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/highuid.h>
31 #include <linux/pagemap.h>
32 #include <linux/quotaops.h>
33 #include <linux/string.h>
34 #include <linux/buffer_head.h>
35 #include <linux/writeback.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
43 * Test whether an inode is a fast symlink.
45 static int ext4_inode_is_fast_symlink(struct inode *inode)
47 int ea_blocks = EXT4_I(inode)->i_file_acl ?
48 (inode->i_sb->s_blocksize >> 9) : 0;
50 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
54 * The ext4 forget function must perform a revoke if we are freeing data
55 * which has been journaled. Metadata (eg. indirect blocks) must be
56 * revoked in all cases.
58 * "bh" may be NULL: a metadata block may have been freed from memory
59 * but there may still be a record of it in the journal, and that record
60 * still needs to be revoked.
62 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
63 struct buffer_head *bh, ext4_fsblk_t blocknr)
69 BUFFER_TRACE(bh, "enter");
71 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
73 bh, is_metadata, inode->i_mode,
74 test_opt(inode->i_sb, DATA_FLAGS));
76 /* Never use the revoke function if we are doing full data
77 * journaling: there is no need to, and a V1 superblock won't
78 * support it. Otherwise, only skip the revoke on un-journaled
81 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
82 (!is_metadata && !ext4_should_journal_data(inode))) {
84 BUFFER_TRACE(bh, "call jbd2_journal_forget");
85 return ext4_journal_forget(handle, bh);
91 * data!=journal && (is_metadata || should_journal_data(inode))
93 BUFFER_TRACE(bh, "call ext4_journal_revoke");
94 err = ext4_journal_revoke(handle, blocknr, bh);
96 ext4_abort(inode->i_sb, __FUNCTION__,
97 "error %d when attempting revoke", err);
98 BUFFER_TRACE(bh, "exit");
103 * Work out how many blocks we need to proceed with the next chunk of a
104 * truncate transaction.
106 static unsigned long blocks_for_truncate(struct inode *inode)
110 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
112 /* Give ourselves just enough room to cope with inodes in which
113 * i_blocks is corrupt: we've seen disk corruptions in the past
114 * which resulted in random data in an inode which looked enough
115 * like a regular file for ext4 to try to delete it. Things
116 * will go a bit crazy if that happens, but at least we should
117 * try not to panic the whole kernel. */
121 /* But we need to bound the transaction so we don't overflow the
123 if (needed > EXT4_MAX_TRANS_DATA)
124 needed = EXT4_MAX_TRANS_DATA;
126 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
130 * Truncate transactions can be complex and absolutely huge. So we need to
131 * be able to restart the transaction at a conventient checkpoint to make
132 * sure we don't overflow the journal.
134 * start_transaction gets us a new handle for a truncate transaction,
135 * and extend_transaction tries to extend the existing one a bit. If
136 * extend fails, we need to propagate the failure up and restart the
137 * transaction in the top-level truncate loop. --sct
139 static handle_t *start_transaction(struct inode *inode)
143 result = ext4_journal_start(inode, blocks_for_truncate(inode));
147 ext4_std_error(inode->i_sb, PTR_ERR(result));
152 * Try to extend this transaction for the purposes of truncation.
154 * Returns 0 if we managed to create more room. If we can't create more
155 * room, and the transaction must be restarted we return 1.
157 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
159 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
161 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
173 jbd_debug(2, "restarting handle %p\n", handle);
174 return ext4_journal_restart(handle, blocks_for_truncate(inode));
178 * Called at the last iput() if i_nlink is zero.
180 void ext4_delete_inode (struct inode * inode)
184 truncate_inode_pages(&inode->i_data, 0);
186 if (is_bad_inode(inode))
189 handle = start_transaction(inode);
190 if (IS_ERR(handle)) {
192 * If we're going to skip the normal cleanup, we still need to
193 * make sure that the in-core orphan linked list is properly
196 ext4_orphan_del(NULL, inode);
204 ext4_truncate(inode);
206 * Kill off the orphan record which ext4_truncate created.
207 * AKPM: I think this can be inside the above `if'.
208 * Note that ext4_orphan_del() has to be able to cope with the
209 * deletion of a non-existent orphan - this is because we don't
210 * know if ext4_truncate() actually created an orphan record.
211 * (Well, we could do this if we need to, but heck - it works)
213 ext4_orphan_del(handle, inode);
214 EXT4_I(inode)->i_dtime = get_seconds();
217 * One subtle ordering requirement: if anything has gone wrong
218 * (transaction abort, IO errors, whatever), then we can still
219 * do these next steps (the fs will already have been marked as
220 * having errors), but we can't free the inode if the mark_dirty
223 if (ext4_mark_inode_dirty(handle, inode))
224 /* If that failed, just do the required in-core inode clear. */
227 ext4_free_inode(handle, inode);
228 ext4_journal_stop(handle);
231 clear_inode(inode); /* We must guarantee clearing of inode... */
237 struct buffer_head *bh;
240 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
242 p->key = *(p->p = v);
246 static int verify_chain(Indirect *from, Indirect *to)
248 while (from <= to && from->key == *from->p)
254 * ext4_block_to_path - parse the block number into array of offsets
255 * @inode: inode in question (we are only interested in its superblock)
256 * @i_block: block number to be parsed
257 * @offsets: array to store the offsets in
258 * @boundary: set this non-zero if the referred-to block is likely to be
259 * followed (on disk) by an indirect block.
261 * To store the locations of file's data ext4 uses a data structure common
262 * for UNIX filesystems - tree of pointers anchored in the inode, with
263 * data blocks at leaves and indirect blocks in intermediate nodes.
264 * This function translates the block number into path in that tree -
265 * return value is the path length and @offsets[n] is the offset of
266 * pointer to (n+1)th node in the nth one. If @block is out of range
267 * (negative or too large) warning is printed and zero returned.
269 * Note: function doesn't find node addresses, so no IO is needed. All
270 * we need to know is the capacity of indirect blocks (taken from the
275 * Portability note: the last comparison (check that we fit into triple
276 * indirect block) is spelled differently, because otherwise on an
277 * architecture with 32-bit longs and 8Kb pages we might get into trouble
278 * if our filesystem had 8Kb blocks. We might use long long, but that would
279 * kill us on x86. Oh, well, at least the sign propagation does not matter -
280 * i_block would have to be negative in the very beginning, so we would not
284 static int ext4_block_to_path(struct inode *inode,
286 ext4_lblk_t offsets[4], int *boundary)
288 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
289 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
290 const long direct_blocks = EXT4_NDIR_BLOCKS,
291 indirect_blocks = ptrs,
292 double_blocks = (1 << (ptrs_bits * 2));
297 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
298 } else if (i_block < direct_blocks) {
299 offsets[n++] = i_block;
300 final = direct_blocks;
301 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
302 offsets[n++] = EXT4_IND_BLOCK;
303 offsets[n++] = i_block;
305 } else if ((i_block -= indirect_blocks) < double_blocks) {
306 offsets[n++] = EXT4_DIND_BLOCK;
307 offsets[n++] = i_block >> ptrs_bits;
308 offsets[n++] = i_block & (ptrs - 1);
310 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
311 offsets[n++] = EXT4_TIND_BLOCK;
312 offsets[n++] = i_block >> (ptrs_bits * 2);
313 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
314 offsets[n++] = i_block & (ptrs - 1);
317 ext4_warning(inode->i_sb, "ext4_block_to_path", "block > big");
320 *boundary = final - 1 - (i_block & (ptrs - 1));
325 * ext4_get_branch - read the chain of indirect blocks leading to data
326 * @inode: inode in question
327 * @depth: depth of the chain (1 - direct pointer, etc.)
328 * @offsets: offsets of pointers in inode/indirect blocks
329 * @chain: place to store the result
330 * @err: here we store the error value
332 * Function fills the array of triples <key, p, bh> and returns %NULL
333 * if everything went OK or the pointer to the last filled triple
334 * (incomplete one) otherwise. Upon the return chain[i].key contains
335 * the number of (i+1)-th block in the chain (as it is stored in memory,
336 * i.e. little-endian 32-bit), chain[i].p contains the address of that
337 * number (it points into struct inode for i==0 and into the bh->b_data
338 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
339 * block for i>0 and NULL for i==0. In other words, it holds the block
340 * numbers of the chain, addresses they were taken from (and where we can
341 * verify that chain did not change) and buffer_heads hosting these
344 * Function stops when it stumbles upon zero pointer (absent block)
345 * (pointer to last triple returned, *@err == 0)
346 * or when it gets an IO error reading an indirect block
347 * (ditto, *@err == -EIO)
348 * or when it notices that chain had been changed while it was reading
349 * (ditto, *@err == -EAGAIN)
350 * or when it reads all @depth-1 indirect blocks successfully and finds
351 * the whole chain, all way to the data (returns %NULL, *err == 0).
353 static Indirect *ext4_get_branch(struct inode *inode, int depth,
354 ext4_lblk_t *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, ext4_lblk_t 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 ext4_lblk_t *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 ext4_lblk_t 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 = ext4_current_time(inode);
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 ext4_lblk_t iblock, unsigned long maxblocks,
790 struct buffer_head *bh_result,
791 int create, int extend_disksize)
794 ext4_lblk_t offsets[4];
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,
809 &blocks_to_boundary);
814 partial = ext4_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(ext4_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 = ext4_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 ext4_init_block_alloc_info(inode);
889 goal = ext4_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 = ext4_blks_to_allocate(partial, indirect_blks,
899 maxblocks, blocks_to_boundary);
901 * Block out ext4_truncate while we alter the tree
903 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
904 offsets + (partial - chain), partial);
907 * The ext4_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 = ext4_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 * ext4_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 (EXT4_RESERVE_TRANS_BLOCKS + 32)
948 static int ext4_get_block(struct inode *inode, sector_t iblock,
949 struct buffer_head *bh_result, int create)
951 handle_t *handle = ext4_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 ext4_journal_stop(handle);
967 handle = ext4_journal_start(inode, DIO_CREDITS);
969 ret = PTR_ERR(handle);
973 if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
975 * Getting low on buffer credits...
977 ret = ext4_journal_extend(handle, DIO_CREDITS);
980 * Couldn't extend the transaction. Start a new one.
982 ret = ext4_journal_restart(handle, DIO_CREDITS);
988 ret = ext4_get_blocks_wrap(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 *ext4_getblk(handle_t *handle, struct inode *inode,
1002 ext4_lblk_t 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 = ext4_get_blocks_wrap(handle, inode, block, 1,
1015 * ext4_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 != NULL);
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 ext4_get_block instead, so it's not a
1043 BUFFER_TRACE(bh, "call get_create_access");
1044 fatal = ext4_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 ext4_journal_dirty_metadata");
1051 err = ext4_journal_dirty_metadata(handle, bh);
1055 BUFFER_TRACE(bh, "not a new buffer");
1068 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1069 ext4_lblk_t block, int create, int *err)
1071 struct buffer_head * bh;
1073 bh = ext4_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 ext4_get_block()
1123 * and the commit_write(). So doing the jbd2_journal_start at the start of
1124 * prepare_write() is the right place.
1126 * Also, this function can nest inside ext4_writepage() ->
1127 * block_write_full_page(). In that case, we *know* that ext4_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, ext4 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 jbd2_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 ext4_journal_get_write_access(handle, bh);
1152 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1153 loff_t pos, unsigned len, unsigned flags,
1154 struct page **pagep, void **fsdata)
1156 struct inode *inode = mapping->host;
1157 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1164 index = pos >> PAGE_CACHE_SHIFT;
1165 from = pos & (PAGE_CACHE_SIZE - 1);
1169 page = __grab_cache_page(mapping, index);
1174 handle = ext4_journal_start(inode, needed_blocks);
1175 if (IS_ERR(handle)) {
1177 page_cache_release(page);
1178 ret = PTR_ERR(handle);
1182 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1185 if (!ret && ext4_should_journal_data(inode)) {
1186 ret = walk_page_buffers(handle, page_buffers(page),
1187 from, to, NULL, do_journal_get_write_access);
1191 ext4_journal_stop(handle);
1193 page_cache_release(page);
1196 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1202 int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1204 int err = jbd2_journal_dirty_data(handle, bh);
1206 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1211 /* For write_end() in data=journal mode */
1212 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1214 if (!buffer_mapped(bh) || buffer_freed(bh))
1216 set_buffer_uptodate(bh);
1217 return ext4_journal_dirty_metadata(handle, bh);
1221 * Generic write_end handler for ordered and writeback ext4 journal modes.
1222 * We can't use generic_write_end, because that unlocks the page and we need to
1223 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1224 * after block_write_end.
1226 static int ext4_generic_write_end(struct file *file,
1227 struct address_space *mapping,
1228 loff_t pos, unsigned len, unsigned copied,
1229 struct page *page, void *fsdata)
1231 struct inode *inode = file->f_mapping->host;
1233 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1235 if (pos+copied > inode->i_size) {
1236 i_size_write(inode, pos+copied);
1237 mark_inode_dirty(inode);
1244 * We need to pick up the new inode size which generic_commit_write gave us
1245 * `file' can be NULL - eg, when called from page_symlink().
1247 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1248 * buffers are managed internally.
1250 static int ext4_ordered_write_end(struct file *file,
1251 struct address_space *mapping,
1252 loff_t pos, unsigned len, unsigned copied,
1253 struct page *page, void *fsdata)
1255 handle_t *handle = ext4_journal_current_handle();
1256 struct inode *inode = file->f_mapping->host;
1260 from = pos & (PAGE_CACHE_SIZE - 1);
1263 ret = walk_page_buffers(handle, page_buffers(page),
1264 from, to, NULL, ext4_journal_dirty_data);
1268 * generic_write_end() will run mark_inode_dirty() if i_size
1269 * changes. So let's piggyback the i_disksize mark_inode_dirty
1274 new_i_size = pos + copied;
1275 if (new_i_size > EXT4_I(inode)->i_disksize)
1276 EXT4_I(inode)->i_disksize = new_i_size;
1277 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1282 ret2 = ext4_journal_stop(handle);
1286 page_cache_release(page);
1288 return ret ? ret : copied;
1291 static int ext4_writeback_write_end(struct file *file,
1292 struct address_space *mapping,
1293 loff_t pos, unsigned len, unsigned copied,
1294 struct page *page, void *fsdata)
1296 handle_t *handle = ext4_journal_current_handle();
1297 struct inode *inode = file->f_mapping->host;
1301 new_i_size = pos + copied;
1302 if (new_i_size > EXT4_I(inode)->i_disksize)
1303 EXT4_I(inode)->i_disksize = new_i_size;
1305 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1310 ret2 = ext4_journal_stop(handle);
1314 page_cache_release(page);
1316 return ret ? ret : copied;
1319 static int ext4_journalled_write_end(struct file *file,
1320 struct address_space *mapping,
1321 loff_t pos, unsigned len, unsigned copied,
1322 struct page *page, void *fsdata)
1324 handle_t *handle = ext4_journal_current_handle();
1325 struct inode *inode = mapping->host;
1330 from = pos & (PAGE_CACHE_SIZE - 1);
1334 if (!PageUptodate(page))
1336 page_zero_new_buffers(page, from+copied, to);
1339 ret = walk_page_buffers(handle, page_buffers(page), from,
1340 to, &partial, write_end_fn);
1342 SetPageUptodate(page);
1343 if (pos+copied > inode->i_size)
1344 i_size_write(inode, pos+copied);
1345 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1346 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1347 EXT4_I(inode)->i_disksize = inode->i_size;
1348 ret2 = ext4_mark_inode_dirty(handle, inode);
1353 ret2 = ext4_journal_stop(handle);
1357 page_cache_release(page);
1359 return ret ? ret : copied;
1363 * bmap() is special. It gets used by applications such as lilo and by
1364 * the swapper to find the on-disk block of a specific piece of data.
1366 * Naturally, this is dangerous if the block concerned is still in the
1367 * journal. If somebody makes a swapfile on an ext4 data-journaling
1368 * filesystem and enables swap, then they may get a nasty shock when the
1369 * data getting swapped to that swapfile suddenly gets overwritten by
1370 * the original zero's written out previously to the journal and
1371 * awaiting writeback in the kernel's buffer cache.
1373 * So, if we see any bmap calls here on a modified, data-journaled file,
1374 * take extra steps to flush any blocks which might be in the cache.
1376 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
1378 struct inode *inode = mapping->host;
1382 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
1384 * This is a REALLY heavyweight approach, but the use of
1385 * bmap on dirty files is expected to be extremely rare:
1386 * only if we run lilo or swapon on a freshly made file
1387 * do we expect this to happen.
1389 * (bmap requires CAP_SYS_RAWIO so this does not
1390 * represent an unprivileged user DOS attack --- we'd be
1391 * in trouble if mortal users could trigger this path at
1394 * NB. EXT4_STATE_JDATA is not set on files other than
1395 * regular files. If somebody wants to bmap a directory
1396 * or symlink and gets confused because the buffer
1397 * hasn't yet been flushed to disk, they deserve
1398 * everything they get.
1401 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1402 journal = EXT4_JOURNAL(inode);
1403 jbd2_journal_lock_updates(journal);
1404 err = jbd2_journal_flush(journal);
1405 jbd2_journal_unlock_updates(journal);
1411 return generic_block_bmap(mapping,block,ext4_get_block);
1414 static int bget_one(handle_t *handle, struct buffer_head *bh)
1420 static int bput_one(handle_t *handle, struct buffer_head *bh)
1426 static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1428 if (buffer_mapped(bh))
1429 return ext4_journal_dirty_data(handle, bh);
1434 * Note that we always start a transaction even if we're not journalling
1435 * data. This is to preserve ordering: any hole instantiation within
1436 * __block_write_full_page -> ext4_get_block() should be journalled
1437 * along with the data so we don't crash and then get metadata which
1438 * refers to old data.
1440 * In all journalling modes block_write_full_page() will start the I/O.
1444 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1449 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1451 * Same applies to ext4_get_block(). We will deadlock on various things like
1452 * lock_journal and i_truncate_mutex.
1454 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1457 * 16May01: If we're reentered then journal_current_handle() will be
1458 * non-zero. We simply *return*.
1460 * 1 July 2001: @@@ FIXME:
1461 * In journalled data mode, a data buffer may be metadata against the
1462 * current transaction. But the same file is part of a shared mapping
1463 * and someone does a writepage() on it.
1465 * We will move the buffer onto the async_data list, but *after* it has
1466 * been dirtied. So there's a small window where we have dirty data on
1469 * Note that this only applies to the last partial page in the file. The
1470 * bit which block_write_full_page() uses prepare/commit for. (That's
1471 * broken code anyway: it's wrong for msync()).
1473 * It's a rare case: affects the final partial page, for journalled data
1474 * where the file is subject to bith write() and writepage() in the same
1475 * transction. To fix it we'll need a custom block_write_full_page().
1476 * We'll probably need that anyway for journalling writepage() output.
1478 * We don't honour synchronous mounts for writepage(). That would be
1479 * disastrous. Any write() or metadata operation will sync the fs for
1482 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1483 * we don't need to open a transaction here.
1485 static int ext4_ordered_writepage(struct page *page,
1486 struct writeback_control *wbc)
1488 struct inode *inode = page->mapping->host;
1489 struct buffer_head *page_bufs;
1490 handle_t *handle = NULL;
1494 J_ASSERT(PageLocked(page));
1497 * We give up here if we're reentered, because it might be for a
1498 * different filesystem.
1500 if (ext4_journal_current_handle())
1503 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1505 if (IS_ERR(handle)) {
1506 ret = PTR_ERR(handle);
1510 if (!page_has_buffers(page)) {
1511 create_empty_buffers(page, inode->i_sb->s_blocksize,
1512 (1 << BH_Dirty)|(1 << BH_Uptodate));
1514 page_bufs = page_buffers(page);
1515 walk_page_buffers(handle, page_bufs, 0,
1516 PAGE_CACHE_SIZE, NULL, bget_one);
1518 ret = block_write_full_page(page, ext4_get_block, wbc);
1521 * The page can become unlocked at any point now, and
1522 * truncate can then come in and change things. So we
1523 * can't touch *page from now on. But *page_bufs is
1524 * safe due to elevated refcount.
1528 * And attach them to the current transaction. But only if
1529 * block_write_full_page() succeeded. Otherwise they are unmapped,
1530 * and generally junk.
1533 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1534 NULL, jbd2_journal_dirty_data_fn);
1538 walk_page_buffers(handle, page_bufs, 0,
1539 PAGE_CACHE_SIZE, NULL, bput_one);
1540 err = ext4_journal_stop(handle);
1546 redirty_page_for_writepage(wbc, page);
1551 static int ext4_writeback_writepage(struct page *page,
1552 struct writeback_control *wbc)
1554 struct inode *inode = page->mapping->host;
1555 handle_t *handle = NULL;
1559 if (ext4_journal_current_handle())
1562 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1563 if (IS_ERR(handle)) {
1564 ret = PTR_ERR(handle);
1568 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1569 ret = nobh_writepage(page, ext4_get_block, wbc);
1571 ret = block_write_full_page(page, ext4_get_block, wbc);
1573 err = ext4_journal_stop(handle);
1579 redirty_page_for_writepage(wbc, page);
1584 static int ext4_journalled_writepage(struct page *page,
1585 struct writeback_control *wbc)
1587 struct inode *inode = page->mapping->host;
1588 handle_t *handle = NULL;
1592 if (ext4_journal_current_handle())
1595 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1596 if (IS_ERR(handle)) {
1597 ret = PTR_ERR(handle);
1601 if (!page_has_buffers(page) || PageChecked(page)) {
1603 * It's mmapped pagecache. Add buffers and journal it. There
1604 * doesn't seem much point in redirtying the page here.
1606 ClearPageChecked(page);
1607 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1610 ext4_journal_stop(handle);
1613 ret = walk_page_buffers(handle, page_buffers(page), 0,
1614 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1616 err = walk_page_buffers(handle, page_buffers(page), 0,
1617 PAGE_CACHE_SIZE, NULL, write_end_fn);
1620 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1624 * It may be a page full of checkpoint-mode buffers. We don't
1625 * really know unless we go poke around in the buffer_heads.
1626 * But block_write_full_page will do the right thing.
1628 ret = block_write_full_page(page, ext4_get_block, wbc);
1630 err = ext4_journal_stop(handle);
1637 redirty_page_for_writepage(wbc, page);
1643 static int ext4_readpage(struct file *file, struct page *page)
1645 return mpage_readpage(page, ext4_get_block);
1649 ext4_readpages(struct file *file, struct address_space *mapping,
1650 struct list_head *pages, unsigned nr_pages)
1652 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
1655 static void ext4_invalidatepage(struct page *page, unsigned long offset)
1657 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1660 * If it's a full truncate we just forget about the pending dirtying
1663 ClearPageChecked(page);
1665 jbd2_journal_invalidatepage(journal, page, offset);
1668 static int ext4_releasepage(struct page *page, gfp_t wait)
1670 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1672 WARN_ON(PageChecked(page));
1673 if (!page_has_buffers(page))
1675 return jbd2_journal_try_to_free_buffers(journal, page, wait);
1679 * If the O_DIRECT write will extend the file then add this inode to the
1680 * orphan list. So recovery will truncate it back to the original size
1681 * if the machine crashes during the write.
1683 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1684 * crashes then stale disk data _may_ be exposed inside the file.
1686 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
1687 const struct iovec *iov, loff_t offset,
1688 unsigned long nr_segs)
1690 struct file *file = iocb->ki_filp;
1691 struct inode *inode = file->f_mapping->host;
1692 struct ext4_inode_info *ei = EXT4_I(inode);
1693 handle_t *handle = NULL;
1696 size_t count = iov_length(iov, nr_segs);
1699 loff_t final_size = offset + count;
1701 handle = ext4_journal_start(inode, DIO_CREDITS);
1702 if (IS_ERR(handle)) {
1703 ret = PTR_ERR(handle);
1706 if (final_size > inode->i_size) {
1707 ret = ext4_orphan_add(handle, inode);
1711 ei->i_disksize = inode->i_size;
1715 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1717 ext4_get_block, NULL);
1720 * Reacquire the handle: ext4_get_block() can restart the transaction
1722 handle = ext4_journal_current_handle();
1728 if (orphan && inode->i_nlink)
1729 ext4_orphan_del(handle, inode);
1730 if (orphan && ret > 0) {
1731 loff_t end = offset + ret;
1732 if (end > inode->i_size) {
1733 ei->i_disksize = end;
1734 i_size_write(inode, end);
1736 * We're going to return a positive `ret'
1737 * here due to non-zero-length I/O, so there's
1738 * no way of reporting error returns from
1739 * ext4_mark_inode_dirty() to userspace. So
1742 ext4_mark_inode_dirty(handle, inode);
1745 err = ext4_journal_stop(handle);
1754 * Pages can be marked dirty completely asynchronously from ext4's journalling
1755 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1756 * much here because ->set_page_dirty is called under VFS locks. The page is
1757 * not necessarily locked.
1759 * We cannot just dirty the page and leave attached buffers clean, because the
1760 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1761 * or jbddirty because all the journalling code will explode.
1763 * So what we do is to mark the page "pending dirty" and next time writepage
1764 * is called, propagate that into the buffers appropriately.
1766 static int ext4_journalled_set_page_dirty(struct page *page)
1768 SetPageChecked(page);
1769 return __set_page_dirty_nobuffers(page);
1772 static const struct address_space_operations ext4_ordered_aops = {
1773 .readpage = ext4_readpage,
1774 .readpages = ext4_readpages,
1775 .writepage = ext4_ordered_writepage,
1776 .sync_page = block_sync_page,
1777 .write_begin = ext4_write_begin,
1778 .write_end = ext4_ordered_write_end,
1780 .invalidatepage = ext4_invalidatepage,
1781 .releasepage = ext4_releasepage,
1782 .direct_IO = ext4_direct_IO,
1783 .migratepage = buffer_migrate_page,
1786 static const struct address_space_operations ext4_writeback_aops = {
1787 .readpage = ext4_readpage,
1788 .readpages = ext4_readpages,
1789 .writepage = ext4_writeback_writepage,
1790 .sync_page = block_sync_page,
1791 .write_begin = ext4_write_begin,
1792 .write_end = ext4_writeback_write_end,
1794 .invalidatepage = ext4_invalidatepage,
1795 .releasepage = ext4_releasepage,
1796 .direct_IO = ext4_direct_IO,
1797 .migratepage = buffer_migrate_page,
1800 static const struct address_space_operations ext4_journalled_aops = {
1801 .readpage = ext4_readpage,
1802 .readpages = ext4_readpages,
1803 .writepage = ext4_journalled_writepage,
1804 .sync_page = block_sync_page,
1805 .write_begin = ext4_write_begin,
1806 .write_end = ext4_journalled_write_end,
1807 .set_page_dirty = ext4_journalled_set_page_dirty,
1809 .invalidatepage = ext4_invalidatepage,
1810 .releasepage = ext4_releasepage,
1813 void ext4_set_aops(struct inode *inode)
1815 if (ext4_should_order_data(inode))
1816 inode->i_mapping->a_ops = &ext4_ordered_aops;
1817 else if (ext4_should_writeback_data(inode))
1818 inode->i_mapping->a_ops = &ext4_writeback_aops;
1820 inode->i_mapping->a_ops = &ext4_journalled_aops;
1824 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1825 * up to the end of the block which corresponds to `from'.
1826 * This required during truncate. We need to physically zero the tail end
1827 * of that block so it doesn't yield old data if the file is later grown.
1829 int ext4_block_truncate_page(handle_t *handle, struct page *page,
1830 struct address_space *mapping, loff_t from)
1832 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1833 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1834 unsigned blocksize, length, pos;
1836 struct inode *inode = mapping->host;
1837 struct buffer_head *bh;
1840 blocksize = inode->i_sb->s_blocksize;
1841 length = blocksize - (offset & (blocksize - 1));
1842 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1845 * For "nobh" option, we can only work if we don't need to
1846 * read-in the page - otherwise we create buffers to do the IO.
1848 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1849 ext4_should_writeback_data(inode) && PageUptodate(page)) {
1850 zero_user_page(page, offset, length, KM_USER0);
1851 set_page_dirty(page);
1855 if (!page_has_buffers(page))
1856 create_empty_buffers(page, blocksize, 0);
1858 /* Find the buffer that contains "offset" */
1859 bh = page_buffers(page);
1861 while (offset >= pos) {
1862 bh = bh->b_this_page;
1868 if (buffer_freed(bh)) {
1869 BUFFER_TRACE(bh, "freed: skip");
1873 if (!buffer_mapped(bh)) {
1874 BUFFER_TRACE(bh, "unmapped");
1875 ext4_get_block(inode, iblock, bh, 0);
1876 /* unmapped? It's a hole - nothing to do */
1877 if (!buffer_mapped(bh)) {
1878 BUFFER_TRACE(bh, "still unmapped");
1883 /* Ok, it's mapped. Make sure it's up-to-date */
1884 if (PageUptodate(page))
1885 set_buffer_uptodate(bh);
1887 if (!buffer_uptodate(bh)) {
1889 ll_rw_block(READ, 1, &bh);
1891 /* Uhhuh. Read error. Complain and punt. */
1892 if (!buffer_uptodate(bh))
1896 if (ext4_should_journal_data(inode)) {
1897 BUFFER_TRACE(bh, "get write access");
1898 err = ext4_journal_get_write_access(handle, bh);
1903 zero_user_page(page, offset, length, KM_USER0);
1905 BUFFER_TRACE(bh, "zeroed end of block");
1908 if (ext4_should_journal_data(inode)) {
1909 err = ext4_journal_dirty_metadata(handle, bh);
1911 if (ext4_should_order_data(inode))
1912 err = ext4_journal_dirty_data(handle, bh);
1913 mark_buffer_dirty(bh);
1918 page_cache_release(page);
1923 * Probably it should be a library function... search for first non-zero word
1924 * or memcmp with zero_page, whatever is better for particular architecture.
1927 static inline int all_zeroes(__le32 *p, __le32 *q)
1936 * ext4_find_shared - find the indirect blocks for partial truncation.
1937 * @inode: inode in question
1938 * @depth: depth of the affected branch
1939 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
1940 * @chain: place to store the pointers to partial indirect blocks
1941 * @top: place to the (detached) top of branch
1943 * This is a helper function used by ext4_truncate().
1945 * When we do truncate() we may have to clean the ends of several
1946 * indirect blocks but leave the blocks themselves alive. Block is
1947 * partially truncated if some data below the new i_size is refered
1948 * from it (and it is on the path to the first completely truncated
1949 * data block, indeed). We have to free the top of that path along
1950 * with everything to the right of the path. Since no allocation
1951 * past the truncation point is possible until ext4_truncate()
1952 * finishes, we may safely do the latter, but top of branch may
1953 * require special attention - pageout below the truncation point
1954 * might try to populate it.
1956 * We atomically detach the top of branch from the tree, store the
1957 * block number of its root in *@top, pointers to buffer_heads of
1958 * partially truncated blocks - in @chain[].bh and pointers to
1959 * their last elements that should not be removed - in
1960 * @chain[].p. Return value is the pointer to last filled element
1963 * The work left to caller to do the actual freeing of subtrees:
1964 * a) free the subtree starting from *@top
1965 * b) free the subtrees whose roots are stored in
1966 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1967 * c) free the subtrees growing from the inode past the @chain[0].
1968 * (no partially truncated stuff there). */
1970 static Indirect *ext4_find_shared(struct inode *inode, int depth,
1971 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
1973 Indirect *partial, *p;
1977 /* Make k index the deepest non-null offest + 1 */
1978 for (k = depth; k > 1 && !offsets[k-1]; k--)
1980 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1981 /* Writer: pointers */
1983 partial = chain + k-1;
1985 * If the branch acquired continuation since we've looked at it -
1986 * fine, it should all survive and (new) top doesn't belong to us.
1988 if (!partial->key && *partial->p)
1991 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1994 * OK, we've found the last block that must survive. The rest of our
1995 * branch should be detached before unlocking. However, if that rest
1996 * of branch is all ours and does not grow immediately from the inode
1997 * it's easier to cheat and just decrement partial->p.
1999 if (p == chain + k - 1 && p > chain) {
2003 /* Nope, don't do this in ext4. Must leave the tree intact */
2010 while(partial > p) {
2011 brelse(partial->bh);
2019 * Zero a number of block pointers in either an inode or an indirect block.
2020 * If we restart the transaction we must again get write access to the
2021 * indirect block for further modification.
2023 * We release `count' blocks on disk, but (last - first) may be greater
2024 * than `count' because there can be holes in there.
2026 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
2027 struct buffer_head *bh, ext4_fsblk_t block_to_free,
2028 unsigned long count, __le32 *first, __le32 *last)
2031 if (try_to_extend_transaction(handle, inode)) {
2033 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2034 ext4_journal_dirty_metadata(handle, bh);
2036 ext4_mark_inode_dirty(handle, inode);
2037 ext4_journal_test_restart(handle, inode);
2039 BUFFER_TRACE(bh, "retaking write access");
2040 ext4_journal_get_write_access(handle, bh);
2045 * Any buffers which are on the journal will be in memory. We find
2046 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
2047 * on them. We've already detached each block from the file, so
2048 * bforget() in jbd2_journal_forget() should be safe.
2050 * AKPM: turn on bforget in jbd2_journal_forget()!!!
2052 for (p = first; p < last; p++) {
2053 u32 nr = le32_to_cpu(*p);
2055 struct buffer_head *bh;
2058 bh = sb_find_get_block(inode->i_sb, nr);
2059 ext4_forget(handle, 0, inode, bh, nr);
2063 ext4_free_blocks(handle, inode, block_to_free, count);
2067 * ext4_free_data - free a list of data blocks
2068 * @handle: handle for this transaction
2069 * @inode: inode we are dealing with
2070 * @this_bh: indirect buffer_head which contains *@first and *@last
2071 * @first: array of block numbers
2072 * @last: points immediately past the end of array
2074 * We are freeing all blocks refered from that array (numbers are stored as
2075 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2077 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2078 * blocks are contiguous then releasing them at one time will only affect one
2079 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2080 * actually use a lot of journal space.
2082 * @this_bh will be %NULL if @first and @last point into the inode's direct
2085 static void ext4_free_data(handle_t *handle, struct inode *inode,
2086 struct buffer_head *this_bh,
2087 __le32 *first, __le32 *last)
2089 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
2090 unsigned long count = 0; /* Number of blocks in the run */
2091 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2094 ext4_fsblk_t nr; /* Current block # */
2095 __le32 *p; /* Pointer into inode/ind
2096 for current block */
2099 if (this_bh) { /* For indirect block */
2100 BUFFER_TRACE(this_bh, "get_write_access");
2101 err = ext4_journal_get_write_access(handle, this_bh);
2102 /* Important: if we can't update the indirect pointers
2103 * to the blocks, we can't free them. */
2108 for (p = first; p < last; p++) {
2109 nr = le32_to_cpu(*p);
2111 /* accumulate blocks to free if they're contiguous */
2114 block_to_free_p = p;
2116 } else if (nr == block_to_free + count) {
2119 ext4_clear_blocks(handle, inode, this_bh,
2121 count, block_to_free_p, p);
2123 block_to_free_p = p;
2130 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
2131 count, block_to_free_p, p);
2134 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2135 ext4_journal_dirty_metadata(handle, this_bh);
2140 * ext4_free_branches - free an array of branches
2141 * @handle: JBD handle for this transaction
2142 * @inode: inode we are dealing with
2143 * @parent_bh: the buffer_head which contains *@first and *@last
2144 * @first: array of block numbers
2145 * @last: pointer immediately past the end of array
2146 * @depth: depth of the branches to free
2148 * We are freeing all blocks refered from these branches (numbers are
2149 * stored as little-endian 32-bit) and updating @inode->i_blocks
2152 static void ext4_free_branches(handle_t *handle, struct inode *inode,
2153 struct buffer_head *parent_bh,
2154 __le32 *first, __le32 *last, int depth)
2159 if (is_handle_aborted(handle))
2163 struct buffer_head *bh;
2164 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2166 while (--p >= first) {
2167 nr = le32_to_cpu(*p);
2169 continue; /* A hole */
2171 /* Go read the buffer for the next level down */
2172 bh = sb_bread(inode->i_sb, nr);
2175 * A read failure? Report error and clear slot
2179 ext4_error(inode->i_sb, "ext4_free_branches",
2180 "Read failure, inode=%lu, block=%llu",
2185 /* This zaps the entire block. Bottom up. */
2186 BUFFER_TRACE(bh, "free child branches");
2187 ext4_free_branches(handle, inode, bh,
2188 (__le32*)bh->b_data,
2189 (__le32*)bh->b_data + addr_per_block,
2193 * We've probably journalled the indirect block several
2194 * times during the truncate. But it's no longer
2195 * needed and we now drop it from the transaction via
2196 * jbd2_journal_revoke().
2198 * That's easy if it's exclusively part of this
2199 * transaction. But if it's part of the committing
2200 * transaction then jbd2_journal_forget() will simply
2201 * brelse() it. That means that if the underlying
2202 * block is reallocated in ext4_get_block(),
2203 * unmap_underlying_metadata() will find this block
2204 * and will try to get rid of it. damn, damn.
2206 * If this block has already been committed to the
2207 * journal, a revoke record will be written. And
2208 * revoke records must be emitted *before* clearing
2209 * this block's bit in the bitmaps.
2211 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
2214 * Everything below this this pointer has been
2215 * released. Now let this top-of-subtree go.
2217 * We want the freeing of this indirect block to be
2218 * atomic in the journal with the updating of the
2219 * bitmap block which owns it. So make some room in
2222 * We zero the parent pointer *after* freeing its
2223 * pointee in the bitmaps, so if extend_transaction()
2224 * for some reason fails to put the bitmap changes and
2225 * the release into the same transaction, recovery
2226 * will merely complain about releasing a free block,
2227 * rather than leaking blocks.
2229 if (is_handle_aborted(handle))
2231 if (try_to_extend_transaction(handle, inode)) {
2232 ext4_mark_inode_dirty(handle, inode);
2233 ext4_journal_test_restart(handle, inode);
2236 ext4_free_blocks(handle, inode, nr, 1);
2240 * The block which we have just freed is
2241 * pointed to by an indirect block: journal it
2243 BUFFER_TRACE(parent_bh, "get_write_access");
2244 if (!ext4_journal_get_write_access(handle,
2247 BUFFER_TRACE(parent_bh,
2248 "call ext4_journal_dirty_metadata");
2249 ext4_journal_dirty_metadata(handle,
2255 /* We have reached the bottom of the tree. */
2256 BUFFER_TRACE(parent_bh, "free data blocks");
2257 ext4_free_data(handle, inode, parent_bh, first, last);
2264 * We block out ext4_get_block() block instantiations across the entire
2265 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2266 * simultaneously on behalf of the same inode.
2268 * As we work through the truncate and commmit bits of it to the journal there
2269 * is one core, guiding principle: the file's tree must always be consistent on
2270 * disk. We must be able to restart the truncate after a crash.
2272 * The file's tree may be transiently inconsistent in memory (although it
2273 * probably isn't), but whenever we close off and commit a journal transaction,
2274 * the contents of (the filesystem + the journal) must be consistent and
2275 * restartable. It's pretty simple, really: bottom up, right to left (although
2276 * left-to-right works OK too).
2278 * Note that at recovery time, journal replay occurs *before* the restart of
2279 * truncate against the orphan inode list.
2281 * The committed inode has the new, desired i_size (which is the same as
2282 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
2283 * that this inode's truncate did not complete and it will again call
2284 * ext4_truncate() to have another go. So there will be instantiated blocks
2285 * to the right of the truncation point in a crashed ext4 filesystem. But
2286 * that's fine - as long as they are linked from the inode, the post-crash
2287 * ext4_truncate() run will find them and release them.
2289 void ext4_truncate(struct inode *inode)
2292 struct ext4_inode_info *ei = EXT4_I(inode);
2293 __le32 *i_data = ei->i_data;
2294 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2295 struct address_space *mapping = inode->i_mapping;
2296 ext4_lblk_t offsets[4];
2301 ext4_lblk_t last_block;
2302 unsigned blocksize = inode->i_sb->s_blocksize;
2305 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2306 S_ISLNK(inode->i_mode)))
2308 if (ext4_inode_is_fast_symlink(inode))
2310 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2314 * We have to lock the EOF page here, because lock_page() nests
2315 * outside jbd2_journal_start().
2317 if ((inode->i_size & (blocksize - 1)) == 0) {
2318 /* Block boundary? Nothing to do */
2321 page = grab_cache_page(mapping,
2322 inode->i_size >> PAGE_CACHE_SHIFT);
2327 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
2328 return ext4_ext_truncate(inode, page);
2330 handle = start_transaction(inode);
2331 if (IS_ERR(handle)) {
2333 clear_highpage(page);
2334 flush_dcache_page(page);
2336 page_cache_release(page);
2338 return; /* AKPM: return what? */
2341 last_block = (inode->i_size + blocksize-1)
2342 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
2345 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2347 n = ext4_block_to_path(inode, last_block, offsets, NULL);
2349 goto out_stop; /* error */
2352 * OK. This truncate is going to happen. We add the inode to the
2353 * orphan list, so that if this truncate spans multiple transactions,
2354 * and we crash, we will resume the truncate when the filesystem
2355 * recovers. It also marks the inode dirty, to catch the new size.
2357 * Implication: the file must always be in a sane, consistent
2358 * truncatable state while each transaction commits.
2360 if (ext4_orphan_add(handle, inode))
2364 * The orphan list entry will now protect us from any crash which
2365 * occurs before the truncate completes, so it is now safe to propagate
2366 * the new, shorter inode size (held for now in i_size) into the
2367 * on-disk inode. We do this via i_disksize, which is the value which
2368 * ext4 *really* writes onto the disk inode.
2370 ei->i_disksize = inode->i_size;
2373 * From here we block out all ext4_get_block() callers who want to
2374 * modify the block allocation tree.
2376 mutex_lock(&ei->truncate_mutex);
2378 if (n == 1) { /* direct blocks */
2379 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2380 i_data + EXT4_NDIR_BLOCKS);
2384 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
2385 /* Kill the top of shared branch (not detached) */
2387 if (partial == chain) {
2388 /* Shared branch grows from the inode */
2389 ext4_free_branches(handle, inode, NULL,
2390 &nr, &nr+1, (chain+n-1) - partial);
2393 * We mark the inode dirty prior to restart,
2394 * and prior to stop. No need for it here.
2397 /* Shared branch grows from an indirect block */
2398 BUFFER_TRACE(partial->bh, "get_write_access");
2399 ext4_free_branches(handle, inode, partial->bh,
2401 partial->p+1, (chain+n-1) - partial);
2404 /* Clear the ends of indirect blocks on the shared branch */
2405 while (partial > chain) {
2406 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
2407 (__le32*)partial->bh->b_data+addr_per_block,
2408 (chain+n-1) - partial);
2409 BUFFER_TRACE(partial->bh, "call brelse");
2410 brelse (partial->bh);
2414 /* Kill the remaining (whole) subtrees */
2415 switch (offsets[0]) {
2417 nr = i_data[EXT4_IND_BLOCK];
2419 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2420 i_data[EXT4_IND_BLOCK] = 0;
2422 case EXT4_IND_BLOCK:
2423 nr = i_data[EXT4_DIND_BLOCK];
2425 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2426 i_data[EXT4_DIND_BLOCK] = 0;
2428 case EXT4_DIND_BLOCK:
2429 nr = i_data[EXT4_TIND_BLOCK];
2431 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2432 i_data[EXT4_TIND_BLOCK] = 0;
2434 case EXT4_TIND_BLOCK:
2438 ext4_discard_reservation(inode);
2440 mutex_unlock(&ei->truncate_mutex);
2441 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2442 ext4_mark_inode_dirty(handle, inode);
2445 * In a multi-transaction truncate, we only make the final transaction
2452 * If this was a simple ftruncate(), and the file will remain alive
2453 * then we need to clear up the orphan record which we created above.
2454 * However, if this was a real unlink then we were called by
2455 * ext4_delete_inode(), and we allow that function to clean up the
2456 * orphan info for us.
2459 ext4_orphan_del(handle, inode);
2461 ext4_journal_stop(handle);
2464 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2465 unsigned long ino, struct ext4_iloc *iloc)
2467 unsigned long desc, group_desc;
2468 ext4_group_t block_group;
2469 unsigned long offset;
2471 struct buffer_head *bh;
2472 struct ext4_group_desc * gdp;
2474 if (!ext4_valid_inum(sb, ino)) {
2476 * This error is already checked for in namei.c unless we are
2477 * looking at an NFS filehandle, in which case no error
2483 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2484 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2485 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
2489 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2490 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2491 bh = EXT4_SB(sb)->s_group_desc[group_desc];
2493 ext4_error (sb, "ext4_get_inode_block",
2494 "Descriptor not loaded");
2498 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2499 desc * EXT4_DESC_SIZE(sb));
2501 * Figure out the offset within the block group inode table
2503 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2504 EXT4_INODE_SIZE(sb);
2505 block = ext4_inode_table(sb, gdp) +
2506 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
2508 iloc->block_group = block_group;
2509 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
2514 * ext4_get_inode_loc returns with an extra refcount against the inode's
2515 * underlying buffer_head on success. If 'in_mem' is true, we have all
2516 * data in memory that is needed to recreate the on-disk version of this
2519 static int __ext4_get_inode_loc(struct inode *inode,
2520 struct ext4_iloc *iloc, int in_mem)
2523 struct buffer_head *bh;
2525 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2529 bh = sb_getblk(inode->i_sb, block);
2531 ext4_error (inode->i_sb, "ext4_get_inode_loc",
2532 "unable to read inode block - "
2533 "inode=%lu, block=%llu",
2534 inode->i_ino, block);
2537 if (!buffer_uptodate(bh)) {
2539 if (buffer_uptodate(bh)) {
2540 /* someone brought it uptodate while we waited */
2546 * If we have all information of the inode in memory and this
2547 * is the only valid inode in the block, we need not read the
2551 struct buffer_head *bitmap_bh;
2552 struct ext4_group_desc *desc;
2553 int inodes_per_buffer;
2554 int inode_offset, i;
2555 ext4_group_t block_group;
2558 block_group = (inode->i_ino - 1) /
2559 EXT4_INODES_PER_GROUP(inode->i_sb);
2560 inodes_per_buffer = bh->b_size /
2561 EXT4_INODE_SIZE(inode->i_sb);
2562 inode_offset = ((inode->i_ino - 1) %
2563 EXT4_INODES_PER_GROUP(inode->i_sb));
2564 start = inode_offset & ~(inodes_per_buffer - 1);
2566 /* Is the inode bitmap in cache? */
2567 desc = ext4_get_group_desc(inode->i_sb,
2572 bitmap_bh = sb_getblk(inode->i_sb,
2573 ext4_inode_bitmap(inode->i_sb, desc));
2578 * If the inode bitmap isn't in cache then the
2579 * optimisation may end up performing two reads instead
2580 * of one, so skip it.
2582 if (!buffer_uptodate(bitmap_bh)) {
2586 for (i = start; i < start + inodes_per_buffer; i++) {
2587 if (i == inode_offset)
2589 if (ext4_test_bit(i, bitmap_bh->b_data))
2593 if (i == start + inodes_per_buffer) {
2594 /* all other inodes are free, so skip I/O */
2595 memset(bh->b_data, 0, bh->b_size);
2596 set_buffer_uptodate(bh);
2604 * There are other valid inodes in the buffer, this inode
2605 * has in-inode xattrs, or we don't have this inode in memory.
2606 * Read the block from disk.
2609 bh->b_end_io = end_buffer_read_sync;
2610 submit_bh(READ_META, bh);
2612 if (!buffer_uptodate(bh)) {
2613 ext4_error(inode->i_sb, "ext4_get_inode_loc",
2614 "unable to read inode block - "
2615 "inode=%lu, block=%llu",
2616 inode->i_ino, block);
2626 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
2628 /* We have all inode data except xattrs in memory here. */
2629 return __ext4_get_inode_loc(inode, iloc,
2630 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
2633 void ext4_set_inode_flags(struct inode *inode)
2635 unsigned int flags = EXT4_I(inode)->i_flags;
2637 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2638 if (flags & EXT4_SYNC_FL)
2639 inode->i_flags |= S_SYNC;
2640 if (flags & EXT4_APPEND_FL)
2641 inode->i_flags |= S_APPEND;
2642 if (flags & EXT4_IMMUTABLE_FL)
2643 inode->i_flags |= S_IMMUTABLE;
2644 if (flags & EXT4_NOATIME_FL)
2645 inode->i_flags |= S_NOATIME;
2646 if (flags & EXT4_DIRSYNC_FL)
2647 inode->i_flags |= S_DIRSYNC;
2650 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2651 void ext4_get_inode_flags(struct ext4_inode_info *ei)
2653 unsigned int flags = ei->vfs_inode.i_flags;
2655 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
2656 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
2658 ei->i_flags |= EXT4_SYNC_FL;
2659 if (flags & S_APPEND)
2660 ei->i_flags |= EXT4_APPEND_FL;
2661 if (flags & S_IMMUTABLE)
2662 ei->i_flags |= EXT4_IMMUTABLE_FL;
2663 if (flags & S_NOATIME)
2664 ei->i_flags |= EXT4_NOATIME_FL;
2665 if (flags & S_DIRSYNC)
2666 ei->i_flags |= EXT4_DIRSYNC_FL;
2669 void ext4_read_inode(struct inode * inode)
2671 struct ext4_iloc iloc;
2672 struct ext4_inode *raw_inode;
2673 struct ext4_inode_info *ei = EXT4_I(inode);
2674 struct buffer_head *bh;
2677 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2678 ei->i_acl = EXT4_ACL_NOT_CACHED;
2679 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
2681 ei->i_block_alloc_info = NULL;
2683 if (__ext4_get_inode_loc(inode, &iloc, 0))
2686 raw_inode = ext4_raw_inode(&iloc);
2687 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2688 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2689 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2690 if(!(test_opt (inode->i_sb, NO_UID32))) {
2691 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2692 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2694 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2695 inode->i_size = le32_to_cpu(raw_inode->i_size);
2698 ei->i_dir_start_lookup = 0;
2699 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2700 /* We now have enough fields to check if the inode was active or not.
2701 * This is needed because nfsd might try to access dead inodes
2702 * the test is that same one that e2fsck uses
2703 * NeilBrown 1999oct15
2705 if (inode->i_nlink == 0) {
2706 if (inode->i_mode == 0 ||
2707 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
2708 /* this inode is deleted */
2712 /* The only unlinked inodes we let through here have
2713 * valid i_mode and are being read by the orphan
2714 * recovery code: that's fine, we're about to complete
2715 * the process of deleting those. */
2717 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2718 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2719 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2720 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2721 cpu_to_le32(EXT4_OS_HURD))
2723 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
2724 if (!S_ISREG(inode->i_mode)) {
2725 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2728 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2730 ei->i_disksize = inode->i_size;
2731 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2732 ei->i_block_group = iloc.block_group;
2734 * NOTE! The in-memory inode i_data array is in little-endian order
2735 * even on big-endian machines: we do NOT byteswap the block numbers!
2737 for (block = 0; block < EXT4_N_BLOCKS; block++)
2738 ei->i_data[block] = raw_inode->i_block[block];
2739 INIT_LIST_HEAD(&ei->i_orphan);
2741 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2742 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
2744 * When mke2fs creates big inodes it does not zero out
2745 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
2746 * so ignore those first few inodes.
2748 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2749 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2750 EXT4_INODE_SIZE(inode->i_sb)) {
2754 if (ei->i_extra_isize == 0) {
2755 /* The extra space is currently unused. Use it. */
2756 ei->i_extra_isize = sizeof(struct ext4_inode) -
2757 EXT4_GOOD_OLD_INODE_SIZE;
2759 __le32 *magic = (void *)raw_inode +
2760 EXT4_GOOD_OLD_INODE_SIZE +
2762 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2763 ei->i_state |= EXT4_STATE_XATTR;
2766 ei->i_extra_isize = 0;
2768 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
2769 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
2770 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
2771 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
2773 if (S_ISREG(inode->i_mode)) {
2774 inode->i_op = &ext4_file_inode_operations;
2775 inode->i_fop = &ext4_file_operations;
2776 ext4_set_aops(inode);
2777 } else if (S_ISDIR(inode->i_mode)) {
2778 inode->i_op = &ext4_dir_inode_operations;
2779 inode->i_fop = &ext4_dir_operations;
2780 } else if (S_ISLNK(inode->i_mode)) {
2781 if (ext4_inode_is_fast_symlink(inode))
2782 inode->i_op = &ext4_fast_symlink_inode_operations;
2784 inode->i_op = &ext4_symlink_inode_operations;
2785 ext4_set_aops(inode);
2788 inode->i_op = &ext4_special_inode_operations;
2789 if (raw_inode->i_block[0])
2790 init_special_inode(inode, inode->i_mode,
2791 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2793 init_special_inode(inode, inode->i_mode,
2794 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2797 ext4_set_inode_flags(inode);
2801 make_bad_inode(inode);
2806 * Post the struct inode info into an on-disk inode location in the
2807 * buffer-cache. This gobbles the caller's reference to the
2808 * buffer_head in the inode location struct.
2810 * The caller must have write access to iloc->bh.
2812 static int ext4_do_update_inode(handle_t *handle,
2813 struct inode *inode,
2814 struct ext4_iloc *iloc)
2816 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2817 struct ext4_inode_info *ei = EXT4_I(inode);
2818 struct buffer_head *bh = iloc->bh;
2819 int err = 0, rc, block;
2821 /* For fields not not tracking in the in-memory inode,
2822 * initialise them to zero for new inodes. */
2823 if (ei->i_state & EXT4_STATE_NEW)
2824 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2826 ext4_get_inode_flags(ei);
2827 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2828 if(!(test_opt(inode->i_sb, NO_UID32))) {
2829 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2830 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2832 * Fix up interoperability with old kernels. Otherwise, old inodes get
2833 * re-used with the upper 16 bits of the uid/gid intact
2836 raw_inode->i_uid_high =
2837 cpu_to_le16(high_16_bits(inode->i_uid));
2838 raw_inode->i_gid_high =
2839 cpu_to_le16(high_16_bits(inode->i_gid));
2841 raw_inode->i_uid_high = 0;
2842 raw_inode->i_gid_high = 0;
2845 raw_inode->i_uid_low =
2846 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2847 raw_inode->i_gid_low =
2848 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2849 raw_inode->i_uid_high = 0;
2850 raw_inode->i_gid_high = 0;
2852 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2853 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2855 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
2856 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
2857 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
2858 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
2860 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2861 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2862 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2863 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2864 cpu_to_le32(EXT4_OS_HURD))
2865 raw_inode->i_file_acl_high =
2866 cpu_to_le16(ei->i_file_acl >> 32);
2867 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2868 if (!S_ISREG(inode->i_mode)) {
2869 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2871 raw_inode->i_size_high =
2872 cpu_to_le32(ei->i_disksize >> 32);
2873 if (ei->i_disksize > 0x7fffffffULL) {
2874 struct super_block *sb = inode->i_sb;
2875 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2876 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2877 EXT4_SB(sb)->s_es->s_rev_level ==
2878 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
2879 /* If this is the first large file
2880 * created, add a flag to the superblock.
2882 err = ext4_journal_get_write_access(handle,
2883 EXT4_SB(sb)->s_sbh);
2886 ext4_update_dynamic_rev(sb);
2887 EXT4_SET_RO_COMPAT_FEATURE(sb,
2888 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
2891 err = ext4_journal_dirty_metadata(handle,
2892 EXT4_SB(sb)->s_sbh);
2896 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2897 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2898 if (old_valid_dev(inode->i_rdev)) {
2899 raw_inode->i_block[0] =
2900 cpu_to_le32(old_encode_dev(inode->i_rdev));
2901 raw_inode->i_block[1] = 0;
2903 raw_inode->i_block[0] = 0;
2904 raw_inode->i_block[1] =
2905 cpu_to_le32(new_encode_dev(inode->i_rdev));
2906 raw_inode->i_block[2] = 0;
2908 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
2909 raw_inode->i_block[block] = ei->i_data[block];
2911 if (ei->i_extra_isize)
2912 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2914 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2915 rc = ext4_journal_dirty_metadata(handle, bh);
2918 ei->i_state &= ~EXT4_STATE_NEW;
2922 ext4_std_error(inode->i_sb, err);
2927 * ext4_write_inode()
2929 * We are called from a few places:
2931 * - Within generic_file_write() for O_SYNC files.
2932 * Here, there will be no transaction running. We wait for any running
2933 * trasnaction to commit.
2935 * - Within sys_sync(), kupdate and such.
2936 * We wait on commit, if tol to.
2938 * - Within prune_icache() (PF_MEMALLOC == true)
2939 * Here we simply return. We can't afford to block kswapd on the
2942 * In all cases it is actually safe for us to return without doing anything,
2943 * because the inode has been copied into a raw inode buffer in
2944 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2947 * Note that we are absolutely dependent upon all inode dirtiers doing the
2948 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2949 * which we are interested.
2951 * It would be a bug for them to not do this. The code:
2953 * mark_inode_dirty(inode)
2955 * inode->i_size = expr;
2957 * is in error because a kswapd-driven write_inode() could occur while
2958 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2959 * will no longer be on the superblock's dirty inode list.
2961 int ext4_write_inode(struct inode *inode, int wait)
2963 if (current->flags & PF_MEMALLOC)
2966 if (ext4_journal_current_handle()) {
2967 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
2975 return ext4_force_commit(inode->i_sb);
2981 * Called from notify_change.
2983 * We want to trap VFS attempts to truncate the file as soon as
2984 * possible. In particular, we want to make sure that when the VFS
2985 * shrinks i_size, we put the inode on the orphan list and modify
2986 * i_disksize immediately, so that during the subsequent flushing of
2987 * dirty pages and freeing of disk blocks, we can guarantee that any
2988 * commit will leave the blocks being flushed in an unused state on
2989 * disk. (On recovery, the inode will get truncated and the blocks will
2990 * be freed, so we have a strong guarantee that no future commit will
2991 * leave these blocks visible to the user.)
2993 * Called with inode->sem down.
2995 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
2997 struct inode *inode = dentry->d_inode;
2999 const unsigned int ia_valid = attr->ia_valid;
3001 error = inode_change_ok(inode, attr);
3005 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3006 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3009 /* (user+group)*(old+new) structure, inode write (sb,
3010 * inode block, ? - but truncate inode update has it) */
3011 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
3012 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
3013 if (IS_ERR(handle)) {
3014 error = PTR_ERR(handle);
3017 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3019 ext4_journal_stop(handle);
3022 /* Update corresponding info in inode so that everything is in
3023 * one transaction */
3024 if (attr->ia_valid & ATTR_UID)
3025 inode->i_uid = attr->ia_uid;
3026 if (attr->ia_valid & ATTR_GID)
3027 inode->i_gid = attr->ia_gid;
3028 error = ext4_mark_inode_dirty(handle, inode);
3029 ext4_journal_stop(handle);
3032 if (S_ISREG(inode->i_mode) &&
3033 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3036 handle = ext4_journal_start(inode, 3);
3037 if (IS_ERR(handle)) {
3038 error = PTR_ERR(handle);
3042 error = ext4_orphan_add(handle, inode);
3043 EXT4_I(inode)->i_disksize = attr->ia_size;
3044 rc = ext4_mark_inode_dirty(handle, inode);
3047 ext4_journal_stop(handle);
3050 rc = inode_setattr(inode, attr);
3052 /* If inode_setattr's call to ext4_truncate failed to get a
3053 * transaction handle at all, we need to clean up the in-core
3054 * orphan list manually. */
3056 ext4_orphan_del(NULL, inode);
3058 if (!rc && (ia_valid & ATTR_MODE))
3059 rc = ext4_acl_chmod(inode);
3062 ext4_std_error(inode->i_sb, error);
3070 * How many blocks doth make a writepage()?
3072 * With N blocks per page, it may be:
3077 * N+5 bitmap blocks (from the above)
3078 * N+5 group descriptor summary blocks
3081 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3083 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3085 * With ordered or writeback data it's the same, less the N data blocks.
3087 * If the inode's direct blocks can hold an integral number of pages then a
3088 * page cannot straddle two indirect blocks, and we can only touch one indirect
3089 * and dindirect block, and the "5" above becomes "3".
3091 * This still overestimates under most circumstances. If we were to pass the
3092 * start and end offsets in here as well we could do block_to_path() on each
3093 * block and work out the exact number of indirects which are touched. Pah.
3096 int ext4_writepage_trans_blocks(struct inode *inode)
3098 int bpp = ext4_journal_blocks_per_page(inode);
3099 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
3102 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3103 return ext4_ext_writepage_trans_blocks(inode, bpp);
3105 if (ext4_should_journal_data(inode))
3106 ret = 3 * (bpp + indirects) + 2;
3108 ret = 2 * (bpp + indirects) + 2;
3111 /* We know that structure was already allocated during DQUOT_INIT so
3112 * we will be updating only the data blocks + inodes */
3113 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
3120 * The caller must have previously called ext4_reserve_inode_write().
3121 * Give this, we know that the caller already has write access to iloc->bh.
3123 int ext4_mark_iloc_dirty(handle_t *handle,
3124 struct inode *inode, struct ext4_iloc *iloc)
3128 /* the do_update_inode consumes one bh->b_count */
3131 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3132 err = ext4_do_update_inode(handle, inode, iloc);
3138 * On success, We end up with an outstanding reference count against
3139 * iloc->bh. This _must_ be cleaned up later.
3143 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3144 struct ext4_iloc *iloc)
3148 err = ext4_get_inode_loc(inode, iloc);
3150 BUFFER_TRACE(iloc->bh, "get_write_access");
3151 err = ext4_journal_get_write_access(handle, iloc->bh);
3158 ext4_std_error(inode->i_sb, err);
3163 * Expand an inode by new_extra_isize bytes.
3164 * Returns 0 on success or negative error number on failure.
3166 int ext4_expand_extra_isize(struct inode *inode, unsigned int new_extra_isize,
3167 struct ext4_iloc iloc, handle_t *handle)
3169 struct ext4_inode *raw_inode;
3170 struct ext4_xattr_ibody_header *header;
3171 struct ext4_xattr_entry *entry;
3173 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
3176 raw_inode = ext4_raw_inode(&iloc);
3178 header = IHDR(inode, raw_inode);
3179 entry = IFIRST(header);
3181 /* No extended attributes present */
3182 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
3183 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
3184 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
3186 EXT4_I(inode)->i_extra_isize = new_extra_isize;
3190 /* try to expand with EAs present */
3191 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
3196 * What we do here is to mark the in-core inode as clean with respect to inode
3197 * dirtiness (it may still be data-dirty).
3198 * This means that the in-core inode may be reaped by prune_icache
3199 * without having to perform any I/O. This is a very good thing,
3200 * because *any* task may call prune_icache - even ones which
3201 * have a transaction open against a different journal.
3203 * Is this cheating? Not really. Sure, we haven't written the
3204 * inode out, but prune_icache isn't a user-visible syncing function.
3205 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3206 * we start and wait on commits.
3208 * Is this efficient/effective? Well, we're being nice to the system
3209 * by cleaning up our inodes proactively so they can be reaped
3210 * without I/O. But we are potentially leaving up to five seconds'
3211 * worth of inodes floating about which prune_icache wants us to
3212 * write out. One way to fix that would be to get prune_icache()
3213 * to do a write_super() to free up some memory. It has the desired
3216 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
3218 struct ext4_iloc iloc;
3219 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3220 static unsigned int mnt_count;
3224 err = ext4_reserve_inode_write(handle, inode, &iloc);
3225 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
3226 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
3228 * We need extra buffer credits since we may write into EA block
3229 * with this same handle. If journal_extend fails, then it will
3230 * only result in a minor loss of functionality for that inode.
3231 * If this is felt to be critical, then e2fsck should be run to
3232 * force a large enough s_min_extra_isize.
3234 if ((jbd2_journal_extend(handle,
3235 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
3236 ret = ext4_expand_extra_isize(inode,
3237 sbi->s_want_extra_isize,
3240 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
3242 le16_to_cpu(sbi->s_es->s_mnt_count)) {
3243 ext4_warning(inode->i_sb, __FUNCTION__,
3244 "Unable to expand inode %lu. Delete"
3245 " some EAs or run e2fsck.",
3248 le16_to_cpu(sbi->s_es->s_mnt_count);
3254 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3259 * ext4_dirty_inode() is called from __mark_inode_dirty()
3261 * We're really interested in the case where a file is being extended.
3262 * i_size has been changed by generic_commit_write() and we thus need
3263 * to include the updated inode in the current transaction.
3265 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3266 * are allocated to the file.
3268 * If the inode is marked synchronous, we don't honour that here - doing
3269 * so would cause a commit on atime updates, which we don't bother doing.
3270 * We handle synchronous inodes at the highest possible level.
3272 void ext4_dirty_inode(struct inode *inode)
3274 handle_t *current_handle = ext4_journal_current_handle();
3277 handle = ext4_journal_start(inode, 2);
3280 if (current_handle &&
3281 current_handle->h_transaction != handle->h_transaction) {
3282 /* This task has a transaction open against a different fs */
3283 printk(KERN_EMERG "%s: transactions do not match!\n",
3286 jbd_debug(5, "marking dirty. outer handle=%p\n",
3288 ext4_mark_inode_dirty(handle, inode);
3290 ext4_journal_stop(handle);
3297 * Bind an inode's backing buffer_head into this transaction, to prevent
3298 * it from being flushed to disk early. Unlike
3299 * ext4_reserve_inode_write, this leaves behind no bh reference and
3300 * returns no iloc structure, so the caller needs to repeat the iloc
3301 * lookup to mark the inode dirty later.
3303 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
3305 struct ext4_iloc iloc;
3309 err = ext4_get_inode_loc(inode, &iloc);
3311 BUFFER_TRACE(iloc.bh, "get_write_access");
3312 err = jbd2_journal_get_write_access(handle, iloc.bh);
3314 err = ext4_journal_dirty_metadata(handle,
3319 ext4_std_error(inode->i_sb, err);
3324 int ext4_change_inode_journal_flag(struct inode *inode, int val)
3331 * We have to be very careful here: changing a data block's
3332 * journaling status dynamically is dangerous. If we write a
3333 * data block to the journal, change the status and then delete
3334 * that block, we risk forgetting to revoke the old log record
3335 * from the journal and so a subsequent replay can corrupt data.
3336 * So, first we make sure that the journal is empty and that
3337 * nobody is changing anything.
3340 journal = EXT4_JOURNAL(inode);
3341 if (is_journal_aborted(journal))
3344 jbd2_journal_lock_updates(journal);
3345 jbd2_journal_flush(journal);
3348 * OK, there are no updates running now, and all cached data is
3349 * synced to disk. We are now in a completely consistent state
3350 * which doesn't have anything in the journal, and we know that
3351 * no filesystem updates are running, so it is safe to modify
3352 * the inode's in-core data-journaling state flag now.
3356 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
3358 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3359 ext4_set_aops(inode);
3361 jbd2_journal_unlock_updates(journal);
3363 /* Finally we can mark the inode as dirty. */
3365 handle = ext4_journal_start(inode, 1);
3367 return PTR_ERR(handle);
3369 err = ext4_mark_inode_dirty(handle, inode);
3371 ext4_journal_stop(handle);
3372 ext4_std_error(inode->i_sb, err);