2 * linux/fs/ext4/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 ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/ext4_jbd2.h>
29 #include <linux/jbd2.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>
45 * Test whether an inode is a fast symlink.
47 static int ext4_inode_is_fast_symlink(struct inode *inode)
49 int ea_blocks = EXT4_I(inode)->i_file_acl ?
50 (inode->i_sb->s_blocksize >> 9) : 0;
52 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
56 * The ext4 forget function must perform a revoke if we are freeing data
57 * which has been journaled. Metadata (eg. indirect blocks) must be
58 * revoked in all cases.
60 * "bh" may be NULL: a metadata block may have been freed from memory
61 * but there may still be a record of it in the journal, and that record
62 * still needs to be revoked.
64 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
65 struct buffer_head *bh, ext4_fsblk_t blocknr)
71 BUFFER_TRACE(bh, "enter");
73 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
75 bh, is_metadata, inode->i_mode,
76 test_opt(inode->i_sb, DATA_FLAGS));
78 /* Never use the revoke function if we are doing full data
79 * journaling: there is no need to, and a V1 superblock won't
80 * support it. Otherwise, only skip the revoke on un-journaled
83 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
84 (!is_metadata && !ext4_should_journal_data(inode))) {
86 BUFFER_TRACE(bh, "call jbd2_journal_forget");
87 return ext4_journal_forget(handle, bh);
93 * data!=journal && (is_metadata || should_journal_data(inode))
95 BUFFER_TRACE(bh, "call ext4_journal_revoke");
96 err = ext4_journal_revoke(handle, blocknr, bh);
98 ext4_abort(inode->i_sb, __FUNCTION__,
99 "error %d when attempting revoke", err);
100 BUFFER_TRACE(bh, "exit");
105 * Work out how many blocks we need to proceed with the next chunk of a
106 * truncate transaction.
108 static unsigned long blocks_for_truncate(struct inode *inode)
110 unsigned long needed;
112 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
114 /* Give ourselves just enough room to cope with inodes in which
115 * i_blocks is corrupt: we've seen disk corruptions in the past
116 * which resulted in random data in an inode which looked enough
117 * like a regular file for ext4 to try to delete it. Things
118 * will go a bit crazy if that happens, but at least we should
119 * try not to panic the whole kernel. */
123 /* But we need to bound the transaction so we don't overflow the
125 if (needed > EXT4_MAX_TRANS_DATA)
126 needed = EXT4_MAX_TRANS_DATA;
128 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
132 * Truncate transactions can be complex and absolutely huge. So we need to
133 * be able to restart the transaction at a conventient checkpoint to make
134 * sure we don't overflow the journal.
136 * start_transaction gets us a new handle for a truncate transaction,
137 * and extend_transaction tries to extend the existing one a bit. If
138 * extend fails, we need to propagate the failure up and restart the
139 * transaction in the top-level truncate loop. --sct
141 static handle_t *start_transaction(struct inode *inode)
145 result = ext4_journal_start(inode, blocks_for_truncate(inode));
149 ext4_std_error(inode->i_sb, PTR_ERR(result));
154 * Try to extend this transaction for the purposes of truncation.
156 * Returns 0 if we managed to create more room. If we can't create more
157 * room, and the transaction must be restarted we return 1.
159 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
161 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
163 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
169 * Restart the transaction associated with *handle. This does a commit,
170 * so before we call here everything must be consistently dirtied against
173 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
175 jbd_debug(2, "restarting handle %p\n", handle);
176 return ext4_journal_restart(handle, blocks_for_truncate(inode));
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_delete_inode (struct inode * inode)
186 truncate_inode_pages(&inode->i_data, 0);
188 if (is_bad_inode(inode))
191 handle = start_transaction(inode);
192 if (IS_ERR(handle)) {
194 * If we're going to skip the normal cleanup, we still need to
195 * make sure that the in-core orphan linked list is properly
198 ext4_orphan_del(NULL, inode);
206 ext4_truncate(inode);
208 * Kill off the orphan record which ext4_truncate created.
209 * AKPM: I think this can be inside the above `if'.
210 * Note that ext4_orphan_del() has to be able to cope with the
211 * deletion of a non-existent orphan - this is because we don't
212 * know if ext4_truncate() actually created an orphan record.
213 * (Well, we could do this if we need to, but heck - it works)
215 ext4_orphan_del(handle, inode);
216 EXT4_I(inode)->i_dtime = get_seconds();
219 * One subtle ordering requirement: if anything has gone wrong
220 * (transaction abort, IO errors, whatever), then we can still
221 * do these next steps (the fs will already have been marked as
222 * having errors), but we can't free the inode if the mark_dirty
225 if (ext4_mark_inode_dirty(handle, inode))
226 /* If that failed, just do the required in-core inode clear. */
229 ext4_free_inode(handle, inode);
230 ext4_journal_stop(handle);
233 clear_inode(inode); /* We must guarantee clearing of inode... */
239 struct buffer_head *bh;
242 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
244 p->key = *(p->p = v);
248 static int verify_chain(Indirect *from, Indirect *to)
250 while (from <= to && from->key == *from->p)
256 * ext4_block_to_path - parse the block number into array of offsets
257 * @inode: inode in question (we are only interested in its superblock)
258 * @i_block: block number to be parsed
259 * @offsets: array to store the offsets in
260 * @boundary: set this non-zero if the referred-to block is likely to be
261 * followed (on disk) by an indirect block.
263 * To store the locations of file's data ext4 uses a data structure common
264 * for UNIX filesystems - tree of pointers anchored in the inode, with
265 * data blocks at leaves and indirect blocks in intermediate nodes.
266 * This function translates the block number into path in that tree -
267 * return value is the path length and @offsets[n] is the offset of
268 * pointer to (n+1)th node in the nth one. If @block is out of range
269 * (negative or too large) warning is printed and zero returned.
271 * Note: function doesn't find node addresses, so no IO is needed. All
272 * we need to know is the capacity of indirect blocks (taken from the
277 * Portability note: the last comparison (check that we fit into triple
278 * indirect block) is spelled differently, because otherwise on an
279 * architecture with 32-bit longs and 8Kb pages we might get into trouble
280 * if our filesystem had 8Kb blocks. We might use long long, but that would
281 * kill us on x86. Oh, well, at least the sign propagation does not matter -
282 * i_block would have to be negative in the very beginning, so we would not
286 static int ext4_block_to_path(struct inode *inode,
287 long i_block, int offsets[4], int *boundary)
289 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
290 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
291 const long direct_blocks = EXT4_NDIR_BLOCKS,
292 indirect_blocks = ptrs,
293 double_blocks = (1 << (ptrs_bits * 2));
298 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
299 } else if (i_block < direct_blocks) {
300 offsets[n++] = i_block;
301 final = direct_blocks;
302 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
303 offsets[n++] = EXT4_IND_BLOCK;
304 offsets[n++] = i_block;
306 } else if ((i_block -= indirect_blocks) < double_blocks) {
307 offsets[n++] = EXT4_DIND_BLOCK;
308 offsets[n++] = i_block >> ptrs_bits;
309 offsets[n++] = i_block & (ptrs - 1);
311 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
312 offsets[n++] = EXT4_TIND_BLOCK;
313 offsets[n++] = i_block >> (ptrs_bits * 2);
314 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
315 offsets[n++] = i_block & (ptrs - 1);
318 ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
321 *boundary = final - 1 - (i_block & (ptrs - 1));
326 * ext4_get_branch - read the chain of indirect blocks leading to data
327 * @inode: inode in question
328 * @depth: depth of the chain (1 - direct pointer, etc.)
329 * @offsets: offsets of pointers in inode/indirect blocks
330 * @chain: place to store the result
331 * @err: here we store the error value
333 * Function fills the array of triples <key, p, bh> and returns %NULL
334 * if everything went OK or the pointer to the last filled triple
335 * (incomplete one) otherwise. Upon the return chain[i].key contains
336 * the number of (i+1)-th block in the chain (as it is stored in memory,
337 * i.e. little-endian 32-bit), chain[i].p contains the address of that
338 * number (it points into struct inode for i==0 and into the bh->b_data
339 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
340 * block for i>0 and NULL for i==0. In other words, it holds the block
341 * numbers of the chain, addresses they were taken from (and where we can
342 * verify that chain did not change) and buffer_heads hosting these
345 * Function stops when it stumbles upon zero pointer (absent block)
346 * (pointer to last triple returned, *@err == 0)
347 * or when it gets an IO error reading an indirect block
348 * (ditto, *@err == -EIO)
349 * or when it notices that chain had been changed while it was reading
350 * (ditto, *@err == -EAGAIN)
351 * or when it reads all @depth-1 indirect blocks successfully and finds
352 * the whole chain, all way to the data (returns %NULL, *err == 0).
354 static Indirect *ext4_get_branch(struct inode *inode, int depth, int *offsets,
355 Indirect chain[4], int *err)
357 struct super_block *sb = inode->i_sb;
359 struct buffer_head *bh;
362 /* i_data is not going away, no lock needed */
363 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
367 bh = sb_bread(sb, le32_to_cpu(p->key));
370 /* Reader: pointers */
371 if (!verify_chain(chain, p))
373 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
391 * ext4_find_near - find a place for allocation with sufficient locality
393 * @ind: descriptor of indirect block.
395 * This function returns the prefered place for block allocation.
396 * It is used when heuristic for sequential allocation fails.
398 * + if there is a block to the left of our position - allocate near it.
399 * + if pointer will live in indirect block - allocate near that block.
400 * + if pointer will live in inode - allocate in the same
403 * In the latter case we colour the starting block by the callers PID to
404 * prevent it from clashing with concurrent allocations for a different inode
405 * in the same block group. The PID is used here so that functionally related
406 * files will be close-by on-disk.
408 * Caller must make sure that @ind is valid and will stay that way.
410 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
412 struct ext4_inode_info *ei = EXT4_I(inode);
413 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
415 ext4_fsblk_t bg_start;
416 ext4_grpblk_t colour;
418 /* Try to find previous block */
419 for (p = ind->p - 1; p >= start; p--) {
421 return le32_to_cpu(*p);
424 /* No such thing, so let's try location of indirect block */
426 return ind->bh->b_blocknr;
429 * It is going to be referred to from the inode itself? OK, just put it
430 * into the same cylinder group then.
432 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
433 colour = (current->pid % 16) *
434 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
435 return bg_start + colour;
439 * ext4_find_goal - find a prefered place for allocation.
441 * @block: block we want
442 * @chain: chain of indirect blocks
443 * @partial: pointer to the last triple within a chain
444 * @goal: place to store the result.
446 * Normally this function find the prefered place for block allocation,
447 * stores it in *@goal and returns zero.
450 static ext4_fsblk_t ext4_find_goal(struct inode *inode, long block,
451 Indirect chain[4], Indirect *partial)
453 struct ext4_block_alloc_info *block_i;
455 block_i = EXT4_I(inode)->i_block_alloc_info;
458 * try the heuristic for sequential allocation,
459 * failing that at least try to get decent locality.
461 if (block_i && (block == block_i->last_alloc_logical_block + 1)
462 && (block_i->last_alloc_physical_block != 0)) {
463 return block_i->last_alloc_physical_block + 1;
466 return ext4_find_near(inode, partial);
470 * ext4_blks_to_allocate: Look up the block map and count the number
471 * of direct blocks need to be allocated for the given branch.
473 * @branch: chain of indirect blocks
474 * @k: number of blocks need for indirect blocks
475 * @blks: number of data blocks to be mapped.
476 * @blocks_to_boundary: the offset in the indirect block
478 * return the total number of blocks to be allocate, including the
479 * direct and indirect blocks.
481 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
482 int blocks_to_boundary)
484 unsigned long count = 0;
487 * Simple case, [t,d]Indirect block(s) has not allocated yet
488 * then it's clear blocks on that path have not allocated
491 /* right now we don't handle cross boundary allocation */
492 if (blks < blocks_to_boundary + 1)
495 count += blocks_to_boundary + 1;
500 while (count < blks && count <= blocks_to_boundary &&
501 le32_to_cpu(*(branch[0].p + count)) == 0) {
508 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
509 * @indirect_blks: the number of blocks need to allocate for indirect
512 * @new_blocks: on return it will store the new block numbers for
513 * the indirect blocks(if needed) and the first direct block,
514 * @blks: on return it will store the total number of allocated
517 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
518 ext4_fsblk_t goal, int indirect_blks, int blks,
519 ext4_fsblk_t new_blocks[4], int *err)
522 unsigned long count = 0;
524 ext4_fsblk_t current_block = 0;
528 * Here we try to allocate the requested multiple blocks at once,
529 * on a best-effort basis.
530 * To build a branch, we should allocate blocks for
531 * the indirect blocks(if not allocated yet), and at least
532 * the first direct block of this branch. That's the
533 * minimum number of blocks need to allocate(required)
535 target = blks + indirect_blks;
539 /* allocating blocks for indirect blocks and direct blocks */
540 current_block = ext4_new_blocks(handle,inode,goal,&count,err);
545 /* allocate blocks for indirect blocks */
546 while (index < indirect_blks && count) {
547 new_blocks[index++] = current_block++;
555 /* save the new block number for the first direct block */
556 new_blocks[index] = current_block;
558 /* total number of blocks allocated for direct blocks */
563 for (i = 0; i <index; i++)
564 ext4_free_blocks(handle, inode, new_blocks[i], 1);
569 * ext4_alloc_branch - allocate and set up a chain of blocks.
571 * @indirect_blks: number of allocated indirect blocks
572 * @blks: number of allocated direct blocks
573 * @offsets: offsets (in the blocks) to store the pointers to next.
574 * @branch: place to store the chain in.
576 * This function allocates blocks, zeroes out all but the last one,
577 * links them into chain and (if we are synchronous) writes them to disk.
578 * In other words, it prepares a branch that can be spliced onto the
579 * inode. It stores the information about that chain in the branch[], in
580 * the same format as ext4_get_branch() would do. We are calling it after
581 * we had read the existing part of chain and partial points to the last
582 * triple of that (one with zero ->key). Upon the exit we have the same
583 * picture as after the successful ext4_get_block(), except that in one
584 * place chain is disconnected - *branch->p is still zero (we did not
585 * set the last link), but branch->key contains the number that should
586 * be placed into *branch->p to fill that gap.
588 * If allocation fails we free all blocks we've allocated (and forget
589 * their buffer_heads) and return the error value the from failed
590 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
591 * as described above and return 0.
593 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
594 int indirect_blks, int *blks, ext4_fsblk_t goal,
595 int *offsets, Indirect *branch)
597 int blocksize = inode->i_sb->s_blocksize;
600 struct buffer_head *bh;
602 ext4_fsblk_t new_blocks[4];
603 ext4_fsblk_t current_block;
605 num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
606 *blks, new_blocks, &err);
610 branch[0].key = cpu_to_le32(new_blocks[0]);
612 * metadata blocks and data blocks are allocated.
614 for (n = 1; n <= indirect_blks; n++) {
616 * Get buffer_head for parent block, zero it out
617 * and set the pointer to new one, then send
620 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
623 BUFFER_TRACE(bh, "call get_create_access");
624 err = ext4_journal_get_create_access(handle, bh);
631 memset(bh->b_data, 0, blocksize);
632 branch[n].p = (__le32 *) bh->b_data + offsets[n];
633 branch[n].key = cpu_to_le32(new_blocks[n]);
634 *branch[n].p = branch[n].key;
635 if ( n == indirect_blks) {
636 current_block = new_blocks[n];
638 * End of chain, update the last new metablock of
639 * the chain to point to the new allocated
640 * data blocks numbers
642 for (i=1; i < num; i++)
643 *(branch[n].p + i) = cpu_to_le32(++current_block);
645 BUFFER_TRACE(bh, "marking uptodate");
646 set_buffer_uptodate(bh);
649 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
650 err = ext4_journal_dirty_metadata(handle, bh);
657 /* Allocation failed, free what we already allocated */
658 for (i = 1; i <= n ; i++) {
659 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
660 ext4_journal_forget(handle, branch[i].bh);
662 for (i = 0; i <indirect_blks; i++)
663 ext4_free_blocks(handle, inode, new_blocks[i], 1);
665 ext4_free_blocks(handle, inode, new_blocks[i], num);
671 * ext4_splice_branch - splice the allocated branch onto inode.
673 * @block: (logical) number of block we are adding
674 * @chain: chain of indirect blocks (with a missing link - see
676 * @where: location of missing link
677 * @num: number of indirect blocks we are adding
678 * @blks: number of direct blocks we are adding
680 * This function fills the missing link and does all housekeeping needed in
681 * inode (->i_blocks, etc.). In case of success we end up with the full
682 * chain to new block and return 0.
684 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
685 long block, Indirect *where, int num, int blks)
689 struct ext4_block_alloc_info *block_i;
690 ext4_fsblk_t current_block;
692 block_i = EXT4_I(inode)->i_block_alloc_info;
694 * If we're splicing into a [td]indirect block (as opposed to the
695 * inode) then we need to get write access to the [td]indirect block
699 BUFFER_TRACE(where->bh, "get_write_access");
700 err = ext4_journal_get_write_access(handle, where->bh);
706 *where->p = where->key;
709 * Update the host buffer_head or inode to point to more just allocated
710 * direct blocks blocks
712 if (num == 0 && blks > 1) {
713 current_block = le32_to_cpu(where->key) + 1;
714 for (i = 1; i < blks; i++)
715 *(where->p + i ) = cpu_to_le32(current_block++);
719 * update the most recently allocated logical & physical block
720 * in i_block_alloc_info, to assist find the proper goal block for next
724 block_i->last_alloc_logical_block = block + blks - 1;
725 block_i->last_alloc_physical_block =
726 le32_to_cpu(where[num].key) + blks - 1;
729 /* We are done with atomic stuff, now do the rest of housekeeping */
731 inode->i_ctime = CURRENT_TIME_SEC;
732 ext4_mark_inode_dirty(handle, inode);
734 /* had we spliced it onto indirect block? */
737 * If we spliced it onto an indirect block, we haven't
738 * altered the inode. Note however that if it is being spliced
739 * onto an indirect block at the very end of the file (the
740 * file is growing) then we *will* alter the inode to reflect
741 * the new i_size. But that is not done here - it is done in
742 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
744 jbd_debug(5, "splicing indirect only\n");
745 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
746 err = ext4_journal_dirty_metadata(handle, where->bh);
751 * OK, we spliced it into the inode itself on a direct block.
752 * Inode was dirtied above.
754 jbd_debug(5, "splicing direct\n");
759 for (i = 1; i <= num; i++) {
760 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
761 ext4_journal_forget(handle, where[i].bh);
762 ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
764 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
770 * Allocation strategy is simple: if we have to allocate something, we will
771 * have to go the whole way to leaf. So let's do it before attaching anything
772 * to tree, set linkage between the newborn blocks, write them if sync is
773 * required, recheck the path, free and repeat if check fails, otherwise
774 * set the last missing link (that will protect us from any truncate-generated
775 * removals - all blocks on the path are immune now) and possibly force the
776 * write on the parent block.
777 * That has a nice additional property: no special recovery from the failed
778 * allocations is needed - we simply release blocks and do not touch anything
779 * reachable from inode.
781 * `handle' can be NULL if create == 0.
783 * The BKL may not be held on entry here. Be sure to take it early.
784 * return > 0, # of blocks mapped or allocated.
785 * return = 0, if plain lookup failed.
786 * return < 0, error case.
788 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
789 sector_t iblock, unsigned long maxblocks,
790 struct buffer_head *bh_result,
791 int create, int extend_disksize)
799 int blocks_to_boundary = 0;
801 struct ext4_inode_info *ei = EXT4_I(inode);
803 ext4_fsblk_t first_block = 0;
806 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
807 J_ASSERT(handle != NULL || create == 0);
808 depth = ext4_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
813 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
815 /* Simplest case - block found, no allocation needed */
817 first_block = le32_to_cpu(chain[depth - 1].key);
818 clear_buffer_new(bh_result);
821 while (count < maxblocks && count <= blocks_to_boundary) {
824 if (!verify_chain(chain, partial)) {
826 * Indirect block might be removed by
827 * truncate while we were reading it.
828 * Handling of that case: forget what we've
829 * got now. Flag the err as EAGAIN, so it
836 blk = le32_to_cpu(*(chain[depth-1].p + count));
838 if (blk == first_block + count)
847 /* Next simple case - plain lookup or failed read of indirect block */
848 if (!create || err == -EIO)
851 mutex_lock(&ei->truncate_mutex);
854 * If the indirect block is missing while we are reading
855 * the chain(ext4_get_branch() returns -EAGAIN err), or
856 * if the chain has been changed after we grab the semaphore,
857 * (either because another process truncated this branch, or
858 * another get_block allocated this branch) re-grab the chain to see if
859 * the request block has been allocated or not.
861 * Since we already block the truncate/other get_block
862 * at this point, we will have the current copy of the chain when we
863 * splice the branch into the tree.
865 if (err == -EAGAIN || !verify_chain(chain, partial)) {
866 while (partial > chain) {
870 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
873 mutex_unlock(&ei->truncate_mutex);
876 clear_buffer_new(bh_result);
882 * Okay, we need to do block allocation. Lazily initialize the block
883 * allocation info here if necessary
885 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
886 ext4_init_block_alloc_info(inode);
888 goal = ext4_find_goal(inode, iblock, chain, partial);
890 /* the number of blocks need to allocate for [d,t]indirect blocks */
891 indirect_blks = (chain + depth) - partial - 1;
894 * Next look up the indirect map to count the totoal number of
895 * direct blocks to allocate for this branch.
897 count = ext4_blks_to_allocate(partial, indirect_blks,
898 maxblocks, blocks_to_boundary);
900 * Block out ext4_truncate while we alter the tree
902 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
903 offsets + (partial - chain), partial);
906 * The ext4_splice_branch call will free and forget any buffers
907 * on the new chain if there is a failure, but that risks using
908 * up transaction credits, especially for bitmaps where the
909 * credits cannot be returned. Can we handle this somehow? We
910 * may need to return -EAGAIN upwards in the worst case. --sct
913 err = ext4_splice_branch(handle, inode, iblock,
914 partial, indirect_blks, count);
916 * i_disksize growing is protected by truncate_mutex. Don't forget to
917 * protect it if you're about to implement concurrent
918 * ext4_get_block() -bzzz
920 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
921 ei->i_disksize = inode->i_size;
922 mutex_unlock(&ei->truncate_mutex);
926 set_buffer_new(bh_result);
928 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
929 if (count > blocks_to_boundary)
930 set_buffer_boundary(bh_result);
932 /* Clean up and exit */
933 partial = chain + depth - 1; /* the whole chain */
935 while (partial > chain) {
936 BUFFER_TRACE(partial->bh, "call brelse");
940 BUFFER_TRACE(bh_result, "returned");
945 #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
947 static int ext4_get_block(struct inode *inode, sector_t iblock,
948 struct buffer_head *bh_result, int create)
950 handle_t *handle = journal_current_handle();
952 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
955 goto get_block; /* A read */
958 goto get_block; /* A single block get */
960 if (handle->h_transaction->t_state == T_LOCKED) {
962 * Huge direct-io writes can hold off commits for long
963 * periods of time. Let this commit run.
965 ext4_journal_stop(handle);
966 handle = ext4_journal_start(inode, DIO_CREDITS);
968 ret = PTR_ERR(handle);
972 if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
974 * Getting low on buffer credits...
976 ret = ext4_journal_extend(handle, DIO_CREDITS);
979 * Couldn't extend the transaction. Start a new one.
981 ret = ext4_journal_restart(handle, DIO_CREDITS);
987 ret = ext4_get_blocks_wrap(handle, inode, iblock,
988 max_blocks, bh_result, create, 0);
990 bh_result->b_size = (ret << inode->i_blkbits);
998 * `handle' can be NULL if create is zero
1000 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1001 long block, int create, int *errp)
1003 struct buffer_head dummy;
1006 J_ASSERT(handle != NULL || create == 0);
1009 dummy.b_blocknr = -1000;
1010 buffer_trace_init(&dummy.b_history);
1011 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1014 * ext4_get_blocks_handle() returns number of blocks
1015 * mapped. 0 in case of a HOLE.
1023 if (!err && buffer_mapped(&dummy)) {
1024 struct buffer_head *bh;
1025 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1030 if (buffer_new(&dummy)) {
1031 J_ASSERT(create != 0);
1032 J_ASSERT(handle != 0);
1035 * Now that we do not always journal data, we should
1036 * keep in mind whether this should always journal the
1037 * new buffer as metadata. For now, regular file
1038 * writes use ext4_get_block instead, so it's not a
1042 BUFFER_TRACE(bh, "call get_create_access");
1043 fatal = ext4_journal_get_create_access(handle, bh);
1044 if (!fatal && !buffer_uptodate(bh)) {
1045 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1046 set_buffer_uptodate(bh);
1049 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1050 err = ext4_journal_dirty_metadata(handle, bh);
1054 BUFFER_TRACE(bh, "not a new buffer");
1067 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1068 int block, int create, int *err)
1070 struct buffer_head * bh;
1072 bh = ext4_getblk(handle, inode, block, create, err);
1075 if (buffer_uptodate(bh))
1077 ll_rw_block(READ_META, 1, &bh);
1079 if (buffer_uptodate(bh))
1086 static int walk_page_buffers( handle_t *handle,
1087 struct buffer_head *head,
1091 int (*fn)( handle_t *handle,
1092 struct buffer_head *bh))
1094 struct buffer_head *bh;
1095 unsigned block_start, block_end;
1096 unsigned blocksize = head->b_size;
1098 struct buffer_head *next;
1100 for ( bh = head, block_start = 0;
1101 ret == 0 && (bh != head || !block_start);
1102 block_start = block_end, bh = next)
1104 next = bh->b_this_page;
1105 block_end = block_start + blocksize;
1106 if (block_end <= from || block_start >= to) {
1107 if (partial && !buffer_uptodate(bh))
1111 err = (*fn)(handle, bh);
1119 * To preserve ordering, it is essential that the hole instantiation and
1120 * the data write be encapsulated in a single transaction. We cannot
1121 * close off a transaction and start a new one between the ext4_get_block()
1122 * and the commit_write(). So doing the jbd2_journal_start at the start of
1123 * prepare_write() is the right place.
1125 * Also, this function can nest inside ext4_writepage() ->
1126 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1127 * has generated enough buffer credits to do the whole page. So we won't
1128 * block on the journal in that case, which is good, because the caller may
1131 * By accident, ext4 can be reentered when a transaction is open via
1132 * quota file writes. If we were to commit the transaction while thus
1133 * reentered, there can be a deadlock - we would be holding a quota
1134 * lock, and the commit would never complete if another thread had a
1135 * transaction open and was blocking on the quota lock - a ranking
1138 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1139 * will _not_ run commit under these circumstances because handle->h_ref
1140 * is elevated. We'll still have enough credits for the tiny quotafile
1143 static int do_journal_get_write_access(handle_t *handle,
1144 struct buffer_head *bh)
1146 if (!buffer_mapped(bh) || buffer_freed(bh))
1148 return ext4_journal_get_write_access(handle, bh);
1151 static int ext4_prepare_write(struct file *file, struct page *page,
1152 unsigned from, unsigned to)
1154 struct inode *inode = page->mapping->host;
1155 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1160 handle = ext4_journal_start(inode, needed_blocks);
1161 if (IS_ERR(handle)) {
1162 ret = PTR_ERR(handle);
1165 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1166 ret = nobh_prepare_write(page, from, to, ext4_get_block);
1168 ret = block_prepare_write(page, from, to, ext4_get_block);
1170 goto prepare_write_failed;
1172 if (ext4_should_journal_data(inode)) {
1173 ret = walk_page_buffers(handle, page_buffers(page),
1174 from, to, NULL, do_journal_get_write_access);
1176 prepare_write_failed:
1178 ext4_journal_stop(handle);
1179 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1185 int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1187 int err = jbd2_journal_dirty_data(handle, bh);
1189 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1194 /* For commit_write() in data=journal mode */
1195 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1197 if (!buffer_mapped(bh) || buffer_freed(bh))
1199 set_buffer_uptodate(bh);
1200 return ext4_journal_dirty_metadata(handle, bh);
1204 * We need to pick up the new inode size which generic_commit_write gave us
1205 * `file' can be NULL - eg, when called from page_symlink().
1207 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1208 * buffers are managed internally.
1210 static int ext4_ordered_commit_write(struct file *file, struct page *page,
1211 unsigned from, unsigned to)
1213 handle_t *handle = ext4_journal_current_handle();
1214 struct inode *inode = page->mapping->host;
1217 ret = walk_page_buffers(handle, page_buffers(page),
1218 from, to, NULL, ext4_journal_dirty_data);
1222 * generic_commit_write() will run mark_inode_dirty() if i_size
1223 * changes. So let's piggyback the i_disksize mark_inode_dirty
1228 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1229 if (new_i_size > EXT4_I(inode)->i_disksize)
1230 EXT4_I(inode)->i_disksize = new_i_size;
1231 ret = generic_commit_write(file, page, from, to);
1233 ret2 = ext4_journal_stop(handle);
1239 static int ext4_writeback_commit_write(struct file *file, struct page *page,
1240 unsigned from, unsigned to)
1242 handle_t *handle = ext4_journal_current_handle();
1243 struct inode *inode = page->mapping->host;
1247 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1248 if (new_i_size > EXT4_I(inode)->i_disksize)
1249 EXT4_I(inode)->i_disksize = new_i_size;
1251 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1252 ret = nobh_commit_write(file, page, from, to);
1254 ret = generic_commit_write(file, page, from, to);
1256 ret2 = ext4_journal_stop(handle);
1262 static int ext4_journalled_commit_write(struct file *file,
1263 struct page *page, unsigned from, unsigned to)
1265 handle_t *handle = ext4_journal_current_handle();
1266 struct inode *inode = page->mapping->host;
1272 * Here we duplicate the generic_commit_write() functionality
1274 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1276 ret = walk_page_buffers(handle, page_buffers(page), from,
1277 to, &partial, commit_write_fn);
1279 SetPageUptodate(page);
1280 if (pos > inode->i_size)
1281 i_size_write(inode, pos);
1282 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1283 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1284 EXT4_I(inode)->i_disksize = inode->i_size;
1285 ret2 = ext4_mark_inode_dirty(handle, inode);
1289 ret2 = ext4_journal_stop(handle);
1296 * bmap() is special. It gets used by applications such as lilo and by
1297 * the swapper to find the on-disk block of a specific piece of data.
1299 * Naturally, this is dangerous if the block concerned is still in the
1300 * journal. If somebody makes a swapfile on an ext4 data-journaling
1301 * filesystem and enables swap, then they may get a nasty shock when the
1302 * data getting swapped to that swapfile suddenly gets overwritten by
1303 * the original zero's written out previously to the journal and
1304 * awaiting writeback in the kernel's buffer cache.
1306 * So, if we see any bmap calls here on a modified, data-journaled file,
1307 * take extra steps to flush any blocks which might be in the cache.
1309 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
1311 struct inode *inode = mapping->host;
1315 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
1317 * This is a REALLY heavyweight approach, but the use of
1318 * bmap on dirty files is expected to be extremely rare:
1319 * only if we run lilo or swapon on a freshly made file
1320 * do we expect this to happen.
1322 * (bmap requires CAP_SYS_RAWIO so this does not
1323 * represent an unprivileged user DOS attack --- we'd be
1324 * in trouble if mortal users could trigger this path at
1327 * NB. EXT4_STATE_JDATA is not set on files other than
1328 * regular files. If somebody wants to bmap a directory
1329 * or symlink and gets confused because the buffer
1330 * hasn't yet been flushed to disk, they deserve
1331 * everything they get.
1334 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1335 journal = EXT4_JOURNAL(inode);
1336 jbd2_journal_lock_updates(journal);
1337 err = jbd2_journal_flush(journal);
1338 jbd2_journal_unlock_updates(journal);
1344 return generic_block_bmap(mapping,block,ext4_get_block);
1347 static int bget_one(handle_t *handle, struct buffer_head *bh)
1353 static int bput_one(handle_t *handle, struct buffer_head *bh)
1359 static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1361 if (buffer_mapped(bh))
1362 return ext4_journal_dirty_data(handle, bh);
1367 * Note that we always start a transaction even if we're not journalling
1368 * data. This is to preserve ordering: any hole instantiation within
1369 * __block_write_full_page -> ext4_get_block() should be journalled
1370 * along with the data so we don't crash and then get metadata which
1371 * refers to old data.
1373 * In all journalling modes block_write_full_page() will start the I/O.
1377 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1382 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1384 * Same applies to ext4_get_block(). We will deadlock on various things like
1385 * lock_journal and i_truncate_mutex.
1387 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1390 * 16May01: If we're reentered then journal_current_handle() will be
1391 * non-zero. We simply *return*.
1393 * 1 July 2001: @@@ FIXME:
1394 * In journalled data mode, a data buffer may be metadata against the
1395 * current transaction. But the same file is part of a shared mapping
1396 * and someone does a writepage() on it.
1398 * We will move the buffer onto the async_data list, but *after* it has
1399 * been dirtied. So there's a small window where we have dirty data on
1402 * Note that this only applies to the last partial page in the file. The
1403 * bit which block_write_full_page() uses prepare/commit for. (That's
1404 * broken code anyway: it's wrong for msync()).
1406 * It's a rare case: affects the final partial page, for journalled data
1407 * where the file is subject to bith write() and writepage() in the same
1408 * transction. To fix it we'll need a custom block_write_full_page().
1409 * We'll probably need that anyway for journalling writepage() output.
1411 * We don't honour synchronous mounts for writepage(). That would be
1412 * disastrous. Any write() or metadata operation will sync the fs for
1415 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1416 * we don't need to open a transaction here.
1418 static int ext4_ordered_writepage(struct page *page,
1419 struct writeback_control *wbc)
1421 struct inode *inode = page->mapping->host;
1422 struct buffer_head *page_bufs;
1423 handle_t *handle = NULL;
1427 J_ASSERT(PageLocked(page));
1430 * We give up here if we're reentered, because it might be for a
1431 * different filesystem.
1433 if (ext4_journal_current_handle())
1436 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1438 if (IS_ERR(handle)) {
1439 ret = PTR_ERR(handle);
1443 if (!page_has_buffers(page)) {
1444 create_empty_buffers(page, inode->i_sb->s_blocksize,
1445 (1 << BH_Dirty)|(1 << BH_Uptodate));
1447 page_bufs = page_buffers(page);
1448 walk_page_buffers(handle, page_bufs, 0,
1449 PAGE_CACHE_SIZE, NULL, bget_one);
1451 ret = block_write_full_page(page, ext4_get_block, wbc);
1454 * The page can become unlocked at any point now, and
1455 * truncate can then come in and change things. So we
1456 * can't touch *page from now on. But *page_bufs is
1457 * safe due to elevated refcount.
1461 * And attach them to the current transaction. But only if
1462 * block_write_full_page() succeeded. Otherwise they are unmapped,
1463 * and generally junk.
1466 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1467 NULL, jbd2_journal_dirty_data_fn);
1471 walk_page_buffers(handle, page_bufs, 0,
1472 PAGE_CACHE_SIZE, NULL, bput_one);
1473 err = ext4_journal_stop(handle);
1479 redirty_page_for_writepage(wbc, page);
1484 static int ext4_writeback_writepage(struct page *page,
1485 struct writeback_control *wbc)
1487 struct inode *inode = page->mapping->host;
1488 handle_t *handle = NULL;
1492 if (ext4_journal_current_handle())
1495 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1496 if (IS_ERR(handle)) {
1497 ret = PTR_ERR(handle);
1501 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1502 ret = nobh_writepage(page, ext4_get_block, wbc);
1504 ret = block_write_full_page(page, ext4_get_block, wbc);
1506 err = ext4_journal_stop(handle);
1512 redirty_page_for_writepage(wbc, page);
1517 static int ext4_journalled_writepage(struct page *page,
1518 struct writeback_control *wbc)
1520 struct inode *inode = page->mapping->host;
1521 handle_t *handle = NULL;
1525 if (ext4_journal_current_handle())
1528 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1529 if (IS_ERR(handle)) {
1530 ret = PTR_ERR(handle);
1534 if (!page_has_buffers(page) || PageChecked(page)) {
1536 * It's mmapped pagecache. Add buffers and journal it. There
1537 * doesn't seem much point in redirtying the page here.
1539 ClearPageChecked(page);
1540 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1543 ext4_journal_stop(handle);
1546 ret = walk_page_buffers(handle, page_buffers(page), 0,
1547 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1549 err = walk_page_buffers(handle, page_buffers(page), 0,
1550 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1553 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1557 * It may be a page full of checkpoint-mode buffers. We don't
1558 * really know unless we go poke around in the buffer_heads.
1559 * But block_write_full_page will do the right thing.
1561 ret = block_write_full_page(page, ext4_get_block, wbc);
1563 err = ext4_journal_stop(handle);
1570 redirty_page_for_writepage(wbc, page);
1576 static int ext4_readpage(struct file *file, struct page *page)
1578 return mpage_readpage(page, ext4_get_block);
1582 ext4_readpages(struct file *file, struct address_space *mapping,
1583 struct list_head *pages, unsigned nr_pages)
1585 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
1588 static void ext4_invalidatepage(struct page *page, unsigned long offset)
1590 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1593 * If it's a full truncate we just forget about the pending dirtying
1596 ClearPageChecked(page);
1598 jbd2_journal_invalidatepage(journal, page, offset);
1601 static int ext4_releasepage(struct page *page, gfp_t wait)
1603 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1605 WARN_ON(PageChecked(page));
1606 if (!page_has_buffers(page))
1608 return jbd2_journal_try_to_free_buffers(journal, page, wait);
1612 * If the O_DIRECT write will extend the file then add this inode to the
1613 * orphan list. So recovery will truncate it back to the original size
1614 * if the machine crashes during the write.
1616 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1617 * crashes then stale disk data _may_ be exposed inside the file.
1619 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
1620 const struct iovec *iov, loff_t offset,
1621 unsigned long nr_segs)
1623 struct file *file = iocb->ki_filp;
1624 struct inode *inode = file->f_mapping->host;
1625 struct ext4_inode_info *ei = EXT4_I(inode);
1626 handle_t *handle = NULL;
1629 size_t count = iov_length(iov, nr_segs);
1632 loff_t final_size = offset + count;
1634 handle = ext4_journal_start(inode, DIO_CREDITS);
1635 if (IS_ERR(handle)) {
1636 ret = PTR_ERR(handle);
1639 if (final_size > inode->i_size) {
1640 ret = ext4_orphan_add(handle, inode);
1644 ei->i_disksize = inode->i_size;
1648 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1650 ext4_get_block, NULL);
1653 * Reacquire the handle: ext4_get_block() can restart the transaction
1655 handle = journal_current_handle();
1661 if (orphan && inode->i_nlink)
1662 ext4_orphan_del(handle, inode);
1663 if (orphan && ret > 0) {
1664 loff_t end = offset + ret;
1665 if (end > inode->i_size) {
1666 ei->i_disksize = end;
1667 i_size_write(inode, end);
1669 * We're going to return a positive `ret'
1670 * here due to non-zero-length I/O, so there's
1671 * no way of reporting error returns from
1672 * ext4_mark_inode_dirty() to userspace. So
1675 ext4_mark_inode_dirty(handle, inode);
1678 err = ext4_journal_stop(handle);
1687 * Pages can be marked dirty completely asynchronously from ext4's journalling
1688 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1689 * much here because ->set_page_dirty is called under VFS locks. The page is
1690 * not necessarily locked.
1692 * We cannot just dirty the page and leave attached buffers clean, because the
1693 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1694 * or jbddirty because all the journalling code will explode.
1696 * So what we do is to mark the page "pending dirty" and next time writepage
1697 * is called, propagate that into the buffers appropriately.
1699 static int ext4_journalled_set_page_dirty(struct page *page)
1701 SetPageChecked(page);
1702 return __set_page_dirty_nobuffers(page);
1705 static const struct address_space_operations ext4_ordered_aops = {
1706 .readpage = ext4_readpage,
1707 .readpages = ext4_readpages,
1708 .writepage = ext4_ordered_writepage,
1709 .sync_page = block_sync_page,
1710 .prepare_write = ext4_prepare_write,
1711 .commit_write = ext4_ordered_commit_write,
1713 .invalidatepage = ext4_invalidatepage,
1714 .releasepage = ext4_releasepage,
1715 .direct_IO = ext4_direct_IO,
1716 .migratepage = buffer_migrate_page,
1719 static const struct address_space_operations ext4_writeback_aops = {
1720 .readpage = ext4_readpage,
1721 .readpages = ext4_readpages,
1722 .writepage = ext4_writeback_writepage,
1723 .sync_page = block_sync_page,
1724 .prepare_write = ext4_prepare_write,
1725 .commit_write = ext4_writeback_commit_write,
1727 .invalidatepage = ext4_invalidatepage,
1728 .releasepage = ext4_releasepage,
1729 .direct_IO = ext4_direct_IO,
1730 .migratepage = buffer_migrate_page,
1733 static const struct address_space_operations ext4_journalled_aops = {
1734 .readpage = ext4_readpage,
1735 .readpages = ext4_readpages,
1736 .writepage = ext4_journalled_writepage,
1737 .sync_page = block_sync_page,
1738 .prepare_write = ext4_prepare_write,
1739 .commit_write = ext4_journalled_commit_write,
1740 .set_page_dirty = ext4_journalled_set_page_dirty,
1742 .invalidatepage = ext4_invalidatepage,
1743 .releasepage = ext4_releasepage,
1746 void ext4_set_aops(struct inode *inode)
1748 if (ext4_should_order_data(inode))
1749 inode->i_mapping->a_ops = &ext4_ordered_aops;
1750 else if (ext4_should_writeback_data(inode))
1751 inode->i_mapping->a_ops = &ext4_writeback_aops;
1753 inode->i_mapping->a_ops = &ext4_journalled_aops;
1757 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1758 * up to the end of the block which corresponds to `from'.
1759 * This required during truncate. We need to physically zero the tail end
1760 * of that block so it doesn't yield old data if the file is later grown.
1762 int ext4_block_truncate_page(handle_t *handle, struct page *page,
1763 struct address_space *mapping, loff_t from)
1765 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1766 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1767 unsigned blocksize, iblock, length, pos;
1768 struct inode *inode = mapping->host;
1769 struct buffer_head *bh;
1773 blocksize = inode->i_sb->s_blocksize;
1774 length = blocksize - (offset & (blocksize - 1));
1775 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1778 * For "nobh" option, we can only work if we don't need to
1779 * read-in the page - otherwise we create buffers to do the IO.
1781 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1782 ext4_should_writeback_data(inode) && PageUptodate(page)) {
1783 kaddr = kmap_atomic(page, KM_USER0);
1784 memset(kaddr + offset, 0, length);
1785 flush_dcache_page(page);
1786 kunmap_atomic(kaddr, KM_USER0);
1787 set_page_dirty(page);
1791 if (!page_has_buffers(page))
1792 create_empty_buffers(page, blocksize, 0);
1794 /* Find the buffer that contains "offset" */
1795 bh = page_buffers(page);
1797 while (offset >= pos) {
1798 bh = bh->b_this_page;
1804 if (buffer_freed(bh)) {
1805 BUFFER_TRACE(bh, "freed: skip");
1809 if (!buffer_mapped(bh)) {
1810 BUFFER_TRACE(bh, "unmapped");
1811 ext4_get_block(inode, iblock, bh, 0);
1812 /* unmapped? It's a hole - nothing to do */
1813 if (!buffer_mapped(bh)) {
1814 BUFFER_TRACE(bh, "still unmapped");
1819 /* Ok, it's mapped. Make sure it's up-to-date */
1820 if (PageUptodate(page))
1821 set_buffer_uptodate(bh);
1823 if (!buffer_uptodate(bh)) {
1825 ll_rw_block(READ, 1, &bh);
1827 /* Uhhuh. Read error. Complain and punt. */
1828 if (!buffer_uptodate(bh))
1832 if (ext4_should_journal_data(inode)) {
1833 BUFFER_TRACE(bh, "get write access");
1834 err = ext4_journal_get_write_access(handle, bh);
1839 kaddr = kmap_atomic(page, KM_USER0);
1840 memset(kaddr + offset, 0, length);
1841 flush_dcache_page(page);
1842 kunmap_atomic(kaddr, KM_USER0);
1844 BUFFER_TRACE(bh, "zeroed end of block");
1847 if (ext4_should_journal_data(inode)) {
1848 err = ext4_journal_dirty_metadata(handle, bh);
1850 if (ext4_should_order_data(inode))
1851 err = ext4_journal_dirty_data(handle, bh);
1852 mark_buffer_dirty(bh);
1857 page_cache_release(page);
1862 * Probably it should be a library function... search for first non-zero word
1863 * or memcmp with zero_page, whatever is better for particular architecture.
1866 static inline int all_zeroes(__le32 *p, __le32 *q)
1875 * ext4_find_shared - find the indirect blocks for partial truncation.
1876 * @inode: inode in question
1877 * @depth: depth of the affected branch
1878 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
1879 * @chain: place to store the pointers to partial indirect blocks
1880 * @top: place to the (detached) top of branch
1882 * This is a helper function used by ext4_truncate().
1884 * When we do truncate() we may have to clean the ends of several
1885 * indirect blocks but leave the blocks themselves alive. Block is
1886 * partially truncated if some data below the new i_size is refered
1887 * from it (and it is on the path to the first completely truncated
1888 * data block, indeed). We have to free the top of that path along
1889 * with everything to the right of the path. Since no allocation
1890 * past the truncation point is possible until ext4_truncate()
1891 * finishes, we may safely do the latter, but top of branch may
1892 * require special attention - pageout below the truncation point
1893 * might try to populate it.
1895 * We atomically detach the top of branch from the tree, store the
1896 * block number of its root in *@top, pointers to buffer_heads of
1897 * partially truncated blocks - in @chain[].bh and pointers to
1898 * their last elements that should not be removed - in
1899 * @chain[].p. Return value is the pointer to last filled element
1902 * The work left to caller to do the actual freeing of subtrees:
1903 * a) free the subtree starting from *@top
1904 * b) free the subtrees whose roots are stored in
1905 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1906 * c) free the subtrees growing from the inode past the @chain[0].
1907 * (no partially truncated stuff there). */
1909 static Indirect *ext4_find_shared(struct inode *inode, int depth,
1910 int offsets[4], Indirect chain[4], __le32 *top)
1912 Indirect *partial, *p;
1916 /* Make k index the deepest non-null offest + 1 */
1917 for (k = depth; k > 1 && !offsets[k-1]; k--)
1919 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1920 /* Writer: pointers */
1922 partial = chain + k-1;
1924 * If the branch acquired continuation since we've looked at it -
1925 * fine, it should all survive and (new) top doesn't belong to us.
1927 if (!partial->key && *partial->p)
1930 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1933 * OK, we've found the last block that must survive. The rest of our
1934 * branch should be detached before unlocking. However, if that rest
1935 * of branch is all ours and does not grow immediately from the inode
1936 * it's easier to cheat and just decrement partial->p.
1938 if (p == chain + k - 1 && p > chain) {
1942 /* Nope, don't do this in ext4. Must leave the tree intact */
1949 while(partial > p) {
1950 brelse(partial->bh);
1958 * Zero a number of block pointers in either an inode or an indirect block.
1959 * If we restart the transaction we must again get write access to the
1960 * indirect block for further modification.
1962 * We release `count' blocks on disk, but (last - first) may be greater
1963 * than `count' because there can be holes in there.
1965 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
1966 struct buffer_head *bh, ext4_fsblk_t block_to_free,
1967 unsigned long count, __le32 *first, __le32 *last)
1970 if (try_to_extend_transaction(handle, inode)) {
1972 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1973 ext4_journal_dirty_metadata(handle, bh);
1975 ext4_mark_inode_dirty(handle, inode);
1976 ext4_journal_test_restart(handle, inode);
1978 BUFFER_TRACE(bh, "retaking write access");
1979 ext4_journal_get_write_access(handle, bh);
1984 * Any buffers which are on the journal will be in memory. We find
1985 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
1986 * on them. We've already detached each block from the file, so
1987 * bforget() in jbd2_journal_forget() should be safe.
1989 * AKPM: turn on bforget in jbd2_journal_forget()!!!
1991 for (p = first; p < last; p++) {
1992 u32 nr = le32_to_cpu(*p);
1994 struct buffer_head *bh;
1997 bh = sb_find_get_block(inode->i_sb, nr);
1998 ext4_forget(handle, 0, inode, bh, nr);
2002 ext4_free_blocks(handle, inode, block_to_free, count);
2006 * ext4_free_data - free a list of data blocks
2007 * @handle: handle for this transaction
2008 * @inode: inode we are dealing with
2009 * @this_bh: indirect buffer_head which contains *@first and *@last
2010 * @first: array of block numbers
2011 * @last: points immediately past the end of array
2013 * We are freeing all blocks refered from that array (numbers are stored as
2014 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2016 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2017 * blocks are contiguous then releasing them at one time will only affect one
2018 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2019 * actually use a lot of journal space.
2021 * @this_bh will be %NULL if @first and @last point into the inode's direct
2024 static void ext4_free_data(handle_t *handle, struct inode *inode,
2025 struct buffer_head *this_bh,
2026 __le32 *first, __le32 *last)
2028 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
2029 unsigned long count = 0; /* Number of blocks in the run */
2030 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2033 ext4_fsblk_t nr; /* Current block # */
2034 __le32 *p; /* Pointer into inode/ind
2035 for current block */
2038 if (this_bh) { /* For indirect block */
2039 BUFFER_TRACE(this_bh, "get_write_access");
2040 err = ext4_journal_get_write_access(handle, this_bh);
2041 /* Important: if we can't update the indirect pointers
2042 * to the blocks, we can't free them. */
2047 for (p = first; p < last; p++) {
2048 nr = le32_to_cpu(*p);
2050 /* accumulate blocks to free if they're contiguous */
2053 block_to_free_p = p;
2055 } else if (nr == block_to_free + count) {
2058 ext4_clear_blocks(handle, inode, this_bh,
2060 count, block_to_free_p, p);
2062 block_to_free_p = p;
2069 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
2070 count, block_to_free_p, p);
2073 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2074 ext4_journal_dirty_metadata(handle, this_bh);
2079 * ext4_free_branches - free an array of branches
2080 * @handle: JBD handle for this transaction
2081 * @inode: inode we are dealing with
2082 * @parent_bh: the buffer_head which contains *@first and *@last
2083 * @first: array of block numbers
2084 * @last: pointer immediately past the end of array
2085 * @depth: depth of the branches to free
2087 * We are freeing all blocks refered from these branches (numbers are
2088 * stored as little-endian 32-bit) and updating @inode->i_blocks
2091 static void ext4_free_branches(handle_t *handle, struct inode *inode,
2092 struct buffer_head *parent_bh,
2093 __le32 *first, __le32 *last, int depth)
2098 if (is_handle_aborted(handle))
2102 struct buffer_head *bh;
2103 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2105 while (--p >= first) {
2106 nr = le32_to_cpu(*p);
2108 continue; /* A hole */
2110 /* Go read the buffer for the next level down */
2111 bh = sb_bread(inode->i_sb, nr);
2114 * A read failure? Report error and clear slot
2118 ext4_error(inode->i_sb, "ext4_free_branches",
2119 "Read failure, inode=%lu, block=%llu",
2124 /* This zaps the entire block. Bottom up. */
2125 BUFFER_TRACE(bh, "free child branches");
2126 ext4_free_branches(handle, inode, bh,
2127 (__le32*)bh->b_data,
2128 (__le32*)bh->b_data + addr_per_block,
2132 * We've probably journalled the indirect block several
2133 * times during the truncate. But it's no longer
2134 * needed and we now drop it from the transaction via
2135 * jbd2_journal_revoke().
2137 * That's easy if it's exclusively part of this
2138 * transaction. But if it's part of the committing
2139 * transaction then jbd2_journal_forget() will simply
2140 * brelse() it. That means that if the underlying
2141 * block is reallocated in ext4_get_block(),
2142 * unmap_underlying_metadata() will find this block
2143 * and will try to get rid of it. damn, damn.
2145 * If this block has already been committed to the
2146 * journal, a revoke record will be written. And
2147 * revoke records must be emitted *before* clearing
2148 * this block's bit in the bitmaps.
2150 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
2153 * Everything below this this pointer has been
2154 * released. Now let this top-of-subtree go.
2156 * We want the freeing of this indirect block to be
2157 * atomic in the journal with the updating of the
2158 * bitmap block which owns it. So make some room in
2161 * We zero the parent pointer *after* freeing its
2162 * pointee in the bitmaps, so if extend_transaction()
2163 * for some reason fails to put the bitmap changes and
2164 * the release into the same transaction, recovery
2165 * will merely complain about releasing a free block,
2166 * rather than leaking blocks.
2168 if (is_handle_aborted(handle))
2170 if (try_to_extend_transaction(handle, inode)) {
2171 ext4_mark_inode_dirty(handle, inode);
2172 ext4_journal_test_restart(handle, inode);
2175 ext4_free_blocks(handle, inode, nr, 1);
2179 * The block which we have just freed is
2180 * pointed to by an indirect block: journal it
2182 BUFFER_TRACE(parent_bh, "get_write_access");
2183 if (!ext4_journal_get_write_access(handle,
2186 BUFFER_TRACE(parent_bh,
2187 "call ext4_journal_dirty_metadata");
2188 ext4_journal_dirty_metadata(handle,
2194 /* We have reached the bottom of the tree. */
2195 BUFFER_TRACE(parent_bh, "free data blocks");
2196 ext4_free_data(handle, inode, parent_bh, first, last);
2203 * We block out ext4_get_block() block instantiations across the entire
2204 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2205 * simultaneously on behalf of the same inode.
2207 * As we work through the truncate and commmit bits of it to the journal there
2208 * is one core, guiding principle: the file's tree must always be consistent on
2209 * disk. We must be able to restart the truncate after a crash.
2211 * The file's tree may be transiently inconsistent in memory (although it
2212 * probably isn't), but whenever we close off and commit a journal transaction,
2213 * the contents of (the filesystem + the journal) must be consistent and
2214 * restartable. It's pretty simple, really: bottom up, right to left (although
2215 * left-to-right works OK too).
2217 * Note that at recovery time, journal replay occurs *before* the restart of
2218 * truncate against the orphan inode list.
2220 * The committed inode has the new, desired i_size (which is the same as
2221 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
2222 * that this inode's truncate did not complete and it will again call
2223 * ext4_truncate() to have another go. So there will be instantiated blocks
2224 * to the right of the truncation point in a crashed ext4 filesystem. But
2225 * that's fine - as long as they are linked from the inode, the post-crash
2226 * ext4_truncate() run will find them and release them.
2228 void ext4_truncate(struct inode *inode)
2231 struct ext4_inode_info *ei = EXT4_I(inode);
2232 __le32 *i_data = ei->i_data;
2233 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2234 struct address_space *mapping = inode->i_mapping;
2241 unsigned blocksize = inode->i_sb->s_blocksize;
2244 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2245 S_ISLNK(inode->i_mode)))
2247 if (ext4_inode_is_fast_symlink(inode))
2249 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2253 * We have to lock the EOF page here, because lock_page() nests
2254 * outside jbd2_journal_start().
2256 if ((inode->i_size & (blocksize - 1)) == 0) {
2257 /* Block boundary? Nothing to do */
2260 page = grab_cache_page(mapping,
2261 inode->i_size >> PAGE_CACHE_SHIFT);
2266 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2267 return ext4_ext_truncate(inode, page);
2269 handle = start_transaction(inode);
2270 if (IS_ERR(handle)) {
2272 clear_highpage(page);
2273 flush_dcache_page(page);
2275 page_cache_release(page);
2277 return; /* AKPM: return what? */
2280 last_block = (inode->i_size + blocksize-1)
2281 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
2284 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2286 n = ext4_block_to_path(inode, last_block, offsets, NULL);
2288 goto out_stop; /* error */
2291 * OK. This truncate is going to happen. We add the inode to the
2292 * orphan list, so that if this truncate spans multiple transactions,
2293 * and we crash, we will resume the truncate when the filesystem
2294 * recovers. It also marks the inode dirty, to catch the new size.
2296 * Implication: the file must always be in a sane, consistent
2297 * truncatable state while each transaction commits.
2299 if (ext4_orphan_add(handle, inode))
2303 * The orphan list entry will now protect us from any crash which
2304 * occurs before the truncate completes, so it is now safe to propagate
2305 * the new, shorter inode size (held for now in i_size) into the
2306 * on-disk inode. We do this via i_disksize, which is the value which
2307 * ext4 *really* writes onto the disk inode.
2309 ei->i_disksize = inode->i_size;
2312 * From here we block out all ext4_get_block() callers who want to
2313 * modify the block allocation tree.
2315 mutex_lock(&ei->truncate_mutex);
2317 if (n == 1) { /* direct blocks */
2318 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2319 i_data + EXT4_NDIR_BLOCKS);
2323 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
2324 /* Kill the top of shared branch (not detached) */
2326 if (partial == chain) {
2327 /* Shared branch grows from the inode */
2328 ext4_free_branches(handle, inode, NULL,
2329 &nr, &nr+1, (chain+n-1) - partial);
2332 * We mark the inode dirty prior to restart,
2333 * and prior to stop. No need for it here.
2336 /* Shared branch grows from an indirect block */
2337 BUFFER_TRACE(partial->bh, "get_write_access");
2338 ext4_free_branches(handle, inode, partial->bh,
2340 partial->p+1, (chain+n-1) - partial);
2343 /* Clear the ends of indirect blocks on the shared branch */
2344 while (partial > chain) {
2345 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
2346 (__le32*)partial->bh->b_data+addr_per_block,
2347 (chain+n-1) - partial);
2348 BUFFER_TRACE(partial->bh, "call brelse");
2349 brelse (partial->bh);
2353 /* Kill the remaining (whole) subtrees */
2354 switch (offsets[0]) {
2356 nr = i_data[EXT4_IND_BLOCK];
2358 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2359 i_data[EXT4_IND_BLOCK] = 0;
2361 case EXT4_IND_BLOCK:
2362 nr = i_data[EXT4_DIND_BLOCK];
2364 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2365 i_data[EXT4_DIND_BLOCK] = 0;
2367 case EXT4_DIND_BLOCK:
2368 nr = i_data[EXT4_TIND_BLOCK];
2370 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2371 i_data[EXT4_TIND_BLOCK] = 0;
2373 case EXT4_TIND_BLOCK:
2377 ext4_discard_reservation(inode);
2379 mutex_unlock(&ei->truncate_mutex);
2380 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
2381 ext4_mark_inode_dirty(handle, inode);
2384 * In a multi-transaction truncate, we only make the final transaction
2391 * If this was a simple ftruncate(), and the file will remain alive
2392 * then we need to clear up the orphan record which we created above.
2393 * However, if this was a real unlink then we were called by
2394 * ext4_delete_inode(), and we allow that function to clean up the
2395 * orphan info for us.
2398 ext4_orphan_del(handle, inode);
2400 ext4_journal_stop(handle);
2403 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2404 unsigned long ino, struct ext4_iloc *iloc)
2406 unsigned long desc, group_desc, block_group;
2407 unsigned long offset;
2409 struct buffer_head *bh;
2410 struct ext4_group_desc * gdp;
2412 if (!ext4_valid_inum(sb, ino)) {
2414 * This error is already checked for in namei.c unless we are
2415 * looking at an NFS filehandle, in which case no error
2421 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2422 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2423 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
2427 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2428 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2429 bh = EXT4_SB(sb)->s_group_desc[group_desc];
2431 ext4_error (sb, "ext4_get_inode_block",
2432 "Descriptor not loaded");
2436 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2437 desc * EXT4_DESC_SIZE(sb));
2439 * Figure out the offset within the block group inode table
2441 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2442 EXT4_INODE_SIZE(sb);
2443 block = ext4_inode_table(sb, gdp) +
2444 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
2446 iloc->block_group = block_group;
2447 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
2452 * ext4_get_inode_loc returns with an extra refcount against the inode's
2453 * underlying buffer_head on success. If 'in_mem' is true, we have all
2454 * data in memory that is needed to recreate the on-disk version of this
2457 static int __ext4_get_inode_loc(struct inode *inode,
2458 struct ext4_iloc *iloc, int in_mem)
2461 struct buffer_head *bh;
2463 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2467 bh = sb_getblk(inode->i_sb, block);
2469 ext4_error (inode->i_sb, "ext4_get_inode_loc",
2470 "unable to read inode block - "
2471 "inode=%lu, block=%llu",
2472 inode->i_ino, block);
2475 if (!buffer_uptodate(bh)) {
2477 if (buffer_uptodate(bh)) {
2478 /* someone brought it uptodate while we waited */
2484 * If we have all information of the inode in memory and this
2485 * is the only valid inode in the block, we need not read the
2489 struct buffer_head *bitmap_bh;
2490 struct ext4_group_desc *desc;
2491 int inodes_per_buffer;
2492 int inode_offset, i;
2496 block_group = (inode->i_ino - 1) /
2497 EXT4_INODES_PER_GROUP(inode->i_sb);
2498 inodes_per_buffer = bh->b_size /
2499 EXT4_INODE_SIZE(inode->i_sb);
2500 inode_offset = ((inode->i_ino - 1) %
2501 EXT4_INODES_PER_GROUP(inode->i_sb));
2502 start = inode_offset & ~(inodes_per_buffer - 1);
2504 /* Is the inode bitmap in cache? */
2505 desc = ext4_get_group_desc(inode->i_sb,
2510 bitmap_bh = sb_getblk(inode->i_sb,
2511 ext4_inode_bitmap(inode->i_sb, desc));
2516 * If the inode bitmap isn't in cache then the
2517 * optimisation may end up performing two reads instead
2518 * of one, so skip it.
2520 if (!buffer_uptodate(bitmap_bh)) {
2524 for (i = start; i < start + inodes_per_buffer; i++) {
2525 if (i == inode_offset)
2527 if (ext4_test_bit(i, bitmap_bh->b_data))
2531 if (i == start + inodes_per_buffer) {
2532 /* all other inodes are free, so skip I/O */
2533 memset(bh->b_data, 0, bh->b_size);
2534 set_buffer_uptodate(bh);
2542 * There are other valid inodes in the buffer, this inode
2543 * has in-inode xattrs, or we don't have this inode in memory.
2544 * Read the block from disk.
2547 bh->b_end_io = end_buffer_read_sync;
2548 submit_bh(READ_META, bh);
2550 if (!buffer_uptodate(bh)) {
2551 ext4_error(inode->i_sb, "ext4_get_inode_loc",
2552 "unable to read inode block - "
2553 "inode=%lu, block=%llu",
2554 inode->i_ino, block);
2564 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
2566 /* We have all inode data except xattrs in memory here. */
2567 return __ext4_get_inode_loc(inode, iloc,
2568 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
2571 void ext4_set_inode_flags(struct inode *inode)
2573 unsigned int flags = EXT4_I(inode)->i_flags;
2575 inode->i_flags &= ~(S_IMMUTABLE | S_IUNLINK | S_BARRIER |
2576 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
2578 if (flags & EXT4_IMMUTABLE_FL)
2579 inode->i_flags |= S_IMMUTABLE;
2580 if (flags & EXT4_IUNLINK_FL)
2581 inode->i_flags |= S_IUNLINK;
2582 if (flags & EXT4_BARRIER_FL)
2583 inode->i_flags |= S_BARRIER;
2585 if (flags & EXT4_SYNC_FL)
2586 inode->i_flags |= S_SYNC;
2587 if (flags & EXT4_APPEND_FL)
2588 inode->i_flags |= S_APPEND;
2589 if (flags & EXT4_NOATIME_FL)
2590 inode->i_flags |= S_NOATIME;
2591 if (flags & EXT4_DIRSYNC_FL)
2592 inode->i_flags |= S_DIRSYNC;
2595 int ext4_sync_flags(struct inode *inode)
2597 unsigned int oldflags, newflags;
2600 oldflags = EXT4_I(inode)->i_flags;
2601 newflags = oldflags & ~(EXT4_IMMUTABLE_FL |
2602 EXT4_IUNLINK_FL | EXT4_BARRIER_FL);
2604 if (IS_IMMUTABLE(inode))
2605 newflags |= EXT4_IMMUTABLE_FL;
2606 if (IS_IUNLINK(inode))
2607 newflags |= EXT4_IUNLINK_FL;
2608 if (IS_BARRIER(inode))
2609 newflags |= EXT4_BARRIER_FL;
2611 if (oldflags ^ newflags) {
2613 struct ext4_iloc iloc;
2615 handle = ext4_journal_start(inode, 1);
2617 return PTR_ERR(handle);
2620 err = ext4_reserve_inode_write(handle, inode, &iloc);
2624 EXT4_I(inode)->i_flags = newflags;
2625 inode->i_ctime = CURRENT_TIME;
2627 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2629 ext4_journal_stop(handle);
2634 void ext4_read_inode(struct inode * inode)
2636 struct ext4_iloc iloc;
2637 struct ext4_inode *raw_inode;
2638 struct ext4_inode_info *ei = EXT4_I(inode);
2639 struct buffer_head *bh;
2644 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2645 ei->i_acl = EXT4_ACL_NOT_CACHED;
2646 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
2648 ei->i_block_alloc_info = NULL;
2650 if (__ext4_get_inode_loc(inode, &iloc, 0))
2653 raw_inode = ext4_raw_inode(&iloc);
2654 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2655 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2656 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2657 if(!(test_opt (inode->i_sb, NO_UID32))) {
2658 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2659 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2661 inode->i_uid = INOTAG_UID(DX_TAG(inode), uid, gid);
2662 inode->i_gid = INOTAG_GID(DX_TAG(inode), uid, gid);
2663 inode->i_tag = INOTAG_TAG(DX_TAG(inode), uid, gid,
2664 le16_to_cpu(raw_inode->i_raw_tag));
2666 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2667 inode->i_size = le32_to_cpu(raw_inode->i_size);
2668 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2669 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2670 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2671 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2674 ei->i_dir_start_lookup = 0;
2675 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2676 /* We now have enough fields to check if the inode was active or not.
2677 * This is needed because nfsd might try to access dead inodes
2678 * the test is that same one that e2fsck uses
2679 * NeilBrown 1999oct15
2681 if (inode->i_nlink == 0) {
2682 if (inode->i_mode == 0 ||
2683 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
2684 /* this inode is deleted */
2688 /* The only unlinked inodes we let through here have
2689 * valid i_mode and are being read by the orphan
2690 * recovery code: that's fine, we're about to complete
2691 * the process of deleting those. */
2693 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2694 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2695 #ifdef EXT4_FRAGMENTS
2696 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2697 ei->i_frag_no = raw_inode->i_frag;
2698 ei->i_frag_size = raw_inode->i_fsize;
2700 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2701 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2702 cpu_to_le32(EXT4_OS_HURD))
2704 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
2705 if (!S_ISREG(inode->i_mode)) {
2706 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2709 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2711 ei->i_disksize = inode->i_size;
2712 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2713 ei->i_block_group = iloc.block_group;
2715 * NOTE! The in-memory inode i_data array is in little-endian order
2716 * even on big-endian machines: we do NOT byteswap the block numbers!
2718 for (block = 0; block < EXT4_N_BLOCKS; block++)
2719 ei->i_data[block] = raw_inode->i_block[block];
2720 INIT_LIST_HEAD(&ei->i_orphan);
2722 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2723 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
2725 * When mke2fs creates big inodes it does not zero out
2726 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
2727 * so ignore those first few inodes.
2729 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2730 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2731 EXT4_INODE_SIZE(inode->i_sb))
2733 if (ei->i_extra_isize == 0) {
2734 /* The extra space is currently unused. Use it. */
2735 ei->i_extra_isize = sizeof(struct ext4_inode) -
2736 EXT4_GOOD_OLD_INODE_SIZE;
2738 __le32 *magic = (void *)raw_inode +
2739 EXT4_GOOD_OLD_INODE_SIZE +
2741 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2742 ei->i_state |= EXT4_STATE_XATTR;
2745 ei->i_extra_isize = 0;
2747 if (S_ISREG(inode->i_mode)) {
2748 inode->i_op = &ext4_file_inode_operations;
2749 inode->i_fop = &ext4_file_operations;
2750 ext4_set_aops(inode);
2751 } else if (S_ISDIR(inode->i_mode)) {
2752 inode->i_op = &ext4_dir_inode_operations;
2753 inode->i_fop = &ext4_dir_operations;
2754 } else if (S_ISLNK(inode->i_mode)) {
2755 if (ext4_inode_is_fast_symlink(inode))
2756 inode->i_op = &ext4_fast_symlink_inode_operations;
2758 inode->i_op = &ext4_symlink_inode_operations;
2759 ext4_set_aops(inode);
2762 inode->i_op = &ext4_special_inode_operations;
2763 if (raw_inode->i_block[0])
2764 init_special_inode(inode, inode->i_mode,
2765 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2767 init_special_inode(inode, inode->i_mode,
2768 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2771 ext4_set_inode_flags(inode);
2775 make_bad_inode(inode);
2780 * Post the struct inode info into an on-disk inode location in the
2781 * buffer-cache. This gobbles the caller's reference to the
2782 * buffer_head in the inode location struct.
2784 * The caller must have write access to iloc->bh.
2786 static int ext4_do_update_inode(handle_t *handle,
2787 struct inode *inode,
2788 struct ext4_iloc *iloc)
2790 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2791 struct ext4_inode_info *ei = EXT4_I(inode);
2792 struct buffer_head *bh = iloc->bh;
2793 uid_t uid = TAGINO_UID(DX_TAG(inode), inode->i_uid, inode->i_tag);
2794 gid_t gid = TAGINO_GID(DX_TAG(inode), inode->i_gid, inode->i_tag);
2795 int err = 0, rc, block;
2797 /* For fields not not tracking in the in-memory inode,
2798 * initialise them to zero for new inodes. */
2799 if (ei->i_state & EXT4_STATE_NEW)
2800 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2802 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2803 if(!(test_opt(inode->i_sb, NO_UID32))) {
2804 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2805 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2807 * Fix up interoperability with old kernels. Otherwise, old inodes get
2808 * re-used with the upper 16 bits of the uid/gid intact
2811 raw_inode->i_uid_high =
2812 cpu_to_le16(high_16_bits(uid));
2813 raw_inode->i_gid_high =
2814 cpu_to_le16(high_16_bits(gid));
2816 raw_inode->i_uid_high = 0;
2817 raw_inode->i_gid_high = 0;
2820 raw_inode->i_uid_low =
2821 cpu_to_le16(fs_high2lowuid(uid));
2822 raw_inode->i_gid_low =
2823 cpu_to_le16(fs_high2lowgid(gid));
2824 raw_inode->i_uid_high = 0;
2825 raw_inode->i_gid_high = 0;
2827 #ifdef CONFIG_TAGGING_INTERN
2828 raw_inode->i_raw_tag = cpu_to_le16(inode->i_tag);
2830 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2831 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2832 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2833 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2834 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2835 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2836 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2837 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2838 #ifdef EXT4_FRAGMENTS
2839 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2840 raw_inode->i_frag = ei->i_frag_no;
2841 raw_inode->i_fsize = ei->i_frag_size;
2843 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2844 cpu_to_le32(EXT4_OS_HURD))
2845 raw_inode->i_file_acl_high =
2846 cpu_to_le16(ei->i_file_acl >> 32);
2847 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2848 if (!S_ISREG(inode->i_mode)) {
2849 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2851 raw_inode->i_size_high =
2852 cpu_to_le32(ei->i_disksize >> 32);
2853 if (ei->i_disksize > 0x7fffffffULL) {
2854 struct super_block *sb = inode->i_sb;
2855 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2856 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2857 EXT4_SB(sb)->s_es->s_rev_level ==
2858 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
2859 /* If this is the first large file
2860 * created, add a flag to the superblock.
2862 err = ext4_journal_get_write_access(handle,
2863 EXT4_SB(sb)->s_sbh);
2866 ext4_update_dynamic_rev(sb);
2867 EXT4_SET_RO_COMPAT_FEATURE(sb,
2868 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
2871 err = ext4_journal_dirty_metadata(handle,
2872 EXT4_SB(sb)->s_sbh);
2876 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2877 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2878 if (old_valid_dev(inode->i_rdev)) {
2879 raw_inode->i_block[0] =
2880 cpu_to_le32(old_encode_dev(inode->i_rdev));
2881 raw_inode->i_block[1] = 0;
2883 raw_inode->i_block[0] = 0;
2884 raw_inode->i_block[1] =
2885 cpu_to_le32(new_encode_dev(inode->i_rdev));
2886 raw_inode->i_block[2] = 0;
2888 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
2889 raw_inode->i_block[block] = ei->i_data[block];
2891 if (ei->i_extra_isize)
2892 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2894 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2895 rc = ext4_journal_dirty_metadata(handle, bh);
2898 ei->i_state &= ~EXT4_STATE_NEW;
2902 ext4_std_error(inode->i_sb, err);
2907 * ext4_write_inode()
2909 * We are called from a few places:
2911 * - Within generic_file_write() for O_SYNC files.
2912 * Here, there will be no transaction running. We wait for any running
2913 * trasnaction to commit.
2915 * - Within sys_sync(), kupdate and such.
2916 * We wait on commit, if tol to.
2918 * - Within prune_icache() (PF_MEMALLOC == true)
2919 * Here we simply return. We can't afford to block kswapd on the
2922 * In all cases it is actually safe for us to return without doing anything,
2923 * because the inode has been copied into a raw inode buffer in
2924 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2927 * Note that we are absolutely dependent upon all inode dirtiers doing the
2928 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2929 * which we are interested.
2931 * It would be a bug for them to not do this. The code:
2933 * mark_inode_dirty(inode)
2935 * inode->i_size = expr;
2937 * is in error because a kswapd-driven write_inode() could occur while
2938 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2939 * will no longer be on the superblock's dirty inode list.
2941 int ext4_write_inode(struct inode *inode, int wait)
2943 if (current->flags & PF_MEMALLOC)
2946 if (ext4_journal_current_handle()) {
2947 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2955 return ext4_force_commit(inode->i_sb);
2961 * Called from notify_change.
2963 * We want to trap VFS attempts to truncate the file as soon as
2964 * possible. In particular, we want to make sure that when the VFS
2965 * shrinks i_size, we put the inode on the orphan list and modify
2966 * i_disksize immediately, so that during the subsequent flushing of
2967 * dirty pages and freeing of disk blocks, we can guarantee that any
2968 * commit will leave the blocks being flushed in an unused state on
2969 * disk. (On recovery, the inode will get truncated and the blocks will
2970 * be freed, so we have a strong guarantee that no future commit will
2971 * leave these blocks visible to the user.)
2973 * Called with inode->sem down.
2975 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
2977 struct inode *inode = dentry->d_inode;
2979 const unsigned int ia_valid = attr->ia_valid;
2981 error = inode_change_ok(inode, attr);
2985 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2986 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
2987 (ia_valid & ATTR_TAG && attr->ia_tag != inode->i_tag)) {
2990 /* (user+group)*(old+new) structure, inode write (sb,
2991 * inode block, ? - but truncate inode update has it) */
2992 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
2993 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
2994 if (IS_ERR(handle)) {
2995 error = PTR_ERR(handle);
2998 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3000 ext4_journal_stop(handle);
3003 /* Update corresponding info in inode so that everything is in
3004 * one transaction */
3005 if (attr->ia_valid & ATTR_UID)
3006 inode->i_uid = attr->ia_uid;
3007 if (attr->ia_valid & ATTR_GID)
3008 inode->i_gid = attr->ia_gid;
3009 if ((attr->ia_valid & ATTR_TAG) && IS_TAGGED(inode))
3010 inode->i_tag = attr->ia_tag;
3011 error = ext4_mark_inode_dirty(handle, inode);
3012 ext4_journal_stop(handle);
3015 if (S_ISREG(inode->i_mode) &&
3016 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3019 handle = ext4_journal_start(inode, 3);
3020 if (IS_ERR(handle)) {
3021 error = PTR_ERR(handle);
3025 error = ext4_orphan_add(handle, inode);
3026 EXT4_I(inode)->i_disksize = attr->ia_size;
3027 rc = ext4_mark_inode_dirty(handle, inode);
3030 ext4_journal_stop(handle);
3033 rc = inode_setattr(inode, attr);
3035 /* If inode_setattr's call to ext4_truncate failed to get a
3036 * transaction handle at all, we need to clean up the in-core
3037 * orphan list manually. */
3039 ext4_orphan_del(NULL, inode);
3041 if (!rc && (ia_valid & ATTR_MODE))
3042 rc = ext4_acl_chmod(inode);
3045 ext4_std_error(inode->i_sb, error);
3053 * How many blocks doth make a writepage()?
3055 * With N blocks per page, it may be:
3060 * N+5 bitmap blocks (from the above)
3061 * N+5 group descriptor summary blocks
3064 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3066 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3068 * With ordered or writeback data it's the same, less the N data blocks.
3070 * If the inode's direct blocks can hold an integral number of pages then a
3071 * page cannot straddle two indirect blocks, and we can only touch one indirect
3072 * and dindirect block, and the "5" above becomes "3".
3074 * This still overestimates under most circumstances. If we were to pass the
3075 * start and end offsets in here as well we could do block_to_path() on each
3076 * block and work out the exact number of indirects which are touched. Pah.
3079 int ext4_writepage_trans_blocks(struct inode *inode)
3081 int bpp = ext4_journal_blocks_per_page(inode);
3082 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
3085 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3086 return ext4_ext_writepage_trans_blocks(inode, bpp);
3088 if (ext4_should_journal_data(inode))
3089 ret = 3 * (bpp + indirects) + 2;
3091 ret = 2 * (bpp + indirects) + 2;
3094 /* We know that structure was already allocated during DQUOT_INIT so
3095 * we will be updating only the data blocks + inodes */
3096 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
3103 * The caller must have previously called ext4_reserve_inode_write().
3104 * Give this, we know that the caller already has write access to iloc->bh.
3106 int ext4_mark_iloc_dirty(handle_t *handle,
3107 struct inode *inode, struct ext4_iloc *iloc)
3111 /* the do_update_inode consumes one bh->b_count */
3114 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3115 err = ext4_do_update_inode(handle, inode, iloc);
3121 * On success, We end up with an outstanding reference count against
3122 * iloc->bh. This _must_ be cleaned up later.
3126 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3127 struct ext4_iloc *iloc)
3131 err = ext4_get_inode_loc(inode, iloc);
3133 BUFFER_TRACE(iloc->bh, "get_write_access");
3134 err = ext4_journal_get_write_access(handle, iloc->bh);
3141 ext4_std_error(inode->i_sb, err);
3146 * What we do here is to mark the in-core inode as clean with respect to inode
3147 * dirtiness (it may still be data-dirty).
3148 * This means that the in-core inode may be reaped by prune_icache
3149 * without having to perform any I/O. This is a very good thing,
3150 * because *any* task may call prune_icache - even ones which
3151 * have a transaction open against a different journal.
3153 * Is this cheating? Not really. Sure, we haven't written the
3154 * inode out, but prune_icache isn't a user-visible syncing function.
3155 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3156 * we start and wait on commits.
3158 * Is this efficient/effective? Well, we're being nice to the system
3159 * by cleaning up our inodes proactively so they can be reaped
3160 * without I/O. But we are potentially leaving up to five seconds'
3161 * worth of inodes floating about which prune_icache wants us to
3162 * write out. One way to fix that would be to get prune_icache()
3163 * to do a write_super() to free up some memory. It has the desired
3166 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
3168 struct ext4_iloc iloc;
3172 err = ext4_reserve_inode_write(handle, inode, &iloc);
3174 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3179 * ext4_dirty_inode() is called from __mark_inode_dirty()
3181 * We're really interested in the case where a file is being extended.
3182 * i_size has been changed by generic_commit_write() and we thus need
3183 * to include the updated inode in the current transaction.
3185 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3186 * are allocated to the file.
3188 * If the inode is marked synchronous, we don't honour that here - doing
3189 * so would cause a commit on atime updates, which we don't bother doing.
3190 * We handle synchronous inodes at the highest possible level.
3192 void ext4_dirty_inode(struct inode *inode)
3194 handle_t *current_handle = ext4_journal_current_handle();
3197 handle = ext4_journal_start(inode, 2);
3200 if (current_handle &&
3201 current_handle->h_transaction != handle->h_transaction) {
3202 /* This task has a transaction open against a different fs */
3203 printk(KERN_EMERG "%s: transactions do not match!\n",
3206 jbd_debug(5, "marking dirty. outer handle=%p\n",
3208 ext4_mark_inode_dirty(handle, inode);
3210 ext4_journal_stop(handle);
3217 * Bind an inode's backing buffer_head into this transaction, to prevent
3218 * it from being flushed to disk early. Unlike
3219 * ext4_reserve_inode_write, this leaves behind no bh reference and
3220 * returns no iloc structure, so the caller needs to repeat the iloc
3221 * lookup to mark the inode dirty later.
3223 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
3225 struct ext4_iloc iloc;
3229 err = ext4_get_inode_loc(inode, &iloc);
3231 BUFFER_TRACE(iloc.bh, "get_write_access");
3232 err = jbd2_journal_get_write_access(handle, iloc.bh);
3234 err = ext4_journal_dirty_metadata(handle,
3239 ext4_std_error(inode->i_sb, err);
3244 int ext4_change_inode_journal_flag(struct inode *inode, int val)
3251 * We have to be very careful here: changing a data block's
3252 * journaling status dynamically is dangerous. If we write a
3253 * data block to the journal, change the status and then delete
3254 * that block, we risk forgetting to revoke the old log record
3255 * from the journal and so a subsequent replay can corrupt data.
3256 * So, first we make sure that the journal is empty and that
3257 * nobody is changing anything.
3260 journal = EXT4_JOURNAL(inode);
3261 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3264 jbd2_journal_lock_updates(journal);
3265 jbd2_journal_flush(journal);
3268 * OK, there are no updates running now, and all cached data is
3269 * synced to disk. We are now in a completely consistent state
3270 * which doesn't have anything in the journal, and we know that
3271 * no filesystem updates are running, so it is safe to modify
3272 * the inode's in-core data-journaling state flag now.
3276 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
3278 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3279 ext4_set_aops(inode);
3281 jbd2_journal_unlock_updates(journal);
3283 /* Finally we can mark the inode as dirty. */
3285 handle = ext4_journal_start(inode, 1);
3287 return PTR_ERR(handle);
3289 err = ext4_mark_inode_dirty(handle, inode);
3291 ext4_journal_stop(handle);
3292 ext4_std_error(inode->i_sb, err);