3_journal_get_write_access(handle, where->bh);
770 *where->p = where->key;
773 * Update the host buffer_head or inode to point to more just allocated
774 * direct blocks blocks
776 if (num == 0 && blks > 1) {
777 current_block = le32_to_cpu(where->key) + 1;
778 for (i = 1; i < blks; i++)
779 *(where->p + i ) = cpu_to_le32(current_block++);
783 * update the most recently allocated logical & physical block
784 * in i_block_alloc_info, to assist find the proper goal block for next
788 block_i->last_alloc_logical_block = block + blks - 1;
789 block_i->last_alloc_physical_block =
790 le32_to_cpu(where[num].key) + blks - 1;
793 /* We are done with atomic stuff, now do the rest of housekeeping */
794 now = CURRENT_TIME_SEC;
795 if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) {
796 inode->i_ctime = now;
797 ext3_mark_inode_dirty(handle, inode);
799 /* ext3_mark_inode_dirty already updated i_sync_tid */
800 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
802 /* had we spliced it onto indirect block? */
805 * If we spliced it onto an indirect block, we haven't
806 * altered the inode. Note however that if it is being spliced
807 * onto an indirect block at the very end of the file (the
808 * file is growing) then we *will* alter the inode to reflect
809 * the new i_size. But that is not done here - it is done in
810 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
812 jbd_debug(5, "splicing indirect only\n");
813 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
814 err = ext3_journal_dirty_metadata(handle, where->bh);
819 * OK, we spliced it into the inode itself on a direct block.
820 * Inode was dirtied above.
822 jbd_debug(5, "splicing direct\n");
827 for (i = 1; i <= num; i++) {
828 BUFFER_TRACE(where[i].bh, "call journal_forget");
829 ext3_journal_forget(handle, where[i].bh);
830 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
832 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
838 * Allocation strategy is simple: if we have to allocate something, we will
839 * have to go the whole way to leaf. So let's do it before attaching anything
840 * to tree, set linkage between the newborn blocks, write them if sync is
841 * required, recheck the path, free and repeat if check fails, otherwise
842 * set the last missing link (that will protect us from any truncate-generated
843 * removals - all blocks on the path are immune now) and possibly force the
844 * write on the parent block.
845 * That has a nice additional property: no special recovery from the failed
846 * allocations is needed - we simply release blocks and do not touch anything
847 * reachable from inode.
849 * `handle' can be NULL if create == 0.
851 * The BKL may not be held on entry here. Be sure to take it early.
852 * return > 0, # of blocks mapped or allocated.
853 * return = 0, if plain lookup failed.
854 * return < 0, error case.
856 int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
857 sector_t iblock, unsigned long maxblocks,
858 struct buffer_head *bh_result,
867 int blocks_to_boundary = 0;
869 struct ext3_inode_info *ei = EXT3_I(inode);
871 ext3_fsblk_t first_block = 0;
874 trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create);
875 J_ASSERT(handle != NULL || create == 0);
876 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
881 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
883 /* Simplest case - block found, no allocation needed */
885 first_block = le32_to_cpu(chain[depth - 1].key);
886 clear_buffer_new(bh_result);
889 while (count < maxblocks && count <= blocks_to_boundary) {
892 if (!verify_chain(chain, chain + depth - 1)) {
894 * Indirect block might be removed by
895 * truncate while we were reading it.
896 * Handling of that case: forget what we've
897 * got now. Flag the err as EAGAIN, so it
904 blk = le32_to_cpu(*(chain[depth-1].p + count));
906 if (blk == first_block + count)
915 /* Next simple case - plain lookup or failed read of indirect block */
916 if (!create || err == -EIO)
920 * Block out ext3_truncate while we alter the tree
922 mutex_lock(&ei->truncate_mutex);
925 * If the indirect block is missing while we are reading
926 * the chain(ext3_get_branch() returns -EAGAIN err), or
927 * if the chain has been changed after we grab the semaphore,
928 * (either because another process truncated this branch, or
929 * another get_block allocated this branch) re-grab the chain to see if
930 * the request block has been allocated or not.
932 * Since we already block the truncate/other get_block
933 * at this point, we will have the current copy of the chain when we
934 * splice the branch into the tree.
936 if (err == -EAGAIN || !verify_chain(chain, partial)) {
937 while (partial > chain) {
941 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
944 mutex_unlock(&ei->truncate_mutex);
947 clear_buffer_new(bh_result);
953 * Okay, we need to do block allocation. Lazily initialize the block
954 * allocation info here if necessary
956 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
957 ext3_init_block_alloc_info(inode);
959 goal = ext3_find_goal(inode, iblock, partial);
961 /* the number of blocks need to allocate for [d,t]indirect blocks */
962 indirect_blks = (chain + depth) - partial - 1;
965 * Next look up the indirect map to count the totoal number of
966 * direct blocks to allocate for this branch.
968 count = ext3_blks_to_allocate(partial, indirect_blks,
969 maxblocks, blocks_to_boundary);
970 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
971 offsets + (partial - chain), partial);
974 * The ext3_splice_branch call will free and forget any buffers
975 * on the new chain if there is a failure, but that risks using
976 * up transaction credits, especially for bitmaps where the
977 * credits cannot be returned. Can we handle this somehow? We
978 * may need to return -EAGAIN upwards in the worst case. --sct
981 err = ext3_splice_branch(handle, inode, iblock,
982 partial, indirect_blks, count);
983 mutex_unlock(&ei->truncate_mutex);
987 set_buffer_new(bh_result);
989 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
990 if (count > blocks_to_boundary)
991 set_buffer_boundary(bh_result);
993 /* Clean up and exit */
994 partial = chain + depth - 1; /* the whole chain */
996 while (partial > chain) {
997 BUFFER_TRACE(partial->bh, "call brelse");
1001 BUFFER_TRACE(bh_result, "returned");
1003 trace_ext3_get_blocks_exit(inode, iblock,
1004 depth ? le32_to_cpu(chain[depth-1].key) : 0,
1009 /* Maximum number of blocks we map for direct IO at once. */
1010 #define DIO_MAX_BLOCKS 4096
1012 * Number of credits we need for writing DIO_MAX_BLOCKS:
1013 * We need sb + group descriptor + bitmap + inode -> 4
1014 * For B blocks with A block pointers per block we need:
1015 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1016 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1018 #define DIO_CREDITS 25
1020 static int ext3_get_block(struct inode *inode, sector_t iblock,
1021 struct buffer_head *bh_result, int create)
1023 handle_t *handle = ext3_journal_current_handle();
1024 int ret = 0, started = 0;
1025 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1027 if (create && !handle) { /* Direct IO write... */
1028 if (max_blocks > DIO_MAX_BLOCKS)
1029 max_blocks = DIO_MAX_BLOCKS;
1030 handle = ext3_journal_start(inode, DIO_CREDITS +
1031 EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb));
1032 if (IS_ERR(handle)) {
1033 ret = PTR_ERR(handle);
1039 ret = ext3_get_blocks_handle(handle, inode, iblock,
1040 max_blocks, bh_result, create);
1042 bh_result->b_size = (ret << inode->i_blkbits);
1046 ext3_journal_stop(handle);
1051 int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1054 return generic_block_fiemap(inode, fieinfo, start, len,
1059 * `handle' can be NULL if create is zero
1061 struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1062 long block, int create, int *errp)
1064 struct buffer_head dummy;
1067 J_ASSERT(handle != NULL || create == 0);
1070 dummy.b_blocknr = -1000;
1071 buffer_trace_init(&dummy.b_history);
1072 err = ext3_get_blocks_handle(handle, inode, block, 1,
1075 * ext3_get_blocks_handle() returns number of blocks
1076 * mapped. 0 in case of a HOLE.
1083 if (!err && buffer_mapped(&dummy)) {
1084 struct buffer_head *bh;
1085 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1086 if (unlikely(!bh)) {
1090 if (buffer_new(&dummy)) {
1091 J_ASSERT(create != 0);
1092 J_ASSERT(handle != NULL);
1095 * Now that we do not always journal data, we should
1096 * keep in mind whether this should always journal the
1097 * new buffer as metadata. For now, regular file
1098 * writes use ext3_get_block instead, so it's not a
1102 BUFFER_TRACE(bh, "call get_create_access");
1103 fatal = ext3_journal_get_create_access(handle, bh);
1104 if (!fatal && !buffer_uptodate(bh)) {
1105 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1106 set_buffer_uptodate(bh);
1109 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1110 err = ext3_journal_dirty_metadata(handle, bh);
1114 BUFFER_TRACE(bh, "not a new buffer");
1127 struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1128 int block, int create, int *err)
1130 struct buffer_head * bh;
1132 bh = ext3_getblk(handle, inode, block, create, err);
1135 if (bh_uptodate_or_lock(bh))
1138 bh->b_end_io = end_buffer_read_sync;
1139 submit_bh(READ | REQ_META | REQ_PRIO, bh);
1141 if (buffer_uptodate(bh))
1148 static int walk_page_buffers( handle_t *handle,
1149 struct buffer_head *head,
1153 int (*fn)( handle_t *handle,
1154 struct buffer_head *bh))
1156 struct buffer_head *bh;
1157 unsigned block_start, block_end;
1158 unsigned blocksize = head->b_size;
1160 struct buffer_head *next;
1162 for ( bh = head, block_start = 0;
1163 ret == 0 && (bh != head || !block_start);
1164 block_start = block_end, bh = next)
1166 next = bh->b_this_page;
1167 block_end = block_start + blocksize;
1168 if (block_end <= from || block_start >= to) {
1169 if (partial && !buffer_uptodate(bh))
1173 err = (*fn)(handle, bh);
1181 * To preserve ordering, it is essential that the hole instantiation and
1182 * the data write be encapsulated in a single transaction. We cannot
1183 * close off a transaction and start a new one between the ext3_get_block()
1184 * and the commit_write(). So doing the journal_start at the start of
1185 * prepare_write() is the right place.
1187 * Also, this function can nest inside ext3_writepage() ->
1188 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1189 * has generated enough buffer credits to do the whole page. So we won't
1190 * block on the journal in that case, which is good, because the caller may
1193 * By accident, ext3 can be reentered when a transaction is open via
1194 * quota file writes. If we were to commit the transaction while thus
1195 * reentered, there can be a deadlock - we would be holding a quota
1196 * lock, and the commit would never complete if another thread had a
1197 * transaction open and was blocking on the quota lock - a ranking
1200 * So what we do is to rely on the fact that journal_stop/journal_start
1201 * will _not_ run commit under these circumstances because handle->h_ref
1202 * is elevated. We'll still have enough credits for the tiny quotafile
1205 static int do_journal_get_write_access(handle_t *handle,
1206 struct buffer_head *bh)
1208 int dirty = buffer_dirty(bh);
1211 if (!buffer_mapped(bh) || buffer_freed(bh))
1214 * __block_prepare_write() could have dirtied some buffers. Clean
1215 * the dirty bit as jbd2_journal_get_write_access() could complain
1216 * otherwise about fs integrity issues. Setting of the dirty bit
1217 * by __block_prepare_write() isn't a real problem here as we clear
1218 * the bit before releasing a page lock and thus writeback cannot
1219 * ever write the buffer.
1222 clear_buffer_dirty(bh);
1223 ret = ext3_journal_get_write_access(handle, bh);
1225 ret = ext3_journal_dirty_metadata(handle, bh);
1230 * Truncate blocks that were not used by write. We have to truncate the
1231 * pagecache as well so that corresponding buffers get properly unmapped.
1233 static void ext3_truncate_failed_write(struct inode *inode)
1235 truncate_inode_pages(inode->i_mapping, inode->i_size);
1236 ext3_truncate(inode);
1240 * Truncate blocks that were not used by direct IO write. We have to zero out
1241 * the last file block as well because direct IO might have written to it.
1243 static void ext3_truncate_failed_direct_write(struct inode *inode)
1245 ext3_block_truncate_page(inode, inode->i_size);
1246 ext3_truncate(inode);
1249 static int ext3_write_begin(struct file *file, struct address_space *mapping,
1250 loff_t pos, unsigned len, unsigned flags,
1251 struct page **pagep, void **fsdata)
1253 struct inode *inode = mapping->host;
1260 /* Reserve one block more for addition to orphan list in case
1261 * we allocate blocks but write fails for some reason */
1262 int needed_blocks = ext3_writepage_trans_blocks(inode) + 1;
1264 trace_ext3_write_begin(inode, pos, len, flags);
1266 index = pos >> PAGE_CACHE_SHIFT;
1267 from = pos & (PAGE_CACHE_SIZE - 1);
1271 page = grab_cache_page_write_begin(mapping, index, flags);
1276 handle = ext3_journal_start(inode, needed_blocks);
1277 if (IS_ERR(handle)) {
1279 page_cache_release(page);
1280 ret = PTR_ERR(handle);
1283 ret = __block_write_begin(page, pos, len, ext3_get_block);
1285 goto write_begin_failed;
1287 if (ext3_should_journal_data(inode)) {
1288 ret = walk_page_buffers(handle, page_buffers(page),
1289 from, to, NULL, do_journal_get_write_access);
1294 * block_write_begin may have instantiated a few blocks
1295 * outside i_size. Trim these off again. Don't need
1296 * i_size_read because we hold i_mutex.
1298 * Add inode to orphan list in case we crash before truncate
1299 * finishes. Do this only if ext3_can_truncate() agrees so
1300 * that orphan processing code is happy.
1302 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1303 ext3_orphan_add(handle, inode);
1304 ext3_journal_stop(handle);
1306 page_cache_release(page);
1307 if (pos + len > inode->i_size)
1308 ext3_truncate_failed_write(inode);
1310 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1317 int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1319 int err = journal_dirty_data(handle, bh);
1321 ext3_journal_abort_handle(__func__, __func__,
1326 /* For ordered writepage and write_end functions */
1327 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1330 * Write could have mapped the buffer but it didn't copy the data in
1331 * yet. So avoid filing such buffer into a transaction.
1333 if (buffer_mapped(bh) && buffer_uptodate(bh))
1334 return ext3_journal_dirty_data(handle, bh);
1338 /* For write_end() in data=journal mode */
1339 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1341 if (!buffer_mapped(bh) || buffer_freed(bh))
1343 set_buffer_uptodate(bh);
1344 return ext3_journal_dirty_metadata(handle, bh);
1348 * This is nasty and subtle: ext3_write_begin() could have allocated blocks
1349 * for the whole page but later we failed to copy the data in. Update inode
1350 * size according to what we managed to copy. The rest is going to be
1351 * truncated in write_end function.
1353 static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied)
1355 /* What matters to us is i_disksize. We don't write i_size anywhere */
1356 if (pos + copied > inode->i_size)
1357 i_size_write(inode, pos + copied);
1358 if (pos + copied > EXT3_I(inode)->i_disksize) {
1359 EXT3_I(inode)->i_disksize = pos + copied;
1360 mark_inode_dirty(inode);
1365 * We need to pick up the new inode size which generic_commit_write gave us
1366 * `file' can be NULL - eg, when called from page_symlink().
1368 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1369 * buffers are managed internally.
1371 static int ext3_ordered_write_end(struct file *file,
1372 struct address_space *mapping,
1373 loff_t pos, unsigned len, unsigned copied,
1374 struct page *page, void *fsdata)
1376 handle_t *handle = ext3_journal_current_handle();
1377 struct inode *inode = file->f_mapping->host;
1381 trace_ext3_ordered_write_end(inode, pos, len, copied);
1382 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1384 from = pos & (PAGE_CACHE_SIZE - 1);
1386 ret = walk_page_buffers(handle, page_buffers(page),
1387 from, to, NULL, journal_dirty_data_fn);
1390 update_file_sizes(inode, pos, copied);
1392 * There may be allocated blocks outside of i_size because
1393 * we failed to copy some data. Prepare for truncate.
1395 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1396 ext3_orphan_add(handle, inode);
1397 ret2 = ext3_journal_stop(handle);
1401 page_cache_release(page);
1403 if (pos + len > inode->i_size)
1404 ext3_truncate_failed_write(inode);
1405 return ret ? ret : copied;
1408 static int ext3_writeback_write_end(struct file *file,
1409 struct address_space *mapping,
1410 loff_t pos, unsigned len, unsigned copied,
1411 struct page *page, void *fsdata)
1413 handle_t *handle = ext3_journal_current_handle();
1414 struct inode *inode = file->f_mapping->host;
1417 trace_ext3_writeback_write_end(inode, pos, len, copied);
1418 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1419 update_file_sizes(inode, pos, copied);
1421 * There may be allocated blocks outside of i_size because
1422 * we failed to copy some data. Prepare for truncate.
1424 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1425 ext3_orphan_add(handle, inode);
1426 ret = ext3_journal_stop(handle);
1428 page_cache_release(page);
1430 if (pos + len > inode->i_size)
1431 ext3_truncate_failed_write(inode);
1432 return ret ? ret : copied;
1435 static int ext3_journalled_write_end(struct file *file,
1436 struct address_space *mapping,
1437 loff_t pos, unsigned len, unsigned copied,
1438 struct page *page, void *fsdata)
1440 handle_t *handle = ext3_journal_current_handle();
1441 struct inode *inode = mapping->host;
1442 struct ext3_inode_info *ei = EXT3_I(inode);
1447 trace_ext3_journalled_write_end(inode, pos, len, copied);
1448 from = pos & (PAGE_CACHE_SIZE - 1);
1452 if (!PageUptodate(page))
1454 page_zero_new_buffers(page, from + copied, to);
1458 ret = walk_page_buffers(handle, page_buffers(page), from,
1459 to, &partial, write_end_fn);
1461 SetPageUptodate(page);
1463 if (pos + copied > inode->i_size)
1464 i_size_write(inode, pos + copied);
1466 * There may be allocated blocks outside of i_size because
1467 * we failed to copy some data. Prepare for truncate.
1469 if (pos + len > inode->i_size && ext3_can_truncate(inode))
1470 ext3_orphan_add(handle, inode);
1471 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1472 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
1473 if (inode->i_size > ei->i_disksize) {
1474 ei->i_disksize = inode->i_size;
1475 ret2 = ext3_mark_inode_dirty(handle, inode);
1480 ret2 = ext3_journal_stop(handle);
1484 page_cache_release(page);
1486 if (pos + len > inode->i_size)
1487 ext3_truncate_failed_write(inode);
1488 return ret ? ret : copied;
1492 * bmap() is special. It gets used by applications such as lilo and by
1493 * the swapper to find the on-disk block of a specific piece of data.
1495 * Naturally, this is dangerous if the block concerned is still in the
1496 * journal. If somebody makes a swapfile on an ext3 data-journaling
1497 * filesystem and enables swap, then they may get a nasty shock when the
1498 * data getting swapped to that swapfile suddenly gets overwritten by
1499 * the original zero's written out previously to the journal and
1500 * awaiting writeback in the kernel's buffer cache.
1502 * So, if we see any bmap calls here on a modified, data-journaled file,
1503 * take extra steps to flush any blocks which might be in the cache.
1505 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1507 struct inode *inode = mapping->host;
1511 if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) {
1513 * This is a REALLY heavyweight approach, but the use of
1514 * bmap on dirty files is expected to be extremely rare:
1515 * only if we run lilo or swapon on a freshly made file
1516 * do we expect this to happen.
1518 * (bmap requires CAP_SYS_RAWIO so this does not
1519 * represent an unprivileged user DOS attack --- we'd be
1520 * in trouble if mortal users could trigger this path at
1523 * NB. EXT3_STATE_JDATA is not set on files other than
1524 * regular files. If somebody wants to bmap a directory
1525 * or symlink and gets confused because the buffer
1526 * hasn't yet been flushed to disk, they deserve
1527 * everything they get.
1530 ext3_clear_inode_state(inode, EXT3_STATE_JDATA);
1531 journal = EXT3_JOURNAL(inode);
1532 journal_lock_updates(journal);
1533 err = journal_flush(journal);
1534 journal_unlock_updates(journal);
1540 return generic_block_bmap(mapping,block,ext3_get_block);
1543 static int bget_one(handle_t *handle, struct buffer_head *bh)
1549 static int bput_one(handle_t *handle, struct buffer_head *bh)
1555 static int buffer_unmapped(handle_t *handle, struct buffer_head *bh)
1557 return !buffer_mapped(bh);
1561 * Note that we always start a transaction even if we're not journalling
1562 * data. This is to preserve ordering: any hole instantiation within
1563 * __block_write_full_page -> ext3_get_block() should be journalled
1564 * along with the data so we don't crash and then get metadata which
1565 * refers to old data.
1567 * In all journalling modes block_write_full_page() will start the I/O.
1571 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1576 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1578 * Same applies to ext3_get_block(). We will deadlock on various things like
1579 * lock_journal and i_truncate_mutex.
1581 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1584 * 16May01: If we're reentered then journal_current_handle() will be
1585 * non-zero. We simply *return*.
1587 * 1 July 2001: @@@ FIXME:
1588 * In journalled data mode, a data buffer may be metadata against the
1589 * current transaction. But the same file is part of a shared mapping
1590 * and someone does a writepage() on it.
1592 * We will move the buffer onto the async_data list, but *after* it has
1593 * been dirtied. So there's a small window where we have dirty data on
1596 * Note that this only applies to the last partial page in the file. The
1597 * bit which block_write_full_page() uses prepare/commit for. (That's
1598 * broken code anyway: it's wrong for msync()).
1600 * It's a rare case: affects the final partial page, for journalled data
1601 * where the file is subject to bith write() and writepage() in the same
1602 * transction. To fix it we'll need a custom block_write_full_page().
1603 * We'll probably need that anyway for journalling writepage() output.
1605 * We don't honour synchronous mounts for writepage(). That would be
1606 * disastrous. Any write() or metadata operation will sync the fs for
1609 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1610 * we don't need to open a transaction here.
1612 static int ext3_ordered_writepage(struct page *page,
1613 struct writeback_control *wbc)
1615 struct inode *inode = page->mapping->host;
1616 struct buffer_head *page_bufs;
1617 handle_t *handle = NULL;
1621 J_ASSERT(PageLocked(page));
1623 * We don't want to warn for emergency remount. The condition is
1624 * ordered to avoid dereferencing inode->i_sb in non-error case to
1627 WARN_ON_ONCE(IS_RDONLY(inode) &&
1628 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1631 * We give up here if we're reentered, because it might be for a
1632 * different filesystem.
1634 if (ext3_journal_current_handle())
1637 trace_ext3_ordered_writepage(page);
1638 if (!page_has_buffers(page)) {
1639 create_empty_buffers(page, inode->i_sb->s_blocksize,
1640 (1 << BH_Dirty)|(1 << BH_Uptodate));
1641 page_bufs = page_buffers(page);
1643 page_bufs = page_buffers(page);
1644 if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE,
1645 NULL, buffer_unmapped)) {
1646 /* Provide NULL get_block() to catch bugs if buffers
1647 * weren't really mapped */
1648 return block_write_full_page(page, NULL, wbc);
1651 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1653 if (IS_ERR(handle)) {
1654 ret = PTR_ERR(handle);
1658 walk_page_buffers(handle, page_bufs, 0,
1659 PAGE_CACHE_SIZE, NULL, bget_one);
1661 ret = block_write_full_page(page, ext3_get_block, wbc);
1664 * The page can become unlocked at any point now, and
1665 * truncate can then come in and change things. So we
1666 * can't touch *page from now on. But *page_bufs is
1667 * safe due to elevated refcount.
1671 * And attach them to the current transaction. But only if
1672 * block_write_full_page() succeeded. Otherwise they are unmapped,
1673 * and generally junk.
1676 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1677 NULL, journal_dirty_data_fn);
1681 walk_page_buffers(handle, page_bufs, 0,
1682 PAGE_CACHE_SIZE, NULL, bput_one);
1683 err = ext3_journal_stop(handle);
1689 redirty_page_for_writepage(wbc, page);
1694 static int ext3_writeback_writepage(struct page *page,
1695 struct writeback_control *wbc)
1697 struct inode *inode = page->mapping->host;
1698 handle_t *handle = NULL;
1702 J_ASSERT(PageLocked(page));
1704 * We don't want to warn for emergency remount. The condition is
1705 * ordered to avoid dereferencing inode->i_sb in non-error case to
1708 WARN_ON_ONCE(IS_RDONLY(inode) &&
1709 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1711 if (ext3_journal_current_handle())
1714 trace_ext3_writeback_writepage(page);
1715 if (page_has_buffers(page)) {
1716 if (!walk_page_buffers(NULL, page_buffers(page), 0,
1717 PAGE_CACHE_SIZE, NULL, buffer_unmapped)) {
1718 /* Provide NULL get_block() to catch bugs if buffers
1719 * weren't really mapped */
1720 return block_write_full_page(page, NULL, wbc);
1724 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1725 if (IS_ERR(handle)) {
1726 ret = PTR_ERR(handle);
1730 ret = block_write_full_page(page, ext3_get_block, wbc);
1732 err = ext3_journal_stop(handle);
1738 redirty_page_for_writepage(wbc, page);
1743 static int ext3_journalled_writepage(struct page *page,
1744 struct writeback_control *wbc)
1746 struct inode *inode = page->mapping->host;
1747 handle_t *handle = NULL;
1751 J_ASSERT(PageLocked(page));
1753 * We don't want to warn for emergency remount. The condition is
1754 * ordered to avoid dereferencing inode->i_sb in non-error case to
1757 WARN_ON_ONCE(IS_RDONLY(inode) &&
1758 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS));
1760 if (ext3_journal_current_handle())
1763 trace_ext3_journalled_writepage(page);
1764 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1765 if (IS_ERR(handle)) {
1766 ret = PTR_ERR(handle);
1770 if (!page_has_buffers(page) || PageChecked(page)) {
1772 * It's mmapped pagecache. Add buffers and journal it. There
1773 * doesn't seem much point in redirtying the page here.
1775 ClearPageChecked(page);
1776 ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE,
1779 ext3_journal_stop(handle);
1782 ret = walk_page_buffers(handle, page_buffers(page), 0,
1783 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1785 err = walk_page_buffers(handle, page_buffers(page), 0,
1786 PAGE_CACHE_SIZE, NULL, write_end_fn);
1789 ext3_set_inode_state(inode, EXT3_STATE_JDATA);
1790 atomic_set(&EXT3_I(inode)->i_datasync_tid,
1791 handle->h_transaction->t_tid);
1795 * It may be a page full of checkpoint-mode buffers. We don't
1796 * really know unless we go poke around in the buffer_heads.
1797 * But block_write_full_page will do the right thing.
1799 ret = block_write_full_page(page, ext3_get_block, wbc);
1801 err = ext3_journal_stop(handle);
1808 redirty_page_for_writepage(wbc, page);
1814 static int ext3_readpage(struct file *file, struct page *page)
1816 trace_ext3_readpage(page);
1817 return mpage_readpage(page, ext3_get_block);
1821 ext3_readpages(struct file *file, struct address_space *mapping,
1822 struct list_head *pages, unsigned nr_pages)
1824 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1827 static void ext3_invalidatepage(struct page *page, unsigned long offset)
1829 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1831 trace_ext3_invalidatepage(page, offset);
1834 * If it's a full truncate we just forget about the pending dirtying
1837 ClearPageChecked(page);
1839 journal_invalidatepage(journal, page, offset);
1842 static int ext3_releasepage(struct page *page, gfp_t wait)
1844 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1846 trace_ext3_releasepage(page);
1847 WARN_ON(PageChecked(page));
1848 if (!page_has_buffers(page))
1850 return journal_try_to_free_buffers(journal, page, wait);
1854 * If the O_DIRECT write will extend the file then add this inode to the
1855 * orphan list. So recovery will truncate it back to the original size
1856 * if the machine crashes during the write.
1858 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1859 * crashes then stale disk data _may_ be exposed inside the file. But current
1860 * VFS code falls back into buffered path in that case so we are safe.
1862 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1863 const struct iovec *iov, loff_t offset,
1864 unsigned long nr_segs)
1866 struct file *file = iocb->ki_filp;
1867 struct inode *inode = file->f_mapping->host;
1868 struct ext3_inode_info *ei = EXT3_I(inode);
1872 size_t count = iov_length(iov, nr_segs);
1875 trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
1878 loff_t final_size = offset + count;
1880 if (final_size > inode->i_size) {
1881 /* Credits for sb + inode write */
1882 handle = ext3_journal_start(inode, 2);
1883 if (IS_ERR(handle)) {
1884 ret = PTR_ERR(handle);
1887 ret = ext3_orphan_add(handle, inode);
1889 ext3_journal_stop(handle);
1893 ei->i_disksize = inode->i_size;
1894 ext3_journal_stop(handle);
1899 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
1902 * In case of error extending write may have instantiated a few
1903 * blocks outside i_size. Trim these off again.
1905 if (unlikely((rw & WRITE) && ret < 0)) {
1906 loff_t isize = i_size_read(inode);
1907 loff_t end = offset + iov_length(iov, nr_segs);
1910 ext3_truncate_failed_direct_write(inode);
1912 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1918 /* Credits for sb + inode write */
1919 handle = ext3_journal_start(inode, 2);
1920 if (IS_ERR(handle)) {
1921 /* This is really bad luck. We've written the data
1922 * but cannot extend i_size. Truncate allocated blocks
1923 * and pretend the write failed... */
1924 ext3_truncate_failed_direct_write(inode);
1925 ret = PTR_ERR(handle);
1929 ext3_orphan_del(handle, inode);
1931 loff_t end = offset + ret;
1932 if (end > inode->i_size) {
1933 ei->i_disksize = end;
1934 i_size_write(inode, end);
1936 * We're going to return a positive `ret'
1937 * here due to non-zero-length I/O, so there's
1938 * no way of reporting error returns from
1939 * ext3_mark_inode_dirty() to userspace. So
1942 ext3_mark_inode_dirty(handle, inode);
1945 err = ext3_journal_stop(handle);
1950 trace_ext3_direct_IO_exit(inode, offset,
1951 iov_length(iov, nr_segs), rw, ret);
1956 * Pages can be marked dirty completely asynchronously from ext3's journalling
1957 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1958 * much here because ->set_page_dirty is called under VFS locks. The page is
1959 * not necessarily locked.
1961 * We cannot just dirty the page and leave attached buffers clean, because the
1962 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1963 * or jbddirty because all the journalling code will explode.
1965 * So what we do is to mark the page "pending dirty" and next time writepage
1966 * is called, propagate that into the buffers appropriately.
1968 static int ext3_journalled_set_page_dirty(struct page *page)
1970 SetPageChecked(page);
1971 return __set_page_dirty_nobuffers(page);
1974 static const struct address_space_operations ext3_ordered_aops = {
1975 .readpage = ext3_readpage,
1976 .readpages = ext3_readpages,
1977 .writepage = ext3_ordered_writepage,
1978 .write_begin = ext3_write_begin,
1979 .write_end = ext3_ordered_write_end,
1981 .invalidatepage = ext3_invalidatepage,
1982 .releasepage = ext3_releasepage,
1983 .direct_IO = ext3_direct_IO,
1984 .migratepage = buffer_migrate_page,
1985 .is_partially_uptodate = block_is_partially_uptodate,
1986 .error_remove_page = generic_error_remove_page,
1989 static const struct address_space_operations ext3_writeback_aops = {
1990 .readpage = ext3_readpage,
1991 .readpages = ext3_readpages,
1992 .writepage = ext3_writeback_writepage,
1993 .write_begin = ext3_write_begin,
1994 .write_end = ext3_writeback_write_end,
1996 .invalidatepage = ext3_invalidatepage,
1997 .releasepage = ext3_releasepage,
1998 .direct_IO = ext3_direct_IO,
1999 .migratepage = buffer_migrate_page,
2000 .is_partially_uptodate = block_is_partially_uptodate,
2001 .error_remove_page = generic_error_remove_page,
2004 static const struct address_space_operations ext3_journalled_aops = {
2005 .readpage = ext3_readpage,
2006 .readpages = ext3_readpages,
2007 .writepage = ext3_journalled_writepage,
2008 .write_begin = ext3_write_begin,
2009 .write_end = ext3_journalled_write_end,
2010 .set_page_dirty = ext3_journalled_set_page_dirty,
2012 .invalidatepage = ext3_invalidatepage,
2013 .releasepage = ext3_releasepage,
2014 .is_partially_uptodate = block_is_partially_uptodate,
2015 .error_remove_page = generic_error_remove_page,
2018 void ext3_set_aops(struct inode *inode)
2020 if (ext3_should_order_data(inode))
2021 inode->i_mapping->a_ops = &ext3_ordered_aops;
2022 else if (ext3_should_writeback_data(inode))
2023 inode->i_mapping->a_ops = &ext3_writeback_aops;
2025 inode->i_mapping->a_ops = &ext3_journalled_aops;
2029 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
2030 * up to the end of the block which corresponds to `from'.
2031 * This required during truncate. We need to physically zero the tail end
2032 * of that block so it doesn't yield old data if the file is later grown.
2034 static int ext3_block_truncate_page(struct inode *inode, loff_t from)
2036 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2037 unsigned offset = from & (PAGE_CACHE_SIZE - 1);
2038 unsigned blocksize, iblock, length, pos;
2040 handle_t *handle = NULL;
2041 struct buffer_head *bh;
2044 /* Truncated on block boundary - nothing to do */
2045 blocksize = inode->i_sb->s_blocksize;
2046 if ((from & (blocksize - 1)) == 0)
2049 page = grab_cache_page(inode->i_mapping, index);
2052 length = blocksize - (offset & (blocksize - 1));
2053 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2055 if (!page_has_buffers(page))
2056 create_empty_buffers(page, blocksize, 0);
2058 /* Find the buffer that contains "offset" */
2059 bh = page_buffers(page);
2061 while (offset >= pos) {
2062 bh = bh->b_this_page;
2068 if (buffer_freed(bh)) {
2069 BUFFER_TRACE(bh, "freed: skip");
2073 if (!buffer_mapped(bh)) {
2074 BUFFER_TRACE(bh, "unmapped");
2075 ext3_get_block(inode, iblock, bh, 0);
2076 /* unmapped? It's a hole - nothing to do */
2077 if (!buffer_mapped(bh)) {
2078 BUFFER_TRACE(bh, "still unmapped");
2083 /* Ok, it's mapped. Make sure it's up-to-date */
2084 if (PageUptodate(page))
2085 set_buffer_uptodate(bh);
2087 if (!bh_uptodate_or_lock(bh)) {
2088 err = bh_submit_read(bh);
2089 /* Uhhuh. Read error. Complain and punt. */
2094 /* data=writeback mode doesn't need transaction to zero-out data */
2095 if (!ext3_should_writeback_data(inode)) {
2096 /* We journal at most one block */
2097 handle = ext3_journal_start(inode, 1);
2098 if (IS_ERR(handle)) {
2099 clear_highpage(page);
2100 flush_dcache_page(page);
2101 err = PTR_ERR(handle);
2106 if (ext3_should_journal_data(inode)) {
2107 BUFFER_TRACE(bh, "get write access");
2108 err = ext3_journal_get_write_access(handle, bh);
2113 zero_user(page, offset, length);
2114 BUFFER_TRACE(bh, "zeroed end of block");
2117 if (ext3_should_journal_data(inode)) {
2118 err = ext3_journal_dirty_metadata(handle, bh);
2120 if (ext3_should_order_data(inode))
2121 err = ext3_journal_dirty_data(handle, bh);
2122 mark_buffer_dirty(bh);
2126 ext3_journal_stop(handle);
2130 page_cache_release(page);
2135 * Probably it should be a library function... search for first non-zero word
2136 * or memcmp with zero_page, whatever is better for particular architecture.
2139 static inline int all_zeroes(__le32 *p, __le32 *q)
2148 * ext3_find_shared - find the indirect blocks for partial truncation.
2149 * @inode: inode in question
2150 * @depth: depth of the affected branch
2151 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
2152 * @chain: place to store the pointers to partial indirect blocks
2153 * @top: place to the (detached) top of branch
2155 * This is a helper function used by ext3_truncate().
2157 * When we do truncate() we may have to clean the ends of several
2158 * indirect blocks but leave the blocks themselves alive. Block is
2159 * partially truncated if some data below the new i_size is referred
2160 * from it (and it is on the path to the first completely truncated
2161 * data block, indeed). We have to free the top of that path along
2162 * with everything to the right of the path. Since no allocation
2163 * past the truncation point is possible until ext3_truncate()
2164 * finishes, we may safely do the latter, but top of branch may
2165 * require special attention - pageout below the truncation point
2166 * might try to populate it.
2168 * We atomically detach the top of branch from the tree, store the
2169 * block number of its root in *@top, pointers to buffer_heads of
2170 * partially truncated blocks - in @chain[].bh and pointers to
2171 * their last elements that should not be removed - in
2172 * @chain[].p. Return value is the pointer to last filled element
2175 * The work left to caller to do the actual freeing of subtrees:
2176 * a) free the subtree starting from *@top
2177 * b) free the subtrees whose roots are stored in
2178 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2179 * c) free the subtrees growing from the inode past the @chain[0].
2180 * (no partially truncated stuff there). */
2182 static Indirect *ext3_find_shared(struct inode *inode, int depth,
2183 int offsets[4], Indirect chain[4], __le32 *top)
2185 Indirect *partial, *p;
2189 /* Make k index the deepest non-null offset + 1 */
2190 for (k = depth; k > 1 && !offsets[k-1]; k--)
2192 partial = ext3_get_branch(inode, k, offsets, chain, &err);
2193 /* Writer: pointers */
2195 partial = chain + k-1;
2197 * If the branch acquired continuation since we've looked at it -
2198 * fine, it should all survive and (new) top doesn't belong to us.
2200 if (!partial->key && *partial->p)
2203 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2206 * OK, we've found the last block that must survive. The rest of our
2207 * branch should be detached before unlocking. However, if that rest
2208 * of branch is all ours and does not grow immediately from the inode
2209 * it's easier to cheat and just decrement partial->p.
2211 if (p == chain + k - 1 && p > chain) {
2215 /* Nope, don't do this in ext3. Must leave the tree intact */
2222 while(partial > p) {
2223 brelse(partial->bh);
2231 * Zero a number of block pointers in either an inode or an indirect block.
2232 * If we restart the transaction we must again get write access to the
2233 * indirect block for further modification.
2235 * We release `count' blocks on disk, but (last - first) may be greater
2236 * than `count' because there can be holes in there.
2238 static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
2239 struct buffer_head *bh, ext3_fsblk_t block_to_free,
2240 unsigned long count, __le32 *first, __le32 *last)
2243 if (try_to_extend_transaction(handle, inode)) {
2245 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2246 if (ext3_journal_dirty_metadata(handle, bh))
2249 ext3_mark_inode_dirty(handle, inode);
2250 truncate_restart_transaction(handle, inode);
2252 BUFFER_TRACE(bh, "retaking write access");
2253 if (ext3_journal_get_write_access(handle, bh))
2259 * Any buffers which are on the journal will be in memory. We find
2260 * them on the hash table so journal_revoke() will run journal_forget()
2261 * on them. We've already detached each block from the file, so
2262 * bforget() in journal_forget() should be safe.
2264 * AKPM: turn on bforget in journal_forget()!!!
2266 for (p = first; p < last; p++) {
2267 u32 nr = le32_to_cpu(*p);
2269 struct buffer_head *bh;
2272 bh = sb_find_get_block(inode->i_sb, nr);
2273 ext3_forget(handle, 0, inode, bh, nr);
2277 ext3_free_blocks(handle, inode, block_to_free, count);
2281 * ext3_free_data - free a list of data blocks
2282 * @handle: handle for this transaction
2283 * @inode: inode we are dealing with
2284 * @this_bh: indirect buffer_head which contains *@first and *@last
2285 * @first: array of block numbers
2286 * @last: points immediately past the end of array
2288 * We are freeing all blocks referred from that array (numbers are stored as
2289 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2291 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2292 * blocks are contiguous then releasing them at one time will only affect one
2293 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2294 * actually use a lot of journal space.
2296 * @this_bh will be %NULL if @first and @last point into the inode's direct
2299 static void ext3_free_data(handle_t *handle, struct inode *inode,
2300 struct buffer_head *this_bh,
2301 __le32 *first, __le32 *last)
2303 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
2304 unsigned long count = 0; /* Number of blocks in the run */
2305 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2308 ext3_fsblk_t nr; /* Current block # */
2309 __le32 *p; /* Pointer into inode/ind
2310 for current block */
2313 if (this_bh) { /* For indirect block */
2314 BUFFER_TRACE(this_bh, "get_write_access");
2315 err = ext3_journal_get_write_access(handle, this_bh);
2316 /* Important: if we can't update the indirect pointers
2317 * to the blocks, we can't free them. */
2322 for (p = first; p < last; p++) {
2323 nr = le32_to_cpu(*p);
2325 /* accumulate blocks to free if they're contiguous */
2328 block_to_free_p = p;
2330 } else if (nr == block_to_free + count) {
2333 ext3_clear_blocks(handle, inode, this_bh,
2335 count, block_to_free_p, p);
2337 block_to_free_p = p;
2344 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2345 count, block_to_free_p, p);
2348 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2351 * The buffer head should have an attached journal head at this
2352 * point. However, if the data is corrupted and an indirect
2353 * block pointed to itself, it would have been detached when
2354 * the block was cleared. Check for this instead of OOPSing.
2357 ext3_journal_dirty_metadata(handle, this_bh);
2359 ext3_error(inode->i_sb, "ext3_free_data",
2360 "circular indirect block detected, "
2361 "inode=%lu, block=%llu",
2363 (unsigned long long)this_bh->b_blocknr);
2368 * ext3_free_branches - free an array of branches
2369 * @handle: JBD handle for this transaction
2370 * @inode: inode we are dealing with
2371 * @parent_bh: the buffer_head which contains *@first and *@last
2372 * @first: array of block numbers
2373 * @last: pointer immediately past the end of array
2374 * @depth: depth of the branches to free
2376 * We are freeing all blocks referred from these branches (numbers are
2377 * stored as little-endian 32-bit) and updating @inode->i_blocks
2380 static void ext3_free_branches(handle_t *handle, struct inode *inode,
2381 struct buffer_head *parent_bh,
2382 __le32 *first, __le32 *last, int depth)
2387 if (is_handle_aborted(handle))
2391 struct buffer_head *bh;
2392 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2394 while (--p >= first) {
2395 nr = le32_to_cpu(*p);
2397 continue; /* A hole */
2399 /* Go read the buffer for the next level down */
2400 bh = sb_bread(inode->i_sb, nr);
2403 * A read failure? Report error and clear slot
2407 ext3_error(inode->i_sb, "ext3_free_branches",
2408 "Read failure, inode=%lu, block="E3FSBLK,
2413 /* This zaps the entire block. Bottom up. */
2414 BUFFER_TRACE(bh, "free child branches");
2415 ext3_free_branches(handle, inode, bh,
2416 (__le32*)bh->b_data,
2417 (__le32*)bh->b_data + addr_per_block,
2421 * Everything below this this pointer has been
2422 * released. Now let this top-of-subtree go.
2424 * We want the freeing of this indirect block to be
2425 * atomic in the journal with the updating of the
2426 * bitmap block which owns it. So make some room in
2429 * We zero the parent pointer *after* freeing its
2430 * pointee in the bitmaps, so if extend_transaction()
2431 * for some reason fails to put the bitmap changes and
2432 * the release into the same transaction, recovery
2433 * will merely complain about releasing a free block,
2434 * rather than leaking blocks.
2436 if (is_handle_aborted(handle))
2438 if (try_to_extend_transaction(handle, inode)) {
2439 ext3_mark_inode_dirty(handle, inode);
2440 truncate_restart_transaction(handle, inode);
2444 * We've probably journalled the indirect block several
2445 * times during the truncate. But it's no longer
2446 * needed and we now drop it from the transaction via
2449 * That's easy if it's exclusively part of this
2450 * transaction. But if it's part of the committing
2451 * transaction then journal_forget() will simply
2452 * brelse() it. That means that if the underlying
2453 * block is reallocated in ext3_get_block(),
2454 * unmap_underlying_metadata() will find this block
2455 * and will try to get rid of it. damn, damn. Thus
2456 * we don't allow a block to be reallocated until
2457 * a transaction freeing it has fully committed.
2459 * We also have to make sure journal replay after a
2460 * crash does not overwrite non-journaled data blocks
2461 * with old metadata when the block got reallocated for
2462 * data. Thus we have to store a revoke record for a
2463 * block in the same transaction in which we free the
2466 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2468 ext3_free_blocks(handle, inode, nr, 1);
2472 * The block which we have just freed is
2473 * pointed to by an indirect block: journal it
2475 BUFFER_TRACE(parent_bh, "get_write_access");
2476 if (!ext3_journal_get_write_access(handle,
2479 BUFFER_TRACE(parent_bh,
2480 "call ext3_journal_dirty_metadata");
2481 ext3_journal_dirty_metadata(handle,
2487 /* We have reached the bottom of the tree. */
2488 BUFFER_TRACE(parent_bh, "free data blocks");
2489 ext3_free_data(handle, inode, parent_bh, first, last);
2493 int ext3_can_truncate(struct inode *inode)
2495 if (S_ISREG(inode->i_mode))
2497 if (S_ISDIR(inode->i_mode))
2499 if (S_ISLNK(inode->i_mode))
2500 return !ext3_inode_is_fast_symlink(inode);
2507 * We block out ext3_get_block() block instantiations across the entire
2508 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2509 * simultaneously on behalf of the same inode.
2511 * As we work through the truncate and commit bits of it to the journal there
2512 * is one core, guiding principle: the file's tree must always be consistent on
2513 * disk. We must be able to restart the truncate after a crash.
2515 * The file's tree may be transiently inconsistent in memory (although it
2516 * probably isn't), but whenever we close off and commit a journal transaction,
2517 * the contents of (the filesystem + the journal) must be consistent and
2518 * restartable. It's pretty simple, really: bottom up, right to left (although
2519 * left-to-right works OK too).
2521 * Note that at recovery time, journal replay occurs *before* the restart of
2522 * truncate against the orphan inode list.
2524 * The committed inode has the new, desired i_size (which is the same as
2525 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2526 * that this inode's truncate did not complete and it will again call
2527 * ext3_truncate() to have another go. So there will be instantiated blocks
2528 * to the right of the truncation point in a crashed ext3 filesystem. But
2529 * that's fine - as long as they are linked from the inode, the post-crash
2530 * ext3_truncate() run will find them and release them.
2532 void ext3_truncate(struct inode *inode)
2535 struct ext3_inode_info *ei = EXT3_I(inode);
2536 __le32 *i_data = ei->i_data;
2537 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2544 unsigned blocksize = inode->i_sb->s_blocksize;
2546 trace_ext3_truncate_enter(inode);
2548 if (!ext3_can_truncate(inode))
2551 if (inode->i_size == 0 && ext3_should_writeback_data(inode))
2552 ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE);
2554 handle = start_transaction(inode);
2558 last_block = (inode->i_size + blocksize-1)
2559 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2560 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2562 goto out_stop; /* error */
2565 * OK. This truncate is going to happen. We add the inode to the
2566 * orphan list, so that if this truncate spans multiple transactions,
2567 * and we crash, we will resume the truncate when the filesystem
2568 * recovers. It also marks the inode dirty, to catch the new size.
2570 * Implication: the file must always be in a sane, consistent
2571 * truncatable state while each transaction commits.
2573 if (ext3_orphan_add(handle, inode))
2577 * The orphan list entry will now protect us from any crash which
2578 * occurs before the truncate completes, so it is now safe to propagate
2579 * the new, shorter inode size (held for now in i_size) into the
2580 * on-disk inode. We do this via i_disksize, which is the value which
2581 * ext3 *really* writes onto the disk inode.
2583 ei->i_disksize = inode->i_size;
2586 * From here we block out all ext3_get_block() callers who want to
2587 * modify the block allocation tree.
2589 mutex_lock(&ei->truncate_mutex);
2591 if (n == 1) { /* direct blocks */
2592 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2593 i_data + EXT3_NDIR_BLOCKS);
2597 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2598 /* Kill the top of shared branch (not detached) */
2600 if (partial == chain) {
2601 /* Shared branch grows from the inode */
2602 ext3_free_branches(handle, inode, NULL,
2603 &nr, &nr+1, (chain+n-1) - partial);
2606 * We mark the inode dirty prior to restart,
2607 * and prior to stop. No need for it here.
2610 /* Shared branch grows from an indirect block */
2611 ext3_free_branches(handle, inode, partial->bh,
2613 partial->p+1, (chain+n-1) - partial);
2616 /* Clear the ends of indirect blocks on the shared branch */
2617 while (partial > chain) {
2618 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2619 (__le32*)partial->bh->b_data+addr_per_block,
2620 (chain+n-1) - partial);
2621 BUFFER_TRACE(partial->bh, "call brelse");
2622 brelse (partial->bh);
2626 /* Kill the remaining (whole) subtrees */
2627 switch (offsets[0]) {
2629 nr = i_data[EXT3_IND_BLOCK];
2631 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2632 i_data[EXT3_IND_BLOCK] = 0;
2634 case EXT3_IND_BLOCK:
2635 nr = i_data[EXT3_DIND_BLOCK];
2637 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2638 i_data[EXT3_DIND_BLOCK] = 0;
2640 case EXT3_DIND_BLOCK:
2641 nr = i_data[EXT3_TIND_BLOCK];
2643 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2644 i_data[EXT3_TIND_BLOCK] = 0;
2646 case EXT3_TIND_BLOCK:
2650 ext3_discard_reservation(inode);
2652 mutex_unlock(&ei->truncate_mutex);
2653 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2654 ext3_mark_inode_dirty(handle, inode);
2657 * In a multi-transaction truncate, we only make the final transaction
2664 * If this was a simple ftruncate(), and the file will remain alive
2665 * then we need to clear up the orphan record which we created above.
2666 * However, if this was a real unlink then we were called by
2667 * ext3_evict_inode(), and we allow that function to clean up the
2668 * orphan info for us.
2671 ext3_orphan_del(handle, inode);
2673 ext3_journal_stop(handle);
2674 trace_ext3_truncate_exit(inode);
2678 * Delete the inode from orphan list so that it doesn't stay there
2679 * forever and trigger assertion on umount.
2682 ext3_orphan_del(NULL, inode);
2683 trace_ext3_truncate_exit(inode);
2686 static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2687 unsigned long ino, struct ext3_iloc *iloc)
2689 unsigned long block_group;
2690 unsigned long offset;
2692 struct ext3_group_desc *gdp;
2694 if (!ext3_valid_inum(sb, ino)) {
2696 * This error is already checked for in namei.c unless we are
2697 * looking at an NFS filehandle, in which case no error
2703 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2704 gdp = ext3_get_group_desc(sb, block_group, NULL);
2708 * Figure out the offset within the block group inode table
2710 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2711 EXT3_INODE_SIZE(sb);
2712 block = le32_to_cpu(gdp->bg_inode_table) +
2713 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2715 iloc->block_group = block_group;
2716 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2721 * ext3_get_inode_loc returns with an extra refcount against the inode's
2722 * underlying buffer_head on success. If 'in_mem' is true, we have all
2723 * data in memory that is needed to recreate the on-disk version of this
2726 static int __ext3_get_inode_loc(struct inode *inode,
2727 struct ext3_iloc *iloc, int in_mem)
2730 struct buffer_head *bh;
2732 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2736 bh = sb_getblk(inode->i_sb, block);
2737 if (unlikely(!bh)) {
2738 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2739 "unable to read inode block - "
2740 "inode=%lu, block="E3FSBLK,
2741 inode->i_ino, block);
2744 if (!buffer_uptodate(bh)) {
2748 * If the buffer has the write error flag, we have failed
2749 * to write out another inode in the same block. In this
2750 * case, we don't have to read the block because we may
2751 * read the old inode data successfully.
2753 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
2754 set_buffer_uptodate(bh);
2756 if (buffer_uptodate(bh)) {
2757 /* someone brought it uptodate while we waited */
2763 * If we have all information of the inode in memory and this
2764 * is the only valid inode in the block, we need not read the
2768 struct buffer_head *bitmap_bh;
2769 struct ext3_group_desc *desc;
2770 int inodes_per_buffer;
2771 int inode_offset, i;
2775 block_group = (inode->i_ino - 1) /
2776 EXT3_INODES_PER_GROUP(inode->i_sb);
2777 inodes_per_buffer = bh->b_size /
2778 EXT3_INODE_SIZE(inode->i_sb);
2779 inode_offset = ((inode->i_ino - 1) %
2780 EXT3_INODES_PER_GROUP(inode->i_sb));
2781 start = inode_offset & ~(inodes_per_buffer - 1);
2783 /* Is the inode bitmap in cache? */
2784 desc = ext3_get_group_desc(inode->i_sb,
2789 bitmap_bh = sb_getblk(inode->i_sb,
2790 le32_to_cpu(desc->bg_inode_bitmap));
2791 if (unlikely(!bitmap_bh))
2795 * If the inode bitmap isn't in cache then the
2796 * optimisation may end up performing two reads instead
2797 * of one, so skip it.
2799 if (!buffer_uptodate(bitmap_bh)) {
2803 for (i = start; i < start + inodes_per_buffer; i++) {
2804 if (i == inode_offset)
2806 if (ext3_test_bit(i, bitmap_bh->b_data))
2810 if (i == start + inodes_per_buffer) {
2811 /* all other inodes are free, so skip I/O */
2812 memset(bh->b_data, 0, bh->b_size);
2813 set_buffer_uptodate(bh);
2821 * There are other valid inodes in the buffer, this inode
2822 * has in-inode xattrs, or we don't have this inode in memory.
2823 * Read the block from disk.
2825 trace_ext3_load_inode(inode);
2827 bh->b_end_io = end_buffer_read_sync;
2828 submit_bh(READ | REQ_META | REQ_PRIO, bh);
2830 if (!buffer_uptodate(bh)) {
2831 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2832 "unable to read inode block - "
2833 "inode=%lu, block="E3FSBLK,
2834 inode->i_ino, block);
2844 int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2846 /* We have all inode data except xattrs in memory here. */
2847 return __ext3_get_inode_loc(inode, iloc,
2848 !ext3_test_inode_state(inode, EXT3_STATE_XATTR));
2851 void ext3_set_inode_flags(struct inode *inode)
2853 unsigned int flags = EXT3_I(inode)->i_flags;
2855 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2856 if (flags & EXT3_SYNC_FL)
2857 inode->i_flags |= S_SYNC;
2858 if (flags & EXT3_APPEND_FL)
2859 inode->i_flags |= S_APPEND;
2860 if (flags & EXT3_IMMUTABLE_FL)
2861 inode->i_flags |= S_IMMUTABLE;
2862 if (flags & EXT3_NOATIME_FL)
2863 inode->i_flags |= S_NOATIME;
2864 if (flags & EXT3_DIRSYNC_FL)
2865 inode->i_flags |= S_DIRSYNC;
2868 /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */
2869 void ext3_get_inode_flags(struct ext3_inode_info *ei)
2871 unsigned int flags = ei->vfs_inode.i_flags;
2873 ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL|
2874 EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL);
2876 ei->i_flags |= EXT3_SYNC_FL;
2877 if (flags & S_APPEND)
2878 ei->i_flags |= EXT3_APPEND_FL;
2879 if (flags & S_IMMUTABLE)
2880 ei->i_flags |= EXT3_IMMUTABLE_FL;
2881 if (flags & S_NOATIME)
2882 ei->i_flags |= EXT3_NOATIME_FL;
2883 if (flags & S_DIRSYNC)
2884 ei->i_flags |= EXT3_DIRSYNC_FL;
2887 struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
2889 struct ext3_iloc iloc;
2890 struct ext3_inode *raw_inode;
2891 struct ext3_inode_info *ei;
2892 struct buffer_head *bh;
2893 struct inode *inode;
2894 journal_t *journal = EXT3_SB(sb)->s_journal;
2895 transaction_t *transaction;
2901 inode = iget_locked(sb, ino);
2903 return ERR_PTR(-ENOMEM);
2904 if (!(inode->i_state & I_NEW))
2908 ei->i_block_alloc_info = NULL;
2910 ret = __ext3_get_inode_loc(inode, &iloc, 0);
2914 raw_inode = ext3_raw_inode(&iloc);
2915 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2916 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2917 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2918 if(!(test_opt (inode->i_sb, NO_UID32))) {
2919 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2920 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2922 i_uid_write(inode, i_uid);
2923 i_gid_write(inode, i_gid);
2924 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
2925 inode->i_size = le32_to_cpu(raw_inode->i_size);
2926 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
2927 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
2928 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
2929 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2931 ei->i_state_flags = 0;
2932 ei->i_dir_start_lookup = 0;
2933 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2934 /* We now have enough fields to check if the inode was active or not.
2935 * This is needed because nfsd might try to access dead inodes
2936 * the test is that same one that e2fsck uses
2937 * NeilBrown 1999oct15
2939 if (inode->i_nlink == 0) {
2940 if (inode->i_mode == 0 ||
2941 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2942 /* this inode is deleted */
2947 /* The only unlinked inodes we let through here have
2948 * valid i_mode and are being read by the orphan
2949 * recovery code: that's fine, we're about to complete
2950 * the process of deleting those. */
2952 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2953 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2954 #ifdef EXT3_FRAGMENTS
2955 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2956 ei->i_frag_no = raw_inode->i_frag;
2957 ei->i_frag_size = raw_inode->i_fsize;
2959 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2960 if (!S_ISREG(inode->i_mode)) {
2961 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2964 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2966 ei->i_disksize = inode->i_size;
2967 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2968 ei->i_block_group = iloc.block_group;
2970 * NOTE! The in-memory inode i_data array is in little-endian order
2971 * even on big-endian machines: we do NOT byteswap the block numbers!
2973 for (block = 0; block < EXT3_N_BLOCKS; block++)
2974 ei->i_data[block] = raw_inode->i_block[block];
2975 INIT_LIST_HEAD(&ei->i_orphan);
2978 * Set transaction id's of transactions that have to be committed
2979 * to finish f[data]sync. We set them to currently running transaction
2980 * as we cannot be sure that the inode or some of its metadata isn't
2981 * part of the transaction - the inode could have been reclaimed and
2982 * now it is reread from disk.
2987 spin_lock(&journal->j_state_lock);
2988 if (journal->j_running_transaction)
2989 transaction = journal->j_running_transaction;
2991 transaction = journal->j_committing_transaction;
2993 tid = transaction->t_tid;
2995 tid = journal->j_commit_sequence;
2996 spin_unlock(&journal->j_state_lock);
2997 atomic_set(&ei->i_sync_tid, tid);
2998 atomic_set(&ei->i_datasync_tid, tid);
3001 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
3002 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
3004 * When mke2fs creates big inodes it does not zero out
3005 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
3006 * so ignore those first few inodes.
3008 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3009 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3010 EXT3_INODE_SIZE(inode->i_sb)) {
3015 if (ei->i_extra_isize == 0) {
3016 /* The extra space is currently unused. Use it. */
3017 ei->i_extra_isize = sizeof(struct ext3_inode) -
3018 EXT3_GOOD_OLD_INODE_SIZE;
3020 __le32 *magic = (void *)raw_inode +
3021 EXT3_GOOD_OLD_INODE_SIZE +
3023 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
3024 ext3_set_inode_state(inode, EXT3_STATE_XATTR);
3027 ei->i_extra_isize = 0;
3029 if (S_ISREG(inode->i_mode)) {
3030 inode->i_op = &ext3_file_inode_operations;
3031 inode->i_fop = &ext3_file_operations;
3032 ext3_set_aops(inode);
3033 } else if (S_ISDIR(inode->i_mode)) {
3034 inode->i_op = &ext3_dir_inode_operations;
3035 inode->i_fop = &ext3_dir_operations;
3036 } else if (S_ISLNK(inode->i_mode)) {
3037 if (ext3_inode_is_fast_symlink(inode)) {
3038 inode->i_op = &ext3_fast_symlink_inode_operations;
3039 nd_terminate_link(ei->i_data, inode->i_size,
3040 sizeof(ei->i_data) - 1);
3042 inode->i_op = &ext3_symlink_inode_operations;
3043 ext3_set_aops(inode);
3046 inode->i_op = &ext3_special_inode_operations;
3047 if (raw_inode->i_block[0])
3048 init_special_inode(inode, inode->i_mode,
3049 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3051 init_special_inode(inode, inode->i_mode,
3052 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3055 ext3_set_inode_flags(inode);
3056 unlock_new_inode(inode);
3061 return ERR_PTR(ret);
3065 * Post the struct inode info into an on-disk inode location in the
3066 * buffer-cache. This gobbles the caller's reference to the
3067 * buffer_head in the inode location struct.
3069 * The caller must have write access to iloc->bh.
3071 static int ext3_do_update_inode(handle_t *handle,
3072 struct inode *inode,
3073 struct ext3_iloc *iloc)
3075 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
3076 struct ext3_inode_info *ei = EXT3_I(inode);
3077 struct buffer_head *bh = iloc->bh;
3078 int err = 0, rc, block;
3079 int need_datasync = 0;
3085 /* we can't allow multiple procs in here at once, its a bit racey */
3088 /* For fields not not tracking in the in-memory inode,
3089 * initialise them to zero for new inodes. */
3090 if (ext3_test_inode_state(inode, EXT3_STATE_NEW))
3091 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
3093 ext3_get_inode_flags(ei);
3094 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3095 i_uid = i_uid_read(inode);
3096 i_gid = i_gid_read(inode);
3097 if(!(test_opt(inode->i_sb, NO_UID32))) {
3098 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
3099 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
3101 * Fix up interoperability with old kernels. Otherwise, old inodes get
3102 * re-used with the upper 16 bits of the uid/gid intact
3105 raw_inode->i_uid_high =
3106 cpu_to_le16(high_16_bits(i_uid));
3107 raw_inode->i_gid_high =
3108 cpu_to_le16(high_16_bits(i_gid));
3110 raw_inode->i_uid_high = 0;
3111 raw_inode->i_gid_high = 0;
3114 raw_inode->i_uid_low =
3115 cpu_to_le16(fs_high2lowuid(i_uid));
3116 raw_inode->i_gid_low =
3117 cpu_to_le16(fs_high2lowgid(i_gid));
3118 raw_inode->i_uid_high = 0;
3119 raw_inode->i_gid_high = 0;
3121 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3122 disksize = cpu_to_le32(ei->i_disksize);
3123 if (disksize != raw_inode->i_size) {
3125 raw_inode->i_size = disksize;
3127 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
3128 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
3129 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
3130 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
3131 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3132 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
3133 #ifdef EXT3_FRAGMENTS
3134 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
3135 raw_inode->i_frag = ei->i_frag_no;
3136 raw_inode->i_fsize = ei->i_frag_size;
3138 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
3139 if (!S_ISREG(inode->i_mode)) {
3140 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
3142 disksize = cpu_to_le32(ei->i_disksize >> 32);
3143 if (disksize != raw_inode->i_size_high) {
3144 raw_inode->i_size_high = disksize;
3147 if (ei->i_disksize > 0x7fffffffULL) {
3148 struct super_block *sb = inode->i_sb;
3149 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
3150 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
3151 EXT3_SB(sb)->s_es->s_rev_level ==
3152 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
3153 /* If this is the first large file
3154 * created, add a flag to the superblock.
3157 err = ext3_journal_get_write_access(handle,
3158 EXT3_SB(sb)->s_sbh);
3162 ext3_update_dynamic_rev(sb);
3163 EXT3_SET_RO_COMPAT_FEATURE(sb,
3164 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
3166 err = ext3_journal_dirty_metadata(handle,
3167 EXT3_SB(sb)->s_sbh);
3168 /* get our lock and start over */
3173 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3174 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3175 if (old_valid_dev(inode->i_rdev)) {
3176 raw_inode->i_block[0] =
3177 cpu_to_le32(old_encode_dev(inode->i_rdev));
3178 raw_inode->i_block[1] = 0;
3180 raw_inode->i_block[0] = 0;
3181 raw_inode->i_block[1] =
3182 cpu_to_le32(new_encode_dev(inode->i_rdev));
3183 raw_inode->i_block[2] = 0;
3185 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
3186 raw_inode->i_block[block] = ei->i_data[block];
3188 if (ei->i_extra_isize)
3189 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3191 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
3193 rc = ext3_journal_dirty_metadata(handle, bh);
3196 ext3_clear_inode_state(inode, EXT3_STATE_NEW);
3198 atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
3200 atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
3203 ext3_std_error(inode->i_sb, err);
3208 * ext3_write_inode()
3210 * We are called from a few places:
3212 * - Within generic_file_write() for O_SYNC files.
3213 * Here, there will be no transaction running. We wait for any running
3214 * transaction to commit.
3216 * - Within sys_sync(), kupdate and such.
3217 * We wait on commit, if tol to.
3219 * - Within prune_icache() (PF_MEMALLOC == true)
3220 * Here we simply return. We can't afford to block kswapd on the
3223 * In all cases it is actually safe for us to return without doing anything,
3224 * because the inode has been copied into a raw inode buffer in
3225 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3228 * Note that we are absolutely dependent upon all inode dirtiers doing the
3229 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3230 * which we are interested.
3232 * It would be a bug for them to not do this. The code:
3234 * mark_inode_dirty(inode)
3236 * inode->i_size = expr;
3238 * is in error because a kswapd-driven write_inode() could occur while
3239 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3240 * will no longer be on the superblock's dirty inode list.
3242 int ext3_write_inode(struct inode *inode, struct writeback_control *wbc)
3244 if (current->flags & PF_MEMALLOC)
3247 if (ext3_journal_current_handle()) {
3248 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3253 if (wbc->sync_mode != WB_SYNC_ALL)
3256 return ext3_force_commit(inode->i_sb);
3262 * Called from notify_change.
3264 * We want to trap VFS attempts to truncate the file as soon as
3265 * possible. In particular, we want to make sure that when the VFS
3266 * shrinks i_size, we put the inode on the orphan list and modify
3267 * i_disksize immediately, so that during the subsequent flushing of
3268 * dirty pages and freeing of disk blocks, we can guarantee that any
3269 * commit will leave the blocks being flushed in an unused state on
3270 * disk. (On recovery, the inode will get truncated and the blocks will
3271 * be freed, so we have a strong guarantee that no future commit will
3272 * leave these blocks visible to the user.)
3274 * Called with inode->sem down.
3276 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
3278 struct inode *inode = dentry->d_inode;
3280 const unsigned int ia_valid = attr->ia_valid;
3282 error = inode_change_ok(inode, attr);
3286 if (is_quota_modification(inode, attr))
3287 dquot_initialize(inode);
3288 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3289 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3292 /* (user+group)*(old+new) structure, inode write (sb,
3293 * inode block, ? - but truncate inode update has it) */
3294 handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
3295 EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3);
3296 if (IS_ERR(handle)) {
3297 error = PTR_ERR(handle);
3300 error = dquot_transfer(inode, attr);
3302 ext3_journal_stop(handle);
3305 /* Update corresponding info in inode so that everything is in
3306 * one transaction */
3307 if (attr->ia_valid & ATTR_UID)
3308 inode->i_uid = attr->ia_uid;
3309 if (attr->ia_valid & ATTR_GID)
3310 inode->i_gid = attr->ia_gid;
3311 error = ext3_mark_inode_dirty(handle, inode);
3312 ext3_journal_stop(handle);
3315 if (attr->ia_valid & ATTR_SIZE)
3316 inode_dio_wait(inode);
3318 if (S_ISREG(inode->i_mode) &&
3319 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3322 handle = ext3_journal_start(inode, 3);
3323 if (IS_ERR(handle)) {
3324 error = PTR_ERR(handle);
3328 error = ext3_orphan_add(handle, inode);
3330 ext3_journal_stop(handle);
3333 EXT3_I(inode)->i_disksize = attr->ia_size;
3334 error = ext3_mark_inode_dirty(handle, inode);
3335 ext3_journal_stop(handle);
3337 /* Some hard fs error must have happened. Bail out. */
3338 ext3_orphan_del(NULL, inode);
3341 rc = ext3_block_truncate_page(inode, attr->ia_size);
3343 /* Cleanup orphan list and exit */
3344 handle = ext3_journal_start(inode, 3);
3345 if (IS_ERR(handle)) {
3346 ext3_orphan_del(NULL, inode);
3349 ext3_orphan_del(handle, inode);
3350 ext3_journal_stop(handle);
3355 if ((attr->ia_valid & ATTR_SIZE) &&
3356 attr->ia_size != i_size_read(inode)) {
3357 truncate_setsize(inode, attr->ia_size);
3358 ext3_truncate(inode);
3361 setattr_copy(inode, attr);
3362 mark_inode_dirty(inode);
3364 if (ia_valid & ATTR_MODE)
3365 rc = ext3_acl_chmod(inode);
3368 ext3_std_error(inode->i_sb, error);
3376 * How many blocks doth make a writepage()?
3378 * With N blocks per page, it may be:
3383 * N+5 bitmap blocks (from the above)
3384 * N+5 group descriptor summary blocks
3387 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3389 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3391 * With ordered or writeback data it's the same, less the N data blocks.
3393 * If the inode's direct blocks can hold an integral number of pages then a
3394 * page cannot straddle two indirect blocks, and we can only touch one indirect
3395 * and dindirect block, and the "5" above becomes "3".
3397 * This still overestimates under most circumstances. If we were to pass the
3398 * start and end offsets in here as well we could do block_to_path() on each
3399 * block and work out the exact number of indirects which are touched. Pah.
3402 static int ext3_writepage_trans_blocks(struct inode *inode)
3404 int bpp = ext3_journal_blocks_per_page(inode);
3405 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3408 if (ext3_should_journal_data(inode))
3409 ret = 3 * (bpp + indirects) + 2;
3411 ret = 2 * (bpp + indirects) + indirects + 2;
3414 /* We know that structure was already allocated during dquot_initialize so
3415 * we will be updating only the data blocks + inodes */
3416 ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
3423 * The caller must have previously called ext3_reserve_inode_write().
3424 * Give this, we know that the caller already has write access to iloc->bh.
3426 int ext3_mark_iloc_dirty(handle_t *handle,
3427 struct inode *inode, struct ext3_iloc *iloc)
3431 /* the do_update_inode consumes one bh->b_count */
3434 /* ext3_do_update_inode() does journal_dirty_metadata */
3435 err = ext3_do_update_inode(handle, inode, iloc);
3441 * On success, We end up with an outstanding reference count against
3442 * iloc->bh. This _must_ be cleaned up later.
3446 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3447 struct ext3_iloc *iloc)
3451 err = ext3_get_inode_loc(inode, iloc);
3453 BUFFER_TRACE(iloc->bh, "get_write_access");
3454 err = ext3_journal_get_write_access(handle, iloc->bh);
3461 ext3_std_error(inode->i_sb, err);
3466 * What we do here is to mark the in-core inode as clean with respect to inode
3467 * dirtiness (it may still be data-dirty).
3468 * This means that the in-core inode may be reaped by prune_icache
3469 * without having to perform any I/O. This is a very good thing,
3470 * because *any* task may call prune_icache - even ones which
3471 * have a transaction open against a different journal.
3473 * Is this cheating? Not really. Sure, we haven't written the
3474 * inode out, but prune_icache isn't a user-visible syncing function.
3475 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3476 * we start and wait on commits.
3478 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3480 struct ext3_iloc iloc;
3484 trace_ext3_mark_inode_dirty(inode, _RET_IP_);
3485 err = ext3_reserve_inode_write(handle, inode, &iloc);
3487 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3492 * ext3_dirty_inode() is called from __mark_inode_dirty()
3494 * We're really interested in the case where a file is being extended.
3495 * i_size has been changed by generic_commit_write() and we thus need
3496 * to include the updated inode in the current transaction.
3498 * Also, dquot_alloc_space() will always dirty the inode when blocks
3499 * are allocated to the file.
3501 * If the inode is marked synchronous, we don't honour that here - doing
3502 * so would cause a commit on atime updates, which we don't bother doing.
3503 * We handle synchronous inodes at the highest possible level.
3505 void ext3_dirty_inode(struct inode *inode, int flags)
3507 handle_t *current_handle = ext3_journal_current_handle();
3510 handle = ext3_journal_start(inode, 2);
3513 if (current_handle &&
3514 current_handle->h_transaction != handle->h_transaction) {
3515 /* This task has a transaction open against a different fs */
3516 printk(KERN_EMERG "%s: transactions do not match!\n",
3519 jbd_debug(5, "marking dirty. outer handle=%p\n",
3521 ext3_mark_inode_dirty(handle, inode);
3523 ext3_journal_stop(handle);
3530 * Bind an inode's backing buffer_head into this transaction, to prevent
3531 * it from being flushed to disk early. Unlike
3532 * ext3_reserve_inode_write, this leaves behind no bh reference and
3533 * returns no iloc structure, so the caller needs to repeat the iloc
3534 * lookup to mark the inode dirty later.
3536 static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3538 struct ext3_iloc iloc;
3542 err = ext3_get_inode_loc(inode, &iloc);
3544 BUFFER_TRACE(iloc.bh, "get_write_access");
3545 err = journal_get_write_access(handle, iloc.bh);
3547 err = ext3_journal_dirty_metadata(handle,
3552 ext3_std_error(inode->i_sb, err);
3557 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3564 * We have to be very careful here: changing a data block's
3565 * journaling status dynamically is dangerous. If we write a
3566 * data block to the journal, change the status and then delete
3567 * that block, we risk forgetting to revoke the old log record
3568 * from the journal and so a subsequent replay can corrupt data.
3569 * So, first we make sure that the journal is empty and that
3570 * nobody is changing anything.
3573 journal = EXT3_JOURNAL(inode);
3574 if (is_journal_aborted(journal))
3577 journal_lock_updates(journal);
3578 journal_flush(journal);
3581 * OK, there are no updates running now, and all cached data is
3582 * synced to disk. We are now in a completely consistent state
3583 * which doesn't have anything in the journal, and we know that
3584 * no filesystem updates are running, so it is safe to modify
3585 * the inode's in-core data-journaling state flag now.
3589 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3591 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3592 ext3_set_aops(inode);
3594 journal_unlock_updates(journal);
3596 /* Finally we can mark the inode as dirty. */
3598 handle = ext3_journal_start(inode, 1);
3600 return PTR_ERR(handle);
3602 err = ext3_mark_inode_dirty(handle, inode);
3604 ext3_journal_stop(handle);
3605 ext3_std_error(inode->i_sb, err);