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/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode *inode,
49 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
53 static void ext4_invalidatepage(struct page *page, unsigned long offset);
56 * Test whether an inode is a fast symlink.
58 static int ext4_inode_is_fast_symlink(struct inode *inode)
60 int ea_blocks = EXT4_I(inode)->i_file_acl ?
61 (inode->i_sb->s_blocksize >> 9) : 0;
63 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
67 * The ext4 forget function must perform a revoke if we are freeing data
68 * which has been journaled. Metadata (eg. indirect blocks) must be
69 * revoked in all cases.
71 * "bh" may be NULL: a metadata block may have been freed from memory
72 * but there may still be a record of it in the journal, and that record
73 * still needs to be revoked.
75 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
76 struct buffer_head *bh, ext4_fsblk_t blocknr)
82 BUFFER_TRACE(bh, "enter");
84 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
86 bh, is_metadata, inode->i_mode,
87 test_opt(inode->i_sb, DATA_FLAGS));
89 /* Never use the revoke function if we are doing full data
90 * journaling: there is no need to, and a V1 superblock won't
91 * support it. Otherwise, only skip the revoke on un-journaled
94 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
95 (!is_metadata && !ext4_should_journal_data(inode))) {
97 BUFFER_TRACE(bh, "call jbd2_journal_forget");
98 return ext4_journal_forget(handle, bh);
104 * data!=journal && (is_metadata || should_journal_data(inode))
106 BUFFER_TRACE(bh, "call ext4_journal_revoke");
107 err = ext4_journal_revoke(handle, blocknr, bh);
109 ext4_abort(inode->i_sb, __func__,
110 "error %d when attempting revoke", err);
111 BUFFER_TRACE(bh, "exit");
116 * Work out how many blocks we need to proceed with the next chunk of a
117 * truncate transaction.
119 static unsigned long blocks_for_truncate(struct inode *inode)
123 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
125 /* Give ourselves just enough room to cope with inodes in which
126 * i_blocks is corrupt: we've seen disk corruptions in the past
127 * which resulted in random data in an inode which looked enough
128 * like a regular file for ext4 to try to delete it. Things
129 * will go a bit crazy if that happens, but at least we should
130 * try not to panic the whole kernel. */
134 /* But we need to bound the transaction so we don't overflow the
136 if (needed > EXT4_MAX_TRANS_DATA)
137 needed = EXT4_MAX_TRANS_DATA;
139 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
143 * Truncate transactions can be complex and absolutely huge. So we need to
144 * be able to restart the transaction at a conventient checkpoint to make
145 * sure we don't overflow the journal.
147 * start_transaction gets us a new handle for a truncate transaction,
148 * and extend_transaction tries to extend the existing one a bit. If
149 * extend fails, we need to propagate the failure up and restart the
150 * transaction in the top-level truncate loop. --sct
152 static handle_t *start_transaction(struct inode *inode)
156 result = ext4_journal_start(inode, blocks_for_truncate(inode));
160 ext4_std_error(inode->i_sb, PTR_ERR(result));
165 * Try to extend this transaction for the purposes of truncation.
167 * Returns 0 if we managed to create more room. If we can't create more
168 * room, and the transaction must be restarted we return 1.
170 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
172 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
174 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
180 * Restart the transaction associated with *handle. This does a commit,
181 * so before we call here everything must be consistently dirtied against
184 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
186 jbd_debug(2, "restarting handle %p\n", handle);
187 return ext4_journal_restart(handle, blocks_for_truncate(inode));
191 * Called at the last iput() if i_nlink is zero.
193 void ext4_delete_inode(struct inode *inode)
198 if (ext4_should_order_data(inode))
199 ext4_begin_ordered_truncate(inode, 0);
200 truncate_inode_pages(&inode->i_data, 0);
202 if (is_bad_inode(inode))
205 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
206 if (IS_ERR(handle)) {
207 ext4_std_error(inode->i_sb, PTR_ERR(handle));
209 * If we're going to skip the normal cleanup, we still need to
210 * make sure that the in-core orphan linked list is properly
213 ext4_orphan_del(NULL, inode);
220 err = ext4_mark_inode_dirty(handle, inode);
222 ext4_warning(inode->i_sb, __func__,
223 "couldn't mark inode dirty (err %d)", err);
227 ext4_truncate(inode);
230 * ext4_ext_truncate() doesn't reserve any slop when it
231 * restarts journal transactions; therefore there may not be
232 * enough credits left in the handle to remove the inode from
233 * the orphan list and set the dtime field.
235 if (handle->h_buffer_credits < 3) {
236 err = ext4_journal_extend(handle, 3);
238 err = ext4_journal_restart(handle, 3);
240 ext4_warning(inode->i_sb, __func__,
241 "couldn't extend journal (err %d)", err);
243 ext4_journal_stop(handle);
249 * Kill off the orphan record which ext4_truncate created.
250 * AKPM: I think this can be inside the above `if'.
251 * Note that ext4_orphan_del() has to be able to cope with the
252 * deletion of a non-existent orphan - this is because we don't
253 * know if ext4_truncate() actually created an orphan record.
254 * (Well, we could do this if we need to, but heck - it works)
256 ext4_orphan_del(handle, inode);
257 EXT4_I(inode)->i_dtime = get_seconds();
260 * One subtle ordering requirement: if anything has gone wrong
261 * (transaction abort, IO errors, whatever), then we can still
262 * do these next steps (the fs will already have been marked as
263 * having errors), but we can't free the inode if the mark_dirty
266 if (ext4_mark_inode_dirty(handle, inode))
267 /* If that failed, just do the required in-core inode clear. */
270 ext4_free_inode(handle, inode);
271 ext4_journal_stop(handle);
274 clear_inode(inode); /* We must guarantee clearing of inode... */
280 struct buffer_head *bh;
283 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
285 p->key = *(p->p = v);
290 * ext4_block_to_path - parse the block number into array of offsets
291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
297 * To store the locations of file's data ext4 uses a data structure common
298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
320 static int ext4_block_to_path(struct inode *inode,
322 ext4_lblk_t offsets[4], int *boundary)
324 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
325 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
326 const long direct_blocks = EXT4_NDIR_BLOCKS,
327 indirect_blocks = ptrs,
328 double_blocks = (1 << (ptrs_bits * 2));
333 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
334 } else if (i_block < direct_blocks) {
335 offsets[n++] = i_block;
336 final = direct_blocks;
337 } else if ((i_block -= direct_blocks) < indirect_blocks) {
338 offsets[n++] = EXT4_IND_BLOCK;
339 offsets[n++] = i_block;
341 } else if ((i_block -= indirect_blocks) < double_blocks) {
342 offsets[n++] = EXT4_DIND_BLOCK;
343 offsets[n++] = i_block >> ptrs_bits;
344 offsets[n++] = i_block & (ptrs - 1);
346 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
347 offsets[n++] = EXT4_TIND_BLOCK;
348 offsets[n++] = i_block >> (ptrs_bits * 2);
349 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
350 offsets[n++] = i_block & (ptrs - 1);
353 ext4_warning(inode->i_sb, "ext4_block_to_path",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks);
359 *boundary = final - 1 - (i_block & (ptrs - 1));
364 * ext4_get_branch - read the chain of indirect blocks leading to data
365 * @inode: inode in question
366 * @depth: depth of the chain (1 - direct pointer, etc.)
367 * @offsets: offsets of pointers in inode/indirect blocks
368 * @chain: place to store the result
369 * @err: here we store the error value
371 * Function fills the array of triples <key, p, bh> and returns %NULL
372 * if everything went OK or the pointer to the last filled triple
373 * (incomplete one) otherwise. Upon the return chain[i].key contains
374 * the number of (i+1)-th block in the chain (as it is stored in memory,
375 * i.e. little-endian 32-bit), chain[i].p contains the address of that
376 * number (it points into struct inode for i==0 and into the bh->b_data
377 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
378 * block for i>0 and NULL for i==0. In other words, it holds the block
379 * numbers of the chain, addresses they were taken from (and where we can
380 * verify that chain did not change) and buffer_heads hosting these
383 * Function stops when it stumbles upon zero pointer (absent block)
384 * (pointer to last triple returned, *@err == 0)
385 * or when it gets an IO error reading an indirect block
386 * (ditto, *@err == -EIO)
387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
390 * Need to be called with
391 * down_read(&EXT4_I(inode)->i_data_sem)
393 static Indirect *ext4_get_branch(struct inode *inode, int depth,
394 ext4_lblk_t *offsets,
395 Indirect chain[4], int *err)
397 struct super_block *sb = inode->i_sb;
399 struct buffer_head *bh;
402 /* i_data is not going away, no lock needed */
403 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
407 bh = sb_bread(sb, le32_to_cpu(p->key));
410 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
424 * ext4_find_near - find a place for allocation with sufficient locality
426 * @ind: descriptor of indirect block.
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
441 * Caller must make sure that @ind is valid and will stay that way.
443 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
445 struct ext4_inode_info *ei = EXT4_I(inode);
446 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
448 ext4_fsblk_t bg_start;
449 ext4_fsblk_t last_block;
450 ext4_grpblk_t colour;
452 /* Try to find previous block */
453 for (p = ind->p - 1; p >= start; p--) {
455 return le32_to_cpu(*p);
458 /* No such thing, so let's try location of indirect block */
460 return ind->bh->b_blocknr;
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
466 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
467 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
469 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
470 colour = (current->pid % 16) *
471 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
473 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
474 return bg_start + colour;
478 * ext4_find_goal - find a preferred place for allocation.
480 * @block: block we want
481 * @partial: pointer to the last triple within a chain
483 * Normally this function find the preferred place for block allocation,
486 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
489 struct ext4_block_alloc_info *block_i;
491 block_i = EXT4_I(inode)->i_block_alloc_info;
494 * try the heuristic for sequential allocation,
495 * failing that at least try to get decent locality.
497 if (block_i && (block == block_i->last_alloc_logical_block + 1)
498 && (block_i->last_alloc_physical_block != 0)) {
499 return block_i->last_alloc_physical_block + 1;
502 return ext4_find_near(inode, partial);
506 * ext4_blks_to_allocate: Look up the block map and count the number
507 * of direct blocks need to be allocated for the given branch.
509 * @branch: chain of indirect blocks
510 * @k: number of blocks need for indirect blocks
511 * @blks: number of data blocks to be mapped.
512 * @blocks_to_boundary: the offset in the indirect block
514 * return the total number of blocks to be allocate, including the
515 * direct and indirect blocks.
517 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
518 int blocks_to_boundary)
520 unsigned long count = 0;
523 * Simple case, [t,d]Indirect block(s) has not allocated yet
524 * then it's clear blocks on that path have not allocated
527 /* right now we don't handle cross boundary allocation */
528 if (blks < blocks_to_boundary + 1)
531 count += blocks_to_boundary + 1;
536 while (count < blks && count <= blocks_to_boundary &&
537 le32_to_cpu(*(branch[0].p + count)) == 0) {
544 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
545 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @new_blocks: on return it will store the new block numbers for
549 * the indirect blocks(if needed) and the first direct block,
550 * @blks: on return it will store the total number of allocated
553 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
554 ext4_lblk_t iblock, ext4_fsblk_t goal,
555 int indirect_blks, int blks,
556 ext4_fsblk_t new_blocks[4], int *err)
559 unsigned long count = 0, blk_allocated = 0;
561 ext4_fsblk_t current_block = 0;
565 * Here we try to allocate the requested multiple blocks at once,
566 * on a best-effort basis.
567 * To build a branch, we should allocate blocks for
568 * the indirect blocks(if not allocated yet), and at least
569 * the first direct block of this branch. That's the
570 * minimum number of blocks need to allocate(required)
572 /* first we try to allocate the indirect blocks */
573 target = indirect_blks;
576 /* allocating blocks for indirect blocks and direct blocks */
577 current_block = ext4_new_meta_blocks(handle, inode,
583 /* allocate blocks for indirect blocks */
584 while (index < indirect_blks && count) {
585 new_blocks[index++] = current_block++;
590 * save the new block number
591 * for the first direct block
593 new_blocks[index] = current_block;
594 printk(KERN_INFO "%s returned more blocks than "
595 "requested\n", __func__);
601 target = blks - count ;
602 blk_allocated = count;
605 /* Now allocate data blocks */
607 /* allocating blocks for data blocks */
608 current_block = ext4_new_blocks(handle, inode, iblock,
610 if (*err && (target == blks)) {
612 * if the allocation failed and we didn't allocate
618 if (target == blks) {
620 * save the new block number
621 * for the first direct block
623 new_blocks[index] = current_block;
625 blk_allocated += count;
628 /* total number of blocks allocated for direct blocks */
633 for (i = 0; i < index; i++)
634 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
639 * ext4_alloc_branch - allocate and set up a chain of blocks.
641 * @indirect_blks: number of allocated indirect blocks
642 * @blks: number of allocated direct blocks
643 * @offsets: offsets (in the blocks) to store the pointers to next.
644 * @branch: place to store the chain in.
646 * This function allocates blocks, zeroes out all but the last one,
647 * links them into chain and (if we are synchronous) writes them to disk.
648 * In other words, it prepares a branch that can be spliced onto the
649 * inode. It stores the information about that chain in the branch[], in
650 * the same format as ext4_get_branch() would do. We are calling it after
651 * we had read the existing part of chain and partial points to the last
652 * triple of that (one with zero ->key). Upon the exit we have the same
653 * picture as after the successful ext4_get_block(), except that in one
654 * place chain is disconnected - *branch->p is still zero (we did not
655 * set the last link), but branch->key contains the number that should
656 * be placed into *branch->p to fill that gap.
658 * If allocation fails we free all blocks we've allocated (and forget
659 * their buffer_heads) and return the error value the from failed
660 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
661 * as described above and return 0.
663 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
664 ext4_lblk_t iblock, int indirect_blks,
665 int *blks, ext4_fsblk_t goal,
666 ext4_lblk_t *offsets, Indirect *branch)
668 int blocksize = inode->i_sb->s_blocksize;
671 struct buffer_head *bh;
673 ext4_fsblk_t new_blocks[4];
674 ext4_fsblk_t current_block;
676 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
677 *blks, new_blocks, &err);
681 branch[0].key = cpu_to_le32(new_blocks[0]);
683 * metadata blocks and data blocks are allocated.
685 for (n = 1; n <= indirect_blks; n++) {
687 * Get buffer_head for parent block, zero it out
688 * and set the pointer to new one, then send
691 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
694 BUFFER_TRACE(bh, "call get_create_access");
695 err = ext4_journal_get_create_access(handle, bh);
702 memset(bh->b_data, 0, blocksize);
703 branch[n].p = (__le32 *) bh->b_data + offsets[n];
704 branch[n].key = cpu_to_le32(new_blocks[n]);
705 *branch[n].p = branch[n].key;
706 if (n == indirect_blks) {
707 current_block = new_blocks[n];
709 * End of chain, update the last new metablock of
710 * the chain to point to the new allocated
711 * data blocks numbers
713 for (i=1; i < num; i++)
714 *(branch[n].p + i) = cpu_to_le32(++current_block);
716 BUFFER_TRACE(bh, "marking uptodate");
717 set_buffer_uptodate(bh);
720 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
721 err = ext4_journal_dirty_metadata(handle, bh);
728 /* Allocation failed, free what we already allocated */
729 for (i = 1; i <= n ; i++) {
730 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
731 ext4_journal_forget(handle, branch[i].bh);
733 for (i = 0; i < indirect_blks; i++)
734 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
736 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
742 * ext4_splice_branch - splice the allocated branch onto inode.
744 * @block: (logical) number of block we are adding
745 * @chain: chain of indirect blocks (with a missing link - see
747 * @where: location of missing link
748 * @num: number of indirect blocks we are adding
749 * @blks: number of direct blocks we are adding
751 * This function fills the missing link and does all housekeeping needed in
752 * inode (->i_blocks, etc.). In case of success we end up with the full
753 * chain to new block and return 0.
755 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
756 ext4_lblk_t block, Indirect *where, int num, int blks)
760 struct ext4_block_alloc_info *block_i;
761 ext4_fsblk_t current_block;
763 block_i = EXT4_I(inode)->i_block_alloc_info;
765 * If we're splicing into a [td]indirect block (as opposed to the
766 * inode) then we need to get write access to the [td]indirect block
770 BUFFER_TRACE(where->bh, "get_write_access");
771 err = ext4_journal_get_write_access(handle, where->bh);
777 *where->p = where->key;
780 * Update the host buffer_head or inode to point to more just allocated
781 * direct blocks blocks
783 if (num == 0 && blks > 1) {
784 current_block = le32_to_cpu(where->key) + 1;
785 for (i = 1; i < blks; i++)
786 *(where->p + i) = cpu_to_le32(current_block++);
790 * update the most recently allocated logical & physical block
791 * in i_block_alloc_info, to assist find the proper goal block for next
795 block_i->last_alloc_logical_block = block + blks - 1;
796 block_i->last_alloc_physical_block =
797 le32_to_cpu(where[num].key) + blks - 1;
800 /* We are done with atomic stuff, now do the rest of housekeeping */
802 inode->i_ctime = ext4_current_time(inode);
803 ext4_mark_inode_dirty(handle, inode);
805 /* had we spliced it onto indirect block? */
808 * If we spliced it onto an indirect block, we haven't
809 * altered the inode. Note however that if it is being spliced
810 * onto an indirect block at the very end of the file (the
811 * file is growing) then we *will* alter the inode to reflect
812 * the new i_size. But that is not done here - it is done in
813 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
815 jbd_debug(5, "splicing indirect only\n");
816 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
817 err = ext4_journal_dirty_metadata(handle, where->bh);
822 * OK, we spliced it into the inode itself on a direct block.
823 * Inode was dirtied above.
825 jbd_debug(5, "splicing direct\n");
830 for (i = 1; i <= num; i++) {
831 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
832 ext4_journal_forget(handle, where[i].bh);
833 ext4_free_blocks(handle, inode,
834 le32_to_cpu(where[i-1].key), 1, 0);
836 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
842 * Allocation strategy is simple: if we have to allocate something, we will
843 * have to go the whole way to leaf. So let's do it before attaching anything
844 * to tree, set linkage between the newborn blocks, write them if sync is
845 * required, recheck the path, free and repeat if check fails, otherwise
846 * set the last missing link (that will protect us from any truncate-generated
847 * removals - all blocks on the path are immune now) and possibly force the
848 * write on the parent block.
849 * That has a nice additional property: no special recovery from the failed
850 * allocations is needed - we simply release blocks and do not touch anything
851 * reachable from inode.
853 * `handle' can be NULL if create == 0.
855 * return > 0, # of blocks mapped or allocated.
856 * return = 0, if plain lookup failed.
857 * return < 0, error case.
860 * Need to be called with
861 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
862 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
864 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
865 ext4_lblk_t iblock, unsigned long maxblocks,
866 struct buffer_head *bh_result,
867 int create, int extend_disksize)
870 ext4_lblk_t offsets[4];
875 int blocks_to_boundary = 0;
877 struct ext4_inode_info *ei = EXT4_I(inode);
879 ext4_fsblk_t first_block = 0;
883 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
884 J_ASSERT(handle != NULL || create == 0);
885 depth = ext4_block_to_path(inode, iblock, offsets,
886 &blocks_to_boundary);
891 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
893 /* Simplest case - block found, no allocation needed */
895 first_block = le32_to_cpu(chain[depth - 1].key);
896 clear_buffer_new(bh_result);
899 while (count < maxblocks && count <= blocks_to_boundary) {
902 blk = le32_to_cpu(*(chain[depth-1].p + count));
904 if (blk == first_block + count)
912 /* Next simple case - plain lookup or failed read of indirect block */
913 if (!create || err == -EIO)
917 * Okay, we need to do block allocation. Lazily initialize the block
918 * allocation info here if necessary
920 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
921 ext4_init_block_alloc_info(inode);
923 goal = ext4_find_goal(inode, iblock, partial);
925 /* the number of blocks need to allocate for [d,t]indirect blocks */
926 indirect_blks = (chain + depth) - partial - 1;
929 * Next look up the indirect map to count the totoal number of
930 * direct blocks to allocate for this branch.
932 count = ext4_blks_to_allocate(partial, indirect_blks,
933 maxblocks, blocks_to_boundary);
935 * Block out ext4_truncate while we alter the tree
937 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
939 offsets + (partial - chain), partial);
942 * The ext4_splice_branch call will free and forget any buffers
943 * on the new chain if there is a failure, but that risks using
944 * up transaction credits, especially for bitmaps where the
945 * credits cannot be returned. Can we handle this somehow? We
946 * may need to return -EAGAIN upwards in the worst case. --sct
949 err = ext4_splice_branch(handle, inode, iblock,
950 partial, indirect_blks, count);
952 * i_disksize growing is protected by i_data_sem. Don't forget to
953 * protect it if you're about to implement concurrent
954 * ext4_get_block() -bzzz
956 if (!err && extend_disksize) {
957 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
958 if (disksize > i_size_read(inode))
959 disksize = i_size_read(inode);
960 if (disksize > ei->i_disksize)
961 ei->i_disksize = disksize;
966 set_buffer_new(bh_result);
968 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
969 if (count > blocks_to_boundary)
970 set_buffer_boundary(bh_result);
972 /* Clean up and exit */
973 partial = chain + depth - 1; /* the whole chain */
975 while (partial > chain) {
976 BUFFER_TRACE(partial->bh, "call brelse");
980 BUFFER_TRACE(bh_result, "returned");
986 * Calculate the number of metadata blocks need to reserve
987 * to allocate @blocks for non extent file based file
989 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
991 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
992 int ind_blks, dind_blks, tind_blks;
994 /* number of new indirect blocks needed */
995 ind_blks = (blocks + icap - 1) / icap;
997 dind_blks = (ind_blks + icap - 1) / icap;
1001 return ind_blks + dind_blks + tind_blks;
1005 * Calculate the number of metadata blocks need to reserve
1006 * to allocate given number of blocks
1008 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1013 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1014 return ext4_ext_calc_metadata_amount(inode, blocks);
1016 return ext4_indirect_calc_metadata_amount(inode, blocks);
1019 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1021 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1022 int total, mdb, mdb_free;
1024 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1025 /* recalculate the number of metablocks still need to be reserved */
1026 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1027 mdb = ext4_calc_metadata_amount(inode, total);
1029 /* figure out how many metablocks to release */
1030 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1031 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1033 /* Account for allocated meta_blocks */
1034 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1036 /* update fs free blocks counter for truncate case */
1037 percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
1039 /* update per-inode reservations */
1040 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1041 EXT4_I(inode)->i_reserved_data_blocks -= used;
1043 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1044 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1045 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1046 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1050 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1051 * and returns if the blocks are already mapped.
1053 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1054 * and store the allocated blocks in the result buffer head and mark it
1057 * If file type is extents based, it will call ext4_ext_get_blocks(),
1058 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1061 * On success, it returns the number of blocks being mapped or allocate.
1062 * if create==0 and the blocks are pre-allocated and uninitialized block,
1063 * the result buffer head is unmapped. If the create ==1, it will make sure
1064 * the buffer head is mapped.
1066 * It returns 0 if plain look up failed (blocks have not been allocated), in
1067 * that casem, buffer head is unmapped
1069 * It returns the error in case of allocation failure.
1071 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1072 unsigned long max_blocks, struct buffer_head *bh,
1073 int create, int extend_disksize, int flag)
1077 clear_buffer_mapped(bh);
1080 * Try to see if we can get the block without requesting
1081 * for new file system block.
1083 down_read((&EXT4_I(inode)->i_data_sem));
1084 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1085 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1088 retval = ext4_get_blocks_handle(handle,
1089 inode, block, max_blocks, bh, 0, 0);
1091 up_read((&EXT4_I(inode)->i_data_sem));
1093 /* If it is only a block(s) look up */
1098 * Returns if the blocks have already allocated
1100 * Note that if blocks have been preallocated
1101 * ext4_ext_get_block() returns th create = 0
1102 * with buffer head unmapped.
1104 if (retval > 0 && buffer_mapped(bh))
1108 * New blocks allocate and/or writing to uninitialized extent
1109 * will possibly result in updating i_data, so we take
1110 * the write lock of i_data_sem, and call get_blocks()
1111 * with create == 1 flag.
1113 down_write((&EXT4_I(inode)->i_data_sem));
1116 * if the caller is from delayed allocation writeout path
1117 * we have already reserved fs blocks for allocation
1118 * let the underlying get_block() function know to
1119 * avoid double accounting
1122 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1124 * We need to check for EXT4 here because migrate
1125 * could have changed the inode type in between
1127 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1128 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1129 bh, create, extend_disksize);
1131 retval = ext4_get_blocks_handle(handle, inode, block,
1132 max_blocks, bh, create, extend_disksize);
1134 if (retval > 0 && buffer_new(bh)) {
1136 * We allocated new blocks which will result in
1137 * i_data's format changing. Force the migrate
1138 * to fail by clearing migrate flags
1140 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1146 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1148 * Update reserved blocks/metadata blocks
1149 * after successful block allocation
1150 * which were deferred till now
1152 if ((retval > 0) && buffer_delay(bh))
1153 ext4_da_update_reserve_space(inode, retval);
1156 up_write((&EXT4_I(inode)->i_data_sem));
1160 /* Maximum number of blocks we map for direct IO at once. */
1161 #define DIO_MAX_BLOCKS 4096
1163 static int ext4_get_block(struct inode *inode, sector_t iblock,
1164 struct buffer_head *bh_result, int create)
1166 handle_t *handle = ext4_journal_current_handle();
1167 int ret = 0, started = 0;
1168 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1171 if (create && !handle) {
1172 /* Direct IO write... */
1173 if (max_blocks > DIO_MAX_BLOCKS)
1174 max_blocks = DIO_MAX_BLOCKS;
1175 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1176 handle = ext4_journal_start(inode, dio_credits);
1177 if (IS_ERR(handle)) {
1178 ret = PTR_ERR(handle);
1184 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1185 max_blocks, bh_result, create, 0, 0);
1187 bh_result->b_size = (ret << inode->i_blkbits);
1191 ext4_journal_stop(handle);
1197 * `handle' can be NULL if create is zero
1199 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1200 ext4_lblk_t block, int create, int *errp)
1202 struct buffer_head dummy;
1205 J_ASSERT(handle != NULL || create == 0);
1208 dummy.b_blocknr = -1000;
1209 buffer_trace_init(&dummy.b_history);
1210 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1211 &dummy, create, 1, 0);
1213 * ext4_get_blocks_handle() returns number of blocks
1214 * mapped. 0 in case of a HOLE.
1222 if (!err && buffer_mapped(&dummy)) {
1223 struct buffer_head *bh;
1224 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1229 if (buffer_new(&dummy)) {
1230 J_ASSERT(create != 0);
1231 J_ASSERT(handle != NULL);
1234 * Now that we do not always journal data, we should
1235 * keep in mind whether this should always journal the
1236 * new buffer as metadata. For now, regular file
1237 * writes use ext4_get_block instead, so it's not a
1241 BUFFER_TRACE(bh, "call get_create_access");
1242 fatal = ext4_journal_get_create_access(handle, bh);
1243 if (!fatal && !buffer_uptodate(bh)) {
1244 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1245 set_buffer_uptodate(bh);
1248 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1249 err = ext4_journal_dirty_metadata(handle, bh);
1253 BUFFER_TRACE(bh, "not a new buffer");
1266 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1267 ext4_lblk_t block, int create, int *err)
1269 struct buffer_head *bh;
1271 bh = ext4_getblk(handle, inode, block, create, err);
1274 if (buffer_uptodate(bh))
1276 ll_rw_block(READ_META, 1, &bh);
1278 if (buffer_uptodate(bh))
1285 static int walk_page_buffers(handle_t *handle,
1286 struct buffer_head *head,
1290 int (*fn)(handle_t *handle,
1291 struct buffer_head *bh))
1293 struct buffer_head *bh;
1294 unsigned block_start, block_end;
1295 unsigned blocksize = head->b_size;
1297 struct buffer_head *next;
1299 for (bh = head, block_start = 0;
1300 ret == 0 && (bh != head || !block_start);
1301 block_start = block_end, bh = next)
1303 next = bh->b_this_page;
1304 block_end = block_start + blocksize;
1305 if (block_end <= from || block_start >= to) {
1306 if (partial && !buffer_uptodate(bh))
1310 err = (*fn)(handle, bh);
1318 * To preserve ordering, it is essential that the hole instantiation and
1319 * the data write be encapsulated in a single transaction. We cannot
1320 * close off a transaction and start a new one between the ext4_get_block()
1321 * and the commit_write(). So doing the jbd2_journal_start at the start of
1322 * prepare_write() is the right place.
1324 * Also, this function can nest inside ext4_writepage() ->
1325 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1326 * has generated enough buffer credits to do the whole page. So we won't
1327 * block on the journal in that case, which is good, because the caller may
1330 * By accident, ext4 can be reentered when a transaction is open via
1331 * quota file writes. If we were to commit the transaction while thus
1332 * reentered, there can be a deadlock - we would be holding a quota
1333 * lock, and the commit would never complete if another thread had a
1334 * transaction open and was blocking on the quota lock - a ranking
1337 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1338 * will _not_ run commit under these circumstances because handle->h_ref
1339 * is elevated. We'll still have enough credits for the tiny quotafile
1342 static int do_journal_get_write_access(handle_t *handle,
1343 struct buffer_head *bh)
1345 if (!buffer_mapped(bh) || buffer_freed(bh))
1347 return ext4_journal_get_write_access(handle, bh);
1350 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1351 loff_t pos, unsigned len, unsigned flags,
1352 struct page **pagep, void **fsdata)
1354 struct inode *inode = mapping->host;
1355 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1362 index = pos >> PAGE_CACHE_SHIFT;
1363 from = pos & (PAGE_CACHE_SIZE - 1);
1367 handle = ext4_journal_start(inode, needed_blocks);
1368 if (IS_ERR(handle)) {
1369 ret = PTR_ERR(handle);
1373 page = __grab_cache_page(mapping, index);
1375 ext4_journal_stop(handle);
1381 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1384 if (!ret && ext4_should_journal_data(inode)) {
1385 ret = walk_page_buffers(handle, page_buffers(page),
1386 from, to, NULL, do_journal_get_write_access);
1391 ext4_journal_stop(handle);
1392 page_cache_release(page);
1395 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1401 /* For write_end() in data=journal mode */
1402 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1404 if (!buffer_mapped(bh) || buffer_freed(bh))
1406 set_buffer_uptodate(bh);
1407 return ext4_journal_dirty_metadata(handle, bh);
1411 * We need to pick up the new inode size which generic_commit_write gave us
1412 * `file' can be NULL - eg, when called from page_symlink().
1414 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1415 * buffers are managed internally.
1417 static int ext4_ordered_write_end(struct file *file,
1418 struct address_space *mapping,
1419 loff_t pos, unsigned len, unsigned copied,
1420 struct page *page, void *fsdata)
1422 handle_t *handle = ext4_journal_current_handle();
1423 struct inode *inode = mapping->host;
1426 ret = ext4_jbd2_file_inode(handle, inode);
1430 * generic_write_end() will run mark_inode_dirty() if i_size
1431 * changes. So let's piggyback the i_disksize mark_inode_dirty
1436 new_i_size = pos + copied;
1437 if (new_i_size > EXT4_I(inode)->i_disksize)
1438 EXT4_I(inode)->i_disksize = new_i_size;
1439 ret2 = generic_write_end(file, mapping, pos, len, copied,
1445 ret2 = ext4_journal_stop(handle);
1449 return ret ? ret : copied;
1452 static int ext4_writeback_write_end(struct file *file,
1453 struct address_space *mapping,
1454 loff_t pos, unsigned len, unsigned copied,
1455 struct page *page, void *fsdata)
1457 handle_t *handle = ext4_journal_current_handle();
1458 struct inode *inode = mapping->host;
1462 new_i_size = pos + copied;
1463 if (new_i_size > EXT4_I(inode)->i_disksize)
1464 EXT4_I(inode)->i_disksize = new_i_size;
1466 ret2 = generic_write_end(file, mapping, pos, len, copied,
1472 ret2 = ext4_journal_stop(handle);
1476 return ret ? ret : copied;
1479 static int ext4_journalled_write_end(struct file *file,
1480 struct address_space *mapping,
1481 loff_t pos, unsigned len, unsigned copied,
1482 struct page *page, void *fsdata)
1484 handle_t *handle = ext4_journal_current_handle();
1485 struct inode *inode = mapping->host;
1490 from = pos & (PAGE_CACHE_SIZE - 1);
1494 if (!PageUptodate(page))
1496 page_zero_new_buffers(page, from+copied, to);
1499 ret = walk_page_buffers(handle, page_buffers(page), from,
1500 to, &partial, write_end_fn);
1502 SetPageUptodate(page);
1503 if (pos+copied > inode->i_size)
1504 i_size_write(inode, pos+copied);
1505 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1506 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1507 EXT4_I(inode)->i_disksize = inode->i_size;
1508 ret2 = ext4_mark_inode_dirty(handle, inode);
1514 ret2 = ext4_journal_stop(handle);
1517 page_cache_release(page);
1519 return ret ? ret : copied;
1522 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1524 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1525 unsigned long md_needed, mdblocks, total = 0;
1528 * recalculate the amount of metadata blocks to reserve
1529 * in order to allocate nrblocks
1530 * worse case is one extent per block
1532 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1533 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1534 mdblocks = ext4_calc_metadata_amount(inode, total);
1535 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1537 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1538 total = md_needed + nrblocks;
1540 if (ext4_claim_free_blocks(sbi, total)) {
1541 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1544 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1545 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1547 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1548 return 0; /* success */
1551 static void ext4_da_release_space(struct inode *inode, int to_free)
1553 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1554 int total, mdb, mdb_free, release;
1557 return; /* Nothing to release, exit */
1559 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1561 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1563 * if there is no reserved blocks, but we try to free some
1564 * then the counter is messed up somewhere.
1565 * but since this function is called from invalidate
1566 * page, it's harmless to return without any action
1568 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1569 "blocks for inode %lu, but there is no reserved "
1570 "data blocks\n", to_free, inode->i_ino);
1571 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1575 /* recalculate the number of metablocks still need to be reserved */
1576 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1577 mdb = ext4_calc_metadata_amount(inode, total);
1579 /* figure out how many metablocks to release */
1580 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1581 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1583 release = to_free + mdb_free;
1585 /* update fs free blocks counter for truncate case */
1586 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1588 /* update per-inode reservations */
1589 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1590 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1592 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1593 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1594 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1597 static void ext4_da_page_release_reservation(struct page *page,
1598 unsigned long offset)
1601 struct buffer_head *head, *bh;
1602 unsigned int curr_off = 0;
1604 head = page_buffers(page);
1607 unsigned int next_off = curr_off + bh->b_size;
1609 if ((offset <= curr_off) && (buffer_delay(bh))) {
1611 clear_buffer_delay(bh);
1613 curr_off = next_off;
1614 } while ((bh = bh->b_this_page) != head);
1615 ext4_da_release_space(page->mapping->host, to_release);
1619 * Delayed allocation stuff
1622 struct mpage_da_data {
1623 struct inode *inode;
1624 struct buffer_head lbh; /* extent of blocks */
1625 unsigned long first_page, next_page; /* extent of pages */
1626 get_block_t *get_block;
1627 struct writeback_control *wbc;
1633 * mpage_da_submit_io - walks through extent of pages and try to write
1634 * them with writepage() call back
1636 * @mpd->inode: inode
1637 * @mpd->first_page: first page of the extent
1638 * @mpd->next_page: page after the last page of the extent
1639 * @mpd->get_block: the filesystem's block mapper function
1641 * By the time mpage_da_submit_io() is called we expect all blocks
1642 * to be allocated. this may be wrong if allocation failed.
1644 * As pages are already locked by write_cache_pages(), we can't use it
1646 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1648 struct address_space *mapping = mpd->inode->i_mapping;
1649 int ret = 0, err, nr_pages, i;
1650 unsigned long index, end;
1651 struct pagevec pvec;
1653 BUG_ON(mpd->next_page <= mpd->first_page);
1654 pagevec_init(&pvec, 0);
1655 index = mpd->first_page;
1656 end = mpd->next_page - 1;
1658 while (index <= end) {
1659 /* XXX: optimize tail */
1660 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1663 for (i = 0; i < nr_pages; i++) {
1664 struct page *page = pvec.pages[i];
1666 index = page->index;
1671 err = mapping->a_ops->writepage(page, mpd->wbc);
1673 mpd->pages_written++;
1675 * In error case, we have to continue because
1676 * remaining pages are still locked
1677 * XXX: unlock and re-dirty them?
1682 pagevec_release(&pvec);
1688 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1690 * @mpd->inode - inode to walk through
1691 * @exbh->b_blocknr - first block on a disk
1692 * @exbh->b_size - amount of space in bytes
1693 * @logical - first logical block to start assignment with
1695 * the function goes through all passed space and put actual disk
1696 * block numbers into buffer heads, dropping BH_Delay
1698 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1699 struct buffer_head *exbh)
1701 struct inode *inode = mpd->inode;
1702 struct address_space *mapping = inode->i_mapping;
1703 int blocks = exbh->b_size >> inode->i_blkbits;
1704 sector_t pblock = exbh->b_blocknr, cur_logical;
1705 struct buffer_head *head, *bh;
1707 struct pagevec pvec;
1710 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1711 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1712 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1714 pagevec_init(&pvec, 0);
1716 while (index <= end) {
1717 /* XXX: optimize tail */
1718 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1721 for (i = 0; i < nr_pages; i++) {
1722 struct page *page = pvec.pages[i];
1724 index = page->index;
1729 BUG_ON(!PageLocked(page));
1730 BUG_ON(PageWriteback(page));
1731 BUG_ON(!page_has_buffers(page));
1733 bh = page_buffers(page);
1736 /* skip blocks out of the range */
1738 if (cur_logical >= logical)
1741 } while ((bh = bh->b_this_page) != head);
1744 if (cur_logical >= logical + blocks)
1746 if (buffer_delay(bh)) {
1747 bh->b_blocknr = pblock;
1748 clear_buffer_delay(bh);
1749 bh->b_bdev = inode->i_sb->s_bdev;
1750 } else if (buffer_unwritten(bh)) {
1751 bh->b_blocknr = pblock;
1752 clear_buffer_unwritten(bh);
1753 set_buffer_mapped(bh);
1755 bh->b_bdev = inode->i_sb->s_bdev;
1756 } else if (buffer_mapped(bh))
1757 BUG_ON(bh->b_blocknr != pblock);
1761 } while ((bh = bh->b_this_page) != head);
1763 pagevec_release(&pvec);
1769 * __unmap_underlying_blocks - just a helper function to unmap
1770 * set of blocks described by @bh
1772 static inline void __unmap_underlying_blocks(struct inode *inode,
1773 struct buffer_head *bh)
1775 struct block_device *bdev = inode->i_sb->s_bdev;
1778 blocks = bh->b_size >> inode->i_blkbits;
1779 for (i = 0; i < blocks; i++)
1780 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1783 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1784 sector_t logical, long blk_cnt)
1788 struct pagevec pvec;
1789 struct inode *inode = mpd->inode;
1790 struct address_space *mapping = inode->i_mapping;
1792 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1793 end = (logical + blk_cnt - 1) >>
1794 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1795 while (index <= end) {
1796 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1799 for (i = 0; i < nr_pages; i++) {
1800 struct page *page = pvec.pages[i];
1801 index = page->index;
1806 BUG_ON(!PageLocked(page));
1807 BUG_ON(PageWriteback(page));
1808 block_invalidatepage(page, 0);
1809 ClearPageUptodate(page);
1817 * mpage_da_map_blocks - go through given space
1819 * @mpd->lbh - bh describing space
1820 * @mpd->get_block - the filesystem's block mapper function
1822 * The function skips space we know is already mapped to disk blocks.
1825 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1828 struct buffer_head *lbh = &mpd->lbh;
1829 sector_t next = lbh->b_blocknr;
1830 struct buffer_head new;
1833 * We consider only non-mapped and non-allocated blocks
1835 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1838 new.b_state = lbh->b_state;
1840 new.b_size = lbh->b_size;
1843 * If we didn't accumulate anything
1844 * to write simply return
1848 err = mpd->get_block(mpd->inode, next, &new, 1);
1851 /* If get block returns with error
1852 * we simply return. Later writepage
1853 * will redirty the page and writepages
1854 * will find the dirty page again
1859 * get block failure will cause us
1860 * to loop in writepages. Because
1861 * a_ops->writepage won't be able to
1862 * make progress. The page will be redirtied
1863 * by writepage and writepages will again
1864 * try to write the same.
1866 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1867 "at logical offset %llu with max blocks "
1868 "%zd with error %d\n",
1869 __func__, mpd->inode->i_ino,
1870 (unsigned long long)next,
1871 lbh->b_size >> mpd->inode->i_blkbits, err);
1872 printk(KERN_EMERG "This should not happen.!! "
1873 "Data will be lost\n");
1874 /* invlaidate all the pages */
1875 ext4_da_block_invalidatepages(mpd, next,
1876 lbh->b_size >> mpd->inode->i_blkbits);
1879 BUG_ON(new.b_size == 0);
1881 if (buffer_new(&new))
1882 __unmap_underlying_blocks(mpd->inode, &new);
1885 * If blocks are delayed marked, we need to
1886 * put actual blocknr and drop delayed bit
1888 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1889 mpage_put_bnr_to_bhs(mpd, next, &new);
1894 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1895 (1 << BH_Delay) | (1 << BH_Unwritten))
1898 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1900 * @mpd->lbh - extent of blocks
1901 * @logical - logical number of the block in the file
1902 * @bh - bh of the block (used to access block's state)
1904 * the function is used to collect contig. blocks in same state
1906 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1907 sector_t logical, struct buffer_head *bh)
1910 size_t b_size = bh->b_size;
1911 struct buffer_head *lbh = &mpd->lbh;
1912 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1914 /* check if thereserved journal credits might overflow */
1915 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1916 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1918 * With non-extent format we are limited by the journal
1919 * credit available. Total credit needed to insert
1920 * nrblocks contiguous blocks is dependent on the
1921 * nrblocks. So limit nrblocks.
1924 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1925 EXT4_MAX_TRANS_DATA) {
1927 * Adding the new buffer_head would make it cross the
1928 * allowed limit for which we have journal credit
1929 * reserved. So limit the new bh->b_size
1931 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1932 mpd->inode->i_blkbits;
1933 /* we will do mpage_da_submit_io in the next loop */
1937 * First block in the extent
1939 if (lbh->b_size == 0) {
1940 lbh->b_blocknr = logical;
1941 lbh->b_size = b_size;
1942 lbh->b_state = bh->b_state & BH_FLAGS;
1946 next = lbh->b_blocknr + nrblocks;
1948 * Can we merge the block to our big extent?
1950 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1951 lbh->b_size += b_size;
1957 * We couldn't merge the block to our extent, so we
1958 * need to flush current extent and start new one
1960 if (mpage_da_map_blocks(mpd) == 0)
1961 mpage_da_submit_io(mpd);
1967 * __mpage_da_writepage - finds extent of pages and blocks
1969 * @page: page to consider
1970 * @wbc: not used, we just follow rules
1973 * The function finds extents of pages and scan them for all blocks.
1975 static int __mpage_da_writepage(struct page *page,
1976 struct writeback_control *wbc, void *data)
1978 struct mpage_da_data *mpd = data;
1979 struct inode *inode = mpd->inode;
1980 struct buffer_head *bh, *head, fake;
1985 * Rest of the page in the page_vec
1986 * redirty then and skip then. We will
1987 * try to to write them again after
1988 * starting a new transaction
1990 redirty_page_for_writepage(wbc, page);
1992 return MPAGE_DA_EXTENT_TAIL;
1995 * Can we merge this page to current extent?
1997 if (mpd->next_page != page->index) {
1999 * Nope, we can't. So, we map non-allocated blocks
2000 * and start IO on them using writepage()
2002 if (mpd->next_page != mpd->first_page) {
2003 if (mpage_da_map_blocks(mpd) == 0)
2004 mpage_da_submit_io(mpd);
2006 * skip rest of the page in the page_vec
2009 redirty_page_for_writepage(wbc, page);
2011 return MPAGE_DA_EXTENT_TAIL;
2015 * Start next extent of pages ...
2017 mpd->first_page = page->index;
2022 mpd->lbh.b_size = 0;
2023 mpd->lbh.b_state = 0;
2024 mpd->lbh.b_blocknr = 0;
2027 mpd->next_page = page->index + 1;
2028 logical = (sector_t) page->index <<
2029 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2031 if (!page_has_buffers(page)) {
2033 * There is no attached buffer heads yet (mmap?)
2034 * we treat the page asfull of dirty blocks
2037 bh->b_size = PAGE_CACHE_SIZE;
2039 set_buffer_dirty(bh);
2040 set_buffer_uptodate(bh);
2041 mpage_add_bh_to_extent(mpd, logical, bh);
2043 return MPAGE_DA_EXTENT_TAIL;
2046 * Page with regular buffer heads, just add all dirty ones
2048 head = page_buffers(page);
2051 BUG_ON(buffer_locked(bh));
2052 if (buffer_dirty(bh) &&
2053 (!buffer_mapped(bh) || buffer_delay(bh))) {
2054 mpage_add_bh_to_extent(mpd, logical, bh);
2056 return MPAGE_DA_EXTENT_TAIL;
2059 } while ((bh = bh->b_this_page) != head);
2066 * mpage_da_writepages - walk the list of dirty pages of the given
2067 * address space, allocates non-allocated blocks, maps newly-allocated
2068 * blocks to existing bhs and issue IO them
2070 * @mapping: address space structure to write
2071 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2072 * @get_block: the filesystem's block mapper function.
2074 * This is a library function, which implements the writepages()
2075 * address_space_operation.
2077 static int mpage_da_writepages(struct address_space *mapping,
2078 struct writeback_control *wbc,
2079 get_block_t get_block)
2081 struct mpage_da_data mpd;
2086 return generic_writepages(mapping, wbc);
2089 mpd.inode = mapping->host;
2091 mpd.lbh.b_state = 0;
2092 mpd.lbh.b_blocknr = 0;
2095 mpd.get_block = get_block;
2097 mpd.pages_written = 0;
2099 to_write = wbc->nr_to_write;
2101 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
2104 * Handle last extent of pages
2106 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2107 if (mpage_da_map_blocks(&mpd) == 0)
2108 mpage_da_submit_io(&mpd);
2111 wbc->nr_to_write = to_write - mpd.pages_written;
2116 * this is a special callback for ->write_begin() only
2117 * it's intention is to return mapped block or reserve space
2119 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2120 struct buffer_head *bh_result, int create)
2124 BUG_ON(create == 0);
2125 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2128 * first, we need to know whether the block is allocated already
2129 * preallocated blocks are unmapped but should treated
2130 * the same as allocated blocks.
2132 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2133 if ((ret == 0) && !buffer_delay(bh_result)) {
2134 /* the block isn't (pre)allocated yet, let's reserve space */
2136 * XXX: __block_prepare_write() unmaps passed block,
2139 ret = ext4_da_reserve_space(inode, 1);
2141 /* not enough space to reserve */
2144 map_bh(bh_result, inode->i_sb, 0);
2145 set_buffer_new(bh_result);
2146 set_buffer_delay(bh_result);
2147 } else if (ret > 0) {
2148 bh_result->b_size = (ret << inode->i_blkbits);
2154 #define EXT4_DELALLOC_RSVED 1
2155 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2156 struct buffer_head *bh_result, int create)
2159 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2160 loff_t disksize = EXT4_I(inode)->i_disksize;
2161 handle_t *handle = NULL;
2163 handle = ext4_journal_current_handle();
2165 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2166 bh_result, 0, 0, 0);
2169 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2170 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2174 bh_result->b_size = (ret << inode->i_blkbits);
2177 * Update on-disk size along with block allocation
2178 * we don't use 'extend_disksize' as size may change
2179 * within already allocated block -bzzz
2181 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2182 if (disksize > i_size_read(inode))
2183 disksize = i_size_read(inode);
2184 if (disksize > EXT4_I(inode)->i_disksize) {
2186 * XXX: replace with spinlock if seen contended -bzzz
2188 down_write(&EXT4_I(inode)->i_data_sem);
2189 if (disksize > EXT4_I(inode)->i_disksize)
2190 EXT4_I(inode)->i_disksize = disksize;
2191 up_write(&EXT4_I(inode)->i_data_sem);
2193 if (EXT4_I(inode)->i_disksize == disksize) {
2194 ret = ext4_mark_inode_dirty(handle, inode);
2203 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2206 * unmapped buffer is possible for holes.
2207 * delay buffer is possible with delayed allocation
2209 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2212 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2213 struct buffer_head *bh_result, int create)
2216 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2219 * we don't want to do block allocation in writepage
2220 * so call get_block_wrap with create = 0
2222 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2223 bh_result, 0, 0, 0);
2225 bh_result->b_size = (ret << inode->i_blkbits);
2232 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2233 * get called via journal_submit_inode_data_buffers (no journal handle)
2234 * get called via shrink_page_list via pdflush (no journal handle)
2235 * or grab_page_cache when doing write_begin (have journal handle)
2237 static int ext4_da_writepage(struct page *page,
2238 struct writeback_control *wbc)
2243 struct buffer_head *page_bufs;
2244 struct inode *inode = page->mapping->host;
2246 size = i_size_read(inode);
2247 if (page->index == size >> PAGE_CACHE_SHIFT)
2248 len = size & ~PAGE_CACHE_MASK;
2250 len = PAGE_CACHE_SIZE;
2252 if (page_has_buffers(page)) {
2253 page_bufs = page_buffers(page);
2254 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2255 ext4_bh_unmapped_or_delay)) {
2257 * We don't want to do block allocation
2258 * So redirty the page and return
2259 * We may reach here when we do a journal commit
2260 * via journal_submit_inode_data_buffers.
2261 * If we don't have mapping block we just ignore
2262 * them. We can also reach here via shrink_page_list
2264 redirty_page_for_writepage(wbc, page);
2270 * The test for page_has_buffers() is subtle:
2271 * We know the page is dirty but it lost buffers. That means
2272 * that at some moment in time after write_begin()/write_end()
2273 * has been called all buffers have been clean and thus they
2274 * must have been written at least once. So they are all
2275 * mapped and we can happily proceed with mapping them
2276 * and writing the page.
2278 * Try to initialize the buffer_heads and check whether
2279 * all are mapped and non delay. We don't want to
2280 * do block allocation here.
2282 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2283 ext4_normal_get_block_write);
2285 page_bufs = page_buffers(page);
2286 /* check whether all are mapped and non delay */
2287 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2288 ext4_bh_unmapped_or_delay)) {
2289 redirty_page_for_writepage(wbc, page);
2295 * We can't do block allocation here
2296 * so just redity the page and unlock
2299 redirty_page_for_writepage(wbc, page);
2305 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2306 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2308 ret = block_write_full_page(page,
2309 ext4_normal_get_block_write,
2316 * This is called via ext4_da_writepages() to
2317 * calulate the total number of credits to reserve to fit
2318 * a single extent allocation into a single transaction,
2319 * ext4_da_writpeages() will loop calling this before
2320 * the block allocation.
2323 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2325 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2328 * With non-extent format the journal credit needed to
2329 * insert nrblocks contiguous block is dependent on
2330 * number of contiguous block. So we will limit
2331 * number of contiguous block to a sane value
2333 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2334 (max_blocks > EXT4_MAX_TRANS_DATA))
2335 max_blocks = EXT4_MAX_TRANS_DATA;
2337 return ext4_chunk_trans_blocks(inode, max_blocks);
2340 static int ext4_da_writepages(struct address_space *mapping,
2341 struct writeback_control *wbc)
2343 handle_t *handle = NULL;
2344 loff_t range_start = 0;
2345 struct inode *inode = mapping->host;
2346 int needed_blocks, ret = 0, nr_to_writebump = 0;
2347 long to_write, pages_skipped = 0;
2348 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2351 * No pages to write? This is mainly a kludge to avoid starting
2352 * a transaction for special inodes like journal inode on last iput()
2353 * because that could violate lock ordering on umount
2355 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2358 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2359 * This make sure small files blocks are allocated in
2360 * single attempt. This ensure that small files
2361 * get less fragmented.
2363 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2364 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2365 wbc->nr_to_write = sbi->s_mb_stream_request;
2368 if (!wbc->range_cyclic)
2370 * If range_cyclic is not set force range_cont
2371 * and save the old writeback_index
2373 wbc->range_cont = 1;
2375 range_start = wbc->range_start;
2376 pages_skipped = wbc->pages_skipped;
2379 to_write = wbc->nr_to_write;
2380 while (!ret && to_write > 0) {
2383 * we insert one extent at a time. So we need
2384 * credit needed for single extent allocation.
2385 * journalled mode is currently not supported
2388 BUG_ON(ext4_should_journal_data(inode));
2389 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2391 /* start a new transaction*/
2392 handle = ext4_journal_start(inode, needed_blocks);
2393 if (IS_ERR(handle)) {
2394 ret = PTR_ERR(handle);
2395 printk(KERN_EMERG "%s: jbd2_start: "
2396 "%ld pages, ino %lu; err %d\n", __func__,
2397 wbc->nr_to_write, inode->i_ino, ret);
2399 goto out_writepages;
2401 if (ext4_should_order_data(inode)) {
2403 * With ordered mode we need to add
2404 * the inode to the journal handl
2405 * when we do block allocation.
2407 ret = ext4_jbd2_file_inode(handle, inode);
2409 ext4_journal_stop(handle);
2410 goto out_writepages;
2414 to_write -= wbc->nr_to_write;
2415 ret = mpage_da_writepages(mapping, wbc,
2416 ext4_da_get_block_write);
2417 ext4_journal_stop(handle);
2418 if (ret == MPAGE_DA_EXTENT_TAIL) {
2420 * got one extent now try with
2423 to_write += wbc->nr_to_write;
2425 } else if (wbc->nr_to_write) {
2427 * There is no more writeout needed
2428 * or we requested for a noblocking writeout
2429 * and we found the device congested
2431 to_write += wbc->nr_to_write;
2434 wbc->nr_to_write = to_write;
2437 if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
2438 /* We skipped pages in this loop */
2439 wbc->range_start = range_start;
2440 wbc->nr_to_write = to_write +
2441 wbc->pages_skipped - pages_skipped;
2442 wbc->pages_skipped = pages_skipped;
2447 wbc->nr_to_write = to_write - nr_to_writebump;
2448 wbc->range_start = range_start;
2452 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2453 loff_t pos, unsigned len, unsigned flags,
2454 struct page **pagep, void **fsdata)
2456 int ret, retries = 0;
2460 struct inode *inode = mapping->host;
2463 index = pos >> PAGE_CACHE_SHIFT;
2464 from = pos & (PAGE_CACHE_SIZE - 1);
2469 * With delayed allocation, we don't log the i_disksize update
2470 * if there is delayed block allocation. But we still need
2471 * to journalling the i_disksize update if writes to the end
2472 * of file which has an already mapped buffer.
2474 handle = ext4_journal_start(inode, 1);
2475 if (IS_ERR(handle)) {
2476 ret = PTR_ERR(handle);
2480 page = __grab_cache_page(mapping, index);
2482 ext4_journal_stop(handle);
2488 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2489 ext4_da_get_block_prep);
2492 ext4_journal_stop(handle);
2493 page_cache_release(page);
2496 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2503 * Check if we should update i_disksize
2504 * when write to the end of file but not require block allocation
2506 static int ext4_da_should_update_i_disksize(struct page *page,
2507 unsigned long offset)
2509 struct buffer_head *bh;
2510 struct inode *inode = page->mapping->host;
2514 bh = page_buffers(page);
2515 idx = offset >> inode->i_blkbits;
2517 for (i = 0; i < idx; i++)
2518 bh = bh->b_this_page;
2520 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2525 static int ext4_da_write_end(struct file *file,
2526 struct address_space *mapping,
2527 loff_t pos, unsigned len, unsigned copied,
2528 struct page *page, void *fsdata)
2530 struct inode *inode = mapping->host;
2532 handle_t *handle = ext4_journal_current_handle();
2534 unsigned long start, end;
2536 start = pos & (PAGE_CACHE_SIZE - 1);
2537 end = start + copied - 1;
2540 * generic_write_end() will run mark_inode_dirty() if i_size
2541 * changes. So let's piggyback the i_disksize mark_inode_dirty
2545 new_i_size = pos + copied;
2546 if (new_i_size > EXT4_I(inode)->i_disksize) {
2547 if (ext4_da_should_update_i_disksize(page, end)) {
2548 down_write(&EXT4_I(inode)->i_data_sem);
2549 if (new_i_size > EXT4_I(inode)->i_disksize) {
2551 * Updating i_disksize when extending file
2552 * without needing block allocation
2554 if (ext4_should_order_data(inode))
2555 ret = ext4_jbd2_file_inode(handle,
2558 EXT4_I(inode)->i_disksize = new_i_size;
2560 up_write(&EXT4_I(inode)->i_data_sem);
2563 ret2 = generic_write_end(file, mapping, pos, len, copied,
2568 ret2 = ext4_journal_stop(handle);
2572 return ret ? ret : copied;
2575 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2578 * Drop reserved blocks
2580 BUG_ON(!PageLocked(page));
2581 if (!page_has_buffers(page))
2584 ext4_da_page_release_reservation(page, offset);
2587 ext4_invalidatepage(page, offset);
2594 * bmap() is special. It gets used by applications such as lilo and by
2595 * the swapper to find the on-disk block of a specific piece of data.
2597 * Naturally, this is dangerous if the block concerned is still in the
2598 * journal. If somebody makes a swapfile on an ext4 data-journaling
2599 * filesystem and enables swap, then they may get a nasty shock when the
2600 * data getting swapped to that swapfile suddenly gets overwritten by
2601 * the original zero's written out previously to the journal and
2602 * awaiting writeback in the kernel's buffer cache.
2604 * So, if we see any bmap calls here on a modified, data-journaled file,
2605 * take extra steps to flush any blocks which might be in the cache.
2607 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2609 struct inode *inode = mapping->host;
2613 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2614 test_opt(inode->i_sb, DELALLOC)) {
2616 * With delalloc we want to sync the file
2617 * so that we can make sure we allocate
2620 filemap_write_and_wait(mapping);
2623 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2625 * This is a REALLY heavyweight approach, but the use of
2626 * bmap on dirty files is expected to be extremely rare:
2627 * only if we run lilo or swapon on a freshly made file
2628 * do we expect this to happen.
2630 * (bmap requires CAP_SYS_RAWIO so this does not
2631 * represent an unprivileged user DOS attack --- we'd be
2632 * in trouble if mortal users could trigger this path at
2635 * NB. EXT4_STATE_JDATA is not set on files other than
2636 * regular files. If somebody wants to bmap a directory
2637 * or symlink and gets confused because the buffer
2638 * hasn't yet been flushed to disk, they deserve
2639 * everything they get.
2642 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2643 journal = EXT4_JOURNAL(inode);
2644 jbd2_journal_lock_updates(journal);
2645 err = jbd2_journal_flush(journal);
2646 jbd2_journal_unlock_updates(journal);
2652 return generic_block_bmap(mapping, block, ext4_get_block);
2655 static int bget_one(handle_t *handle, struct buffer_head *bh)
2661 static int bput_one(handle_t *handle, struct buffer_head *bh)
2668 * Note that we don't need to start a transaction unless we're journaling data
2669 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2670 * need to file the inode to the transaction's list in ordered mode because if
2671 * we are writing back data added by write(), the inode is already there and if
2672 * we are writing back data modified via mmap(), noone guarantees in which
2673 * transaction the data will hit the disk. In case we are journaling data, we
2674 * cannot start transaction directly because transaction start ranks above page
2675 * lock so we have to do some magic.
2677 * In all journaling modes block_write_full_page() will start the I/O.
2681 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2686 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2688 * Same applies to ext4_get_block(). We will deadlock on various things like
2689 * lock_journal and i_data_sem
2691 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2694 * 16May01: If we're reentered then journal_current_handle() will be
2695 * non-zero. We simply *return*.
2697 * 1 July 2001: @@@ FIXME:
2698 * In journalled data mode, a data buffer may be metadata against the
2699 * current transaction. But the same file is part of a shared mapping
2700 * and someone does a writepage() on it.
2702 * We will move the buffer onto the async_data list, but *after* it has
2703 * been dirtied. So there's a small window where we have dirty data on
2706 * Note that this only applies to the last partial page in the file. The
2707 * bit which block_write_full_page() uses prepare/commit for. (That's
2708 * broken code anyway: it's wrong for msync()).
2710 * It's a rare case: affects the final partial page, for journalled data
2711 * where the file is subject to bith write() and writepage() in the same
2712 * transction. To fix it we'll need a custom block_write_full_page().
2713 * We'll probably need that anyway for journalling writepage() output.
2715 * We don't honour synchronous mounts for writepage(). That would be
2716 * disastrous. Any write() or metadata operation will sync the fs for
2720 static int __ext4_normal_writepage(struct page *page,
2721 struct writeback_control *wbc)
2723 struct inode *inode = page->mapping->host;
2725 if (test_opt(inode->i_sb, NOBH))
2726 return nobh_writepage(page,
2727 ext4_normal_get_block_write, wbc);
2729 return block_write_full_page(page,
2730 ext4_normal_get_block_write,
2734 static int ext4_normal_writepage(struct page *page,
2735 struct writeback_control *wbc)
2737 struct inode *inode = page->mapping->host;
2738 loff_t size = i_size_read(inode);
2741 J_ASSERT(PageLocked(page));
2742 if (page->index == size >> PAGE_CACHE_SHIFT)
2743 len = size & ~PAGE_CACHE_MASK;
2745 len = PAGE_CACHE_SIZE;
2747 if (page_has_buffers(page)) {
2748 /* if page has buffers it should all be mapped
2749 * and allocated. If there are not buffers attached
2750 * to the page we know the page is dirty but it lost
2751 * buffers. That means that at some moment in time
2752 * after write_begin() / write_end() has been called
2753 * all buffers have been clean and thus they must have been
2754 * written at least once. So they are all mapped and we can
2755 * happily proceed with mapping them and writing the page.
2757 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2758 ext4_bh_unmapped_or_delay));
2761 if (!ext4_journal_current_handle())
2762 return __ext4_normal_writepage(page, wbc);
2764 redirty_page_for_writepage(wbc, page);
2769 static int __ext4_journalled_writepage(struct page *page,
2770 struct writeback_control *wbc)
2772 struct address_space *mapping = page->mapping;
2773 struct inode *inode = mapping->host;
2774 struct buffer_head *page_bufs;
2775 handle_t *handle = NULL;
2779 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2780 ext4_normal_get_block_write);
2784 page_bufs = page_buffers(page);
2785 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2787 /* As soon as we unlock the page, it can go away, but we have
2788 * references to buffers so we are safe */
2791 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2792 if (IS_ERR(handle)) {
2793 ret = PTR_ERR(handle);
2797 ret = walk_page_buffers(handle, page_bufs, 0,
2798 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2800 err = walk_page_buffers(handle, page_bufs, 0,
2801 PAGE_CACHE_SIZE, NULL, write_end_fn);
2804 err = ext4_journal_stop(handle);
2808 walk_page_buffers(handle, page_bufs, 0,
2809 PAGE_CACHE_SIZE, NULL, bput_one);
2810 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2819 static int ext4_journalled_writepage(struct page *page,
2820 struct writeback_control *wbc)
2822 struct inode *inode = page->mapping->host;
2823 loff_t size = i_size_read(inode);
2826 J_ASSERT(PageLocked(page));
2827 if (page->index == size >> PAGE_CACHE_SHIFT)
2828 len = size & ~PAGE_CACHE_MASK;
2830 len = PAGE_CACHE_SIZE;
2832 if (page_has_buffers(page)) {
2833 /* if page has buffers it should all be mapped
2834 * and allocated. If there are not buffers attached
2835 * to the page we know the page is dirty but it lost
2836 * buffers. That means that at some moment in time
2837 * after write_begin() / write_end() has been called
2838 * all buffers have been clean and thus they must have been
2839 * written at least once. So they are all mapped and we can
2840 * happily proceed with mapping them and writing the page.
2842 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2843 ext4_bh_unmapped_or_delay));
2846 if (ext4_journal_current_handle())
2849 if (PageChecked(page)) {
2851 * It's mmapped pagecache. Add buffers and journal it. There
2852 * doesn't seem much point in redirtying the page here.
2854 ClearPageChecked(page);
2855 return __ext4_journalled_writepage(page, wbc);
2858 * It may be a page full of checkpoint-mode buffers. We don't
2859 * really know unless we go poke around in the buffer_heads.
2860 * But block_write_full_page will do the right thing.
2862 return block_write_full_page(page,
2863 ext4_normal_get_block_write,
2867 redirty_page_for_writepage(wbc, page);
2872 static int ext4_readpage(struct file *file, struct page *page)
2874 return mpage_readpage(page, ext4_get_block);
2878 ext4_readpages(struct file *file, struct address_space *mapping,
2879 struct list_head *pages, unsigned nr_pages)
2881 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2884 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2886 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2889 * If it's a full truncate we just forget about the pending dirtying
2892 ClearPageChecked(page);
2894 jbd2_journal_invalidatepage(journal, page, offset);
2897 static int ext4_releasepage(struct page *page, gfp_t wait)
2899 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2901 WARN_ON(PageChecked(page));
2902 if (!page_has_buffers(page))
2904 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2908 * If the O_DIRECT write will extend the file then add this inode to the
2909 * orphan list. So recovery will truncate it back to the original size
2910 * if the machine crashes during the write.
2912 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2913 * crashes then stale disk data _may_ be exposed inside the file. But current
2914 * VFS code falls back into buffered path in that case so we are safe.
2916 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2917 const struct iovec *iov, loff_t offset,
2918 unsigned long nr_segs)
2920 struct file *file = iocb->ki_filp;
2921 struct inode *inode = file->f_mapping->host;
2922 struct ext4_inode_info *ei = EXT4_I(inode);
2926 size_t count = iov_length(iov, nr_segs);
2929 loff_t final_size = offset + count;
2931 if (final_size > inode->i_size) {
2932 /* Credits for sb + inode write */
2933 handle = ext4_journal_start(inode, 2);
2934 if (IS_ERR(handle)) {
2935 ret = PTR_ERR(handle);
2938 ret = ext4_orphan_add(handle, inode);
2940 ext4_journal_stop(handle);
2944 ei->i_disksize = inode->i_size;
2945 ext4_journal_stop(handle);
2949 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2951 ext4_get_block, NULL);
2956 /* Credits for sb + inode write */
2957 handle = ext4_journal_start(inode, 2);
2958 if (IS_ERR(handle)) {
2959 /* This is really bad luck. We've written the data
2960 * but cannot extend i_size. Bail out and pretend
2961 * the write failed... */
2962 ret = PTR_ERR(handle);
2966 ext4_orphan_del(handle, inode);
2968 loff_t end = offset + ret;
2969 if (end > inode->i_size) {
2970 ei->i_disksize = end;
2971 i_size_write(inode, end);
2973 * We're going to return a positive `ret'
2974 * here due to non-zero-length I/O, so there's
2975 * no way of reporting error returns from
2976 * ext4_mark_inode_dirty() to userspace. So
2979 ext4_mark_inode_dirty(handle, inode);
2982 err = ext4_journal_stop(handle);
2991 * Pages can be marked dirty completely asynchronously from ext4's journalling
2992 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2993 * much here because ->set_page_dirty is called under VFS locks. The page is
2994 * not necessarily locked.
2996 * We cannot just dirty the page and leave attached buffers clean, because the
2997 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2998 * or jbddirty because all the journalling code will explode.
3000 * So what we do is to mark the page "pending dirty" and next time writepage
3001 * is called, propagate that into the buffers appropriately.
3003 static int ext4_journalled_set_page_dirty(struct page *page)
3005 SetPageChecked(page);
3006 return __set_page_dirty_nobuffers(page);
3009 static const struct address_space_operations ext4_ordered_aops = {
3010 .readpage = ext4_readpage,
3011 .readpages = ext4_readpages,
3012 .writepage = ext4_normal_writepage,
3013 .sync_page = block_sync_page,
3014 .write_begin = ext4_write_begin,
3015 .write_end = ext4_ordered_write_end,
3017 .invalidatepage = ext4_invalidatepage,
3018 .releasepage = ext4_releasepage,
3019 .direct_IO = ext4_direct_IO,
3020 .migratepage = buffer_migrate_page,
3021 .is_partially_uptodate = block_is_partially_uptodate,
3024 static const struct address_space_operations ext4_writeback_aops = {
3025 .readpage = ext4_readpage,
3026 .readpages = ext4_readpages,
3027 .writepage = ext4_normal_writepage,
3028 .sync_page = block_sync_page,
3029 .write_begin = ext4_write_begin,
3030 .write_end = ext4_writeback_write_end,
3032 .invalidatepage = ext4_invalidatepage,
3033 .releasepage = ext4_releasepage,
3034 .direct_IO = ext4_direct_IO,
3035 .migratepage = buffer_migrate_page,
3036 .is_partially_uptodate = block_is_partially_uptodate,
3039 static const struct address_space_operations ext4_journalled_aops = {
3040 .readpage = ext4_readpage,
3041 .readpages = ext4_readpages,
3042 .writepage = ext4_journalled_writepage,
3043 .sync_page = block_sync_page,
3044 .write_begin = ext4_write_begin,
3045 .write_end = ext4_journalled_write_end,
3046 .set_page_dirty = ext4_journalled_set_page_dirty,
3048 .invalidatepage = ext4_invalidatepage,
3049 .releasepage = ext4_releasepage,
3050 .is_partially_uptodate = block_is_partially_uptodate,
3053 static const struct address_space_operations ext4_da_aops = {
3054 .readpage = ext4_readpage,
3055 .readpages = ext4_readpages,
3056 .writepage = ext4_da_writepage,
3057 .writepages = ext4_da_writepages,
3058 .sync_page = block_sync_page,
3059 .write_begin = ext4_da_write_begin,
3060 .write_end = ext4_da_write_end,
3062 .invalidatepage = ext4_da_invalidatepage,
3063 .releasepage = ext4_releasepage,
3064 .direct_IO = ext4_direct_IO,
3065 .migratepage = buffer_migrate_page,
3066 .is_partially_uptodate = block_is_partially_uptodate,
3069 void ext4_set_aops(struct inode *inode)
3071 if (ext4_should_order_data(inode) &&
3072 test_opt(inode->i_sb, DELALLOC))
3073 inode->i_mapping->a_ops = &ext4_da_aops;
3074 else if (ext4_should_order_data(inode))
3075 inode->i_mapping->a_ops = &ext4_ordered_aops;
3076 else if (ext4_should_writeback_data(inode) &&
3077 test_opt(inode->i_sb, DELALLOC))
3078 inode->i_mapping->a_ops = &ext4_da_aops;
3079 else if (ext4_should_writeback_data(inode))
3080 inode->i_mapping->a_ops = &ext4_writeback_aops;
3082 inode->i_mapping->a_ops = &ext4_journalled_aops;
3086 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3087 * up to the end of the block which corresponds to `from'.
3088 * This required during truncate. We need to physically zero the tail end
3089 * of that block so it doesn't yield old data if the file is later grown.
3091 int ext4_block_truncate_page(handle_t *handle,
3092 struct address_space *mapping, loff_t from)
3094 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3095 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3096 unsigned blocksize, length, pos;
3098 struct inode *inode = mapping->host;
3099 struct buffer_head *bh;
3103 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3107 blocksize = inode->i_sb->s_blocksize;
3108 length = blocksize - (offset & (blocksize - 1));
3109 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3112 * For "nobh" option, we can only work if we don't need to
3113 * read-in the page - otherwise we create buffers to do the IO.
3115 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3116 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3117 zero_user(page, offset, length);
3118 set_page_dirty(page);
3122 if (!page_has_buffers(page))
3123 create_empty_buffers(page, blocksize, 0);
3125 /* Find the buffer that contains "offset" */
3126 bh = page_buffers(page);
3128 while (offset >= pos) {
3129 bh = bh->b_this_page;
3135 if (buffer_freed(bh)) {
3136 BUFFER_TRACE(bh, "freed: skip");
3140 if (!buffer_mapped(bh)) {
3141 BUFFER_TRACE(bh, "unmapped");
3142 ext4_get_block(inode, iblock, bh, 0);
3143 /* unmapped? It's a hole - nothing to do */
3144 if (!buffer_mapped(bh)) {
3145 BUFFER_TRACE(bh, "still unmapped");
3150 /* Ok, it's mapped. Make sure it's up-to-date */
3151 if (PageUptodate(page))
3152 set_buffer_uptodate(bh);
3154 if (!buffer_uptodate(bh)) {
3156 ll_rw_block(READ, 1, &bh);
3158 /* Uhhuh. Read error. Complain and punt. */
3159 if (!buffer_uptodate(bh))
3163 if (ext4_should_journal_data(inode)) {
3164 BUFFER_TRACE(bh, "get write access");
3165 err = ext4_journal_get_write_access(handle, bh);
3170 zero_user(page, offset, length);
3172 BUFFER_TRACE(bh, "zeroed end of block");
3175 if (ext4_should_journal_data(inode)) {
3176 err = ext4_journal_dirty_metadata(handle, bh);
3178 if (ext4_should_order_data(inode))
3179 err = ext4_jbd2_file_inode(handle, inode);
3180 mark_buffer_dirty(bh);
3185 page_cache_release(page);
3190 * Probably it should be a library function... search for first non-zero word
3191 * or memcmp with zero_page, whatever is better for particular architecture.
3194 static inline int all_zeroes(__le32 *p, __le32 *q)
3203 * ext4_find_shared - find the indirect blocks for partial truncation.
3204 * @inode: inode in question
3205 * @depth: depth of the affected branch
3206 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3207 * @chain: place to store the pointers to partial indirect blocks
3208 * @top: place to the (detached) top of branch
3210 * This is a helper function used by ext4_truncate().
3212 * When we do truncate() we may have to clean the ends of several
3213 * indirect blocks but leave the blocks themselves alive. Block is
3214 * partially truncated if some data below the new i_size is refered
3215 * from it (and it is on the path to the first completely truncated
3216 * data block, indeed). We have to free the top of that path along
3217 * with everything to the right of the path. Since no allocation
3218 * past the truncation point is possible until ext4_truncate()
3219 * finishes, we may safely do the latter, but top of branch may
3220 * require special attention - pageout below the truncation point
3221 * might try to populate it.
3223 * We atomically detach the top of branch from the tree, store the
3224 * block number of its root in *@top, pointers to buffer_heads of
3225 * partially truncated blocks - in @chain[].bh and pointers to
3226 * their last elements that should not be removed - in
3227 * @chain[].p. Return value is the pointer to last filled element
3230 * The work left to caller to do the actual freeing of subtrees:
3231 * a) free the subtree starting from *@top
3232 * b) free the subtrees whose roots are stored in
3233 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3234 * c) free the subtrees growing from the inode past the @chain[0].
3235 * (no partially truncated stuff there). */
3237 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3238 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3240 Indirect *partial, *p;
3244 /* Make k index the deepest non-null offest + 1 */
3245 for (k = depth; k > 1 && !offsets[k-1]; k--)
3247 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3248 /* Writer: pointers */
3250 partial = chain + k-1;
3252 * If the branch acquired continuation since we've looked at it -
3253 * fine, it should all survive and (new) top doesn't belong to us.
3255 if (!partial->key && *partial->p)
3258 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3261 * OK, we've found the last block that must survive. The rest of our
3262 * branch should be detached before unlocking. However, if that rest
3263 * of branch is all ours and does not grow immediately from the inode
3264 * it's easier to cheat and just decrement partial->p.
3266 if (p == chain + k - 1 && p > chain) {
3270 /* Nope, don't do this in ext4. Must leave the tree intact */
3277 while (partial > p) {
3278 brelse(partial->bh);
3286 * Zero a number of block pointers in either an inode or an indirect block.
3287 * If we restart the transaction we must again get write access to the
3288 * indirect block for further modification.
3290 * We release `count' blocks on disk, but (last - first) may be greater
3291 * than `count' because there can be holes in there.
3293 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3294 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3295 unsigned long count, __le32 *first, __le32 *last)
3298 if (try_to_extend_transaction(handle, inode)) {
3300 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3301 ext4_journal_dirty_metadata(handle, bh);
3303 ext4_mark_inode_dirty(handle, inode);
3304 ext4_journal_test_restart(handle, inode);
3306 BUFFER_TRACE(bh, "retaking write access");
3307 ext4_journal_get_write_access(handle, bh);
3312 * Any buffers which are on the journal will be in memory. We find
3313 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3314 * on them. We've already detached each block from the file, so
3315 * bforget() in jbd2_journal_forget() should be safe.
3317 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3319 for (p = first; p < last; p++) {
3320 u32 nr = le32_to_cpu(*p);
3322 struct buffer_head *tbh;
3325 tbh = sb_find_get_block(inode->i_sb, nr);
3326 ext4_forget(handle, 0, inode, tbh, nr);
3330 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3334 * ext4_free_data - free a list of data blocks
3335 * @handle: handle for this transaction
3336 * @inode: inode we are dealing with
3337 * @this_bh: indirect buffer_head which contains *@first and *@last
3338 * @first: array of block numbers
3339 * @last: points immediately past the end of array
3341 * We are freeing all blocks refered from that array (numbers are stored as
3342 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3344 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3345 * blocks are contiguous then releasing them at one time will only affect one
3346 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3347 * actually use a lot of journal space.
3349 * @this_bh will be %NULL if @first and @last point into the inode's direct
3352 static void ext4_free_data(handle_t *handle, struct inode *inode,
3353 struct buffer_head *this_bh,
3354 __le32 *first, __le32 *last)
3356 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3357 unsigned long count = 0; /* Number of blocks in the run */
3358 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3361 ext4_fsblk_t nr; /* Current block # */
3362 __le32 *p; /* Pointer into inode/ind
3363 for current block */
3366 if (this_bh) { /* For indirect block */
3367 BUFFER_TRACE(this_bh, "get_write_access");
3368 err = ext4_journal_get_write_access(handle, this_bh);
3369 /* Important: if we can't update the indirect pointers
3370 * to the blocks, we can't free them. */
3375 for (p = first; p < last; p++) {
3376 nr = le32_to_cpu(*p);
3378 /* accumulate blocks to free if they're contiguous */
3381 block_to_free_p = p;
3383 } else if (nr == block_to_free + count) {
3386 ext4_clear_blocks(handle, inode, this_bh,
3388 count, block_to_free_p, p);
3390 block_to_free_p = p;
3397 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3398 count, block_to_free_p, p);
3401 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3404 * The buffer head should have an attached journal head at this
3405 * point. However, if the data is corrupted and an indirect
3406 * block pointed to itself, it would have been detached when
3407 * the block was cleared. Check for this instead of OOPSing.
3410 ext4_journal_dirty_metadata(handle, this_bh);
3412 ext4_error(inode->i_sb, __func__,
3413 "circular indirect block detected, "
3414 "inode=%lu, block=%llu",
3416 (unsigned long long) this_bh->b_blocknr);
3421 * ext4_free_branches - free an array of branches
3422 * @handle: JBD handle for this transaction
3423 * @inode: inode we are dealing with
3424 * @parent_bh: the buffer_head which contains *@first and *@last
3425 * @first: array of block numbers
3426 * @last: pointer immediately past the end of array
3427 * @depth: depth of the branches to free
3429 * We are freeing all blocks refered from these branches (numbers are
3430 * stored as little-endian 32-bit) and updating @inode->i_blocks
3433 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3434 struct buffer_head *parent_bh,
3435 __le32 *first, __le32 *last, int depth)
3440 if (is_handle_aborted(handle))
3444 struct buffer_head *bh;
3445 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3447 while (--p >= first) {
3448 nr = le32_to_cpu(*p);
3450 continue; /* A hole */
3452 /* Go read the buffer for the next level down */
3453 bh = sb_bread(inode->i_sb, nr);
3456 * A read failure? Report error and clear slot
3460 ext4_error(inode->i_sb, "ext4_free_branches",
3461 "Read failure, inode=%lu, block=%llu",
3466 /* This zaps the entire block. Bottom up. */
3467 BUFFER_TRACE(bh, "free child branches");
3468 ext4_free_branches(handle, inode, bh,
3469 (__le32 *) bh->b_data,
3470 (__le32 *) bh->b_data + addr_per_block,
3474 * We've probably journalled the indirect block several
3475 * times during the truncate. But it's no longer
3476 * needed and we now drop it from the transaction via
3477 * jbd2_journal_revoke().
3479 * That's easy if it's exclusively part of this
3480 * transaction. But if it's part of the committing
3481 * transaction then jbd2_journal_forget() will simply
3482 * brelse() it. That means that if the underlying
3483 * block is reallocated in ext4_get_block(),
3484 * unmap_underlying_metadata() will find this block
3485 * and will try to get rid of it. damn, damn.
3487 * If this block has already been committed to the
3488 * journal, a revoke record will be written. And
3489 * revoke records must be emitted *before* clearing
3490 * this block's bit in the bitmaps.
3492 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3495 * Everything below this this pointer has been
3496 * released. Now let this top-of-subtree go.
3498 * We want the freeing of this indirect block to be
3499 * atomic in the journal with the updating of the
3500 * bitmap block which owns it. So make some room in
3503 * We zero the parent pointer *after* freeing its
3504 * pointee in the bitmaps, so if extend_transaction()
3505 * for some reason fails to put the bitmap changes and
3506 * the release into the same transaction, recovery
3507 * will merely complain about releasing a free block,
3508 * rather than leaking blocks.
3510 if (is_handle_aborted(handle))
3512 if (try_to_extend_transaction(handle, inode)) {
3513 ext4_mark_inode_dirty(handle, inode);
3514 ext4_journal_test_restart(handle, inode);
3517 ext4_free_blocks(handle, inode, nr, 1, 1);
3521 * The block which we have just freed is
3522 * pointed to by an indirect block: journal it
3524 BUFFER_TRACE(parent_bh, "get_write_access");
3525 if (!ext4_journal_get_write_access(handle,
3528 BUFFER_TRACE(parent_bh,
3529 "call ext4_journal_dirty_metadata");
3530 ext4_journal_dirty_metadata(handle,
3536 /* We have reached the bottom of the tree. */
3537 BUFFER_TRACE(parent_bh, "free data blocks");
3538 ext4_free_data(handle, inode, parent_bh, first, last);
3542 int ext4_can_truncate(struct inode *inode)
3544 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3546 if (S_ISREG(inode->i_mode))
3548 if (S_ISDIR(inode->i_mode))
3550 if (S_ISLNK(inode->i_mode))
3551 return !ext4_inode_is_fast_symlink(inode);
3558 * We block out ext4_get_block() block instantiations across the entire
3559 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3560 * simultaneously on behalf of the same inode.
3562 * As we work through the truncate and commmit bits of it to the journal there
3563 * is one core, guiding principle: the file's tree must always be consistent on
3564 * disk. We must be able to restart the truncate after a crash.
3566 * The file's tree may be transiently inconsistent in memory (although it
3567 * probably isn't), but whenever we close off and commit a journal transaction,
3568 * the contents of (the filesystem + the journal) must be consistent and
3569 * restartable. It's pretty simple, really: bottom up, right to left (although
3570 * left-to-right works OK too).
3572 * Note that at recovery time, journal replay occurs *before* the restart of
3573 * truncate against the orphan inode list.
3575 * The committed inode has the new, desired i_size (which is the same as
3576 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3577 * that this inode's truncate did not complete and it will again call
3578 * ext4_truncate() to have another go. So there will be instantiated blocks
3579 * to the right of the truncation point in a crashed ext4 filesystem. But
3580 * that's fine - as long as they are linked from the inode, the post-crash
3581 * ext4_truncate() run will find them and release them.
3583 void ext4_truncate(struct inode *inode)
3586 struct ext4_inode_info *ei = EXT4_I(inode);
3587 __le32 *i_data = ei->i_data;
3588 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3589 struct address_space *mapping = inode->i_mapping;
3590 ext4_lblk_t offsets[4];
3595 ext4_lblk_t last_block;
3596 unsigned blocksize = inode->i_sb->s_blocksize;
3598 if (!ext4_can_truncate(inode))
3601 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3602 ext4_ext_truncate(inode);
3606 handle = start_transaction(inode);
3608 return; /* AKPM: return what? */
3610 last_block = (inode->i_size + blocksize-1)
3611 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3613 if (inode->i_size & (blocksize - 1))
3614 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3617 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3619 goto out_stop; /* error */
3622 * OK. This truncate is going to happen. We add the inode to the
3623 * orphan list, so that if this truncate spans multiple transactions,
3624 * and we crash, we will resume the truncate when the filesystem
3625 * recovers. It also marks the inode dirty, to catch the new size.
3627 * Implication: the file must always be in a sane, consistent
3628 * truncatable state while each transaction commits.
3630 if (ext4_orphan_add(handle, inode))
3634 * From here we block out all ext4_get_block() callers who want to
3635 * modify the block allocation tree.
3637 down_write(&ei->i_data_sem);
3639 ext4_discard_reservation(inode);
3642 * The orphan list entry will now protect us from any crash which
3643 * occurs before the truncate completes, so it is now safe to propagate
3644 * the new, shorter inode size (held for now in i_size) into the
3645 * on-disk inode. We do this via i_disksize, which is the value which
3646 * ext4 *really* writes onto the disk inode.
3648 ei->i_disksize = inode->i_size;
3650 if (n == 1) { /* direct blocks */
3651 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3652 i_data + EXT4_NDIR_BLOCKS);
3656 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3657 /* Kill the top of shared branch (not detached) */
3659 if (partial == chain) {
3660 /* Shared branch grows from the inode */
3661 ext4_free_branches(handle, inode, NULL,
3662 &nr, &nr+1, (chain+n-1) - partial);
3665 * We mark the inode dirty prior to restart,
3666 * and prior to stop. No need for it here.
3669 /* Shared branch grows from an indirect block */
3670 BUFFER_TRACE(partial->bh, "get_write_access");
3671 ext4_free_branches(handle, inode, partial->bh,
3673 partial->p+1, (chain+n-1) - partial);
3676 /* Clear the ends of indirect blocks on the shared branch */
3677 while (partial > chain) {
3678 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3679 (__le32*)partial->bh->b_data+addr_per_block,
3680 (chain+n-1) - partial);
3681 BUFFER_TRACE(partial->bh, "call brelse");
3682 brelse (partial->bh);
3686 /* Kill the remaining (whole) subtrees */
3687 switch (offsets[0]) {
3689 nr = i_data[EXT4_IND_BLOCK];
3691 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3692 i_data[EXT4_IND_BLOCK] = 0;
3694 case EXT4_IND_BLOCK:
3695 nr = i_data[EXT4_DIND_BLOCK];
3697 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3698 i_data[EXT4_DIND_BLOCK] = 0;
3700 case EXT4_DIND_BLOCK:
3701 nr = i_data[EXT4_TIND_BLOCK];
3703 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3704 i_data[EXT4_TIND_BLOCK] = 0;
3706 case EXT4_TIND_BLOCK:
3710 up_write(&ei->i_data_sem);
3711 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3712 ext4_mark_inode_dirty(handle, inode);
3715 * In a multi-transaction truncate, we only make the final transaction
3722 * If this was a simple ftruncate(), and the file will remain alive
3723 * then we need to clear up the orphan record which we created above.
3724 * However, if this was a real unlink then we were called by
3725 * ext4_delete_inode(), and we allow that function to clean up the
3726 * orphan info for us.
3729 ext4_orphan_del(handle, inode);
3731 ext4_journal_stop(handle);
3734 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3735 unsigned long ino, struct ext4_iloc *iloc)
3737 ext4_group_t block_group;
3738 unsigned long offset;
3740 struct ext4_group_desc *gdp;
3742 if (!ext4_valid_inum(sb, ino)) {
3744 * This error is already checked for in namei.c unless we are
3745 * looking at an NFS filehandle, in which case no error
3751 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3752 gdp = ext4_get_group_desc(sb, block_group, NULL);
3757 * Figure out the offset within the block group inode table
3759 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3760 EXT4_INODE_SIZE(sb);
3761 block = ext4_inode_table(sb, gdp) +
3762 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3764 iloc->block_group = block_group;
3765 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3770 * ext4_get_inode_loc returns with an extra refcount against the inode's
3771 * underlying buffer_head on success. If 'in_mem' is true, we have all
3772 * data in memory that is needed to recreate the on-disk version of this
3775 static int __ext4_get_inode_loc(struct inode *inode,
3776 struct ext4_iloc *iloc, int in_mem)
3779 struct buffer_head *bh;
3781 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3785 bh = sb_getblk(inode->i_sb, block);
3787 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3788 "unable to read inode block - "
3789 "inode=%lu, block=%llu",
3790 inode->i_ino, block);
3793 if (!buffer_uptodate(bh)) {
3797 * If the buffer has the write error flag, we have failed
3798 * to write out another inode in the same block. In this
3799 * case, we don't have to read the block because we may
3800 * read the old inode data successfully.
3802 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3803 set_buffer_uptodate(bh);
3805 if (buffer_uptodate(bh)) {
3806 /* someone brought it uptodate while we waited */
3812 * If we have all information of the inode in memory and this
3813 * is the only valid inode in the block, we need not read the
3817 struct buffer_head *bitmap_bh;
3818 struct ext4_group_desc *desc;
3819 int inodes_per_buffer;
3820 int inode_offset, i;
3821 ext4_group_t block_group;
3824 block_group = (inode->i_ino - 1) /
3825 EXT4_INODES_PER_GROUP(inode->i_sb);
3826 inodes_per_buffer = bh->b_size /
3827 EXT4_INODE_SIZE(inode->i_sb);
3828 inode_offset = ((inode->i_ino - 1) %
3829 EXT4_INODES_PER_GROUP(inode->i_sb));
3830 start = inode_offset & ~(inodes_per_buffer - 1);
3832 /* Is the inode bitmap in cache? */
3833 desc = ext4_get_group_desc(inode->i_sb,
3838 bitmap_bh = sb_getblk(inode->i_sb,
3839 ext4_inode_bitmap(inode->i_sb, desc));
3844 * If the inode bitmap isn't in cache then the
3845 * optimisation may end up performing two reads instead
3846 * of one, so skip it.
3848 if (!buffer_uptodate(bitmap_bh)) {
3852 for (i = start; i < start + inodes_per_buffer; i++) {
3853 if (i == inode_offset)
3855 if (ext4_test_bit(i, bitmap_bh->b_data))
3859 if (i == start + inodes_per_buffer) {
3860 /* all other inodes are free, so skip I/O */
3861 memset(bh->b_data, 0, bh->b_size);
3862 set_buffer_uptodate(bh);
3870 * There are other valid inodes in the buffer, this inode
3871 * has in-inode xattrs, or we don't have this inode in memory.
3872 * Read the block from disk.
3875 bh->b_end_io = end_buffer_read_sync;
3876 submit_bh(READ_META, bh);
3878 if (!buffer_uptodate(bh)) {
3879 ext4_error(inode->i_sb, "ext4_get_inode_loc",
3880 "unable to read inode block - "
3881 "inode=%lu, block=%llu",
3882 inode->i_ino, block);
3892 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3894 /* We have all inode data except xattrs in memory here. */
3895 return __ext4_get_inode_loc(inode, iloc,
3896 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3899 void ext4_set_inode_flags(struct inode *inode)
3901 unsigned int flags = EXT4_I(inode)->i_flags;
3903 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3904 if (flags & EXT4_SYNC_FL)
3905 inode->i_flags |= S_SYNC;
3906 if (flags & EXT4_APPEND_FL)
3907 inode->i_flags |= S_APPEND;
3908 if (flags & EXT4_IMMUTABLE_FL)
3909 inode->i_flags |= S_IMMUTABLE;
3910 if (flags & EXT4_NOATIME_FL)
3911 inode->i_flags |= S_NOATIME;
3912 if (flags & EXT4_DIRSYNC_FL)
3913 inode->i_flags |= S_DIRSYNC;
3916 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3917 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3919 unsigned int flags = ei->vfs_inode.i_flags;
3921 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3922 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
3924 ei->i_flags |= EXT4_SYNC_FL;
3925 if (flags & S_APPEND)
3926 ei->i_flags |= EXT4_APPEND_FL;
3927 if (flags & S_IMMUTABLE)
3928 ei->i_flags |= EXT4_IMMUTABLE_FL;
3929 if (flags & S_NOATIME)
3930 ei->i_flags |= EXT4_NOATIME_FL;
3931 if (flags & S_DIRSYNC)
3932 ei->i_flags |= EXT4_DIRSYNC_FL;
3934 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3935 struct ext4_inode_info *ei)
3938 struct inode *inode = &(ei->vfs_inode);
3939 struct super_block *sb = inode->i_sb;
3941 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3942 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3943 /* we are using combined 48 bit field */
3944 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3945 le32_to_cpu(raw_inode->i_blocks_lo);
3946 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
3947 /* i_blocks represent file system block size */
3948 return i_blocks << (inode->i_blkbits - 9);
3953 return le32_to_cpu(raw_inode->i_blocks_lo);
3957 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3959 struct ext4_iloc iloc;
3960 struct ext4_inode *raw_inode;
3961 struct ext4_inode_info *ei;
3962 struct buffer_head *bh;
3963 struct inode *inode;
3967 inode = iget_locked(sb, ino);
3969 return ERR_PTR(-ENOMEM);
3970 if (!(inode->i_state & I_NEW))
3974 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3975 ei->i_acl = EXT4_ACL_NOT_CACHED;
3976 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
3978 ei->i_block_alloc_info = NULL;
3980 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3984 raw_inode = ext4_raw_inode(&iloc);
3985 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3986 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3987 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3988 if (!(test_opt(inode->i_sb, NO_UID32))) {
3989 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3990 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3992 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3995 ei->i_dir_start_lookup = 0;
3996 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3997 /* We now have enough fields to check if the inode was active or not.
3998 * This is needed because nfsd might try to access dead inodes
3999 * the test is that same one that e2fsck uses
4000 * NeilBrown 1999oct15
4002 if (inode->i_nlink == 0) {
4003 if (inode->i_mode == 0 ||
4004 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4005 /* this inode is deleted */
4010 /* The only unlinked inodes we let through here have
4011 * valid i_mode and are being read by the orphan
4012 * recovery code: that's fine, we're about to complete
4013 * the process of deleting those. */
4015 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4016 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4017 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4018 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4019 cpu_to_le32(EXT4_OS_HURD)) {
4021 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4023 inode->i_size = ext4_isize(raw_inode);
4024 ei->i_disksize = inode->i_size;
4025 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4026 ei->i_block_group = iloc.block_group;
4028 * NOTE! The in-memory inode i_data array is in little-endian order
4029 * even on big-endian machines: we do NOT byteswap the block numbers!
4031 for (block = 0; block < EXT4_N_BLOCKS; block++)
4032 ei->i_data[block] = raw_inode->i_block[block];
4033 INIT_LIST_HEAD(&ei->i_orphan);
4035 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4036 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4037 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4038 EXT4_INODE_SIZE(inode->i_sb)) {
4043 if (ei->i_extra_isize == 0) {
4044 /* The extra space is currently unused. Use it. */
4045 ei->i_extra_isize = sizeof(struct ext4_inode) -
4046 EXT4_GOOD_OLD_INODE_SIZE;
4048 __le32 *magic = (void *)raw_inode +
4049 EXT4_GOOD_OLD_INODE_SIZE +
4051 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4052 ei->i_state |= EXT4_STATE_XATTR;
4055 ei->i_extra_isize = 0;
4057 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4058 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4059 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4060 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4062 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4063 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4064 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4066 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4069 if (S_ISREG(inode->i_mode)) {
4070 inode->i_op = &ext4_file_inode_operations;
4071 inode->i_fop = &ext4_file_operations;
4072 ext4_set_aops(inode);
4073 } else if (S_ISDIR(inode->i_mode)) {
4074 inode->i_op = &ext4_dir_inode_operations;
4075 inode->i_fop = &ext4_dir_operations;
4076 } else if (S_ISLNK(inode->i_mode)) {
4077 if (ext4_inode_is_fast_symlink(inode))
4078 inode->i_op = &ext4_fast_symlink_inode_operations;
4080 inode->i_op = &ext4_symlink_inode_operations;
4081 ext4_set_aops(inode);
4084 inode->i_op = &ext4_special_inode_operations;
4085 if (raw_inode->i_block[0])
4086 init_special_inode(inode, inode->i_mode,
4087 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4089 init_special_inode(inode, inode->i_mode,
4090 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4093 ext4_set_inode_flags(inode);
4094 unlock_new_inode(inode);
4099 return ERR_PTR(ret);
4102 static int ext4_inode_blocks_set(handle_t *handle,
4103 struct ext4_inode *raw_inode,
4104 struct ext4_inode_info *ei)
4106 struct inode *inode = &(ei->vfs_inode);
4107 u64 i_blocks = inode->i_blocks;
4108 struct super_block *sb = inode->i_sb;
4111 if (i_blocks <= ~0U) {
4113 * i_blocks can be represnted in a 32 bit variable
4114 * as multiple of 512 bytes
4116 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4117 raw_inode->i_blocks_high = 0;
4118 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4119 } else if (i_blocks <= 0xffffffffffffULL) {
4121 * i_blocks can be represented in a 48 bit variable
4122 * as multiple of 512 bytes
4124 err = ext4_update_rocompat_feature(handle, sb,
4125 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4128 /* i_block is stored in the split 48 bit fields */
4129 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4130 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4131 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4134 * i_blocks should be represented in a 48 bit variable
4135 * as multiple of file system block size
4137 err = ext4_update_rocompat_feature(handle, sb,
4138 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4141 ei->i_flags |= EXT4_HUGE_FILE_FL;
4142 /* i_block is stored in file system block size */
4143 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4144 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4145 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4152 * Post the struct inode info into an on-disk inode location in the
4153 * buffer-cache. This gobbles the caller's reference to the
4154 * buffer_head in the inode location struct.
4156 * The caller must have write access to iloc->bh.
4158 static int ext4_do_update_inode(handle_t *handle,
4159 struct inode *inode,
4160 struct ext4_iloc *iloc)
4162 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4163 struct ext4_inode_info *ei = EXT4_I(inode);
4164 struct buffer_head *bh = iloc->bh;
4165 int err = 0, rc, block;
4167 /* For fields not not tracking in the in-memory inode,
4168 * initialise them to zero for new inodes. */
4169 if (ei->i_state & EXT4_STATE_NEW)
4170 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4172 ext4_get_inode_flags(ei);
4173 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4174 if (!(test_opt(inode->i_sb, NO_UID32))) {
4175 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4176 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4178 * Fix up interoperability with old kernels. Otherwise, old inodes get
4179 * re-used with the upper 16 bits of the uid/gid intact
4182 raw_inode->i_uid_high =
4183 cpu_to_le16(high_16_bits(inode->i_uid));
4184 raw_inode->i_gid_high =
4185 cpu_to_le16(high_16_bits(inode->i_gid));
4187 raw_inode->i_uid_high = 0;
4188 raw_inode->i_gid_high = 0;
4191 raw_inode->i_uid_low =
4192 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4193 raw_inode->i_gid_low =
4194 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4195 raw_inode->i_uid_high = 0;
4196 raw_inode->i_gid_high = 0;
4198 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4200 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4201 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4202 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4203 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4205 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4207 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4208 /* clear the migrate flag in the raw_inode */
4209 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4210 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4211 cpu_to_le32(EXT4_OS_HURD))
4212 raw_inode->i_file_acl_high =
4213 cpu_to_le16(ei->i_file_acl >> 32);
4214 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4215 ext4_isize_set(raw_inode, ei->i_disksize);
4216 if (ei->i_disksize > 0x7fffffffULL) {
4217 struct super_block *sb = inode->i_sb;
4218 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4219 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4220 EXT4_SB(sb)->s_es->s_rev_level ==
4221 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4222 /* If this is the first large file
4223 * created, add a flag to the superblock.
4225 err = ext4_journal_get_write_access(handle,
4226 EXT4_SB(sb)->s_sbh);
4229 ext4_update_dynamic_rev(sb);
4230 EXT4_SET_RO_COMPAT_FEATURE(sb,
4231 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4234 err = ext4_journal_dirty_metadata(handle,
4235 EXT4_SB(sb)->s_sbh);
4238 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4239 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4240 if (old_valid_dev(inode->i_rdev)) {
4241 raw_inode->i_block[0] =
4242 cpu_to_le32(old_encode_dev(inode->i_rdev));
4243 raw_inode->i_block[1] = 0;
4245 raw_inode->i_block[0] = 0;
4246 raw_inode->i_block[1] =
4247 cpu_to_le32(new_encode_dev(inode->i_rdev));
4248 raw_inode->i_block[2] = 0;
4250 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4251 raw_inode->i_block[block] = ei->i_data[block];
4253 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4254 if (ei->i_extra_isize) {
4255 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4256 raw_inode->i_version_hi =
4257 cpu_to_le32(inode->i_version >> 32);
4258 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4262 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4263 rc = ext4_journal_dirty_metadata(handle, bh);
4266 ei->i_state &= ~EXT4_STATE_NEW;
4270 ext4_std_error(inode->i_sb, err);
4275 * ext4_write_inode()
4277 * We are called from a few places:
4279 * - Within generic_file_write() for O_SYNC files.
4280 * Here, there will be no transaction running. We wait for any running
4281 * trasnaction to commit.
4283 * - Within sys_sync(), kupdate and such.
4284 * We wait on commit, if tol to.
4286 * - Within prune_icache() (PF_MEMALLOC == true)
4287 * Here we simply return. We can't afford to block kswapd on the
4290 * In all cases it is actually safe for us to return without doing anything,
4291 * because the inode has been copied into a raw inode buffer in
4292 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4295 * Note that we are absolutely dependent upon all inode dirtiers doing the
4296 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4297 * which we are interested.
4299 * It would be a bug for them to not do this. The code:
4301 * mark_inode_dirty(inode)
4303 * inode->i_size = expr;
4305 * is in error because a kswapd-driven write_inode() could occur while
4306 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4307 * will no longer be on the superblock's dirty inode list.
4309 int ext4_write_inode(struct inode *inode, int wait)
4311 if (current->flags & PF_MEMALLOC)
4314 if (ext4_journal_current_handle()) {
4315 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4323 return ext4_force_commit(inode->i_sb);
4329 * Called from notify_change.
4331 * We want to trap VFS attempts to truncate the file as soon as
4332 * possible. In particular, we want to make sure that when the VFS
4333 * shrinks i_size, we put the inode on the orphan list and modify
4334 * i_disksize immediately, so that during the subsequent flushing of
4335 * dirty pages and freeing of disk blocks, we can guarantee that any
4336 * commit will leave the blocks being flushed in an unused state on
4337 * disk. (On recovery, the inode will get truncated and the blocks will
4338 * be freed, so we have a strong guarantee that no future commit will
4339 * leave these blocks visible to the user.)
4341 * Another thing we have to assure is that if we are in ordered mode
4342 * and inode is still attached to the committing transaction, we must
4343 * we start writeout of all the dirty pages which are being truncated.
4344 * This way we are sure that all the data written in the previous
4345 * transaction are already on disk (truncate waits for pages under
4348 * Called with inode->i_mutex down.
4350 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4352 struct inode *inode = dentry->d_inode;
4354 const unsigned int ia_valid = attr->ia_valid;
4356 error = inode_change_ok(inode, attr);
4360 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4361 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4364 /* (user+group)*(old+new) structure, inode write (sb,
4365 * inode block, ? - but truncate inode update has it) */
4366 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4367 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4368 if (IS_ERR(handle)) {
4369 error = PTR_ERR(handle);
4372 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4374 ext4_journal_stop(handle);
4377 /* Update corresponding info in inode so that everything is in
4378 * one transaction */
4379 if (attr->ia_valid & ATTR_UID)
4380 inode->i_uid = attr->ia_uid;
4381 if (attr->ia_valid & ATTR_GID)
4382 inode->i_gid = attr->ia_gid;
4383 error = ext4_mark_inode_dirty(handle, inode);
4384 ext4_journal_stop(handle);
4387 if (attr->ia_valid & ATTR_SIZE) {
4388 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4389 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4391 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4398 if (S_ISREG(inode->i_mode) &&
4399 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4402 handle = ext4_journal_start(inode, 3);
4403 if (IS_ERR(handle)) {
4404 error = PTR_ERR(handle);
4408 error = ext4_orphan_add(handle, inode);
4409 EXT4_I(inode)->i_disksize = attr->ia_size;
4410 rc = ext4_mark_inode_dirty(handle, inode);
4413 ext4_journal_stop(handle);
4415 if (ext4_should_order_data(inode)) {
4416 error = ext4_begin_ordered_truncate(inode,
4419 /* Do as much error cleanup as possible */
4420 handle = ext4_journal_start(inode, 3);
4421 if (IS_ERR(handle)) {
4422 ext4_orphan_del(NULL, inode);
4425 ext4_orphan_del(handle, inode);
4426 ext4_journal_stop(handle);
4432 rc = inode_setattr(inode, attr);
4434 /* If inode_setattr's call to ext4_truncate failed to get a
4435 * transaction handle at all, we need to clean up the in-core
4436 * orphan list manually. */
4438 ext4_orphan_del(NULL, inode);
4440 if (!rc && (ia_valid & ATTR_MODE))
4441 rc = ext4_acl_chmod(inode);
4444 ext4_std_error(inode->i_sb, error);
4450 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4453 struct inode *inode;
4454 unsigned long delalloc_blocks;
4456 inode = dentry->d_inode;
4457 generic_fillattr(inode, stat);
4460 * We can't update i_blocks if the block allocation is delayed
4461 * otherwise in the case of system crash before the real block
4462 * allocation is done, we will have i_blocks inconsistent with
4463 * on-disk file blocks.
4464 * We always keep i_blocks updated together with real
4465 * allocation. But to not confuse with user, stat
4466 * will return the blocks that include the delayed allocation
4467 * blocks for this file.
4469 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4470 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4471 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4473 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4477 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4482 /* if nrblocks are contiguous */
4485 * With N contiguous data blocks, it need at most
4486 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4487 * 2 dindirect blocks
4490 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4491 return indirects + 3;
4494 * if nrblocks are not contiguous, worse case, each block touch
4495 * a indirect block, and each indirect block touch a double indirect
4496 * block, plus a triple indirect block
4498 indirects = nrblocks * 2 + 1;
4502 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4504 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4505 return ext4_indirect_trans_blocks(inode, nrblocks, 0);
4506 return ext4_ext_index_trans_blocks(inode, nrblocks, 0);
4509 * Account for index blocks, block groups bitmaps and block group
4510 * descriptor blocks if modify datablocks and index blocks
4511 * worse case, the indexs blocks spread over different block groups
4513 * If datablocks are discontiguous, they are possible to spread over
4514 * different block groups too. If they are contiugous, with flexbg,
4515 * they could still across block group boundary.
4517 * Also account for superblock, inode, quota and xattr blocks
4519 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4521 int groups, gdpblocks;
4526 * How many index blocks need to touch to modify nrblocks?
4527 * The "Chunk" flag indicating whether the nrblocks is
4528 * physically contiguous on disk
4530 * For Direct IO and fallocate, they calls get_block to allocate
4531 * one single extent at a time, so they could set the "Chunk" flag
4533 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4538 * Now let's see how many group bitmaps and group descriptors need
4548 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4549 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4550 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4551 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4553 /* bitmaps and block group descriptor blocks */
4554 ret += groups + gdpblocks;
4556 /* Blocks for super block, inode, quota and xattr blocks */
4557 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4563 * Calulate the total number of credits to reserve to fit
4564 * the modification of a single pages into a single transaction,
4565 * which may include multiple chunks of block allocations.
4567 * This could be called via ext4_write_begin()
4569 * We need to consider the worse case, when
4570 * one new block per extent.
4572 int ext4_writepage_trans_blocks(struct inode *inode)
4574 int bpp = ext4_journal_blocks_per_page(inode);
4577 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4579 /* Account for data blocks for journalled mode */
4580 if (ext4_should_journal_data(inode))
4586 * Calculate the journal credits for a chunk of data modification.
4588 * This is called from DIO, fallocate or whoever calling
4589 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4591 * journal buffers for data blocks are not included here, as DIO
4592 * and fallocate do no need to journal data buffers.
4594 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4596 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4600 * The caller must have previously called ext4_reserve_inode_write().
4601 * Give this, we know that the caller already has write access to iloc->bh.
4603 int ext4_mark_iloc_dirty(handle_t *handle,
4604 struct inode *inode, struct ext4_iloc *iloc)
4608 if (test_opt(inode->i_sb, I_VERSION))
4609 inode_inc_iversion(inode);
4611 /* the do_update_inode consumes one bh->b_count */
4614 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4615 err = ext4_do_update_inode(handle, inode, iloc);
4621 * On success, We end up with an outstanding reference count against
4622 * iloc->bh. This _must_ be cleaned up later.
4626 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4627 struct ext4_iloc *iloc)
4631 err = ext4_get_inode_loc(inode, iloc);
4633 BUFFER_TRACE(iloc->bh, "get_write_access");
4634 err = ext4_journal_get_write_access(handle, iloc->bh);
4641 ext4_std_error(inode->i_sb, err);
4646 * Expand an inode by new_extra_isize bytes.
4647 * Returns 0 on success or negative error number on failure.
4649 static int ext4_expand_extra_isize(struct inode *inode,
4650 unsigned int new_extra_isize,
4651 struct ext4_iloc iloc,
4654 struct ext4_inode *raw_inode;
4655 struct ext4_xattr_ibody_header *header;
4656 struct ext4_xattr_entry *entry;
4658 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4661 raw_inode = ext4_raw_inode(&iloc);
4663 header = IHDR(inode, raw_inode);
4664 entry = IFIRST(header);
4666 /* No extended attributes present */
4667 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4668 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4669 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4671 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4675 /* try to expand with EAs present */
4676 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4681 * What we do here is to mark the in-core inode as clean with respect to inode
4682 * dirtiness (it may still be data-dirty).
4683 * This means that the in-core inode may be reaped by prune_icache
4684 * without having to perform any I/O. This is a very good thing,
4685 * because *any* task may call prune_icache - even ones which
4686 * have a transaction open against a different journal.
4688 * Is this cheating? Not really. Sure, we haven't written the
4689 * inode out, but prune_icache isn't a user-visible syncing function.
4690 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4691 * we start and wait on commits.
4693 * Is this efficient/effective? Well, we're being nice to the system
4694 * by cleaning up our inodes proactively so they can be reaped
4695 * without I/O. But we are potentially leaving up to five seconds'
4696 * worth of inodes floating about which prune_icache wants us to
4697 * write out. One way to fix that would be to get prune_icache()
4698 * to do a write_super() to free up some memory. It has the desired
4701 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4703 struct ext4_iloc iloc;
4704 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4705 static unsigned int mnt_count;
4709 err = ext4_reserve_inode_write(handle, inode, &iloc);
4710 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4711 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4713 * We need extra buffer credits since we may write into EA block
4714 * with this same handle. If journal_extend fails, then it will
4715 * only result in a minor loss of functionality for that inode.
4716 * If this is felt to be critical, then e2fsck should be run to
4717 * force a large enough s_min_extra_isize.
4719 if ((jbd2_journal_extend(handle,
4720 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4721 ret = ext4_expand_extra_isize(inode,
4722 sbi->s_want_extra_isize,
4725 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4727 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4728 ext4_warning(inode->i_sb, __func__,
4729 "Unable to expand inode %lu. Delete"
4730 " some EAs or run e2fsck.",
4733 le16_to_cpu(sbi->s_es->s_mnt_count);
4739 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4744 * ext4_dirty_inode() is called from __mark_inode_dirty()
4746 * We're really interested in the case where a file is being extended.
4747 * i_size has been changed by generic_commit_write() and we thus need
4748 * to include the updated inode in the current transaction.
4750 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4751 * are allocated to the file.
4753 * If the inode is marked synchronous, we don't honour that here - doing
4754 * so would cause a commit on atime updates, which we don't bother doing.
4755 * We handle synchronous inodes at the highest possible level.
4757 void ext4_dirty_inode(struct inode *inode)
4759 handle_t *current_handle = ext4_journal_current_handle();
4762 handle = ext4_journal_start(inode, 2);
4765 if (current_handle &&
4766 current_handle->h_transaction != handle->h_transaction) {
4767 /* This task has a transaction open against a different fs */
4768 printk(KERN_EMERG "%s: transactions do not match!\n",
4771 jbd_debug(5, "marking dirty. outer handle=%p\n",
4773 ext4_mark_inode_dirty(handle, inode);
4775 ext4_journal_stop(handle);
4782 * Bind an inode's backing buffer_head into this transaction, to prevent
4783 * it from being flushed to disk early. Unlike
4784 * ext4_reserve_inode_write, this leaves behind no bh reference and
4785 * returns no iloc structure, so the caller needs to repeat the iloc
4786 * lookup to mark the inode dirty later.
4788 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4790 struct ext4_iloc iloc;
4794 err = ext4_get_inode_loc(inode, &iloc);
4796 BUFFER_TRACE(iloc.bh, "get_write_access");
4797 err = jbd2_journal_get_write_access(handle, iloc.bh);
4799 err = ext4_journal_dirty_metadata(handle,
4804 ext4_std_error(inode->i_sb, err);
4809 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4816 * We have to be very careful here: changing a data block's
4817 * journaling status dynamically is dangerous. If we write a
4818 * data block to the journal, change the status and then delete
4819 * that block, we risk forgetting to revoke the old log record
4820 * from the journal and so a subsequent replay can corrupt data.
4821 * So, first we make sure that the journal is empty and that
4822 * nobody is changing anything.
4825 journal = EXT4_JOURNAL(inode);
4826 if (is_journal_aborted(journal))
4829 jbd2_journal_lock_updates(journal);
4830 jbd2_journal_flush(journal);
4833 * OK, there are no updates running now, and all cached data is
4834 * synced to disk. We are now in a completely consistent state
4835 * which doesn't have anything in the journal, and we know that
4836 * no filesystem updates are running, so it is safe to modify
4837 * the inode's in-core data-journaling state flag now.
4841 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4843 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4844 ext4_set_aops(inode);
4846 jbd2_journal_unlock_updates(journal);
4848 /* Finally we can mark the inode as dirty. */
4850 handle = ext4_journal_start(inode, 1);
4852 return PTR_ERR(handle);
4854 err = ext4_mark_inode_dirty(handle, inode);
4856 ext4_journal_stop(handle);
4857 ext4_std_error(inode->i_sb, err);
4862 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4864 return !buffer_mapped(bh);
4867 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4872 struct file *file = vma->vm_file;
4873 struct inode *inode = file->f_path.dentry->d_inode;
4874 struct address_space *mapping = inode->i_mapping;
4877 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4878 * get i_mutex because we are already holding mmap_sem.
4880 down_read(&inode->i_alloc_sem);
4881 size = i_size_read(inode);
4882 if (page->mapping != mapping || size <= page_offset(page)
4883 || !PageUptodate(page)) {
4884 /* page got truncated from under us? */
4888 if (PageMappedToDisk(page))
4891 if (page->index == size >> PAGE_CACHE_SHIFT)
4892 len = size & ~PAGE_CACHE_MASK;
4894 len = PAGE_CACHE_SIZE;
4896 if (page_has_buffers(page)) {
4897 /* return if we have all the buffers mapped */
4898 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4903 * OK, we need to fill the hole... Do write_begin write_end
4904 * to do block allocation/reservation.We are not holding
4905 * inode.i__mutex here. That allow * parallel write_begin,
4906 * write_end call. lock_page prevent this from happening
4907 * on the same page though
4909 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4910 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4913 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4914 len, len, page, NULL);
4919 up_read(&inode->i_alloc_sem);