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/module.h>
27 #include <linux/time.h>
28 #include <linux/ext3_jbd.h>
29 #include <linux/jbd.h>
30 #include <linux/smp_lock.h>
31 #include <linux/highuid.h>
32 #include <linux/pagemap.h>
33 #include <linux/quotaops.h>
34 #include <linux/string.h>
35 #include <linux/buffer_head.h>
36 #include <linux/writeback.h>
37 #include <linux/mpage.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/vs_tag.h>
44 static int ext3_writepage_trans_blocks(struct inode *inode);
47 * Test whether an inode is a fast symlink.
49 static int ext3_inode_is_fast_symlink(struct inode *inode)
51 int ea_blocks = EXT3_I(inode)->i_file_acl ?
52 (inode->i_sb->s_blocksize >> 9) : 0;
54 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
58 * The ext3 forget function must perform a revoke if we are freeing data
59 * which has been journaled. Metadata (eg. indirect blocks) must be
60 * revoked in all cases.
62 * "bh" may be NULL: a metadata block may have been freed from memory
63 * but there may still be a record of it in the journal, and that record
64 * still needs to be revoked.
66 int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode,
67 struct buffer_head *bh, ext3_fsblk_t blocknr)
73 BUFFER_TRACE(bh, "enter");
75 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
77 bh, is_metadata, inode->i_mode,
78 test_opt(inode->i_sb, DATA_FLAGS));
80 /* Never use the revoke function if we are doing full data
81 * journaling: there is no need to, and a V1 superblock won't
82 * support it. Otherwise, only skip the revoke on un-journaled
85 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
86 (!is_metadata && !ext3_should_journal_data(inode))) {
88 BUFFER_TRACE(bh, "call journal_forget");
89 return ext3_journal_forget(handle, bh);
95 * data!=journal && (is_metadata || should_journal_data(inode))
97 BUFFER_TRACE(bh, "call ext3_journal_revoke");
98 err = ext3_journal_revoke(handle, blocknr, bh);
100 ext3_abort(inode->i_sb, __FUNCTION__,
101 "error %d when attempting revoke", err);
102 BUFFER_TRACE(bh, "exit");
107 * Work out how many blocks we need to proceed with the next chunk of a
108 * truncate transaction.
110 static unsigned long blocks_for_truncate(struct inode *inode)
112 unsigned long needed;
114 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
116 /* Give ourselves just enough room to cope with inodes in which
117 * i_blocks is corrupt: we've seen disk corruptions in the past
118 * which resulted in random data in an inode which looked enough
119 * like a regular file for ext3 to try to delete it. Things
120 * will go a bit crazy if that happens, but at least we should
121 * try not to panic the whole kernel. */
125 /* But we need to bound the transaction so we don't overflow the
127 if (needed > EXT3_MAX_TRANS_DATA)
128 needed = EXT3_MAX_TRANS_DATA;
130 return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
141 * transaction in the top-level truncate loop. --sct
143 static handle_t *start_transaction(struct inode *inode)
147 result = ext3_journal_start(inode, blocks_for_truncate(inode));
151 ext3_std_error(inode->i_sb, PTR_ERR(result));
156 * Try to extend this transaction for the purposes of truncation.
158 * Returns 0 if we managed to create more room. If we can't create more
159 * room, and the transaction must be restarted we return 1.
161 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
163 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
165 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
171 * Restart the transaction associated with *handle. This does a commit,
172 * so before we call here everything must be consistently dirtied against
175 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
177 jbd_debug(2, "restarting handle %p\n", handle);
178 return ext3_journal_restart(handle, blocks_for_truncate(inode));
182 * Called at the last iput() if i_nlink is zero.
184 void ext3_delete_inode (struct inode * inode)
188 truncate_inode_pages(&inode->i_data, 0);
190 if (is_bad_inode(inode))
193 handle = start_transaction(inode);
194 if (IS_ERR(handle)) {
196 * If we're going to skip the normal cleanup, we still need to
197 * make sure that the in-core orphan linked list is properly
200 ext3_orphan_del(NULL, inode);
208 ext3_truncate(inode);
210 * Kill off the orphan record which ext3_truncate created.
211 * AKPM: I think this can be inside the above `if'.
212 * Note that ext3_orphan_del() has to be able to cope with the
213 * deletion of a non-existent orphan - this is because we don't
214 * know if ext3_truncate() actually created an orphan record.
215 * (Well, we could do this if we need to, but heck - it works)
217 ext3_orphan_del(handle, inode);
218 EXT3_I(inode)->i_dtime = get_seconds();
221 * One subtle ordering requirement: if anything has gone wrong
222 * (transaction abort, IO errors, whatever), then we can still
223 * do these next steps (the fs will already have been marked as
224 * having errors), but we can't free the inode if the mark_dirty
227 if (ext3_mark_inode_dirty(handle, inode))
228 /* If that failed, just do the required in-core inode clear. */
231 ext3_free_inode(handle, inode);
232 ext3_journal_stop(handle);
235 clear_inode(inode); /* We must guarantee clearing of inode... */
241 struct buffer_head *bh;
244 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
246 p->key = *(p->p = v);
250 static int verify_chain(Indirect *from, Indirect *to)
252 while (from <= to && from->key == *from->p)
258 * ext3_block_to_path - parse the block number into array of offsets
259 * @inode: inode in question (we are only interested in its superblock)
260 * @i_block: block number to be parsed
261 * @offsets: array to store the offsets in
262 * @boundary: set this non-zero if the referred-to block is likely to be
263 * followed (on disk) by an indirect block.
265 * To store the locations of file's data ext3 uses a data structure common
266 * for UNIX filesystems - tree of pointers anchored in the inode, with
267 * data blocks at leaves and indirect blocks in intermediate nodes.
268 * This function translates the block number into path in that tree -
269 * return value is the path length and @offsets[n] is the offset of
270 * pointer to (n+1)th node in the nth one. If @block is out of range
271 * (negative or too large) warning is printed and zero returned.
273 * Note: function doesn't find node addresses, so no IO is needed. All
274 * we need to know is the capacity of indirect blocks (taken from the
279 * Portability note: the last comparison (check that we fit into triple
280 * indirect block) is spelled differently, because otherwise on an
281 * architecture with 32-bit longs and 8Kb pages we might get into trouble
282 * if our filesystem had 8Kb blocks. We might use long long, but that would
283 * kill us on x86. Oh, well, at least the sign propagation does not matter -
284 * i_block would have to be negative in the very beginning, so we would not
288 static int ext3_block_to_path(struct inode *inode,
289 long i_block, int offsets[4], int *boundary)
291 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
292 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
293 const long direct_blocks = EXT3_NDIR_BLOCKS,
294 indirect_blocks = ptrs,
295 double_blocks = (1 << (ptrs_bits * 2));
300 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
301 } else if (i_block < direct_blocks) {
302 offsets[n++] = i_block;
303 final = direct_blocks;
304 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
305 offsets[n++] = EXT3_IND_BLOCK;
306 offsets[n++] = i_block;
308 } else if ((i_block -= indirect_blocks) < double_blocks) {
309 offsets[n++] = EXT3_DIND_BLOCK;
310 offsets[n++] = i_block >> ptrs_bits;
311 offsets[n++] = i_block & (ptrs - 1);
313 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
314 offsets[n++] = EXT3_TIND_BLOCK;
315 offsets[n++] = i_block >> (ptrs_bits * 2);
316 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
317 offsets[n++] = i_block & (ptrs - 1);
320 ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big");
323 *boundary = final - 1 - (i_block & (ptrs - 1));
328 * ext3_get_branch - read the chain of indirect blocks leading to data
329 * @inode: inode in question
330 * @depth: depth of the chain (1 - direct pointer, etc.)
331 * @offsets: offsets of pointers in inode/indirect blocks
332 * @chain: place to store the result
333 * @err: here we store the error value
335 * Function fills the array of triples <key, p, bh> and returns %NULL
336 * if everything went OK or the pointer to the last filled triple
337 * (incomplete one) otherwise. Upon the return chain[i].key contains
338 * the number of (i+1)-th block in the chain (as it is stored in memory,
339 * i.e. little-endian 32-bit), chain[i].p contains the address of that
340 * number (it points into struct inode for i==0 and into the bh->b_data
341 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
342 * block for i>0 and NULL for i==0. In other words, it holds the block
343 * numbers of the chain, addresses they were taken from (and where we can
344 * verify that chain did not change) and buffer_heads hosting these
347 * Function stops when it stumbles upon zero pointer (absent block)
348 * (pointer to last triple returned, *@err == 0)
349 * or when it gets an IO error reading an indirect block
350 * (ditto, *@err == -EIO)
351 * or when it notices that chain had been changed while it was reading
352 * (ditto, *@err == -EAGAIN)
353 * or when it reads all @depth-1 indirect blocks successfully and finds
354 * the whole chain, all way to the data (returns %NULL, *err == 0).
356 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
357 Indirect chain[4], int *err)
359 struct super_block *sb = inode->i_sb;
361 struct buffer_head *bh;
364 /* i_data is not going away, no lock needed */
365 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
369 bh = sb_bread(sb, le32_to_cpu(p->key));
372 /* Reader: pointers */
373 if (!verify_chain(chain, p))
375 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
393 * ext3_find_near - find a place for allocation with sufficient locality
395 * @ind: descriptor of indirect block.
397 * This function returns the prefered place for block allocation.
398 * It is used when heuristic for sequential allocation fails.
400 * + if there is a block to the left of our position - allocate near it.
401 * + if pointer will live in indirect block - allocate near that block.
402 * + if pointer will live in inode - allocate in the same
405 * In the latter case we colour the starting block by the callers PID to
406 * prevent it from clashing with concurrent allocations for a different inode
407 * in the same block group. The PID is used here so that functionally related
408 * files will be close-by on-disk.
410 * Caller must make sure that @ind is valid and will stay that way.
412 static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind)
414 struct ext3_inode_info *ei = EXT3_I(inode);
415 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
417 ext3_fsblk_t bg_start;
418 ext3_grpblk_t colour;
420 /* Try to find previous block */
421 for (p = ind->p - 1; p >= start; p--) {
423 return le32_to_cpu(*p);
426 /* No such thing, so let's try location of indirect block */
428 return ind->bh->b_blocknr;
431 * It is going to be referred to from the inode itself? OK, just put it
432 * into the same cylinder group then.
434 bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group);
435 colour = (current->pid % 16) *
436 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
437 return bg_start + colour;
441 * ext3_find_goal - find a prefered place for allocation.
443 * @block: block we want
444 * @chain: chain of indirect blocks
445 * @partial: pointer to the last triple within a chain
446 * @goal: place to store the result.
448 * Normally this function find the prefered place for block allocation,
449 * stores it in *@goal and returns zero.
452 static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block,
453 Indirect chain[4], Indirect *partial)
455 struct ext3_block_alloc_info *block_i;
457 block_i = EXT3_I(inode)->i_block_alloc_info;
460 * try the heuristic for sequential allocation,
461 * failing that at least try to get decent locality.
463 if (block_i && (block == block_i->last_alloc_logical_block + 1)
464 && (block_i->last_alloc_physical_block != 0)) {
465 return block_i->last_alloc_physical_block + 1;
468 return ext3_find_near(inode, partial);
472 * ext3_blks_to_allocate: Look up the block map and count the number
473 * of direct blocks need to be allocated for the given branch.
475 * @branch: chain of indirect blocks
476 * @k: number of blocks need for indirect blocks
477 * @blks: number of data blocks to be mapped.
478 * @blocks_to_boundary: the offset in the indirect block
480 * return the total number of blocks to be allocate, including the
481 * direct and indirect blocks.
483 static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
484 int blocks_to_boundary)
486 unsigned long count = 0;
489 * Simple case, [t,d]Indirect block(s) has not allocated yet
490 * then it's clear blocks on that path have not allocated
493 /* right now we don't handle cross boundary allocation */
494 if (blks < blocks_to_boundary + 1)
497 count += blocks_to_boundary + 1;
502 while (count < blks && count <= blocks_to_boundary &&
503 le32_to_cpu(*(branch[0].p + count)) == 0) {
510 * ext3_alloc_blocks: multiple allocate blocks needed for a branch
511 * @indirect_blks: the number of blocks need to allocate for indirect
514 * @new_blocks: on return it will store the new block numbers for
515 * the indirect blocks(if needed) and the first direct block,
516 * @blks: on return it will store the total number of allocated
519 static int ext3_alloc_blocks(handle_t *handle, struct inode *inode,
520 ext3_fsblk_t goal, int indirect_blks, int blks,
521 ext3_fsblk_t new_blocks[4], int *err)
524 unsigned long count = 0;
526 ext3_fsblk_t current_block = 0;
530 * Here we try to allocate the requested multiple blocks at once,
531 * on a best-effort basis.
532 * To build a branch, we should allocate blocks for
533 * the indirect blocks(if not allocated yet), and at least
534 * the first direct block of this branch. That's the
535 * minimum number of blocks need to allocate(required)
537 target = blks + indirect_blks;
541 /* allocating blocks for indirect blocks and direct blocks */
542 current_block = ext3_new_blocks(handle,inode,goal,&count,err);
547 /* allocate blocks for indirect blocks */
548 while (index < indirect_blks && count) {
549 new_blocks[index++] = current_block++;
557 /* save the new block number for the first direct block */
558 new_blocks[index] = current_block;
560 /* total number of blocks allocated for direct blocks */
565 for (i = 0; i <index; i++)
566 ext3_free_blocks(handle, inode, new_blocks[i], 1);
571 * ext3_alloc_branch - allocate and set up a chain of blocks.
573 * @indirect_blks: number of allocated indirect blocks
574 * @blks: number of allocated direct blocks
575 * @offsets: offsets (in the blocks) to store the pointers to next.
576 * @branch: place to store the chain in.
578 * This function allocates blocks, zeroes out all but the last one,
579 * links them into chain and (if we are synchronous) writes them to disk.
580 * In other words, it prepares a branch that can be spliced onto the
581 * inode. It stores the information about that chain in the branch[], in
582 * the same format as ext3_get_branch() would do. We are calling it after
583 * we had read the existing part of chain and partial points to the last
584 * triple of that (one with zero ->key). Upon the exit we have the same
585 * picture as after the successful ext3_get_block(), except that in one
586 * place chain is disconnected - *branch->p is still zero (we did not
587 * set the last link), but branch->key contains the number that should
588 * be placed into *branch->p to fill that gap.
590 * If allocation fails we free all blocks we've allocated (and forget
591 * their buffer_heads) and return the error value the from failed
592 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
593 * as described above and return 0.
595 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
596 int indirect_blks, int *blks, ext3_fsblk_t goal,
597 int *offsets, Indirect *branch)
599 int blocksize = inode->i_sb->s_blocksize;
602 struct buffer_head *bh;
604 ext3_fsblk_t new_blocks[4];
605 ext3_fsblk_t current_block;
607 num = ext3_alloc_blocks(handle, inode, goal, indirect_blks,
608 *blks, new_blocks, &err);
612 branch[0].key = cpu_to_le32(new_blocks[0]);
614 * metadata blocks and data blocks are allocated.
616 for (n = 1; n <= indirect_blks; n++) {
618 * Get buffer_head for parent block, zero it out
619 * and set the pointer to new one, then send
622 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
625 BUFFER_TRACE(bh, "call get_create_access");
626 err = ext3_journal_get_create_access(handle, bh);
633 memset(bh->b_data, 0, blocksize);
634 branch[n].p = (__le32 *) bh->b_data + offsets[n];
635 branch[n].key = cpu_to_le32(new_blocks[n]);
636 *branch[n].p = branch[n].key;
637 if ( n == indirect_blks) {
638 current_block = new_blocks[n];
640 * End of chain, update the last new metablock of
641 * the chain to point to the new allocated
642 * data blocks numbers
644 for (i=1; i < num; i++)
645 *(branch[n].p + i) = cpu_to_le32(++current_block);
647 BUFFER_TRACE(bh, "marking uptodate");
648 set_buffer_uptodate(bh);
651 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
652 err = ext3_journal_dirty_metadata(handle, bh);
659 /* Allocation failed, free what we already allocated */
660 for (i = 1; i <= n ; i++) {
661 BUFFER_TRACE(branch[i].bh, "call journal_forget");
662 ext3_journal_forget(handle, branch[i].bh);
664 for (i = 0; i <indirect_blks; i++)
665 ext3_free_blocks(handle, inode, new_blocks[i], 1);
667 ext3_free_blocks(handle, inode, new_blocks[i], num);
673 * ext3_splice_branch - splice the allocated branch onto inode.
675 * @block: (logical) number of block we are adding
676 * @chain: chain of indirect blocks (with a missing link - see
678 * @where: location of missing link
679 * @num: number of indirect blocks we are adding
680 * @blks: number of direct blocks we are adding
682 * This function fills the missing link and does all housekeeping needed in
683 * inode (->i_blocks, etc.). In case of success we end up with the full
684 * chain to new block and return 0.
686 static int ext3_splice_branch(handle_t *handle, struct inode *inode,
687 long block, Indirect *where, int num, int blks)
691 struct ext3_block_alloc_info *block_i;
692 ext3_fsblk_t current_block;
694 block_i = EXT3_I(inode)->i_block_alloc_info;
696 * If we're splicing into a [td]indirect block (as opposed to the
697 * inode) then we need to get write access to the [td]indirect block
701 BUFFER_TRACE(where->bh, "get_write_access");
702 err = ext3_journal_get_write_access(handle, where->bh);
708 *where->p = where->key;
711 * Update the host buffer_head or inode to point to more just allocated
712 * direct blocks blocks
714 if (num == 0 && blks > 1) {
715 current_block = le32_to_cpu(where->key) + 1;
716 for (i = 1; i < blks; i++)
717 *(where->p + i ) = cpu_to_le32(current_block++);
721 * update the most recently allocated logical & physical block
722 * in i_block_alloc_info, to assist find the proper goal block for next
726 block_i->last_alloc_logical_block = block + blks - 1;
727 block_i->last_alloc_physical_block =
728 le32_to_cpu(where[num].key) + blks - 1;
731 /* We are done with atomic stuff, now do the rest of housekeeping */
733 inode->i_ctime = CURRENT_TIME_SEC;
734 ext3_mark_inode_dirty(handle, inode);
736 /* had we spliced it onto indirect block? */
739 * If we spliced it onto an indirect block, we haven't
740 * altered the inode. Note however that if it is being spliced
741 * onto an indirect block at the very end of the file (the
742 * file is growing) then we *will* alter the inode to reflect
743 * the new i_size. But that is not done here - it is done in
744 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
746 jbd_debug(5, "splicing indirect only\n");
747 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
748 err = ext3_journal_dirty_metadata(handle, where->bh);
753 * OK, we spliced it into the inode itself on a direct block.
754 * Inode was dirtied above.
756 jbd_debug(5, "splicing direct\n");
761 for (i = 1; i <= num; i++) {
762 BUFFER_TRACE(where[i].bh, "call journal_forget");
763 ext3_journal_forget(handle, where[i].bh);
764 ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
766 ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
772 * Allocation strategy is simple: if we have to allocate something, we will
773 * have to go the whole way to leaf. So let's do it before attaching anything
774 * to tree, set linkage between the newborn blocks, write them if sync is
775 * required, recheck the path, free and repeat if check fails, otherwise
776 * set the last missing link (that will protect us from any truncate-generated
777 * removals - all blocks on the path are immune now) and possibly force the
778 * write on the parent block.
779 * That has a nice additional property: no special recovery from the failed
780 * allocations is needed - we simply release blocks and do not touch anything
781 * reachable from inode.
783 * `handle' can be NULL if create == 0.
785 * The BKL may not be held on entry here. Be sure to take it early.
786 * return > 0, # of blocks mapped or allocated.
787 * return = 0, if plain lookup failed.
788 * return < 0, error case.
790 int ext3_get_blocks_handle(handle_t *handle, struct inode *inode,
791 sector_t iblock, unsigned long maxblocks,
792 struct buffer_head *bh_result,
793 int create, int extend_disksize)
801 int blocks_to_boundary = 0;
803 struct ext3_inode_info *ei = EXT3_I(inode);
805 ext3_fsblk_t first_block = 0;
808 J_ASSERT(handle != NULL || create == 0);
809 depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
814 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
816 /* Simplest case - block found, no allocation needed */
818 first_block = le32_to_cpu(chain[depth - 1].key);
819 clear_buffer_new(bh_result);
822 while (count < maxblocks && count <= blocks_to_boundary) {
825 if (!verify_chain(chain, partial)) {
827 * Indirect block might be removed by
828 * truncate while we were reading it.
829 * Handling of that case: forget what we've
830 * got now. Flag the err as EAGAIN, so it
837 blk = le32_to_cpu(*(chain[depth-1].p + count));
839 if (blk == first_block + count)
848 /* Next simple case - plain lookup or failed read of indirect block */
849 if (!create || err == -EIO)
852 mutex_lock(&ei->truncate_mutex);
855 * If the indirect block is missing while we are reading
856 * the chain(ext3_get_branch() returns -EAGAIN err), or
857 * if the chain has been changed after we grab the semaphore,
858 * (either because another process truncated this branch, or
859 * another get_block allocated this branch) re-grab the chain to see if
860 * the request block has been allocated or not.
862 * Since we already block the truncate/other get_block
863 * at this point, we will have the current copy of the chain when we
864 * splice the branch into the tree.
866 if (err == -EAGAIN || !verify_chain(chain, partial)) {
867 while (partial > chain) {
871 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
874 mutex_unlock(&ei->truncate_mutex);
877 clear_buffer_new(bh_result);
883 * Okay, we need to do block allocation. Lazily initialize the block
884 * allocation info here if necessary
886 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
887 ext3_init_block_alloc_info(inode);
889 goal = ext3_find_goal(inode, iblock, chain, partial);
891 /* the number of blocks need to allocate for [d,t]indirect blocks */
892 indirect_blks = (chain + depth) - partial - 1;
895 * Next look up the indirect map to count the totoal number of
896 * direct blocks to allocate for this branch.
898 count = ext3_blks_to_allocate(partial, indirect_blks,
899 maxblocks, blocks_to_boundary);
901 * Block out ext3_truncate while we alter the tree
903 err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal,
904 offsets + (partial - chain), partial);
907 * The ext3_splice_branch call will free and forget any buffers
908 * on the new chain if there is a failure, but that risks using
909 * up transaction credits, especially for bitmaps where the
910 * credits cannot be returned. Can we handle this somehow? We
911 * may need to return -EAGAIN upwards in the worst case. --sct
914 err = ext3_splice_branch(handle, inode, iblock,
915 partial, indirect_blks, count);
917 * i_disksize growing is protected by truncate_mutex. Don't forget to
918 * protect it if you're about to implement concurrent
919 * ext3_get_block() -bzzz
921 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
922 ei->i_disksize = inode->i_size;
923 mutex_unlock(&ei->truncate_mutex);
927 set_buffer_new(bh_result);
929 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
930 if (count > blocks_to_boundary)
931 set_buffer_boundary(bh_result);
933 /* Clean up and exit */
934 partial = chain + depth - 1; /* the whole chain */
936 while (partial > chain) {
937 BUFFER_TRACE(partial->bh, "call brelse");
941 BUFFER_TRACE(bh_result, "returned");
946 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
948 static int ext3_get_block(struct inode *inode, sector_t iblock,
949 struct buffer_head *bh_result, int create)
951 handle_t *handle = journal_current_handle();
953 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
956 goto get_block; /* A read */
959 goto get_block; /* A single block get */
961 if (handle->h_transaction->t_state == T_LOCKED) {
963 * Huge direct-io writes can hold off commits for long
964 * periods of time. Let this commit run.
966 ext3_journal_stop(handle);
967 handle = ext3_journal_start(inode, DIO_CREDITS);
969 ret = PTR_ERR(handle);
973 if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
975 * Getting low on buffer credits...
977 ret = ext3_journal_extend(handle, DIO_CREDITS);
980 * Couldn't extend the transaction. Start a new one.
982 ret = ext3_journal_restart(handle, DIO_CREDITS);
988 ret = ext3_get_blocks_handle(handle, inode, iblock,
989 max_blocks, bh_result, create, 0);
991 bh_result->b_size = (ret << inode->i_blkbits);
999 * `handle' can be NULL if create is zero
1001 struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode,
1002 long block, int create, int *errp)
1004 struct buffer_head dummy;
1007 J_ASSERT(handle != NULL || create == 0);
1010 dummy.b_blocknr = -1000;
1011 buffer_trace_init(&dummy.b_history);
1012 err = ext3_get_blocks_handle(handle, inode, block, 1,
1015 * ext3_get_blocks_handle() returns number of blocks
1016 * mapped. 0 in case of a HOLE.
1024 if (!err && buffer_mapped(&dummy)) {
1025 struct buffer_head *bh;
1026 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1031 if (buffer_new(&dummy)) {
1032 J_ASSERT(create != 0);
1033 J_ASSERT(handle != 0);
1036 * Now that we do not always journal data, we should
1037 * keep in mind whether this should always journal the
1038 * new buffer as metadata. For now, regular file
1039 * writes use ext3_get_block instead, so it's not a
1043 BUFFER_TRACE(bh, "call get_create_access");
1044 fatal = ext3_journal_get_create_access(handle, bh);
1045 if (!fatal && !buffer_uptodate(bh)) {
1046 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1047 set_buffer_uptodate(bh);
1050 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1051 err = ext3_journal_dirty_metadata(handle, bh);
1055 BUFFER_TRACE(bh, "not a new buffer");
1068 struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode,
1069 int block, int create, int *err)
1071 struct buffer_head * bh;
1073 bh = ext3_getblk(handle, inode, block, create, err);
1076 if (buffer_uptodate(bh))
1078 ll_rw_block(READ_META, 1, &bh);
1080 if (buffer_uptodate(bh))
1087 static int walk_page_buffers( handle_t *handle,
1088 struct buffer_head *head,
1092 int (*fn)( handle_t *handle,
1093 struct buffer_head *bh))
1095 struct buffer_head *bh;
1096 unsigned block_start, block_end;
1097 unsigned blocksize = head->b_size;
1099 struct buffer_head *next;
1101 for ( bh = head, block_start = 0;
1102 ret == 0 && (bh != head || !block_start);
1103 block_start = block_end, bh = next)
1105 next = bh->b_this_page;
1106 block_end = block_start + blocksize;
1107 if (block_end <= from || block_start >= to) {
1108 if (partial && !buffer_uptodate(bh))
1112 err = (*fn)(handle, bh);
1120 * To preserve ordering, it is essential that the hole instantiation and
1121 * the data write be encapsulated in a single transaction. We cannot
1122 * close off a transaction and start a new one between the ext3_get_block()
1123 * and the commit_write(). So doing the journal_start at the start of
1124 * prepare_write() is the right place.
1126 * Also, this function can nest inside ext3_writepage() ->
1127 * block_write_full_page(). In that case, we *know* that ext3_writepage()
1128 * has generated enough buffer credits to do the whole page. So we won't
1129 * block on the journal in that case, which is good, because the caller may
1132 * By accident, ext3 can be reentered when a transaction is open via
1133 * quota file writes. If we were to commit the transaction while thus
1134 * reentered, there can be a deadlock - we would be holding a quota
1135 * lock, and the commit would never complete if another thread had a
1136 * transaction open and was blocking on the quota lock - a ranking
1139 * So what we do is to rely on the fact that journal_stop/journal_start
1140 * will _not_ run commit under these circumstances because handle->h_ref
1141 * is elevated. We'll still have enough credits for the tiny quotafile
1144 static int do_journal_get_write_access(handle_t *handle,
1145 struct buffer_head *bh)
1147 if (!buffer_mapped(bh) || buffer_freed(bh))
1149 return ext3_journal_get_write_access(handle, bh);
1152 static int ext3_prepare_write(struct file *file, struct page *page,
1153 unsigned from, unsigned to)
1155 struct inode *inode = page->mapping->host;
1156 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1161 handle = ext3_journal_start(inode, needed_blocks);
1162 if (IS_ERR(handle)) {
1163 ret = PTR_ERR(handle);
1166 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
1167 ret = nobh_prepare_write(page, from, to, ext3_get_block);
1169 ret = block_prepare_write(page, from, to, ext3_get_block);
1171 goto prepare_write_failed;
1173 if (ext3_should_journal_data(inode)) {
1174 ret = walk_page_buffers(handle, page_buffers(page),
1175 from, to, NULL, do_journal_get_write_access);
1177 prepare_write_failed:
1179 ext3_journal_stop(handle);
1180 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1186 int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1188 int err = journal_dirty_data(handle, bh);
1190 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1195 /* For commit_write() in data=journal mode */
1196 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1198 if (!buffer_mapped(bh) || buffer_freed(bh))
1200 set_buffer_uptodate(bh);
1201 return ext3_journal_dirty_metadata(handle, bh);
1205 * We need to pick up the new inode size which generic_commit_write gave us
1206 * `file' can be NULL - eg, when called from page_symlink().
1208 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1209 * buffers are managed internally.
1211 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1212 unsigned from, unsigned to)
1214 handle_t *handle = ext3_journal_current_handle();
1215 struct inode *inode = page->mapping->host;
1218 ret = walk_page_buffers(handle, page_buffers(page),
1219 from, to, NULL, ext3_journal_dirty_data);
1223 * generic_commit_write() will run mark_inode_dirty() if i_size
1224 * changes. So let's piggyback the i_disksize mark_inode_dirty
1229 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1230 if (new_i_size > EXT3_I(inode)->i_disksize)
1231 EXT3_I(inode)->i_disksize = new_i_size;
1232 ret = generic_commit_write(file, page, from, to);
1234 ret2 = ext3_journal_stop(handle);
1240 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1241 unsigned from, unsigned to)
1243 handle_t *handle = ext3_journal_current_handle();
1244 struct inode *inode = page->mapping->host;
1248 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1249 if (new_i_size > EXT3_I(inode)->i_disksize)
1250 EXT3_I(inode)->i_disksize = new_i_size;
1252 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
1253 ret = nobh_commit_write(file, page, from, to);
1255 ret = generic_commit_write(file, page, from, to);
1257 ret2 = ext3_journal_stop(handle);
1263 static int ext3_journalled_commit_write(struct file *file,
1264 struct page *page, unsigned from, unsigned to)
1266 handle_t *handle = ext3_journal_current_handle();
1267 struct inode *inode = page->mapping->host;
1273 * Here we duplicate the generic_commit_write() functionality
1275 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1277 ret = walk_page_buffers(handle, page_buffers(page), from,
1278 to, &partial, commit_write_fn);
1280 SetPageUptodate(page);
1281 if (pos > inode->i_size)
1282 i_size_write(inode, pos);
1283 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1284 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1285 EXT3_I(inode)->i_disksize = inode->i_size;
1286 ret2 = ext3_mark_inode_dirty(handle, inode);
1290 ret2 = ext3_journal_stop(handle);
1297 * bmap() is special. It gets used by applications such as lilo and by
1298 * the swapper to find the on-disk block of a specific piece of data.
1300 * Naturally, this is dangerous if the block concerned is still in the
1301 * journal. If somebody makes a swapfile on an ext3 data-journaling
1302 * filesystem and enables swap, then they may get a nasty shock when the
1303 * data getting swapped to that swapfile suddenly gets overwritten by
1304 * the original zero's written out previously to the journal and
1305 * awaiting writeback in the kernel's buffer cache.
1307 * So, if we see any bmap calls here on a modified, data-journaled file,
1308 * take extra steps to flush any blocks which might be in the cache.
1310 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1312 struct inode *inode = mapping->host;
1316 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1318 * This is a REALLY heavyweight approach, but the use of
1319 * bmap on dirty files is expected to be extremely rare:
1320 * only if we run lilo or swapon on a freshly made file
1321 * do we expect this to happen.
1323 * (bmap requires CAP_SYS_RAWIO so this does not
1324 * represent an unprivileged user DOS attack --- we'd be
1325 * in trouble if mortal users could trigger this path at
1328 * NB. EXT3_STATE_JDATA is not set on files other than
1329 * regular files. If somebody wants to bmap a directory
1330 * or symlink and gets confused because the buffer
1331 * hasn't yet been flushed to disk, they deserve
1332 * everything they get.
1335 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1336 journal = EXT3_JOURNAL(inode);
1337 journal_lock_updates(journal);
1338 err = journal_flush(journal);
1339 journal_unlock_updates(journal);
1345 return generic_block_bmap(mapping,block,ext3_get_block);
1348 static int bget_one(handle_t *handle, struct buffer_head *bh)
1354 static int bput_one(handle_t *handle, struct buffer_head *bh)
1360 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1362 if (buffer_mapped(bh))
1363 return ext3_journal_dirty_data(handle, bh);
1368 * Note that we always start a transaction even if we're not journalling
1369 * data. This is to preserve ordering: any hole instantiation within
1370 * __block_write_full_page -> ext3_get_block() should be journalled
1371 * along with the data so we don't crash and then get metadata which
1372 * refers to old data.
1374 * In all journalling modes block_write_full_page() will start the I/O.
1378 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1383 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1385 * Same applies to ext3_get_block(). We will deadlock on various things like
1386 * lock_journal and i_truncate_mutex.
1388 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1391 * 16May01: If we're reentered then journal_current_handle() will be
1392 * non-zero. We simply *return*.
1394 * 1 July 2001: @@@ FIXME:
1395 * In journalled data mode, a data buffer may be metadata against the
1396 * current transaction. But the same file is part of a shared mapping
1397 * and someone does a writepage() on it.
1399 * We will move the buffer onto the async_data list, but *after* it has
1400 * been dirtied. So there's a small window where we have dirty data on
1403 * Note that this only applies to the last partial page in the file. The
1404 * bit which block_write_full_page() uses prepare/commit for. (That's
1405 * broken code anyway: it's wrong for msync()).
1407 * It's a rare case: affects the final partial page, for journalled data
1408 * where the file is subject to bith write() and writepage() in the same
1409 * transction. To fix it we'll need a custom block_write_full_page().
1410 * We'll probably need that anyway for journalling writepage() output.
1412 * We don't honour synchronous mounts for writepage(). That would be
1413 * disastrous. Any write() or metadata operation will sync the fs for
1416 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1417 * we don't need to open a transaction here.
1419 static int ext3_ordered_writepage(struct page *page,
1420 struct writeback_control *wbc)
1422 struct inode *inode = page->mapping->host;
1423 struct buffer_head *page_bufs;
1424 handle_t *handle = NULL;
1428 J_ASSERT(PageLocked(page));
1431 * We give up here if we're reentered, because it might be for a
1432 * different filesystem.
1434 if (ext3_journal_current_handle())
1437 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1439 if (IS_ERR(handle)) {
1440 ret = PTR_ERR(handle);
1444 if (!page_has_buffers(page)) {
1445 create_empty_buffers(page, inode->i_sb->s_blocksize,
1446 (1 << BH_Dirty)|(1 << BH_Uptodate));
1448 page_bufs = page_buffers(page);
1449 walk_page_buffers(handle, page_bufs, 0,
1450 PAGE_CACHE_SIZE, NULL, bget_one);
1452 ret = block_write_full_page(page, ext3_get_block, wbc);
1455 * The page can become unlocked at any point now, and
1456 * truncate can then come in and change things. So we
1457 * can't touch *page from now on. But *page_bufs is
1458 * safe due to elevated refcount.
1462 * And attach them to the current transaction. But only if
1463 * block_write_full_page() succeeded. Otherwise they are unmapped,
1464 * and generally junk.
1467 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1468 NULL, journal_dirty_data_fn);
1472 walk_page_buffers(handle, page_bufs, 0,
1473 PAGE_CACHE_SIZE, NULL, bput_one);
1474 err = ext3_journal_stop(handle);
1480 redirty_page_for_writepage(wbc, page);
1485 static int ext3_writeback_writepage(struct page *page,
1486 struct writeback_control *wbc)
1488 struct inode *inode = page->mapping->host;
1489 handle_t *handle = NULL;
1493 if (ext3_journal_current_handle())
1496 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1497 if (IS_ERR(handle)) {
1498 ret = PTR_ERR(handle);
1502 if (test_opt(inode->i_sb, NOBH) && ext3_should_writeback_data(inode))
1503 ret = nobh_writepage(page, ext3_get_block, wbc);
1505 ret = block_write_full_page(page, ext3_get_block, wbc);
1507 err = ext3_journal_stop(handle);
1513 redirty_page_for_writepage(wbc, page);
1518 static int ext3_journalled_writepage(struct page *page,
1519 struct writeback_control *wbc)
1521 struct inode *inode = page->mapping->host;
1522 handle_t *handle = NULL;
1526 if (ext3_journal_current_handle())
1529 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1530 if (IS_ERR(handle)) {
1531 ret = PTR_ERR(handle);
1535 if (!page_has_buffers(page) || PageChecked(page)) {
1537 * It's mmapped pagecache. Add buffers and journal it. There
1538 * doesn't seem much point in redirtying the page here.
1540 ClearPageChecked(page);
1541 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1544 ext3_journal_stop(handle);
1547 ret = walk_page_buffers(handle, page_buffers(page), 0,
1548 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1550 err = walk_page_buffers(handle, page_buffers(page), 0,
1551 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1554 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1558 * It may be a page full of checkpoint-mode buffers. We don't
1559 * really know unless we go poke around in the buffer_heads.
1560 * But block_write_full_page will do the right thing.
1562 ret = block_write_full_page(page, ext3_get_block, wbc);
1564 err = ext3_journal_stop(handle);
1571 redirty_page_for_writepage(wbc, page);
1577 static int ext3_readpage(struct file *file, struct page *page)
1579 return mpage_readpage(page, ext3_get_block);
1583 ext3_readpages(struct file *file, struct address_space *mapping,
1584 struct list_head *pages, unsigned nr_pages)
1586 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1589 static void ext3_invalidatepage(struct page *page, unsigned long offset)
1591 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1594 * If it's a full truncate we just forget about the pending dirtying
1597 ClearPageChecked(page);
1599 journal_invalidatepage(journal, page, offset);
1602 static int ext3_releasepage(struct page *page, gfp_t wait)
1604 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1606 WARN_ON(PageChecked(page));
1607 if (!page_has_buffers(page))
1609 return journal_try_to_free_buffers(journal, page, wait);
1613 * If the O_DIRECT write will extend the file then add this inode to the
1614 * orphan list. So recovery will truncate it back to the original size
1615 * if the machine crashes during the write.
1617 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1618 * crashes then stale disk data _may_ be exposed inside the file.
1620 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1621 const struct iovec *iov, loff_t offset,
1622 unsigned long nr_segs)
1624 struct file *file = iocb->ki_filp;
1625 struct inode *inode = file->f_mapping->host;
1626 struct ext3_inode_info *ei = EXT3_I(inode);
1627 handle_t *handle = NULL;
1630 size_t count = iov_length(iov, nr_segs);
1633 loff_t final_size = offset + count;
1635 handle = ext3_journal_start(inode, DIO_CREDITS);
1636 if (IS_ERR(handle)) {
1637 ret = PTR_ERR(handle);
1640 if (final_size > inode->i_size) {
1641 ret = ext3_orphan_add(handle, inode);
1645 ei->i_disksize = inode->i_size;
1649 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1651 ext3_get_block, NULL);
1654 * Reacquire the handle: ext3_get_block() can restart the transaction
1656 handle = journal_current_handle();
1662 if (orphan && inode->i_nlink)
1663 ext3_orphan_del(handle, inode);
1664 if (orphan && ret > 0) {
1665 loff_t end = offset + ret;
1666 if (end > inode->i_size) {
1667 ei->i_disksize = end;
1668 i_size_write(inode, end);
1670 * We're going to return a positive `ret'
1671 * here due to non-zero-length I/O, so there's
1672 * no way of reporting error returns from
1673 * ext3_mark_inode_dirty() to userspace. So
1676 ext3_mark_inode_dirty(handle, inode);
1679 err = ext3_journal_stop(handle);
1688 * Pages can be marked dirty completely asynchronously from ext3's journalling
1689 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1690 * much here because ->set_page_dirty is called under VFS locks. The page is
1691 * not necessarily locked.
1693 * We cannot just dirty the page and leave attached buffers clean, because the
1694 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1695 * or jbddirty because all the journalling code will explode.
1697 * So what we do is to mark the page "pending dirty" and next time writepage
1698 * is called, propagate that into the buffers appropriately.
1700 static int ext3_journalled_set_page_dirty(struct page *page)
1702 SetPageChecked(page);
1703 return __set_page_dirty_nobuffers(page);
1706 static const struct address_space_operations ext3_ordered_aops = {
1707 .readpage = ext3_readpage,
1708 .readpages = ext3_readpages,
1709 .writepage = ext3_ordered_writepage,
1710 .sync_page = block_sync_page,
1711 .prepare_write = ext3_prepare_write,
1712 .commit_write = ext3_ordered_commit_write,
1714 .invalidatepage = ext3_invalidatepage,
1715 .releasepage = ext3_releasepage,
1716 .direct_IO = ext3_direct_IO,
1717 .migratepage = buffer_migrate_page,
1720 static const struct address_space_operations ext3_writeback_aops = {
1721 .readpage = ext3_readpage,
1722 .readpages = ext3_readpages,
1723 .writepage = ext3_writeback_writepage,
1724 .sync_page = block_sync_page,
1725 .prepare_write = ext3_prepare_write,
1726 .commit_write = ext3_writeback_commit_write,
1728 .invalidatepage = ext3_invalidatepage,
1729 .releasepage = ext3_releasepage,
1730 .direct_IO = ext3_direct_IO,
1731 .migratepage = buffer_migrate_page,
1734 static const struct address_space_operations ext3_journalled_aops = {
1735 .readpage = ext3_readpage,
1736 .readpages = ext3_readpages,
1737 .writepage = ext3_journalled_writepage,
1738 .sync_page = block_sync_page,
1739 .prepare_write = ext3_prepare_write,
1740 .commit_write = ext3_journalled_commit_write,
1741 .set_page_dirty = ext3_journalled_set_page_dirty,
1743 .invalidatepage = ext3_invalidatepage,
1744 .releasepage = ext3_releasepage,
1747 void ext3_set_aops(struct inode *inode)
1749 if (ext3_should_order_data(inode))
1750 inode->i_mapping->a_ops = &ext3_ordered_aops;
1751 else if (ext3_should_writeback_data(inode))
1752 inode->i_mapping->a_ops = &ext3_writeback_aops;
1754 inode->i_mapping->a_ops = &ext3_journalled_aops;
1758 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1759 * up to the end of the block which corresponds to `from'.
1760 * This required during truncate. We need to physically zero the tail end
1761 * of that block so it doesn't yield old data if the file is later grown.
1763 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1764 struct address_space *mapping, loff_t from)
1766 ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1767 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1768 unsigned blocksize, iblock, length, pos;
1769 struct inode *inode = mapping->host;
1770 struct buffer_head *bh;
1774 blocksize = inode->i_sb->s_blocksize;
1775 length = blocksize - (offset & (blocksize - 1));
1776 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1779 * For "nobh" option, we can only work if we don't need to
1780 * read-in the page - otherwise we create buffers to do the IO.
1782 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1783 ext3_should_writeback_data(inode) && PageUptodate(page)) {
1784 kaddr = kmap_atomic(page, KM_USER0);
1785 memset(kaddr + offset, 0, length);
1786 flush_dcache_page(page);
1787 kunmap_atomic(kaddr, KM_USER0);
1788 set_page_dirty(page);
1792 if (!page_has_buffers(page))
1793 create_empty_buffers(page, blocksize, 0);
1795 /* Find the buffer that contains "offset" */
1796 bh = page_buffers(page);
1798 while (offset >= pos) {
1799 bh = bh->b_this_page;
1805 if (buffer_freed(bh)) {
1806 BUFFER_TRACE(bh, "freed: skip");
1810 if (!buffer_mapped(bh)) {
1811 BUFFER_TRACE(bh, "unmapped");
1812 ext3_get_block(inode, iblock, bh, 0);
1813 /* unmapped? It's a hole - nothing to do */
1814 if (!buffer_mapped(bh)) {
1815 BUFFER_TRACE(bh, "still unmapped");
1820 /* Ok, it's mapped. Make sure it's up-to-date */
1821 if (PageUptodate(page))
1822 set_buffer_uptodate(bh);
1824 if (!buffer_uptodate(bh)) {
1826 ll_rw_block(READ, 1, &bh);
1828 /* Uhhuh. Read error. Complain and punt. */
1829 if (!buffer_uptodate(bh))
1833 if (ext3_should_journal_data(inode)) {
1834 BUFFER_TRACE(bh, "get write access");
1835 err = ext3_journal_get_write_access(handle, bh);
1840 kaddr = kmap_atomic(page, KM_USER0);
1841 memset(kaddr + offset, 0, length);
1842 flush_dcache_page(page);
1843 kunmap_atomic(kaddr, KM_USER0);
1845 BUFFER_TRACE(bh, "zeroed end of block");
1848 if (ext3_should_journal_data(inode)) {
1849 err = ext3_journal_dirty_metadata(handle, bh);
1851 if (ext3_should_order_data(inode))
1852 err = ext3_journal_dirty_data(handle, bh);
1853 mark_buffer_dirty(bh);
1858 page_cache_release(page);
1863 * Probably it should be a library function... search for first non-zero word
1864 * or memcmp with zero_page, whatever is better for particular architecture.
1867 static inline int all_zeroes(__le32 *p, __le32 *q)
1876 * ext3_find_shared - find the indirect blocks for partial truncation.
1877 * @inode: inode in question
1878 * @depth: depth of the affected branch
1879 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1880 * @chain: place to store the pointers to partial indirect blocks
1881 * @top: place to the (detached) top of branch
1883 * This is a helper function used by ext3_truncate().
1885 * When we do truncate() we may have to clean the ends of several
1886 * indirect blocks but leave the blocks themselves alive. Block is
1887 * partially truncated if some data below the new i_size is refered
1888 * from it (and it is on the path to the first completely truncated
1889 * data block, indeed). We have to free the top of that path along
1890 * with everything to the right of the path. Since no allocation
1891 * past the truncation point is possible until ext3_truncate()
1892 * finishes, we may safely do the latter, but top of branch may
1893 * require special attention - pageout below the truncation point
1894 * might try to populate it.
1896 * We atomically detach the top of branch from the tree, store the
1897 * block number of its root in *@top, pointers to buffer_heads of
1898 * partially truncated blocks - in @chain[].bh and pointers to
1899 * their last elements that should not be removed - in
1900 * @chain[].p. Return value is the pointer to last filled element
1903 * The work left to caller to do the actual freeing of subtrees:
1904 * a) free the subtree starting from *@top
1905 * b) free the subtrees whose roots are stored in
1906 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1907 * c) free the subtrees growing from the inode past the @chain[0].
1908 * (no partially truncated stuff there). */
1910 static Indirect *ext3_find_shared(struct inode *inode, int depth,
1911 int offsets[4], Indirect chain[4], __le32 *top)
1913 Indirect *partial, *p;
1917 /* Make k index the deepest non-null offest + 1 */
1918 for (k = depth; k > 1 && !offsets[k-1]; k--)
1920 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1921 /* Writer: pointers */
1923 partial = chain + k-1;
1925 * If the branch acquired continuation since we've looked at it -
1926 * fine, it should all survive and (new) top doesn't belong to us.
1928 if (!partial->key && *partial->p)
1931 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1934 * OK, we've found the last block that must survive. The rest of our
1935 * branch should be detached before unlocking. However, if that rest
1936 * of branch is all ours and does not grow immediately from the inode
1937 * it's easier to cheat and just decrement partial->p.
1939 if (p == chain + k - 1 && p > chain) {
1943 /* Nope, don't do this in ext3. Must leave the tree intact */
1950 while(partial > p) {
1951 brelse(partial->bh);
1959 * Zero a number of block pointers in either an inode or an indirect block.
1960 * If we restart the transaction we must again get write access to the
1961 * indirect block for further modification.
1963 * We release `count' blocks on disk, but (last - first) may be greater
1964 * than `count' because there can be holes in there.
1966 static void ext3_clear_blocks(handle_t *handle, struct inode *inode,
1967 struct buffer_head *bh, ext3_fsblk_t block_to_free,
1968 unsigned long count, __le32 *first, __le32 *last)
1971 if (try_to_extend_transaction(handle, inode)) {
1973 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1974 ext3_journal_dirty_metadata(handle, bh);
1976 ext3_mark_inode_dirty(handle, inode);
1977 ext3_journal_test_restart(handle, inode);
1979 BUFFER_TRACE(bh, "retaking write access");
1980 ext3_journal_get_write_access(handle, bh);
1985 * Any buffers which are on the journal will be in memory. We find
1986 * them on the hash table so journal_revoke() will run journal_forget()
1987 * on them. We've already detached each block from the file, so
1988 * bforget() in journal_forget() should be safe.
1990 * AKPM: turn on bforget in journal_forget()!!!
1992 for (p = first; p < last; p++) {
1993 u32 nr = le32_to_cpu(*p);
1995 struct buffer_head *bh;
1998 bh = sb_find_get_block(inode->i_sb, nr);
1999 ext3_forget(handle, 0, inode, bh, nr);
2003 ext3_free_blocks(handle, inode, block_to_free, count);
2007 * ext3_free_data - free a list of data blocks
2008 * @handle: handle for this transaction
2009 * @inode: inode we are dealing with
2010 * @this_bh: indirect buffer_head which contains *@first and *@last
2011 * @first: array of block numbers
2012 * @last: points immediately past the end of array
2014 * We are freeing all blocks refered from that array (numbers are stored as
2015 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2017 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2018 * blocks are contiguous then releasing them at one time will only affect one
2019 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2020 * actually use a lot of journal space.
2022 * @this_bh will be %NULL if @first and @last point into the inode's direct
2025 static void ext3_free_data(handle_t *handle, struct inode *inode,
2026 struct buffer_head *this_bh,
2027 __le32 *first, __le32 *last)
2029 ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */
2030 unsigned long count = 0; /* Number of blocks in the run */
2031 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2034 ext3_fsblk_t nr; /* Current block # */
2035 __le32 *p; /* Pointer into inode/ind
2036 for current block */
2039 if (this_bh) { /* For indirect block */
2040 BUFFER_TRACE(this_bh, "get_write_access");
2041 err = ext3_journal_get_write_access(handle, this_bh);
2042 /* Important: if we can't update the indirect pointers
2043 * to the blocks, we can't free them. */
2048 for (p = first; p < last; p++) {
2049 nr = le32_to_cpu(*p);
2051 /* accumulate blocks to free if they're contiguous */
2054 block_to_free_p = p;
2056 } else if (nr == block_to_free + count) {
2059 ext3_clear_blocks(handle, inode, this_bh,
2061 count, block_to_free_p, p);
2063 block_to_free_p = p;
2070 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
2071 count, block_to_free_p, p);
2074 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
2075 ext3_journal_dirty_metadata(handle, this_bh);
2080 * ext3_free_branches - free an array of branches
2081 * @handle: JBD handle for this transaction
2082 * @inode: inode we are dealing with
2083 * @parent_bh: the buffer_head which contains *@first and *@last
2084 * @first: array of block numbers
2085 * @last: pointer immediately past the end of array
2086 * @depth: depth of the branches to free
2088 * We are freeing all blocks refered from these branches (numbers are
2089 * stored as little-endian 32-bit) and updating @inode->i_blocks
2092 static void ext3_free_branches(handle_t *handle, struct inode *inode,
2093 struct buffer_head *parent_bh,
2094 __le32 *first, __le32 *last, int depth)
2099 if (is_handle_aborted(handle))
2103 struct buffer_head *bh;
2104 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2106 while (--p >= first) {
2107 nr = le32_to_cpu(*p);
2109 continue; /* A hole */
2111 /* Go read the buffer for the next level down */
2112 bh = sb_bread(inode->i_sb, nr);
2115 * A read failure? Report error and clear slot
2119 ext3_error(inode->i_sb, "ext3_free_branches",
2120 "Read failure, inode=%lu, block="E3FSBLK,
2125 /* This zaps the entire block. Bottom up. */
2126 BUFFER_TRACE(bh, "free child branches");
2127 ext3_free_branches(handle, inode, bh,
2128 (__le32*)bh->b_data,
2129 (__le32*)bh->b_data + addr_per_block,
2133 * We've probably journalled the indirect block several
2134 * times during the truncate. But it's no longer
2135 * needed and we now drop it from the transaction via
2138 * That's easy if it's exclusively part of this
2139 * transaction. But if it's part of the committing
2140 * transaction then journal_forget() will simply
2141 * brelse() it. That means that if the underlying
2142 * block is reallocated in ext3_get_block(),
2143 * unmap_underlying_metadata() will find this block
2144 * and will try to get rid of it. damn, damn.
2146 * If this block has already been committed to the
2147 * journal, a revoke record will be written. And
2148 * revoke records must be emitted *before* clearing
2149 * this block's bit in the bitmaps.
2151 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
2154 * Everything below this this pointer has been
2155 * released. Now let this top-of-subtree go.
2157 * We want the freeing of this indirect block to be
2158 * atomic in the journal with the updating of the
2159 * bitmap block which owns it. So make some room in
2162 * We zero the parent pointer *after* freeing its
2163 * pointee in the bitmaps, so if extend_transaction()
2164 * for some reason fails to put the bitmap changes and
2165 * the release into the same transaction, recovery
2166 * will merely complain about releasing a free block,
2167 * rather than leaking blocks.
2169 if (is_handle_aborted(handle))
2171 if (try_to_extend_transaction(handle, inode)) {
2172 ext3_mark_inode_dirty(handle, inode);
2173 ext3_journal_test_restart(handle, inode);
2176 ext3_free_blocks(handle, inode, nr, 1);
2180 * The block which we have just freed is
2181 * pointed to by an indirect block: journal it
2183 BUFFER_TRACE(parent_bh, "get_write_access");
2184 if (!ext3_journal_get_write_access(handle,
2187 BUFFER_TRACE(parent_bh,
2188 "call ext3_journal_dirty_metadata");
2189 ext3_journal_dirty_metadata(handle,
2195 /* We have reached the bottom of the tree. */
2196 BUFFER_TRACE(parent_bh, "free data blocks");
2197 ext3_free_data(handle, inode, parent_bh, first, last);
2204 * We block out ext3_get_block() block instantiations across the entire
2205 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2206 * simultaneously on behalf of the same inode.
2208 * As we work through the truncate and commmit bits of it to the journal there
2209 * is one core, guiding principle: the file's tree must always be consistent on
2210 * disk. We must be able to restart the truncate after a crash.
2212 * The file's tree may be transiently inconsistent in memory (although it
2213 * probably isn't), but whenever we close off and commit a journal transaction,
2214 * the contents of (the filesystem + the journal) must be consistent and
2215 * restartable. It's pretty simple, really: bottom up, right to left (although
2216 * left-to-right works OK too).
2218 * Note that at recovery time, journal replay occurs *before* the restart of
2219 * truncate against the orphan inode list.
2221 * The committed inode has the new, desired i_size (which is the same as
2222 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2223 * that this inode's truncate did not complete and it will again call
2224 * ext3_truncate() to have another go. So there will be instantiated blocks
2225 * to the right of the truncation point in a crashed ext3 filesystem. But
2226 * that's fine - as long as they are linked from the inode, the post-crash
2227 * ext3_truncate() run will find them and release them.
2229 void ext3_truncate(struct inode *inode)
2232 struct ext3_inode_info *ei = EXT3_I(inode);
2233 __le32 *i_data = ei->i_data;
2234 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2235 struct address_space *mapping = inode->i_mapping;
2242 unsigned blocksize = inode->i_sb->s_blocksize;
2245 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2246 S_ISLNK(inode->i_mode)))
2248 if (ext3_inode_is_fast_symlink(inode))
2250 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2254 * We have to lock the EOF page here, because lock_page() nests
2255 * outside journal_start().
2257 if ((inode->i_size & (blocksize - 1)) == 0) {
2258 /* Block boundary? Nothing to do */
2261 page = grab_cache_page(mapping,
2262 inode->i_size >> PAGE_CACHE_SHIFT);
2267 handle = start_transaction(inode);
2268 if (IS_ERR(handle)) {
2270 clear_highpage(page);
2271 flush_dcache_page(page);
2273 page_cache_release(page);
2275 return; /* AKPM: return what? */
2278 last_block = (inode->i_size + blocksize-1)
2279 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2282 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2284 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2286 goto out_stop; /* error */
2289 * OK. This truncate is going to happen. We add the inode to the
2290 * orphan list, so that if this truncate spans multiple transactions,
2291 * and we crash, we will resume the truncate when the filesystem
2292 * recovers. It also marks the inode dirty, to catch the new size.
2294 * Implication: the file must always be in a sane, consistent
2295 * truncatable state while each transaction commits.
2297 if (ext3_orphan_add(handle, inode))
2301 * The orphan list entry will now protect us from any crash which
2302 * occurs before the truncate completes, so it is now safe to propagate
2303 * the new, shorter inode size (held for now in i_size) into the
2304 * on-disk inode. We do this via i_disksize, which is the value which
2305 * ext3 *really* writes onto the disk inode.
2307 ei->i_disksize = inode->i_size;
2310 * From here we block out all ext3_get_block() callers who want to
2311 * modify the block allocation tree.
2313 mutex_lock(&ei->truncate_mutex);
2315 if (n == 1) { /* direct blocks */
2316 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2317 i_data + EXT3_NDIR_BLOCKS);
2321 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2322 /* Kill the top of shared branch (not detached) */
2324 if (partial == chain) {
2325 /* Shared branch grows from the inode */
2326 ext3_free_branches(handle, inode, NULL,
2327 &nr, &nr+1, (chain+n-1) - partial);
2330 * We mark the inode dirty prior to restart,
2331 * and prior to stop. No need for it here.
2334 /* Shared branch grows from an indirect block */
2335 BUFFER_TRACE(partial->bh, "get_write_access");
2336 ext3_free_branches(handle, inode, partial->bh,
2338 partial->p+1, (chain+n-1) - partial);
2341 /* Clear the ends of indirect blocks on the shared branch */
2342 while (partial > chain) {
2343 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2344 (__le32*)partial->bh->b_data+addr_per_block,
2345 (chain+n-1) - partial);
2346 BUFFER_TRACE(partial->bh, "call brelse");
2347 brelse (partial->bh);
2351 /* Kill the remaining (whole) subtrees */
2352 switch (offsets[0]) {
2354 nr = i_data[EXT3_IND_BLOCK];
2356 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2357 i_data[EXT3_IND_BLOCK] = 0;
2359 case EXT3_IND_BLOCK:
2360 nr = i_data[EXT3_DIND_BLOCK];
2362 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2363 i_data[EXT3_DIND_BLOCK] = 0;
2365 case EXT3_DIND_BLOCK:
2366 nr = i_data[EXT3_TIND_BLOCK];
2368 ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2369 i_data[EXT3_TIND_BLOCK] = 0;
2371 case EXT3_TIND_BLOCK:
2375 ext3_discard_reservation(inode);
2377 mutex_unlock(&ei->truncate_mutex);
2378 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2379 ext3_mark_inode_dirty(handle, inode);
2382 * In a multi-transaction truncate, we only make the final transaction
2389 * If this was a simple ftruncate(), and the file will remain alive
2390 * then we need to clear up the orphan record which we created above.
2391 * However, if this was a real unlink then we were called by
2392 * ext3_delete_inode(), and we allow that function to clean up the
2393 * orphan info for us.
2396 ext3_orphan_del(handle, inode);
2398 ext3_journal_stop(handle);
2401 static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb,
2402 unsigned long ino, struct ext3_iloc *iloc)
2404 unsigned long desc, group_desc, block_group;
2405 unsigned long offset;
2407 struct buffer_head *bh;
2408 struct ext3_group_desc * gdp;
2410 if (!ext3_valid_inum(sb, ino)) {
2412 * This error is already checked for in namei.c unless we are
2413 * looking at an NFS filehandle, in which case no error
2419 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2420 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2421 ext3_error(sb,"ext3_get_inode_block","group >= groups count");
2425 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2426 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2427 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2429 ext3_error (sb, "ext3_get_inode_block",
2430 "Descriptor not loaded");
2434 gdp = (struct ext3_group_desc *)bh->b_data;
2436 * Figure out the offset within the block group inode table
2438 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2439 EXT3_INODE_SIZE(sb);
2440 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2441 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2443 iloc->block_group = block_group;
2444 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2449 * ext3_get_inode_loc returns with an extra refcount against the inode's
2450 * underlying buffer_head on success. If 'in_mem' is true, we have all
2451 * data in memory that is needed to recreate the on-disk version of this
2454 static int __ext3_get_inode_loc(struct inode *inode,
2455 struct ext3_iloc *iloc, int in_mem)
2458 struct buffer_head *bh;
2460 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2464 bh = sb_getblk(inode->i_sb, block);
2466 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2467 "unable to read inode block - "
2468 "inode=%lu, block="E3FSBLK,
2469 inode->i_ino, block);
2472 if (!buffer_uptodate(bh)) {
2474 if (buffer_uptodate(bh)) {
2475 /* someone brought it uptodate while we waited */
2481 * If we have all information of the inode in memory and this
2482 * is the only valid inode in the block, we need not read the
2486 struct buffer_head *bitmap_bh;
2487 struct ext3_group_desc *desc;
2488 int inodes_per_buffer;
2489 int inode_offset, i;
2493 block_group = (inode->i_ino - 1) /
2494 EXT3_INODES_PER_GROUP(inode->i_sb);
2495 inodes_per_buffer = bh->b_size /
2496 EXT3_INODE_SIZE(inode->i_sb);
2497 inode_offset = ((inode->i_ino - 1) %
2498 EXT3_INODES_PER_GROUP(inode->i_sb));
2499 start = inode_offset & ~(inodes_per_buffer - 1);
2501 /* Is the inode bitmap in cache? */
2502 desc = ext3_get_group_desc(inode->i_sb,
2507 bitmap_bh = sb_getblk(inode->i_sb,
2508 le32_to_cpu(desc->bg_inode_bitmap));
2513 * If the inode bitmap isn't in cache then the
2514 * optimisation may end up performing two reads instead
2515 * of one, so skip it.
2517 if (!buffer_uptodate(bitmap_bh)) {
2521 for (i = start; i < start + inodes_per_buffer; i++) {
2522 if (i == inode_offset)
2524 if (ext3_test_bit(i, bitmap_bh->b_data))
2528 if (i == start + inodes_per_buffer) {
2529 /* all other inodes are free, so skip I/O */
2530 memset(bh->b_data, 0, bh->b_size);
2531 set_buffer_uptodate(bh);
2539 * There are other valid inodes in the buffer, this inode
2540 * has in-inode xattrs, or we don't have this inode in memory.
2541 * Read the block from disk.
2544 bh->b_end_io = end_buffer_read_sync;
2545 submit_bh(READ_META, bh);
2547 if (!buffer_uptodate(bh)) {
2548 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2549 "unable to read inode block - "
2550 "inode=%lu, block="E3FSBLK,
2551 inode->i_ino, block);
2561 int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc)
2563 /* We have all inode data except xattrs in memory here. */
2564 return __ext3_get_inode_loc(inode, iloc,
2565 !(EXT3_I(inode)->i_state & EXT3_STATE_XATTR));
2568 void ext3_set_inode_flags(struct inode *inode)
2570 unsigned int flags = EXT3_I(inode)->i_flags;
2572 inode->i_flags &= ~(S_IMMUTABLE | S_IUNLINK | S_BARRIER |
2573 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
2575 if (flags & EXT3_IMMUTABLE_FL)
2576 inode->i_flags |= S_IMMUTABLE;
2577 if (flags & EXT3_IUNLINK_FL)
2578 inode->i_flags |= S_IUNLINK;
2579 if (flags & EXT3_BARRIER_FL)
2580 inode->i_flags |= S_BARRIER;
2582 if (flags & EXT3_SYNC_FL)
2583 inode->i_flags |= S_SYNC;
2584 if (flags & EXT3_APPEND_FL)
2585 inode->i_flags |= S_APPEND;
2586 if (flags & EXT3_NOATIME_FL)
2587 inode->i_flags |= S_NOATIME;
2588 if (flags & EXT3_DIRSYNC_FL)
2589 inode->i_flags |= S_DIRSYNC;
2592 int ext3_sync_flags(struct inode *inode)
2594 unsigned int oldflags, newflags;
2597 oldflags = EXT3_I(inode)->i_flags;
2598 newflags = oldflags & ~(EXT3_IMMUTABLE_FL |
2599 EXT3_IUNLINK_FL | EXT3_BARRIER_FL);
2601 if (IS_IMMUTABLE(inode))
2602 newflags |= EXT3_IMMUTABLE_FL;
2603 if (IS_IUNLINK(inode))
2604 newflags |= EXT3_IUNLINK_FL;
2605 if (IS_BARRIER(inode))
2606 newflags |= EXT3_BARRIER_FL;
2608 if (oldflags ^ newflags) {
2610 struct ext3_iloc iloc;
2612 handle = ext3_journal_start(inode, 1);
2614 return PTR_ERR(handle);
2617 err = ext3_reserve_inode_write(handle, inode, &iloc);
2621 EXT3_I(inode)->i_flags = newflags;
2622 inode->i_ctime = CURRENT_TIME;
2624 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2626 ext3_journal_stop(handle);
2631 void ext3_read_inode(struct inode * inode)
2633 struct ext3_iloc iloc;
2634 struct ext3_inode *raw_inode;
2635 struct ext3_inode_info *ei = EXT3_I(inode);
2636 struct buffer_head *bh;
2641 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2642 ei->i_acl = EXT3_ACL_NOT_CACHED;
2643 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2645 ei->i_block_alloc_info = NULL;
2647 if (__ext3_get_inode_loc(inode, &iloc, 0))
2650 raw_inode = ext3_raw_inode(&iloc);
2651 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2652 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2653 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2654 if(!(test_opt (inode->i_sb, NO_UID32))) {
2655 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2656 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2658 inode->i_uid = INOTAG_UID(DX_TAG(inode), uid, gid);
2659 inode->i_gid = INOTAG_GID(DX_TAG(inode), uid, gid);
2660 inode->i_tag = INOTAG_TAG(DX_TAG(inode), uid, gid,
2661 le16_to_cpu(raw_inode->i_raw_tag));
2663 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2664 inode->i_size = le32_to_cpu(raw_inode->i_size);
2665 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2666 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2667 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2668 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2671 ei->i_dir_start_lookup = 0;
2672 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2673 /* We now have enough fields to check if the inode was active or not.
2674 * This is needed because nfsd might try to access dead inodes
2675 * the test is that same one that e2fsck uses
2676 * NeilBrown 1999oct15
2678 if (inode->i_nlink == 0) {
2679 if (inode->i_mode == 0 ||
2680 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2681 /* this inode is deleted */
2685 /* The only unlinked inodes we let through here have
2686 * valid i_mode and are being read by the orphan
2687 * recovery code: that's fine, we're about to complete
2688 * the process of deleting those. */
2690 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2691 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2692 #ifdef EXT3_FRAGMENTS
2693 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2694 ei->i_frag_no = raw_inode->i_frag;
2695 ei->i_frag_size = raw_inode->i_fsize;
2697 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2698 if (!S_ISREG(inode->i_mode)) {
2699 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2702 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2704 ei->i_disksize = inode->i_size;
2705 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2706 ei->i_block_group = iloc.block_group;
2708 * NOTE! The in-memory inode i_data array is in little-endian order
2709 * even on big-endian machines: we do NOT byteswap the block numbers!
2711 for (block = 0; block < EXT3_N_BLOCKS; block++)
2712 ei->i_data[block] = raw_inode->i_block[block];
2713 INIT_LIST_HEAD(&ei->i_orphan);
2715 if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
2716 EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
2718 * When mke2fs creates big inodes it does not zero out
2719 * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE,
2720 * so ignore those first few inodes.
2722 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2723 if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2724 EXT3_INODE_SIZE(inode->i_sb))
2726 if (ei->i_extra_isize == 0) {
2727 /* The extra space is currently unused. Use it. */
2728 ei->i_extra_isize = sizeof(struct ext3_inode) -
2729 EXT3_GOOD_OLD_INODE_SIZE;
2731 __le32 *magic = (void *)raw_inode +
2732 EXT3_GOOD_OLD_INODE_SIZE +
2734 if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC))
2735 ei->i_state |= EXT3_STATE_XATTR;
2738 ei->i_extra_isize = 0;
2740 if (S_ISREG(inode->i_mode)) {
2741 inode->i_op = &ext3_file_inode_operations;
2742 inode->i_fop = &ext3_file_operations;
2743 ext3_set_aops(inode);
2744 } else if (S_ISDIR(inode->i_mode)) {
2745 inode->i_op = &ext3_dir_inode_operations;
2746 inode->i_fop = &ext3_dir_operations;
2747 } else if (S_ISLNK(inode->i_mode)) {
2748 if (ext3_inode_is_fast_symlink(inode))
2749 inode->i_op = &ext3_fast_symlink_inode_operations;
2751 inode->i_op = &ext3_symlink_inode_operations;
2752 ext3_set_aops(inode);
2755 inode->i_op = &ext3_special_inode_operations;
2756 if (raw_inode->i_block[0])
2757 init_special_inode(inode, inode->i_mode,
2758 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2760 init_special_inode(inode, inode->i_mode,
2761 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2764 ext3_set_inode_flags(inode);
2768 make_bad_inode(inode);
2773 * Post the struct inode info into an on-disk inode location in the
2774 * buffer-cache. This gobbles the caller's reference to the
2775 * buffer_head in the inode location struct.
2777 * The caller must have write access to iloc->bh.
2779 static int ext3_do_update_inode(handle_t *handle,
2780 struct inode *inode,
2781 struct ext3_iloc *iloc)
2783 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2784 struct ext3_inode_info *ei = EXT3_I(inode);
2785 struct buffer_head *bh = iloc->bh;
2786 uid_t uid = TAGINO_UID(DX_TAG(inode), inode->i_uid, inode->i_tag);
2787 gid_t gid = TAGINO_GID(DX_TAG(inode), inode->i_gid, inode->i_tag);
2788 int err = 0, rc, block;
2790 /* For fields not not tracking in the in-memory inode,
2791 * initialise them to zero for new inodes. */
2792 if (ei->i_state & EXT3_STATE_NEW)
2793 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2795 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2796 if(!(test_opt(inode->i_sb, NO_UID32))) {
2797 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2798 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2800 * Fix up interoperability with old kernels. Otherwise, old inodes get
2801 * re-used with the upper 16 bits of the uid/gid intact
2804 raw_inode->i_uid_high =
2805 cpu_to_le16(high_16_bits(uid));
2806 raw_inode->i_gid_high =
2807 cpu_to_le16(high_16_bits(gid));
2809 raw_inode->i_uid_high = 0;
2810 raw_inode->i_gid_high = 0;
2813 raw_inode->i_uid_low =
2814 cpu_to_le16(fs_high2lowuid(uid));
2815 raw_inode->i_gid_low =
2816 cpu_to_le16(fs_high2lowgid(gid));
2817 raw_inode->i_uid_high = 0;
2818 raw_inode->i_gid_high = 0;
2820 #ifdef CONFIG_TAGGING_INTERN
2821 raw_inode->i_raw_tag = cpu_to_le16(inode->i_tag);
2823 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2824 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2825 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2826 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2827 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2828 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2829 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2830 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2831 #ifdef EXT3_FRAGMENTS
2832 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2833 raw_inode->i_frag = ei->i_frag_no;
2834 raw_inode->i_fsize = ei->i_frag_size;
2836 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2837 if (!S_ISREG(inode->i_mode)) {
2838 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2840 raw_inode->i_size_high =
2841 cpu_to_le32(ei->i_disksize >> 32);
2842 if (ei->i_disksize > 0x7fffffffULL) {
2843 struct super_block *sb = inode->i_sb;
2844 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2845 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2846 EXT3_SB(sb)->s_es->s_rev_level ==
2847 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2848 /* If this is the first large file
2849 * created, add a flag to the superblock.
2851 err = ext3_journal_get_write_access(handle,
2852 EXT3_SB(sb)->s_sbh);
2855 ext3_update_dynamic_rev(sb);
2856 EXT3_SET_RO_COMPAT_FEATURE(sb,
2857 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2860 err = ext3_journal_dirty_metadata(handle,
2861 EXT3_SB(sb)->s_sbh);
2865 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2866 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2867 if (old_valid_dev(inode->i_rdev)) {
2868 raw_inode->i_block[0] =
2869 cpu_to_le32(old_encode_dev(inode->i_rdev));
2870 raw_inode->i_block[1] = 0;
2872 raw_inode->i_block[0] = 0;
2873 raw_inode->i_block[1] =
2874 cpu_to_le32(new_encode_dev(inode->i_rdev));
2875 raw_inode->i_block[2] = 0;
2877 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2878 raw_inode->i_block[block] = ei->i_data[block];
2880 if (ei->i_extra_isize)
2881 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2883 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2884 rc = ext3_journal_dirty_metadata(handle, bh);
2887 ei->i_state &= ~EXT3_STATE_NEW;
2891 ext3_std_error(inode->i_sb, err);
2896 * ext3_write_inode()
2898 * We are called from a few places:
2900 * - Within generic_file_write() for O_SYNC files.
2901 * Here, there will be no transaction running. We wait for any running
2902 * trasnaction to commit.
2904 * - Within sys_sync(), kupdate and such.
2905 * We wait on commit, if tol to.
2907 * - Within prune_icache() (PF_MEMALLOC == true)
2908 * Here we simply return. We can't afford to block kswapd on the
2911 * In all cases it is actually safe for us to return without doing anything,
2912 * because the inode has been copied into a raw inode buffer in
2913 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2916 * Note that we are absolutely dependent upon all inode dirtiers doing the
2917 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2918 * which we are interested.
2920 * It would be a bug for them to not do this. The code:
2922 * mark_inode_dirty(inode)
2924 * inode->i_size = expr;
2926 * is in error because a kswapd-driven write_inode() could occur while
2927 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2928 * will no longer be on the superblock's dirty inode list.
2930 int ext3_write_inode(struct inode *inode, int wait)
2932 if (current->flags & PF_MEMALLOC)
2935 if (ext3_journal_current_handle()) {
2936 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2944 return ext3_force_commit(inode->i_sb);
2950 * Called from notify_change.
2952 * We want to trap VFS attempts to truncate the file as soon as
2953 * possible. In particular, we want to make sure that when the VFS
2954 * shrinks i_size, we put the inode on the orphan list and modify
2955 * i_disksize immediately, so that during the subsequent flushing of
2956 * dirty pages and freeing of disk blocks, we can guarantee that any
2957 * commit will leave the blocks being flushed in an unused state on
2958 * disk. (On recovery, the inode will get truncated and the blocks will
2959 * be freed, so we have a strong guarantee that no future commit will
2960 * leave these blocks visible to the user.)
2962 * Called with inode->sem down.
2964 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2966 struct inode *inode = dentry->d_inode;
2968 const unsigned int ia_valid = attr->ia_valid;
2970 error = inode_change_ok(inode, attr);
2974 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2975 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
2976 (ia_valid & ATTR_TAG && attr->ia_tag != inode->i_tag)) {
2979 /* (user+group)*(old+new) structure, inode write (sb,
2980 * inode block, ? - but truncate inode update has it) */
2981 handle = ext3_journal_start(inode, 2*(EXT3_QUOTA_INIT_BLOCKS(inode->i_sb)+
2982 EXT3_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
2983 if (IS_ERR(handle)) {
2984 error = PTR_ERR(handle);
2987 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2989 ext3_journal_stop(handle);
2992 /* Update corresponding info in inode so that everything is in
2993 * one transaction */
2994 if (attr->ia_valid & ATTR_UID)
2995 inode->i_uid = attr->ia_uid;
2996 if (attr->ia_valid & ATTR_GID)
2997 inode->i_gid = attr->ia_gid;
2998 if ((attr->ia_valid & ATTR_TAG) && IS_TAGGED(inode))
2999 inode->i_tag = attr->ia_tag;
3000 error = ext3_mark_inode_dirty(handle, inode);
3001 ext3_journal_stop(handle);
3004 if (S_ISREG(inode->i_mode) &&
3005 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3008 handle = ext3_journal_start(inode, 3);
3009 if (IS_ERR(handle)) {
3010 error = PTR_ERR(handle);
3014 error = ext3_orphan_add(handle, inode);
3015 EXT3_I(inode)->i_disksize = attr->ia_size;
3016 rc = ext3_mark_inode_dirty(handle, inode);
3019 ext3_journal_stop(handle);
3022 rc = inode_setattr(inode, attr);
3024 /* If inode_setattr's call to ext3_truncate failed to get a
3025 * transaction handle at all, we need to clean up the in-core
3026 * orphan list manually. */
3028 ext3_orphan_del(NULL, inode);
3030 if (!rc && (ia_valid & ATTR_MODE))
3031 rc = ext3_acl_chmod(inode);
3034 ext3_std_error(inode->i_sb, error);
3042 * How many blocks doth make a writepage()?
3044 * With N blocks per page, it may be:
3049 * N+5 bitmap blocks (from the above)
3050 * N+5 group descriptor summary blocks
3053 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
3055 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
3057 * With ordered or writeback data it's the same, less the N data blocks.
3059 * If the inode's direct blocks can hold an integral number of pages then a
3060 * page cannot straddle two indirect blocks, and we can only touch one indirect
3061 * and dindirect block, and the "5" above becomes "3".
3063 * This still overestimates under most circumstances. If we were to pass the
3064 * start and end offsets in here as well we could do block_to_path() on each
3065 * block and work out the exact number of indirects which are touched. Pah.
3068 static int ext3_writepage_trans_blocks(struct inode *inode)
3070 int bpp = ext3_journal_blocks_per_page(inode);
3071 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
3074 if (ext3_should_journal_data(inode))
3075 ret = 3 * (bpp + indirects) + 2;
3077 ret = 2 * (bpp + indirects) + 2;
3080 /* We know that structure was already allocated during DQUOT_INIT so
3081 * we will be updating only the data blocks + inodes */
3082 ret += 2*EXT3_QUOTA_TRANS_BLOCKS(inode->i_sb);
3089 * The caller must have previously called ext3_reserve_inode_write().
3090 * Give this, we know that the caller already has write access to iloc->bh.
3092 int ext3_mark_iloc_dirty(handle_t *handle,
3093 struct inode *inode, struct ext3_iloc *iloc)
3097 /* the do_update_inode consumes one bh->b_count */
3100 /* ext3_do_update_inode() does journal_dirty_metadata */
3101 err = ext3_do_update_inode(handle, inode, iloc);
3107 * On success, We end up with an outstanding reference count against
3108 * iloc->bh. This _must_ be cleaned up later.
3112 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
3113 struct ext3_iloc *iloc)
3117 err = ext3_get_inode_loc(inode, iloc);
3119 BUFFER_TRACE(iloc->bh, "get_write_access");
3120 err = ext3_journal_get_write_access(handle, iloc->bh);
3127 ext3_std_error(inode->i_sb, err);
3132 * What we do here is to mark the in-core inode as clean with respect to inode
3133 * dirtiness (it may still be data-dirty).
3134 * This means that the in-core inode may be reaped by prune_icache
3135 * without having to perform any I/O. This is a very good thing,
3136 * because *any* task may call prune_icache - even ones which
3137 * have a transaction open against a different journal.
3139 * Is this cheating? Not really. Sure, we haven't written the
3140 * inode out, but prune_icache isn't a user-visible syncing function.
3141 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3142 * we start and wait on commits.
3144 * Is this efficient/effective? Well, we're being nice to the system
3145 * by cleaning up our inodes proactively so they can be reaped
3146 * without I/O. But we are potentially leaving up to five seconds'
3147 * worth of inodes floating about which prune_icache wants us to
3148 * write out. One way to fix that would be to get prune_icache()
3149 * to do a write_super() to free up some memory. It has the desired
3152 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
3154 struct ext3_iloc iloc;
3158 err = ext3_reserve_inode_write(handle, inode, &iloc);
3160 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
3165 * ext3_dirty_inode() is called from __mark_inode_dirty()
3167 * We're really interested in the case where a file is being extended.
3168 * i_size has been changed by generic_commit_write() and we thus need
3169 * to include the updated inode in the current transaction.
3171 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3172 * are allocated to the file.
3174 * If the inode is marked synchronous, we don't honour that here - doing
3175 * so would cause a commit on atime updates, which we don't bother doing.
3176 * We handle synchronous inodes at the highest possible level.
3178 void ext3_dirty_inode(struct inode *inode)
3180 handle_t *current_handle = ext3_journal_current_handle();
3183 handle = ext3_journal_start(inode, 2);
3186 if (current_handle &&
3187 current_handle->h_transaction != handle->h_transaction) {
3188 /* This task has a transaction open against a different fs */
3189 printk(KERN_EMERG "%s: transactions do not match!\n",
3192 jbd_debug(5, "marking dirty. outer handle=%p\n",
3194 ext3_mark_inode_dirty(handle, inode);
3196 ext3_journal_stop(handle);
3203 * Bind an inode's backing buffer_head into this transaction, to prevent
3204 * it from being flushed to disk early. Unlike
3205 * ext3_reserve_inode_write, this leaves behind no bh reference and
3206 * returns no iloc structure, so the caller needs to repeat the iloc
3207 * lookup to mark the inode dirty later.
3209 static int ext3_pin_inode(handle_t *handle, struct inode *inode)
3211 struct ext3_iloc iloc;
3215 err = ext3_get_inode_loc(inode, &iloc);
3217 BUFFER_TRACE(iloc.bh, "get_write_access");
3218 err = journal_get_write_access(handle, iloc.bh);
3220 err = ext3_journal_dirty_metadata(handle,
3225 ext3_std_error(inode->i_sb, err);
3230 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3237 * We have to be very careful here: changing a data block's
3238 * journaling status dynamically is dangerous. If we write a
3239 * data block to the journal, change the status and then delete
3240 * that block, we risk forgetting to revoke the old log record
3241 * from the journal and so a subsequent replay can corrupt data.
3242 * So, first we make sure that the journal is empty and that
3243 * nobody is changing anything.
3246 journal = EXT3_JOURNAL(inode);
3247 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3250 journal_lock_updates(journal);
3251 journal_flush(journal);
3254 * OK, there are no updates running now, and all cached data is
3255 * synced to disk. We are now in a completely consistent state
3256 * which doesn't have anything in the journal, and we know that
3257 * no filesystem updates are running, so it is safe to modify
3258 * the inode's in-core data-journaling state flag now.
3262 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3264 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3265 ext3_set_aops(inode);
3267 journal_unlock_updates(journal);
3269 /* Finally we can mark the inode as dirty. */
3271 handle = ext3_journal_start(inode, 1);
3273 return PTR_ERR(handle);
3275 err = ext3_mark_inode_dirty(handle, inode);
3277 ext3_journal_stop(handle);
3278 ext3_std_error(inode->i_sb, err);