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/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
41 #include "ext4_jbd2.h"
44 #include "ext4_extents.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static inline int ext4_begin_ordered_truncate(struct inode *inode,
53 return jbd2_journal_begin_ordered_truncate(
54 EXT4_SB(inode->i_sb)->s_journal,
55 &EXT4_I(inode)->jinode,
59 static void ext4_invalidatepage(struct page *page, unsigned long offset);
62 * Test whether an inode is a fast symlink.
64 static int ext4_inode_is_fast_symlink(struct inode *inode)
66 int ea_blocks = EXT4_I(inode)->i_file_acl ?
67 (inode->i_sb->s_blocksize >> 9) : 0;
69 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
73 * The ext4 forget function must perform a revoke if we are freeing data
74 * which has been journaled. Metadata (eg. indirect blocks) must be
75 * revoked in all cases.
77 * "bh" may be NULL: a metadata block may have been freed from memory
78 * but there may still be a record of it in the journal, and that record
79 * still needs to be revoked.
81 * If the handle isn't valid we're not journaling, but we still need to
82 * call into ext4_journal_revoke() to put the buffer head.
84 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
85 struct buffer_head *bh, ext4_fsblk_t blocknr)
91 BUFFER_TRACE(bh, "enter");
93 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
95 bh, is_metadata, inode->i_mode,
96 test_opt(inode->i_sb, DATA_FLAGS));
98 /* Never use the revoke function if we are doing full data
99 * journaling: there is no need to, and a V1 superblock won't
100 * support it. Otherwise, only skip the revoke on un-journaled
103 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
104 (!is_metadata && !ext4_should_journal_data(inode))) {
106 BUFFER_TRACE(bh, "call jbd2_journal_forget");
107 return ext4_journal_forget(handle, bh);
113 * data!=journal && (is_metadata || should_journal_data(inode))
115 BUFFER_TRACE(bh, "call ext4_journal_revoke");
116 err = ext4_journal_revoke(handle, blocknr, bh);
118 ext4_abort(inode->i_sb, __func__,
119 "error %d when attempting revoke", err);
120 BUFFER_TRACE(bh, "exit");
125 * Work out how many blocks we need to proceed with the next chunk of a
126 * truncate transaction.
128 static unsigned long blocks_for_truncate(struct inode *inode)
132 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
134 /* Give ourselves just enough room to cope with inodes in which
135 * i_blocks is corrupt: we've seen disk corruptions in the past
136 * which resulted in random data in an inode which looked enough
137 * like a regular file for ext4 to try to delete it. Things
138 * will go a bit crazy if that happens, but at least we should
139 * try not to panic the whole kernel. */
143 /* But we need to bound the transaction so we don't overflow the
145 if (needed > EXT4_MAX_TRANS_DATA)
146 needed = EXT4_MAX_TRANS_DATA;
148 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
152 * Truncate transactions can be complex and absolutely huge. So we need to
153 * be able to restart the transaction at a conventient checkpoint to make
154 * sure we don't overflow the journal.
156 * start_transaction gets us a new handle for a truncate transaction,
157 * and extend_transaction tries to extend the existing one a bit. If
158 * extend fails, we need to propagate the failure up and restart the
159 * transaction in the top-level truncate loop. --sct
161 static handle_t *start_transaction(struct inode *inode)
165 result = ext4_journal_start(inode, blocks_for_truncate(inode));
169 ext4_std_error(inode->i_sb, PTR_ERR(result));
174 * Try to extend this transaction for the purposes of truncation.
176 * Returns 0 if we managed to create more room. If we can't create more
177 * room, and the transaction must be restarted we return 1.
179 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
181 if (!ext4_handle_valid(handle))
183 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
185 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
191 * Restart the transaction associated with *handle. This does a commit,
192 * so before we call here everything must be consistently dirtied against
195 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
201 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
202 * moment, get_block can be called only for blocks inside i_size since
203 * page cache has been already dropped and writes are blocked by
204 * i_mutex. So we can safely drop the i_data_sem here.
206 BUG_ON(EXT4_JOURNAL(inode) == NULL);
207 jbd_debug(2, "restarting handle %p\n", handle);
208 up_write(&EXT4_I(inode)->i_data_sem);
209 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
210 down_write(&EXT4_I(inode)->i_data_sem);
216 * Called at the last iput() if i_nlink is zero.
218 void ext4_delete_inode(struct inode *inode)
223 if (ext4_should_order_data(inode))
224 ext4_begin_ordered_truncate(inode, 0);
225 truncate_inode_pages(&inode->i_data, 0);
227 if (is_bad_inode(inode))
230 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
231 if (IS_ERR(handle)) {
232 ext4_std_error(inode->i_sb, PTR_ERR(handle));
234 * If we're going to skip the normal cleanup, we still need to
235 * make sure that the in-core orphan linked list is properly
238 ext4_orphan_del(NULL, inode);
243 ext4_handle_sync(handle);
245 err = ext4_mark_inode_dirty(handle, inode);
247 ext4_warning(inode->i_sb, __func__,
248 "couldn't mark inode dirty (err %d)", err);
252 ext4_truncate(inode);
255 * ext4_ext_truncate() doesn't reserve any slop when it
256 * restarts journal transactions; therefore there may not be
257 * enough credits left in the handle to remove the inode from
258 * the orphan list and set the dtime field.
260 if (!ext4_handle_has_enough_credits(handle, 3)) {
261 err = ext4_journal_extend(handle, 3);
263 err = ext4_journal_restart(handle, 3);
265 ext4_warning(inode->i_sb, __func__,
266 "couldn't extend journal (err %d)", err);
268 ext4_journal_stop(handle);
274 * Kill off the orphan record which ext4_truncate created.
275 * AKPM: I think this can be inside the above `if'.
276 * Note that ext4_orphan_del() has to be able to cope with the
277 * deletion of a non-existent orphan - this is because we don't
278 * know if ext4_truncate() actually created an orphan record.
279 * (Well, we could do this if we need to, but heck - it works)
281 ext4_orphan_del(handle, inode);
282 EXT4_I(inode)->i_dtime = get_seconds();
285 * One subtle ordering requirement: if anything has gone wrong
286 * (transaction abort, IO errors, whatever), then we can still
287 * do these next steps (the fs will already have been marked as
288 * having errors), but we can't free the inode if the mark_dirty
291 if (ext4_mark_inode_dirty(handle, inode))
292 /* If that failed, just do the required in-core inode clear. */
295 ext4_free_inode(handle, inode);
296 ext4_journal_stop(handle);
299 clear_inode(inode); /* We must guarantee clearing of inode... */
305 struct buffer_head *bh;
308 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
310 p->key = *(p->p = v);
315 * ext4_block_to_path - parse the block number into array of offsets
316 * @inode: inode in question (we are only interested in its superblock)
317 * @i_block: block number to be parsed
318 * @offsets: array to store the offsets in
319 * @boundary: set this non-zero if the referred-to block is likely to be
320 * followed (on disk) by an indirect block.
322 * To store the locations of file's data ext4 uses a data structure common
323 * for UNIX filesystems - tree of pointers anchored in the inode, with
324 * data blocks at leaves and indirect blocks in intermediate nodes.
325 * This function translates the block number into path in that tree -
326 * return value is the path length and @offsets[n] is the offset of
327 * pointer to (n+1)th node in the nth one. If @block is out of range
328 * (negative or too large) warning is printed and zero returned.
330 * Note: function doesn't find node addresses, so no IO is needed. All
331 * we need to know is the capacity of indirect blocks (taken from the
336 * Portability note: the last comparison (check that we fit into triple
337 * indirect block) is spelled differently, because otherwise on an
338 * architecture with 32-bit longs and 8Kb pages we might get into trouble
339 * if our filesystem had 8Kb blocks. We might use long long, but that would
340 * kill us on x86. Oh, well, at least the sign propagation does not matter -
341 * i_block would have to be negative in the very beginning, so we would not
345 static int ext4_block_to_path(struct inode *inode,
347 ext4_lblk_t offsets[4], int *boundary)
349 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
350 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
351 const long direct_blocks = EXT4_NDIR_BLOCKS,
352 indirect_blocks = ptrs,
353 double_blocks = (1 << (ptrs_bits * 2));
357 if (i_block < direct_blocks) {
358 offsets[n++] = i_block;
359 final = direct_blocks;
360 } else if ((i_block -= direct_blocks) < indirect_blocks) {
361 offsets[n++] = EXT4_IND_BLOCK;
362 offsets[n++] = i_block;
364 } else if ((i_block -= indirect_blocks) < double_blocks) {
365 offsets[n++] = EXT4_DIND_BLOCK;
366 offsets[n++] = i_block >> ptrs_bits;
367 offsets[n++] = i_block & (ptrs - 1);
369 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
370 offsets[n++] = EXT4_TIND_BLOCK;
371 offsets[n++] = i_block >> (ptrs_bits * 2);
372 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
373 offsets[n++] = i_block & (ptrs - 1);
376 ext4_warning(inode->i_sb, "ext4_block_to_path",
377 "block %lu > max in inode %lu",
378 i_block + direct_blocks +
379 indirect_blocks + double_blocks, inode->i_ino);
382 *boundary = final - 1 - (i_block & (ptrs - 1));
386 static int __ext4_check_blockref(const char *function, struct inode *inode,
387 __le32 *p, unsigned int max)
392 while (bref < p+max) {
393 blk = le32_to_cpu(*bref++);
395 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
397 ext4_error(inode->i_sb, function,
398 "invalid block reference %u "
399 "in inode #%lu", blk, inode->i_ino);
407 #define ext4_check_indirect_blockref(inode, bh) \
408 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
409 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
411 #define ext4_check_inode_blockref(inode) \
412 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
416 * ext4_get_branch - read the chain of indirect blocks leading to data
417 * @inode: inode in question
418 * @depth: depth of the chain (1 - direct pointer, etc.)
419 * @offsets: offsets of pointers in inode/indirect blocks
420 * @chain: place to store the result
421 * @err: here we store the error value
423 * Function fills the array of triples <key, p, bh> and returns %NULL
424 * if everything went OK or the pointer to the last filled triple
425 * (incomplete one) otherwise. Upon the return chain[i].key contains
426 * the number of (i+1)-th block in the chain (as it is stored in memory,
427 * i.e. little-endian 32-bit), chain[i].p contains the address of that
428 * number (it points into struct inode for i==0 and into the bh->b_data
429 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
430 * block for i>0 and NULL for i==0. In other words, it holds the block
431 * numbers of the chain, addresses they were taken from (and where we can
432 * verify that chain did not change) and buffer_heads hosting these
435 * Function stops when it stumbles upon zero pointer (absent block)
436 * (pointer to last triple returned, *@err == 0)
437 * or when it gets an IO error reading an indirect block
438 * (ditto, *@err == -EIO)
439 * or when it reads all @depth-1 indirect blocks successfully and finds
440 * the whole chain, all way to the data (returns %NULL, *err == 0).
442 * Need to be called with
443 * down_read(&EXT4_I(inode)->i_data_sem)
445 static Indirect *ext4_get_branch(struct inode *inode, int depth,
446 ext4_lblk_t *offsets,
447 Indirect chain[4], int *err)
449 struct super_block *sb = inode->i_sb;
451 struct buffer_head *bh;
454 /* i_data is not going away, no lock needed */
455 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
459 bh = sb_getblk(sb, le32_to_cpu(p->key));
463 if (!bh_uptodate_or_lock(bh)) {
464 if (bh_submit_read(bh) < 0) {
468 /* validate block references */
469 if (ext4_check_indirect_blockref(inode, bh)) {
475 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
489 * ext4_find_near - find a place for allocation with sufficient locality
491 * @ind: descriptor of indirect block.
493 * This function returns the preferred place for block allocation.
494 * It is used when heuristic for sequential allocation fails.
496 * + if there is a block to the left of our position - allocate near it.
497 * + if pointer will live in indirect block - allocate near that block.
498 * + if pointer will live in inode - allocate in the same
501 * In the latter case we colour the starting block by the callers PID to
502 * prevent it from clashing with concurrent allocations for a different inode
503 * in the same block group. The PID is used here so that functionally related
504 * files will be close-by on-disk.
506 * Caller must make sure that @ind is valid and will stay that way.
508 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
510 struct ext4_inode_info *ei = EXT4_I(inode);
511 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
513 ext4_fsblk_t bg_start;
514 ext4_fsblk_t last_block;
515 ext4_grpblk_t colour;
516 ext4_group_t block_group;
517 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
519 /* Try to find previous block */
520 for (p = ind->p - 1; p >= start; p--) {
522 return le32_to_cpu(*p);
525 /* No such thing, so let's try location of indirect block */
527 return ind->bh->b_blocknr;
530 * It is going to be referred to from the inode itself? OK, just put it
531 * into the same cylinder group then.
533 block_group = ei->i_block_group;
534 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
535 block_group &= ~(flex_size-1);
536 if (S_ISREG(inode->i_mode))
539 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
540 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
543 * If we are doing delayed allocation, we don't need take
544 * colour into account.
546 if (test_opt(inode->i_sb, DELALLOC))
549 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
550 colour = (current->pid % 16) *
551 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
553 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
554 return bg_start + colour;
558 * ext4_find_goal - find a preferred place for allocation.
560 * @block: block we want
561 * @partial: pointer to the last triple within a chain
563 * Normally this function find the preferred place for block allocation,
565 * Because this is only used for non-extent files, we limit the block nr
568 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
574 * XXX need to get goal block from mballoc's data structures
577 goal = ext4_find_near(inode, partial);
578 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
583 * ext4_blks_to_allocate: Look up the block map and count the number
584 * of direct blocks need to be allocated for the given branch.
586 * @branch: chain of indirect blocks
587 * @k: number of blocks need for indirect blocks
588 * @blks: number of data blocks to be mapped.
589 * @blocks_to_boundary: the offset in the indirect block
591 * return the total number of blocks to be allocate, including the
592 * direct and indirect blocks.
594 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
595 int blocks_to_boundary)
597 unsigned int count = 0;
600 * Simple case, [t,d]Indirect block(s) has not allocated yet
601 * then it's clear blocks on that path have not allocated
604 /* right now we don't handle cross boundary allocation */
605 if (blks < blocks_to_boundary + 1)
608 count += blocks_to_boundary + 1;
613 while (count < blks && count <= blocks_to_boundary &&
614 le32_to_cpu(*(branch[0].p + count)) == 0) {
621 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
622 * @indirect_blks: the number of blocks need to allocate for indirect
625 * @new_blocks: on return it will store the new block numbers for
626 * the indirect blocks(if needed) and the first direct block,
627 * @blks: on return it will store the total number of allocated
630 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
631 ext4_lblk_t iblock, ext4_fsblk_t goal,
632 int indirect_blks, int blks,
633 ext4_fsblk_t new_blocks[4], int *err)
635 struct ext4_allocation_request ar;
637 unsigned long count = 0, blk_allocated = 0;
639 ext4_fsblk_t current_block = 0;
643 * Here we try to allocate the requested multiple blocks at once,
644 * on a best-effort basis.
645 * To build a branch, we should allocate blocks for
646 * the indirect blocks(if not allocated yet), and at least
647 * the first direct block of this branch. That's the
648 * minimum number of blocks need to allocate(required)
650 /* first we try to allocate the indirect blocks */
651 target = indirect_blks;
654 /* allocating blocks for indirect blocks and direct blocks */
655 current_block = ext4_new_meta_blocks(handle, inode,
660 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
663 /* allocate blocks for indirect blocks */
664 while (index < indirect_blks && count) {
665 new_blocks[index++] = current_block++;
670 * save the new block number
671 * for the first direct block
673 new_blocks[index] = current_block;
674 printk(KERN_INFO "%s returned more blocks than "
675 "requested\n", __func__);
681 target = blks - count ;
682 blk_allocated = count;
685 /* Now allocate data blocks */
686 memset(&ar, 0, sizeof(ar));
691 if (S_ISREG(inode->i_mode))
692 /* enable in-core preallocation only for regular files */
693 ar.flags = EXT4_MB_HINT_DATA;
695 current_block = ext4_mb_new_blocks(handle, &ar, err);
696 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
698 if (*err && (target == blks)) {
700 * if the allocation failed and we didn't allocate
706 if (target == blks) {
708 * save the new block number
709 * for the first direct block
711 new_blocks[index] = current_block;
713 blk_allocated += ar.len;
716 /* total number of blocks allocated for direct blocks */
721 for (i = 0; i < index; i++)
722 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
727 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @indirect_blks: number of allocated indirect blocks
730 * @blks: number of allocated direct blocks
731 * @offsets: offsets (in the blocks) to store the pointers to next.
732 * @branch: place to store the chain in.
734 * This function allocates blocks, zeroes out all but the last one,
735 * links them into chain and (if we are synchronous) writes them to disk.
736 * In other words, it prepares a branch that can be spliced onto the
737 * inode. It stores the information about that chain in the branch[], in
738 * the same format as ext4_get_branch() would do. We are calling it after
739 * we had read the existing part of chain and partial points to the last
740 * triple of that (one with zero ->key). Upon the exit we have the same
741 * picture as after the successful ext4_get_block(), except that in one
742 * place chain is disconnected - *branch->p is still zero (we did not
743 * set the last link), but branch->key contains the number that should
744 * be placed into *branch->p to fill that gap.
746 * If allocation fails we free all blocks we've allocated (and forget
747 * their buffer_heads) and return the error value the from failed
748 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
749 * as described above and return 0.
751 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
752 ext4_lblk_t iblock, int indirect_blks,
753 int *blks, ext4_fsblk_t goal,
754 ext4_lblk_t *offsets, Indirect *branch)
756 int blocksize = inode->i_sb->s_blocksize;
759 struct buffer_head *bh;
761 ext4_fsblk_t new_blocks[4];
762 ext4_fsblk_t current_block;
764 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
765 *blks, new_blocks, &err);
769 branch[0].key = cpu_to_le32(new_blocks[0]);
771 * metadata blocks and data blocks are allocated.
773 for (n = 1; n <= indirect_blks; n++) {
775 * Get buffer_head for parent block, zero it out
776 * and set the pointer to new one, then send
779 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
782 BUFFER_TRACE(bh, "call get_create_access");
783 err = ext4_journal_get_create_access(handle, bh);
785 /* Don't brelse(bh) here; it's done in
786 * ext4_journal_forget() below */
791 memset(bh->b_data, 0, blocksize);
792 branch[n].p = (__le32 *) bh->b_data + offsets[n];
793 branch[n].key = cpu_to_le32(new_blocks[n]);
794 *branch[n].p = branch[n].key;
795 if (n == indirect_blks) {
796 current_block = new_blocks[n];
798 * End of chain, update the last new metablock of
799 * the chain to point to the new allocated
800 * data blocks numbers
802 for (i = 1; i < num; i++)
803 *(branch[n].p + i) = cpu_to_le32(++current_block);
805 BUFFER_TRACE(bh, "marking uptodate");
806 set_buffer_uptodate(bh);
809 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
810 err = ext4_handle_dirty_metadata(handle, inode, bh);
817 /* Allocation failed, free what we already allocated */
818 for (i = 1; i <= n ; i++) {
819 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
820 ext4_journal_forget(handle, branch[i].bh);
822 for (i = 0; i < indirect_blks; i++)
823 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
825 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
831 * ext4_splice_branch - splice the allocated branch onto inode.
833 * @block: (logical) number of block we are adding
834 * @chain: chain of indirect blocks (with a missing link - see
836 * @where: location of missing link
837 * @num: number of indirect blocks we are adding
838 * @blks: number of direct blocks we are adding
840 * This function fills the missing link and does all housekeeping needed in
841 * inode (->i_blocks, etc.). In case of success we end up with the full
842 * chain to new block and return 0.
844 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
845 ext4_lblk_t block, Indirect *where, int num,
850 ext4_fsblk_t current_block;
853 * If we're splicing into a [td]indirect block (as opposed to the
854 * inode) then we need to get write access to the [td]indirect block
858 BUFFER_TRACE(where->bh, "get_write_access");
859 err = ext4_journal_get_write_access(handle, where->bh);
865 *where->p = where->key;
868 * Update the host buffer_head or inode to point to more just allocated
869 * direct blocks blocks
871 if (num == 0 && blks > 1) {
872 current_block = le32_to_cpu(where->key) + 1;
873 for (i = 1; i < blks; i++)
874 *(where->p + i) = cpu_to_le32(current_block++);
877 /* We are done with atomic stuff, now do the rest of housekeeping */
878 /* had we spliced it onto indirect block? */
881 * If we spliced it onto an indirect block, we haven't
882 * altered the inode. Note however that if it is being spliced
883 * onto an indirect block at the very end of the file (the
884 * file is growing) then we *will* alter the inode to reflect
885 * the new i_size. But that is not done here - it is done in
886 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
888 jbd_debug(5, "splicing indirect only\n");
889 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
890 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
895 * OK, we spliced it into the inode itself on a direct block.
897 ext4_mark_inode_dirty(handle, inode);
898 jbd_debug(5, "splicing direct\n");
903 for (i = 1; i <= num; i++) {
904 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
905 ext4_journal_forget(handle, where[i].bh);
906 ext4_free_blocks(handle, inode,
907 le32_to_cpu(where[i-1].key), 1, 0);
909 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
915 * The ext4_ind_get_blocks() function handles non-extents inodes
916 * (i.e., using the traditional indirect/double-indirect i_blocks
917 * scheme) for ext4_get_blocks().
919 * Allocation strategy is simple: if we have to allocate something, we will
920 * have to go the whole way to leaf. So let's do it before attaching anything
921 * to tree, set linkage between the newborn blocks, write them if sync is
922 * required, recheck the path, free and repeat if check fails, otherwise
923 * set the last missing link (that will protect us from any truncate-generated
924 * removals - all blocks on the path are immune now) and possibly force the
925 * write on the parent block.
926 * That has a nice additional property: no special recovery from the failed
927 * allocations is needed - we simply release blocks and do not touch anything
928 * reachable from inode.
930 * `handle' can be NULL if create == 0.
932 * return > 0, # of blocks mapped or allocated.
933 * return = 0, if plain lookup failed.
934 * return < 0, error case.
936 * The ext4_ind_get_blocks() function should be called with
937 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
938 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
939 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
942 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
943 ext4_lblk_t iblock, unsigned int maxblocks,
944 struct buffer_head *bh_result,
948 ext4_lblk_t offsets[4];
953 int blocks_to_boundary = 0;
956 ext4_fsblk_t first_block = 0;
958 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
959 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
960 depth = ext4_block_to_path(inode, iblock, offsets,
961 &blocks_to_boundary);
966 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
968 /* Simplest case - block found, no allocation needed */
970 first_block = le32_to_cpu(chain[depth - 1].key);
971 clear_buffer_new(bh_result);
974 while (count < maxblocks && count <= blocks_to_boundary) {
977 blk = le32_to_cpu(*(chain[depth-1].p + count));
979 if (blk == first_block + count)
987 /* Next simple case - plain lookup or failed read of indirect block */
988 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
992 * Okay, we need to do block allocation.
994 goal = ext4_find_goal(inode, iblock, partial);
996 /* the number of blocks need to allocate for [d,t]indirect blocks */
997 indirect_blks = (chain + depth) - partial - 1;
1000 * Next look up the indirect map to count the totoal number of
1001 * direct blocks to allocate for this branch.
1003 count = ext4_blks_to_allocate(partial, indirect_blks,
1004 maxblocks, blocks_to_boundary);
1006 * Block out ext4_truncate while we alter the tree
1008 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
1010 offsets + (partial - chain), partial);
1013 * The ext4_splice_branch call will free and forget any buffers
1014 * on the new chain if there is a failure, but that risks using
1015 * up transaction credits, especially for bitmaps where the
1016 * credits cannot be returned. Can we handle this somehow? We
1017 * may need to return -EAGAIN upwards in the worst case. --sct
1020 err = ext4_splice_branch(handle, inode, iblock,
1021 partial, indirect_blks, count);
1025 set_buffer_new(bh_result);
1027 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1028 if (count > blocks_to_boundary)
1029 set_buffer_boundary(bh_result);
1031 /* Clean up and exit */
1032 partial = chain + depth - 1; /* the whole chain */
1034 while (partial > chain) {
1035 BUFFER_TRACE(partial->bh, "call brelse");
1036 brelse(partial->bh);
1039 BUFFER_TRACE(bh_result, "returned");
1044 qsize_t ext4_get_reserved_space(struct inode *inode)
1046 unsigned long long total;
1048 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1049 total = EXT4_I(inode)->i_reserved_data_blocks +
1050 EXT4_I(inode)->i_reserved_meta_blocks;
1051 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1056 * Calculate the number of metadata blocks need to reserve
1057 * to allocate @blocks for non extent file based file
1059 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1061 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1062 int ind_blks, dind_blks, tind_blks;
1064 /* number of new indirect blocks needed */
1065 ind_blks = (blocks + icap - 1) / icap;
1067 dind_blks = (ind_blks + icap - 1) / icap;
1071 return ind_blks + dind_blks + tind_blks;
1075 * Calculate the number of metadata blocks need to reserve
1076 * to allocate given number of blocks
1078 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1083 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1084 return ext4_ext_calc_metadata_amount(inode, blocks);
1086 return ext4_indirect_calc_metadata_amount(inode, blocks);
1089 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1091 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1092 int total, mdb, mdb_free;
1094 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1095 /* recalculate the number of metablocks still need to be reserved */
1096 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1097 mdb = ext4_calc_metadata_amount(inode, total);
1099 /* figure out how many metablocks to release */
1100 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1101 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1104 /* Account for allocated meta_blocks */
1105 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1107 /* update fs dirty blocks counter */
1108 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1109 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1110 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1113 /* update per-inode reservations */
1114 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1115 EXT4_I(inode)->i_reserved_data_blocks -= used;
1116 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1119 * free those over-booking quota for metadata blocks
1122 vfs_dq_release_reservation_block(inode, mdb_free);
1125 * If we have done all the pending block allocations and if
1126 * there aren't any writers on the inode, we can discard the
1127 * inode's preallocations.
1129 if (!total && (atomic_read(&inode->i_writecount) == 0))
1130 ext4_discard_preallocations(inode);
1133 static int check_block_validity(struct inode *inode, const char *msg,
1134 sector_t logical, sector_t phys, int len)
1136 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1137 ext4_error(inode->i_sb, msg,
1138 "inode #%lu logical block %llu mapped to %llu "
1139 "(size %d)", inode->i_ino,
1140 (unsigned long long) logical,
1141 (unsigned long long) phys, len);
1148 * The ext4_get_blocks() function tries to look up the requested blocks,
1149 * and returns if the blocks are already mapped.
1151 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1152 * and store the allocated blocks in the result buffer head and mark it
1155 * If file type is extents based, it will call ext4_ext_get_blocks(),
1156 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1159 * On success, it returns the number of blocks being mapped or allocate.
1160 * if create==0 and the blocks are pre-allocated and uninitialized block,
1161 * the result buffer head is unmapped. If the create ==1, it will make sure
1162 * the buffer head is mapped.
1164 * It returns 0 if plain look up failed (blocks have not been allocated), in
1165 * that casem, buffer head is unmapped
1167 * It returns the error in case of allocation failure.
1169 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1170 unsigned int max_blocks, struct buffer_head *bh,
1175 clear_buffer_mapped(bh);
1176 clear_buffer_unwritten(bh);
1179 * Try to see if we can get the block without requesting a new
1180 * file system block.
1182 down_read((&EXT4_I(inode)->i_data_sem));
1183 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1184 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1187 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1190 up_read((&EXT4_I(inode)->i_data_sem));
1192 if (retval > 0 && buffer_mapped(bh)) {
1193 int ret = check_block_validity(inode, "file system corruption",
1194 block, bh->b_blocknr, retval);
1199 /* If it is only a block(s) look up */
1200 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1204 * Returns if the blocks have already allocated
1206 * Note that if blocks have been preallocated
1207 * ext4_ext_get_block() returns th create = 0
1208 * with buffer head unmapped.
1210 if (retval > 0 && buffer_mapped(bh))
1214 * When we call get_blocks without the create flag, the
1215 * BH_Unwritten flag could have gotten set if the blocks
1216 * requested were part of a uninitialized extent. We need to
1217 * clear this flag now that we are committed to convert all or
1218 * part of the uninitialized extent to be an initialized
1219 * extent. This is because we need to avoid the combination
1220 * of BH_Unwritten and BH_Mapped flags being simultaneously
1221 * set on the buffer_head.
1223 clear_buffer_unwritten(bh);
1226 * New blocks allocate and/or writing to uninitialized extent
1227 * will possibly result in updating i_data, so we take
1228 * the write lock of i_data_sem, and call get_blocks()
1229 * with create == 1 flag.
1231 down_write((&EXT4_I(inode)->i_data_sem));
1234 * if the caller is from delayed allocation writeout path
1235 * we have already reserved fs blocks for allocation
1236 * let the underlying get_block() function know to
1237 * avoid double accounting
1239 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1240 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1242 * We need to check for EXT4 here because migrate
1243 * could have changed the inode type in between
1245 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1246 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1249 retval = ext4_ind_get_blocks(handle, inode, block,
1250 max_blocks, bh, flags);
1252 if (retval > 0 && buffer_new(bh)) {
1254 * We allocated new blocks which will result in
1255 * i_data's format changing. Force the migrate
1256 * to fail by clearing migrate flags
1258 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
1262 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1263 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1266 * Update reserved blocks/metadata blocks after successful
1267 * block allocation which had been deferred till now.
1269 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1270 ext4_da_update_reserve_space(inode, retval);
1272 up_write((&EXT4_I(inode)->i_data_sem));
1273 if (retval > 0 && buffer_mapped(bh)) {
1274 int ret = check_block_validity(inode, "file system "
1275 "corruption after allocation",
1276 block, bh->b_blocknr, retval);
1283 /* Maximum number of blocks we map for direct IO at once. */
1284 #define DIO_MAX_BLOCKS 4096
1286 int ext4_get_block(struct inode *inode, sector_t iblock,
1287 struct buffer_head *bh_result, int create)
1289 handle_t *handle = ext4_journal_current_handle();
1290 int ret = 0, started = 0;
1291 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1294 if (create && !handle) {
1295 /* Direct IO write... */
1296 if (max_blocks > DIO_MAX_BLOCKS)
1297 max_blocks = DIO_MAX_BLOCKS;
1298 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1299 handle = ext4_journal_start(inode, dio_credits);
1300 if (IS_ERR(handle)) {
1301 ret = PTR_ERR(handle);
1307 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1308 create ? EXT4_GET_BLOCKS_CREATE : 0);
1310 bh_result->b_size = (ret << inode->i_blkbits);
1314 ext4_journal_stop(handle);
1320 * `handle' can be NULL if create is zero
1322 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1323 ext4_lblk_t block, int create, int *errp)
1325 struct buffer_head dummy;
1329 J_ASSERT(handle != NULL || create == 0);
1332 dummy.b_blocknr = -1000;
1333 buffer_trace_init(&dummy.b_history);
1335 flags |= EXT4_GET_BLOCKS_CREATE;
1336 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1338 * ext4_get_blocks() returns number of blocks mapped. 0 in
1347 if (!err && buffer_mapped(&dummy)) {
1348 struct buffer_head *bh;
1349 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1354 if (buffer_new(&dummy)) {
1355 J_ASSERT(create != 0);
1356 J_ASSERT(handle != NULL);
1359 * Now that we do not always journal data, we should
1360 * keep in mind whether this should always journal the
1361 * new buffer as metadata. For now, regular file
1362 * writes use ext4_get_block instead, so it's not a
1366 BUFFER_TRACE(bh, "call get_create_access");
1367 fatal = ext4_journal_get_create_access(handle, bh);
1368 if (!fatal && !buffer_uptodate(bh)) {
1369 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1370 set_buffer_uptodate(bh);
1373 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1374 err = ext4_handle_dirty_metadata(handle, inode, bh);
1378 BUFFER_TRACE(bh, "not a new buffer");
1391 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1392 ext4_lblk_t block, int create, int *err)
1394 struct buffer_head *bh;
1396 bh = ext4_getblk(handle, inode, block, create, err);
1399 if (buffer_uptodate(bh))
1401 ll_rw_block(READ_META, 1, &bh);
1403 if (buffer_uptodate(bh))
1410 static int walk_page_buffers(handle_t *handle,
1411 struct buffer_head *head,
1415 int (*fn)(handle_t *handle,
1416 struct buffer_head *bh))
1418 struct buffer_head *bh;
1419 unsigned block_start, block_end;
1420 unsigned blocksize = head->b_size;
1422 struct buffer_head *next;
1424 for (bh = head, block_start = 0;
1425 ret == 0 && (bh != head || !block_start);
1426 block_start = block_end, bh = next) {
1427 next = bh->b_this_page;
1428 block_end = block_start + blocksize;
1429 if (block_end <= from || block_start >= to) {
1430 if (partial && !buffer_uptodate(bh))
1434 err = (*fn)(handle, bh);
1442 * To preserve ordering, it is essential that the hole instantiation and
1443 * the data write be encapsulated in a single transaction. We cannot
1444 * close off a transaction and start a new one between the ext4_get_block()
1445 * and the commit_write(). So doing the jbd2_journal_start at the start of
1446 * prepare_write() is the right place.
1448 * Also, this function can nest inside ext4_writepage() ->
1449 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1450 * has generated enough buffer credits to do the whole page. So we won't
1451 * block on the journal in that case, which is good, because the caller may
1454 * By accident, ext4 can be reentered when a transaction is open via
1455 * quota file writes. If we were to commit the transaction while thus
1456 * reentered, there can be a deadlock - we would be holding a quota
1457 * lock, and the commit would never complete if another thread had a
1458 * transaction open and was blocking on the quota lock - a ranking
1461 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1462 * will _not_ run commit under these circumstances because handle->h_ref
1463 * is elevated. We'll still have enough credits for the tiny quotafile
1466 static int do_journal_get_write_access(handle_t *handle,
1467 struct buffer_head *bh)
1469 if (!buffer_mapped(bh) || buffer_freed(bh))
1471 return ext4_journal_get_write_access(handle, bh);
1474 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1475 loff_t pos, unsigned len, unsigned flags,
1476 struct page **pagep, void **fsdata)
1478 struct inode *inode = mapping->host;
1479 int ret, needed_blocks;
1486 trace_ext4_write_begin(inode, pos, len, flags);
1488 * Reserve one block more for addition to orphan list in case
1489 * we allocate blocks but write fails for some reason
1491 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1492 index = pos >> PAGE_CACHE_SHIFT;
1493 from = pos & (PAGE_CACHE_SIZE - 1);
1497 handle = ext4_journal_start(inode, needed_blocks);
1498 if (IS_ERR(handle)) {
1499 ret = PTR_ERR(handle);
1503 /* We cannot recurse into the filesystem as the transaction is already
1505 flags |= AOP_FLAG_NOFS;
1507 page = grab_cache_page_write_begin(mapping, index, flags);
1509 ext4_journal_stop(handle);
1515 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1518 if (!ret && ext4_should_journal_data(inode)) {
1519 ret = walk_page_buffers(handle, page_buffers(page),
1520 from, to, NULL, do_journal_get_write_access);
1525 page_cache_release(page);
1527 * block_write_begin may have instantiated a few blocks
1528 * outside i_size. Trim these off again. Don't need
1529 * i_size_read because we hold i_mutex.
1531 * Add inode to orphan list in case we crash before
1534 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1535 ext4_orphan_add(handle, inode);
1537 ext4_journal_stop(handle);
1538 if (pos + len > inode->i_size) {
1539 ext4_truncate(inode);
1541 * If truncate failed early the inode might
1542 * still be on the orphan list; we need to
1543 * make sure the inode is removed from the
1544 * orphan list in that case.
1547 ext4_orphan_del(NULL, inode);
1551 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1557 /* For write_end() in data=journal mode */
1558 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1560 if (!buffer_mapped(bh) || buffer_freed(bh))
1562 set_buffer_uptodate(bh);
1563 return ext4_handle_dirty_metadata(handle, NULL, bh);
1566 static int ext4_generic_write_end(struct file *file,
1567 struct address_space *mapping,
1568 loff_t pos, unsigned len, unsigned copied,
1569 struct page *page, void *fsdata)
1571 int i_size_changed = 0;
1572 struct inode *inode = mapping->host;
1573 handle_t *handle = ext4_journal_current_handle();
1575 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1578 * No need to use i_size_read() here, the i_size
1579 * cannot change under us because we hold i_mutex.
1581 * But it's important to update i_size while still holding page lock:
1582 * page writeout could otherwise come in and zero beyond i_size.
1584 if (pos + copied > inode->i_size) {
1585 i_size_write(inode, pos + copied);
1589 if (pos + copied > EXT4_I(inode)->i_disksize) {
1590 /* We need to mark inode dirty even if
1591 * new_i_size is less that inode->i_size
1592 * bu greater than i_disksize.(hint delalloc)
1594 ext4_update_i_disksize(inode, (pos + copied));
1598 page_cache_release(page);
1601 * Don't mark the inode dirty under page lock. First, it unnecessarily
1602 * makes the holding time of page lock longer. Second, it forces lock
1603 * ordering of page lock and transaction start for journaling
1607 ext4_mark_inode_dirty(handle, inode);
1613 * We need to pick up the new inode size which generic_commit_write gave us
1614 * `file' can be NULL - eg, when called from page_symlink().
1616 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1617 * buffers are managed internally.
1619 static int ext4_ordered_write_end(struct file *file,
1620 struct address_space *mapping,
1621 loff_t pos, unsigned len, unsigned copied,
1622 struct page *page, void *fsdata)
1624 handle_t *handle = ext4_journal_current_handle();
1625 struct inode *inode = mapping->host;
1628 trace_ext4_ordered_write_end(inode, pos, len, copied);
1629 ret = ext4_jbd2_file_inode(handle, inode);
1632 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1635 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1636 /* if we have allocated more blocks and copied
1637 * less. We will have blocks allocated outside
1638 * inode->i_size. So truncate them
1640 ext4_orphan_add(handle, inode);
1644 ret2 = ext4_journal_stop(handle);
1648 if (pos + len > inode->i_size) {
1649 ext4_truncate(inode);
1651 * If truncate failed early the inode might still be
1652 * on the orphan list; we need to make sure the inode
1653 * is removed from the orphan list in that case.
1656 ext4_orphan_del(NULL, inode);
1660 return ret ? ret : copied;
1663 static int ext4_writeback_write_end(struct file *file,
1664 struct address_space *mapping,
1665 loff_t pos, unsigned len, unsigned copied,
1666 struct page *page, void *fsdata)
1668 handle_t *handle = ext4_journal_current_handle();
1669 struct inode *inode = mapping->host;
1672 trace_ext4_writeback_write_end(inode, pos, len, copied);
1673 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1676 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1677 /* if we have allocated more blocks and copied
1678 * less. We will have blocks allocated outside
1679 * inode->i_size. So truncate them
1681 ext4_orphan_add(handle, inode);
1686 ret2 = ext4_journal_stop(handle);
1690 if (pos + len > inode->i_size) {
1691 ext4_truncate(inode);
1693 * If truncate failed early the inode might still be
1694 * on the orphan list; we need to make sure the inode
1695 * is removed from the orphan list in that case.
1698 ext4_orphan_del(NULL, inode);
1701 return ret ? ret : copied;
1704 static int ext4_journalled_write_end(struct file *file,
1705 struct address_space *mapping,
1706 loff_t pos, unsigned len, unsigned copied,
1707 struct page *page, void *fsdata)
1709 handle_t *handle = ext4_journal_current_handle();
1710 struct inode *inode = mapping->host;
1716 trace_ext4_journalled_write_end(inode, pos, len, copied);
1717 from = pos & (PAGE_CACHE_SIZE - 1);
1721 if (!PageUptodate(page))
1723 page_zero_new_buffers(page, from+copied, to);
1726 ret = walk_page_buffers(handle, page_buffers(page), from,
1727 to, &partial, write_end_fn);
1729 SetPageUptodate(page);
1730 new_i_size = pos + copied;
1731 if (new_i_size > inode->i_size)
1732 i_size_write(inode, pos+copied);
1733 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1734 if (new_i_size > EXT4_I(inode)->i_disksize) {
1735 ext4_update_i_disksize(inode, new_i_size);
1736 ret2 = ext4_mark_inode_dirty(handle, inode);
1742 page_cache_release(page);
1743 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1744 /* if we have allocated more blocks and copied
1745 * less. We will have blocks allocated outside
1746 * inode->i_size. So truncate them
1748 ext4_orphan_add(handle, inode);
1750 ret2 = ext4_journal_stop(handle);
1753 if (pos + len > inode->i_size) {
1754 ext4_truncate(inode);
1756 * If truncate failed early the inode might still be
1757 * on the orphan list; we need to make sure the inode
1758 * is removed from the orphan list in that case.
1761 ext4_orphan_del(NULL, inode);
1764 return ret ? ret : copied;
1767 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1770 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1771 unsigned long md_needed, mdblocks, total = 0;
1774 * recalculate the amount of metadata blocks to reserve
1775 * in order to allocate nrblocks
1776 * worse case is one extent per block
1779 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1780 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1781 mdblocks = ext4_calc_metadata_amount(inode, total);
1782 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1784 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1785 total = md_needed + nrblocks;
1788 * Make quota reservation here to prevent quota overflow
1789 * later. Real quota accounting is done at pages writeout
1792 if (vfs_dq_reserve_block(inode, total)) {
1793 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1797 if (ext4_claim_free_blocks(sbi, total)) {
1798 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1799 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1803 vfs_dq_release_reservation_block(inode, total);
1806 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1807 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1809 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1810 return 0; /* success */
1813 static void ext4_da_release_space(struct inode *inode, int to_free)
1815 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1816 int total, mdb, mdb_free, release;
1819 return; /* Nothing to release, exit */
1821 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1823 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1825 * if there is no reserved blocks, but we try to free some
1826 * then the counter is messed up somewhere.
1827 * but since this function is called from invalidate
1828 * page, it's harmless to return without any action
1830 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1831 "blocks for inode %lu, but there is no reserved "
1832 "data blocks\n", to_free, inode->i_ino);
1833 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1837 /* recalculate the number of metablocks still need to be reserved */
1838 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1839 mdb = ext4_calc_metadata_amount(inode, total);
1841 /* figure out how many metablocks to release */
1842 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1843 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1845 release = to_free + mdb_free;
1847 /* update fs dirty blocks counter for truncate case */
1848 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1850 /* update per-inode reservations */
1851 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1852 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1854 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1855 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1856 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1858 vfs_dq_release_reservation_block(inode, release);
1861 static void ext4_da_page_release_reservation(struct page *page,
1862 unsigned long offset)
1865 struct buffer_head *head, *bh;
1866 unsigned int curr_off = 0;
1868 head = page_buffers(page);
1871 unsigned int next_off = curr_off + bh->b_size;
1873 if ((offset <= curr_off) && (buffer_delay(bh))) {
1875 clear_buffer_delay(bh);
1877 curr_off = next_off;
1878 } while ((bh = bh->b_this_page) != head);
1879 ext4_da_release_space(page->mapping->host, to_release);
1883 * Delayed allocation stuff
1887 * mpage_da_submit_io - walks through extent of pages and try to write
1888 * them with writepage() call back
1890 * @mpd->inode: inode
1891 * @mpd->first_page: first page of the extent
1892 * @mpd->next_page: page after the last page of the extent
1894 * By the time mpage_da_submit_io() is called we expect all blocks
1895 * to be allocated. this may be wrong if allocation failed.
1897 * As pages are already locked by write_cache_pages(), we can't use it
1899 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1902 struct pagevec pvec;
1903 unsigned long index, end;
1904 int ret = 0, err, nr_pages, i;
1905 struct inode *inode = mpd->inode;
1906 struct address_space *mapping = inode->i_mapping;
1908 BUG_ON(mpd->next_page <= mpd->first_page);
1910 * We need to start from the first_page to the next_page - 1
1911 * to make sure we also write the mapped dirty buffer_heads.
1912 * If we look at mpd->b_blocknr we would only be looking
1913 * at the currently mapped buffer_heads.
1915 index = mpd->first_page;
1916 end = mpd->next_page - 1;
1918 pagevec_init(&pvec, 0);
1919 while (index <= end) {
1920 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1923 for (i = 0; i < nr_pages; i++) {
1924 struct page *page = pvec.pages[i];
1926 index = page->index;
1931 BUG_ON(!PageLocked(page));
1932 BUG_ON(PageWriteback(page));
1934 pages_skipped = mpd->wbc->pages_skipped;
1935 err = mapping->a_ops->writepage(page, mpd->wbc);
1936 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1938 * have successfully written the page
1939 * without skipping the same
1941 mpd->pages_written++;
1943 * In error case, we have to continue because
1944 * remaining pages are still locked
1945 * XXX: unlock and re-dirty them?
1950 pagevec_release(&pvec);
1956 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1958 * @mpd->inode - inode to walk through
1959 * @exbh->b_blocknr - first block on a disk
1960 * @exbh->b_size - amount of space in bytes
1961 * @logical - first logical block to start assignment with
1963 * the function goes through all passed space and put actual disk
1964 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1966 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1967 struct buffer_head *exbh)
1969 struct inode *inode = mpd->inode;
1970 struct address_space *mapping = inode->i_mapping;
1971 int blocks = exbh->b_size >> inode->i_blkbits;
1972 sector_t pblock = exbh->b_blocknr, cur_logical;
1973 struct buffer_head *head, *bh;
1975 struct pagevec pvec;
1978 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1979 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1980 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1982 pagevec_init(&pvec, 0);
1984 while (index <= end) {
1985 /* XXX: optimize tail */
1986 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1989 for (i = 0; i < nr_pages; i++) {
1990 struct page *page = pvec.pages[i];
1992 index = page->index;
1997 BUG_ON(!PageLocked(page));
1998 BUG_ON(PageWriteback(page));
1999 BUG_ON(!page_has_buffers(page));
2001 bh = page_buffers(page);
2004 /* skip blocks out of the range */
2006 if (cur_logical >= logical)
2009 } while ((bh = bh->b_this_page) != head);
2012 if (cur_logical >= logical + blocks)
2015 if (buffer_delay(bh) ||
2016 buffer_unwritten(bh)) {
2018 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2020 if (buffer_delay(bh)) {
2021 clear_buffer_delay(bh);
2022 bh->b_blocknr = pblock;
2025 * unwritten already should have
2026 * blocknr assigned. Verify that
2028 clear_buffer_unwritten(bh);
2029 BUG_ON(bh->b_blocknr != pblock);
2032 } else if (buffer_mapped(bh))
2033 BUG_ON(bh->b_blocknr != pblock);
2037 } while ((bh = bh->b_this_page) != head);
2039 pagevec_release(&pvec);
2045 * __unmap_underlying_blocks - just a helper function to unmap
2046 * set of blocks described by @bh
2048 static inline void __unmap_underlying_blocks(struct inode *inode,
2049 struct buffer_head *bh)
2051 struct block_device *bdev = inode->i_sb->s_bdev;
2054 blocks = bh->b_size >> inode->i_blkbits;
2055 for (i = 0; i < blocks; i++)
2056 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2059 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2060 sector_t logical, long blk_cnt)
2064 struct pagevec pvec;
2065 struct inode *inode = mpd->inode;
2066 struct address_space *mapping = inode->i_mapping;
2068 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2069 end = (logical + blk_cnt - 1) >>
2070 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2071 while (index <= end) {
2072 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2075 for (i = 0; i < nr_pages; i++) {
2076 struct page *page = pvec.pages[i];
2077 index = page->index;
2082 BUG_ON(!PageLocked(page));
2083 BUG_ON(PageWriteback(page));
2084 block_invalidatepage(page, 0);
2085 ClearPageUptodate(page);
2092 static void ext4_print_free_blocks(struct inode *inode)
2094 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2095 printk(KERN_EMERG "Total free blocks count %lld\n",
2096 ext4_count_free_blocks(inode->i_sb));
2097 printk(KERN_EMERG "Free/Dirty block details\n");
2098 printk(KERN_EMERG "free_blocks=%lld\n",
2099 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
2100 printk(KERN_EMERG "dirty_blocks=%lld\n",
2101 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2102 printk(KERN_EMERG "Block reservation details\n");
2103 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
2104 EXT4_I(inode)->i_reserved_data_blocks);
2105 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
2106 EXT4_I(inode)->i_reserved_meta_blocks);
2111 * mpage_da_map_blocks - go through given space
2113 * @mpd - bh describing space
2115 * The function skips space we know is already mapped to disk blocks.
2118 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2120 int err, blks, get_blocks_flags;
2121 struct buffer_head new;
2122 sector_t next = mpd->b_blocknr;
2123 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2124 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2125 handle_t *handle = NULL;
2128 * We consider only non-mapped and non-allocated blocks
2130 if ((mpd->b_state & (1 << BH_Mapped)) &&
2131 !(mpd->b_state & (1 << BH_Delay)) &&
2132 !(mpd->b_state & (1 << BH_Unwritten)))
2136 * If we didn't accumulate anything to write simply return
2141 handle = ext4_journal_current_handle();
2145 * Call ext4_get_blocks() to allocate any delayed allocation
2146 * blocks, or to convert an uninitialized extent to be
2147 * initialized (in the case where we have written into
2148 * one or more preallocated blocks).
2150 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2151 * indicate that we are on the delayed allocation path. This
2152 * affects functions in many different parts of the allocation
2153 * call path. This flag exists primarily because we don't
2154 * want to change *many* call functions, so ext4_get_blocks()
2155 * will set the magic i_delalloc_reserved_flag once the
2156 * inode's allocation semaphore is taken.
2158 * If the blocks in questions were delalloc blocks, set
2159 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2160 * variables are updated after the blocks have been allocated.
2163 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2164 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2165 if (mpd->b_state & (1 << BH_Delay))
2166 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2167 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2168 &new, get_blocks_flags);
2172 * If get block returns with error we simply
2173 * return. Later writepage will redirty the page and
2174 * writepages will find the dirty page again
2179 if (err == -ENOSPC &&
2180 ext4_count_free_blocks(mpd->inode->i_sb)) {
2186 * get block failure will cause us to loop in
2187 * writepages, because a_ops->writepage won't be able
2188 * to make progress. The page will be redirtied by
2189 * writepage and writepages will again try to write
2192 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2193 "at logical offset %llu with max blocks "
2194 "%zd with error %d\n",
2195 __func__, mpd->inode->i_ino,
2196 (unsigned long long)next,
2197 mpd->b_size >> mpd->inode->i_blkbits, err);
2198 printk(KERN_EMERG "This should not happen.!! "
2199 "Data will be lost\n");
2200 if (err == -ENOSPC) {
2201 ext4_print_free_blocks(mpd->inode);
2203 /* invalidate all the pages */
2204 ext4_da_block_invalidatepages(mpd, next,
2205 mpd->b_size >> mpd->inode->i_blkbits);
2210 new.b_size = (blks << mpd->inode->i_blkbits);
2212 if (buffer_new(&new))
2213 __unmap_underlying_blocks(mpd->inode, &new);
2216 * If blocks are delayed marked, we need to
2217 * put actual blocknr and drop delayed bit
2219 if ((mpd->b_state & (1 << BH_Delay)) ||
2220 (mpd->b_state & (1 << BH_Unwritten)))
2221 mpage_put_bnr_to_bhs(mpd, next, &new);
2223 if (ext4_should_order_data(mpd->inode)) {
2224 err = ext4_jbd2_file_inode(handle, mpd->inode);
2230 * Update on-disk size along with block allocation.
2232 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2233 if (disksize > i_size_read(mpd->inode))
2234 disksize = i_size_read(mpd->inode);
2235 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2236 ext4_update_i_disksize(mpd->inode, disksize);
2237 return ext4_mark_inode_dirty(handle, mpd->inode);
2243 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2244 (1 << BH_Delay) | (1 << BH_Unwritten))
2247 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2249 * @mpd->lbh - extent of blocks
2250 * @logical - logical number of the block in the file
2251 * @bh - bh of the block (used to access block's state)
2253 * the function is used to collect contig. blocks in same state
2255 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2256 sector_t logical, size_t b_size,
2257 unsigned long b_state)
2260 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2262 /* check if thereserved journal credits might overflow */
2263 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2264 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2266 * With non-extent format we are limited by the journal
2267 * credit available. Total credit needed to insert
2268 * nrblocks contiguous blocks is dependent on the
2269 * nrblocks. So limit nrblocks.
2272 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2273 EXT4_MAX_TRANS_DATA) {
2275 * Adding the new buffer_head would make it cross the
2276 * allowed limit for which we have journal credit
2277 * reserved. So limit the new bh->b_size
2279 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2280 mpd->inode->i_blkbits;
2281 /* we will do mpage_da_submit_io in the next loop */
2285 * First block in the extent
2287 if (mpd->b_size == 0) {
2288 mpd->b_blocknr = logical;
2289 mpd->b_size = b_size;
2290 mpd->b_state = b_state & BH_FLAGS;
2294 next = mpd->b_blocknr + nrblocks;
2296 * Can we merge the block to our big extent?
2298 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2299 mpd->b_size += b_size;
2305 * We couldn't merge the block to our extent, so we
2306 * need to flush current extent and start new one
2308 if (mpage_da_map_blocks(mpd) == 0)
2309 mpage_da_submit_io(mpd);
2314 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2316 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2320 * __mpage_da_writepage - finds extent of pages and blocks
2322 * @page: page to consider
2323 * @wbc: not used, we just follow rules
2326 * The function finds extents of pages and scan them for all blocks.
2328 static int __mpage_da_writepage(struct page *page,
2329 struct writeback_control *wbc, void *data)
2331 struct mpage_da_data *mpd = data;
2332 struct inode *inode = mpd->inode;
2333 struct buffer_head *bh, *head;
2338 * Rest of the page in the page_vec
2339 * redirty then and skip then. We will
2340 * try to write them again after
2341 * starting a new transaction
2343 redirty_page_for_writepage(wbc, page);
2345 return MPAGE_DA_EXTENT_TAIL;
2348 * Can we merge this page to current extent?
2350 if (mpd->next_page != page->index) {
2352 * Nope, we can't. So, we map non-allocated blocks
2353 * and start IO on them using writepage()
2355 if (mpd->next_page != mpd->first_page) {
2356 if (mpage_da_map_blocks(mpd) == 0)
2357 mpage_da_submit_io(mpd);
2359 * skip rest of the page in the page_vec
2362 redirty_page_for_writepage(wbc, page);
2364 return MPAGE_DA_EXTENT_TAIL;
2368 * Start next extent of pages ...
2370 mpd->first_page = page->index;
2380 mpd->next_page = page->index + 1;
2381 logical = (sector_t) page->index <<
2382 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2384 if (!page_has_buffers(page)) {
2385 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2386 (1 << BH_Dirty) | (1 << BH_Uptodate));
2388 return MPAGE_DA_EXTENT_TAIL;
2391 * Page with regular buffer heads, just add all dirty ones
2393 head = page_buffers(page);
2396 BUG_ON(buffer_locked(bh));
2398 * We need to try to allocate
2399 * unmapped blocks in the same page.
2400 * Otherwise we won't make progress
2401 * with the page in ext4_writepage
2403 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2404 mpage_add_bh_to_extent(mpd, logical,
2408 return MPAGE_DA_EXTENT_TAIL;
2409 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2411 * mapped dirty buffer. We need to update
2412 * the b_state because we look at
2413 * b_state in mpage_da_map_blocks. We don't
2414 * update b_size because if we find an
2415 * unmapped buffer_head later we need to
2416 * use the b_state flag of that buffer_head.
2418 if (mpd->b_size == 0)
2419 mpd->b_state = bh->b_state & BH_FLAGS;
2422 } while ((bh = bh->b_this_page) != head);
2429 * This is a special get_blocks_t callback which is used by
2430 * ext4_da_write_begin(). It will either return mapped block or
2431 * reserve space for a single block.
2433 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2434 * We also have b_blocknr = -1 and b_bdev initialized properly
2436 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2437 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2438 * initialized properly.
2440 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2441 struct buffer_head *bh_result, int create)
2444 sector_t invalid_block = ~((sector_t) 0xffff);
2446 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2449 BUG_ON(create == 0);
2450 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2453 * first, we need to know whether the block is allocated already
2454 * preallocated blocks are unmapped but should treated
2455 * the same as allocated blocks.
2457 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2458 if ((ret == 0) && !buffer_delay(bh_result)) {
2459 /* the block isn't (pre)allocated yet, let's reserve space */
2461 * XXX: __block_prepare_write() unmaps passed block,
2464 ret = ext4_da_reserve_space(inode, 1);
2466 /* not enough space to reserve */
2469 map_bh(bh_result, inode->i_sb, invalid_block);
2470 set_buffer_new(bh_result);
2471 set_buffer_delay(bh_result);
2472 } else if (ret > 0) {
2473 bh_result->b_size = (ret << inode->i_blkbits);
2474 if (buffer_unwritten(bh_result)) {
2475 /* A delayed write to unwritten bh should
2476 * be marked new and mapped. Mapped ensures
2477 * that we don't do get_block multiple times
2478 * when we write to the same offset and new
2479 * ensures that we do proper zero out for
2482 set_buffer_new(bh_result);
2483 set_buffer_mapped(bh_result);
2492 * This function is used as a standard get_block_t calback function
2493 * when there is no desire to allocate any blocks. It is used as a
2494 * callback function for block_prepare_write(), nobh_writepage(), and
2495 * block_write_full_page(). These functions should only try to map a
2496 * single block at a time.
2498 * Since this function doesn't do block allocations even if the caller
2499 * requests it by passing in create=1, it is critically important that
2500 * any caller checks to make sure that any buffer heads are returned
2501 * by this function are either all already mapped or marked for
2502 * delayed allocation before calling nobh_writepage() or
2503 * block_write_full_page(). Otherwise, b_blocknr could be left
2504 * unitialized, and the page write functions will be taken by
2507 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2508 struct buffer_head *bh_result, int create)
2511 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2513 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2516 * we don't want to do block allocation in writepage
2517 * so call get_block_wrap with create = 0
2519 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2521 bh_result->b_size = (ret << inode->i_blkbits);
2527 static int bget_one(handle_t *handle, struct buffer_head *bh)
2533 static int bput_one(handle_t *handle, struct buffer_head *bh)
2539 static int __ext4_journalled_writepage(struct page *page,
2540 struct writeback_control *wbc,
2543 struct address_space *mapping = page->mapping;
2544 struct inode *inode = mapping->host;
2545 struct buffer_head *page_bufs;
2546 handle_t *handle = NULL;
2550 page_bufs = page_buffers(page);
2552 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2553 /* As soon as we unlock the page, it can go away, but we have
2554 * references to buffers so we are safe */
2557 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2558 if (IS_ERR(handle)) {
2559 ret = PTR_ERR(handle);
2563 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2564 do_journal_get_write_access);
2566 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2570 err = ext4_journal_stop(handle);
2574 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2575 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2581 * Note that we don't need to start a transaction unless we're journaling data
2582 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2583 * need to file the inode to the transaction's list in ordered mode because if
2584 * we are writing back data added by write(), the inode is already there and if
2585 * we are writing back data modified via mmap(), noone guarantees in which
2586 * transaction the data will hit the disk. In case we are journaling data, we
2587 * cannot start transaction directly because transaction start ranks above page
2588 * lock so we have to do some magic.
2590 * This function can get called via...
2591 * - ext4_da_writepages after taking page lock (have journal handle)
2592 * - journal_submit_inode_data_buffers (no journal handle)
2593 * - shrink_page_list via pdflush (no journal handle)
2594 * - grab_page_cache when doing write_begin (have journal handle)
2596 * We don't do any block allocation in this function. If we have page with
2597 * multiple blocks we need to write those buffer_heads that are mapped. This
2598 * is important for mmaped based write. So if we do with blocksize 1K
2599 * truncate(f, 1024);
2600 * a = mmap(f, 0, 4096);
2602 * truncate(f, 4096);
2603 * we have in the page first buffer_head mapped via page_mkwrite call back
2604 * but other bufer_heads would be unmapped but dirty(dirty done via the
2605 * do_wp_page). So writepage should write the first block. If we modify
2606 * the mmap area beyond 1024 we will again get a page_fault and the
2607 * page_mkwrite callback will do the block allocation and mark the
2608 * buffer_heads mapped.
2610 * We redirty the page if we have any buffer_heads that is either delay or
2611 * unwritten in the page.
2613 * We can get recursively called as show below.
2615 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2618 * But since we don't do any block allocation we should not deadlock.
2619 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2621 static int ext4_writepage(struct page *page,
2622 struct writeback_control *wbc)
2627 struct buffer_head *page_bufs;
2628 struct inode *inode = page->mapping->host;
2630 trace_ext4_writepage(inode, page);
2631 size = i_size_read(inode);
2632 if (page->index == size >> PAGE_CACHE_SHIFT)
2633 len = size & ~PAGE_CACHE_MASK;
2635 len = PAGE_CACHE_SIZE;
2637 if (page_has_buffers(page)) {
2638 page_bufs = page_buffers(page);
2639 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2640 ext4_bh_delay_or_unwritten)) {
2642 * We don't want to do block allocation
2643 * So redirty the page and return
2644 * We may reach here when we do a journal commit
2645 * via journal_submit_inode_data_buffers.
2646 * If we don't have mapping block we just ignore
2647 * them. We can also reach here via shrink_page_list
2649 redirty_page_for_writepage(wbc, page);
2655 * The test for page_has_buffers() is subtle:
2656 * We know the page is dirty but it lost buffers. That means
2657 * that at some moment in time after write_begin()/write_end()
2658 * has been called all buffers have been clean and thus they
2659 * must have been written at least once. So they are all
2660 * mapped and we can happily proceed with mapping them
2661 * and writing the page.
2663 * Try to initialize the buffer_heads and check whether
2664 * all are mapped and non delay. We don't want to
2665 * do block allocation here.
2667 ret = block_prepare_write(page, 0, len,
2668 noalloc_get_block_write);
2670 page_bufs = page_buffers(page);
2671 /* check whether all are mapped and non delay */
2672 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2673 ext4_bh_delay_or_unwritten)) {
2674 redirty_page_for_writepage(wbc, page);
2680 * We can't do block allocation here
2681 * so just redity the page and unlock
2684 redirty_page_for_writepage(wbc, page);
2688 /* now mark the buffer_heads as dirty and uptodate */
2689 block_commit_write(page, 0, len);
2692 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2694 * It's mmapped pagecache. Add buffers and journal it. There
2695 * doesn't seem much point in redirtying the page here.
2697 ClearPageChecked(page);
2698 return __ext4_journalled_writepage(page, wbc, len);
2701 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2702 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2704 ret = block_write_full_page(page, noalloc_get_block_write,
2711 * This is called via ext4_da_writepages() to
2712 * calulate the total number of credits to reserve to fit
2713 * a single extent allocation into a single transaction,
2714 * ext4_da_writpeages() will loop calling this before
2715 * the block allocation.
2718 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2720 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2723 * With non-extent format the journal credit needed to
2724 * insert nrblocks contiguous block is dependent on
2725 * number of contiguous block. So we will limit
2726 * number of contiguous block to a sane value
2728 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2729 (max_blocks > EXT4_MAX_TRANS_DATA))
2730 max_blocks = EXT4_MAX_TRANS_DATA;
2732 return ext4_chunk_trans_blocks(inode, max_blocks);
2735 static int ext4_da_writepages(struct address_space *mapping,
2736 struct writeback_control *wbc)
2739 int range_whole = 0;
2740 handle_t *handle = NULL;
2741 struct mpage_da_data mpd;
2742 struct inode *inode = mapping->host;
2743 int no_nrwrite_index_update;
2744 int pages_written = 0;
2746 int range_cyclic, cycled = 1, io_done = 0;
2747 int needed_blocks, ret = 0, nr_to_writebump = 0;
2748 loff_t range_start = wbc->range_start;
2749 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2751 trace_ext4_da_writepages(inode, wbc);
2754 * No pages to write? This is mainly a kludge to avoid starting
2755 * a transaction for special inodes like journal inode on last iput()
2756 * because that could violate lock ordering on umount
2758 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2762 * If the filesystem has aborted, it is read-only, so return
2763 * right away instead of dumping stack traces later on that
2764 * will obscure the real source of the problem. We test
2765 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2766 * the latter could be true if the filesystem is mounted
2767 * read-only, and in that case, ext4_da_writepages should
2768 * *never* be called, so if that ever happens, we would want
2771 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2775 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2776 * This make sure small files blocks are allocated in
2777 * single attempt. This ensure that small files
2778 * get less fragmented.
2780 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2781 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2782 wbc->nr_to_write = sbi->s_mb_stream_request;
2784 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2787 range_cyclic = wbc->range_cyclic;
2788 if (wbc->range_cyclic) {
2789 index = mapping->writeback_index;
2792 wbc->range_start = index << PAGE_CACHE_SHIFT;
2793 wbc->range_end = LLONG_MAX;
2794 wbc->range_cyclic = 0;
2796 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2799 mpd.inode = mapping->host;
2802 * we don't want write_cache_pages to update
2803 * nr_to_write and writeback_index
2805 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2806 wbc->no_nrwrite_index_update = 1;
2807 pages_skipped = wbc->pages_skipped;
2810 while (!ret && wbc->nr_to_write > 0) {
2813 * we insert one extent at a time. So we need
2814 * credit needed for single extent allocation.
2815 * journalled mode is currently not supported
2818 BUG_ON(ext4_should_journal_data(inode));
2819 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2821 /* start a new transaction*/
2822 handle = ext4_journal_start(inode, needed_blocks);
2823 if (IS_ERR(handle)) {
2824 ret = PTR_ERR(handle);
2825 printk(KERN_CRIT "%s: jbd2_start: "
2826 "%ld pages, ino %lu; err %d\n", __func__,
2827 wbc->nr_to_write, inode->i_ino, ret);
2829 goto out_writepages;
2833 * Now call __mpage_da_writepage to find the next
2834 * contiguous region of logical blocks that need
2835 * blocks to be allocated by ext4. We don't actually
2836 * submit the blocks for I/O here, even though
2837 * write_cache_pages thinks it will, and will set the
2838 * pages as clean for write before calling
2839 * __mpage_da_writepage().
2847 mpd.pages_written = 0;
2849 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2852 * If we have a contigous extent of pages and we
2853 * haven't done the I/O yet, map the blocks and submit
2856 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2857 if (mpage_da_map_blocks(&mpd) == 0)
2858 mpage_da_submit_io(&mpd);
2860 ret = MPAGE_DA_EXTENT_TAIL;
2862 trace_ext4_da_write_pages(inode, &mpd);
2863 wbc->nr_to_write -= mpd.pages_written;
2865 ext4_journal_stop(handle);
2867 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2868 /* commit the transaction which would
2869 * free blocks released in the transaction
2872 jbd2_journal_force_commit_nested(sbi->s_journal);
2873 wbc->pages_skipped = pages_skipped;
2875 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2877 * got one extent now try with
2880 pages_written += mpd.pages_written;
2881 wbc->pages_skipped = pages_skipped;
2884 } else if (wbc->nr_to_write)
2886 * There is no more writeout needed
2887 * or we requested for a noblocking writeout
2888 * and we found the device congested
2892 if (!io_done && !cycled) {
2895 wbc->range_start = index << PAGE_CACHE_SHIFT;
2896 wbc->range_end = mapping->writeback_index - 1;
2899 if (pages_skipped != wbc->pages_skipped)
2900 printk(KERN_EMERG "This should not happen leaving %s "
2901 "with nr_to_write = %ld ret = %d\n",
2902 __func__, wbc->nr_to_write, ret);
2905 index += pages_written;
2906 wbc->range_cyclic = range_cyclic;
2907 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2909 * set the writeback_index so that range_cyclic
2910 * mode will write it back later
2912 mapping->writeback_index = index;
2915 if (!no_nrwrite_index_update)
2916 wbc->no_nrwrite_index_update = 0;
2917 wbc->nr_to_write -= nr_to_writebump;
2918 wbc->range_start = range_start;
2919 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
2923 #define FALL_BACK_TO_NONDELALLOC 1
2924 static int ext4_nonda_switch(struct super_block *sb)
2926 s64 free_blocks, dirty_blocks;
2927 struct ext4_sb_info *sbi = EXT4_SB(sb);
2930 * switch to non delalloc mode if we are running low
2931 * on free block. The free block accounting via percpu
2932 * counters can get slightly wrong with percpu_counter_batch getting
2933 * accumulated on each CPU without updating global counters
2934 * Delalloc need an accurate free block accounting. So switch
2935 * to non delalloc when we are near to error range.
2937 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2938 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2939 if (2 * free_blocks < 3 * dirty_blocks ||
2940 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2942 * free block count is less that 150% of dirty blocks
2943 * or free blocks is less that watermark
2950 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2951 loff_t pos, unsigned len, unsigned flags,
2952 struct page **pagep, void **fsdata)
2954 int ret, retries = 0;
2958 struct inode *inode = mapping->host;
2961 index = pos >> PAGE_CACHE_SHIFT;
2962 from = pos & (PAGE_CACHE_SIZE - 1);
2965 if (ext4_nonda_switch(inode->i_sb)) {
2966 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2967 return ext4_write_begin(file, mapping, pos,
2968 len, flags, pagep, fsdata);
2970 *fsdata = (void *)0;
2971 trace_ext4_da_write_begin(inode, pos, len, flags);
2974 * With delayed allocation, we don't log the i_disksize update
2975 * if there is delayed block allocation. But we still need
2976 * to journalling the i_disksize update if writes to the end
2977 * of file which has an already mapped buffer.
2979 handle = ext4_journal_start(inode, 1);
2980 if (IS_ERR(handle)) {
2981 ret = PTR_ERR(handle);
2984 /* We cannot recurse into the filesystem as the transaction is already
2986 flags |= AOP_FLAG_NOFS;
2988 page = grab_cache_page_write_begin(mapping, index, flags);
2990 ext4_journal_stop(handle);
2996 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2997 ext4_da_get_block_prep);
3000 ext4_journal_stop(handle);
3001 page_cache_release(page);
3003 * block_write_begin may have instantiated a few blocks
3004 * outside i_size. Trim these off again. Don't need
3005 * i_size_read because we hold i_mutex.
3007 if (pos + len > inode->i_size)
3008 ext4_truncate(inode);
3011 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3018 * Check if we should update i_disksize
3019 * when write to the end of file but not require block allocation
3021 static int ext4_da_should_update_i_disksize(struct page *page,
3022 unsigned long offset)
3024 struct buffer_head *bh;
3025 struct inode *inode = page->mapping->host;
3029 bh = page_buffers(page);
3030 idx = offset >> inode->i_blkbits;
3032 for (i = 0; i < idx; i++)
3033 bh = bh->b_this_page;
3035 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3040 static int ext4_da_write_end(struct file *file,
3041 struct address_space *mapping,
3042 loff_t pos, unsigned len, unsigned copied,
3043 struct page *page, void *fsdata)
3045 struct inode *inode = mapping->host;
3047 handle_t *handle = ext4_journal_current_handle();
3049 unsigned long start, end;
3050 int write_mode = (int)(unsigned long)fsdata;
3052 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3053 if (ext4_should_order_data(inode)) {
3054 return ext4_ordered_write_end(file, mapping, pos,
3055 len, copied, page, fsdata);
3056 } else if (ext4_should_writeback_data(inode)) {
3057 return ext4_writeback_write_end(file, mapping, pos,
3058 len, copied, page, fsdata);
3064 trace_ext4_da_write_end(inode, pos, len, copied);
3065 start = pos & (PAGE_CACHE_SIZE - 1);
3066 end = start + copied - 1;
3069 * generic_write_end() will run mark_inode_dirty() if i_size
3070 * changes. So let's piggyback the i_disksize mark_inode_dirty
3074 new_i_size = pos + copied;
3075 if (new_i_size > EXT4_I(inode)->i_disksize) {
3076 if (ext4_da_should_update_i_disksize(page, end)) {
3077 down_write(&EXT4_I(inode)->i_data_sem);
3078 if (new_i_size > EXT4_I(inode)->i_disksize) {
3080 * Updating i_disksize when extending file
3081 * without needing block allocation
3083 if (ext4_should_order_data(inode))
3084 ret = ext4_jbd2_file_inode(handle,
3087 EXT4_I(inode)->i_disksize = new_i_size;
3089 up_write(&EXT4_I(inode)->i_data_sem);
3090 /* We need to mark inode dirty even if
3091 * new_i_size is less that inode->i_size
3092 * bu greater than i_disksize.(hint delalloc)
3094 ext4_mark_inode_dirty(handle, inode);
3097 ret2 = generic_write_end(file, mapping, pos, len, copied,
3102 ret2 = ext4_journal_stop(handle);
3106 return ret ? ret : copied;
3109 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3112 * Drop reserved blocks
3114 BUG_ON(!PageLocked(page));
3115 if (!page_has_buffers(page))
3118 ext4_da_page_release_reservation(page, offset);
3121 ext4_invalidatepage(page, offset);
3127 * Force all delayed allocation blocks to be allocated for a given inode.
3129 int ext4_alloc_da_blocks(struct inode *inode)
3131 trace_ext4_alloc_da_blocks(inode);
3133 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3134 !EXT4_I(inode)->i_reserved_meta_blocks)
3138 * We do something simple for now. The filemap_flush() will
3139 * also start triggering a write of the data blocks, which is
3140 * not strictly speaking necessary (and for users of
3141 * laptop_mode, not even desirable). However, to do otherwise
3142 * would require replicating code paths in:
3144 * ext4_da_writepages() ->
3145 * write_cache_pages() ---> (via passed in callback function)
3146 * __mpage_da_writepage() -->
3147 * mpage_add_bh_to_extent()
3148 * mpage_da_map_blocks()
3150 * The problem is that write_cache_pages(), located in
3151 * mm/page-writeback.c, marks pages clean in preparation for
3152 * doing I/O, which is not desirable if we're not planning on
3155 * We could call write_cache_pages(), and then redirty all of
3156 * the pages by calling redirty_page_for_writeback() but that
3157 * would be ugly in the extreme. So instead we would need to
3158 * replicate parts of the code in the above functions,
3159 * simplifying them becuase we wouldn't actually intend to
3160 * write out the pages, but rather only collect contiguous
3161 * logical block extents, call the multi-block allocator, and
3162 * then update the buffer heads with the block allocations.
3164 * For now, though, we'll cheat by calling filemap_flush(),
3165 * which will map the blocks, and start the I/O, but not
3166 * actually wait for the I/O to complete.
3168 return filemap_flush(inode->i_mapping);
3172 * bmap() is special. It gets used by applications such as lilo and by
3173 * the swapper to find the on-disk block of a specific piece of data.
3175 * Naturally, this is dangerous if the block concerned is still in the
3176 * journal. If somebody makes a swapfile on an ext4 data-journaling
3177 * filesystem and enables swap, then they may get a nasty shock when the
3178 * data getting swapped to that swapfile suddenly gets overwritten by
3179 * the original zero's written out previously to the journal and
3180 * awaiting writeback in the kernel's buffer cache.
3182 * So, if we see any bmap calls here on a modified, data-journaled file,
3183 * take extra steps to flush any blocks which might be in the cache.
3185 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3187 struct inode *inode = mapping->host;
3191 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3192 test_opt(inode->i_sb, DELALLOC)) {
3194 * With delalloc we want to sync the file
3195 * so that we can make sure we allocate
3198 filemap_write_and_wait(mapping);
3201 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3203 * This is a REALLY heavyweight approach, but the use of
3204 * bmap on dirty files is expected to be extremely rare:
3205 * only if we run lilo or swapon on a freshly made file
3206 * do we expect this to happen.
3208 * (bmap requires CAP_SYS_RAWIO so this does not
3209 * represent an unprivileged user DOS attack --- we'd be
3210 * in trouble if mortal users could trigger this path at
3213 * NB. EXT4_STATE_JDATA is not set on files other than
3214 * regular files. If somebody wants to bmap a directory
3215 * or symlink and gets confused because the buffer
3216 * hasn't yet been flushed to disk, they deserve
3217 * everything they get.
3220 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3221 journal = EXT4_JOURNAL(inode);
3222 jbd2_journal_lock_updates(journal);
3223 err = jbd2_journal_flush(journal);
3224 jbd2_journal_unlock_updates(journal);
3230 return generic_block_bmap(mapping, block, ext4_get_block);
3233 static int ext4_readpage(struct file *file, struct page *page)
3235 return mpage_readpage(page, ext4_get_block);
3239 ext4_readpages(struct file *file, struct address_space *mapping,
3240 struct list_head *pages, unsigned nr_pages)
3242 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3245 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3247 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3250 * If it's a full truncate we just forget about the pending dirtying
3253 ClearPageChecked(page);
3256 jbd2_journal_invalidatepage(journal, page, offset);
3258 block_invalidatepage(page, offset);
3261 static int ext4_releasepage(struct page *page, gfp_t wait)
3263 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3265 WARN_ON(PageChecked(page));
3266 if (!page_has_buffers(page))
3269 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3271 return try_to_free_buffers(page);
3275 * If the O_DIRECT write will extend the file then add this inode to the
3276 * orphan list. So recovery will truncate it back to the original size
3277 * if the machine crashes during the write.
3279 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3280 * crashes then stale disk data _may_ be exposed inside the file. But current
3281 * VFS code falls back into buffered path in that case so we are safe.
3283 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3284 const struct iovec *iov, loff_t offset,
3285 unsigned long nr_segs)
3287 struct file *file = iocb->ki_filp;
3288 struct inode *inode = file->f_mapping->host;
3289 struct ext4_inode_info *ei = EXT4_I(inode);
3293 size_t count = iov_length(iov, nr_segs);
3296 loff_t final_size = offset + count;
3298 if (final_size > inode->i_size) {
3299 /* Credits for sb + inode write */
3300 handle = ext4_journal_start(inode, 2);
3301 if (IS_ERR(handle)) {
3302 ret = PTR_ERR(handle);
3305 ret = ext4_orphan_add(handle, inode);
3307 ext4_journal_stop(handle);
3311 ei->i_disksize = inode->i_size;
3312 ext4_journal_stop(handle);
3316 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3318 ext4_get_block, NULL);
3323 /* Credits for sb + inode write */
3324 handle = ext4_journal_start(inode, 2);
3325 if (IS_ERR(handle)) {
3326 /* This is really bad luck. We've written the data
3327 * but cannot extend i_size. Bail out and pretend
3328 * the write failed... */
3329 ret = PTR_ERR(handle);
3333 ext4_orphan_del(handle, inode);
3335 loff_t end = offset + ret;
3336 if (end > inode->i_size) {
3337 ei->i_disksize = end;
3338 i_size_write(inode, end);
3340 * We're going to return a positive `ret'
3341 * here due to non-zero-length I/O, so there's
3342 * no way of reporting error returns from
3343 * ext4_mark_inode_dirty() to userspace. So
3346 ext4_mark_inode_dirty(handle, inode);
3349 err = ext4_journal_stop(handle);
3358 * Pages can be marked dirty completely asynchronously from ext4's journalling
3359 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3360 * much here because ->set_page_dirty is called under VFS locks. The page is
3361 * not necessarily locked.
3363 * We cannot just dirty the page and leave attached buffers clean, because the
3364 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3365 * or jbddirty because all the journalling code will explode.
3367 * So what we do is to mark the page "pending dirty" and next time writepage
3368 * is called, propagate that into the buffers appropriately.
3370 static int ext4_journalled_set_page_dirty(struct page *page)
3372 SetPageChecked(page);
3373 return __set_page_dirty_nobuffers(page);
3376 static const struct address_space_operations ext4_ordered_aops = {
3377 .readpage = ext4_readpage,
3378 .readpages = ext4_readpages,
3379 .writepage = ext4_writepage,
3380 .sync_page = block_sync_page,
3381 .write_begin = ext4_write_begin,
3382 .write_end = ext4_ordered_write_end,
3384 .invalidatepage = ext4_invalidatepage,
3385 .releasepage = ext4_releasepage,
3386 .direct_IO = ext4_direct_IO,
3387 .migratepage = buffer_migrate_page,
3388 .is_partially_uptodate = block_is_partially_uptodate,
3389 .error_remove_page = generic_error_remove_page,
3392 static const struct address_space_operations ext4_writeback_aops = {
3393 .readpage = ext4_readpage,
3394 .readpages = ext4_readpages,
3395 .writepage = ext4_writepage,
3396 .sync_page = block_sync_page,
3397 .write_begin = ext4_write_begin,
3398 .write_end = ext4_writeback_write_end,
3400 .invalidatepage = ext4_invalidatepage,
3401 .releasepage = ext4_releasepage,
3402 .direct_IO = ext4_direct_IO,
3403 .migratepage = buffer_migrate_page,
3404 .is_partially_uptodate = block_is_partially_uptodate,
3405 .error_remove_page = generic_error_remove_page,
3408 static const struct address_space_operations ext4_journalled_aops = {
3409 .readpage = ext4_readpage,
3410 .readpages = ext4_readpages,
3411 .writepage = ext4_writepage,
3412 .sync_page = block_sync_page,
3413 .write_begin = ext4_write_begin,
3414 .write_end = ext4_journalled_write_end,
3415 .set_page_dirty = ext4_journalled_set_page_dirty,
3417 .invalidatepage = ext4_invalidatepage,
3418 .releasepage = ext4_releasepage,
3419 .is_partially_uptodate = block_is_partially_uptodate,
3420 .error_remove_page = generic_error_remove_page,
3423 static const struct address_space_operations ext4_da_aops = {
3424 .readpage = ext4_readpage,
3425 .readpages = ext4_readpages,
3426 .writepage = ext4_writepage,
3427 .writepages = ext4_da_writepages,
3428 .sync_page = block_sync_page,
3429 .write_begin = ext4_da_write_begin,
3430 .write_end = ext4_da_write_end,
3432 .invalidatepage = ext4_da_invalidatepage,
3433 .releasepage = ext4_releasepage,
3434 .direct_IO = ext4_direct_IO,
3435 .migratepage = buffer_migrate_page,
3436 .is_partially_uptodate = block_is_partially_uptodate,
3437 .error_remove_page = generic_error_remove_page,
3440 void ext4_set_aops(struct inode *inode)
3442 if (ext4_should_order_data(inode) &&
3443 test_opt(inode->i_sb, DELALLOC))
3444 inode->i_mapping->a_ops = &ext4_da_aops;
3445 else if (ext4_should_order_data(inode))
3446 inode->i_mapping->a_ops = &ext4_ordered_aops;
3447 else if (ext4_should_writeback_data(inode) &&
3448 test_opt(inode->i_sb, DELALLOC))
3449 inode->i_mapping->a_ops = &ext4_da_aops;
3450 else if (ext4_should_writeback_data(inode))
3451 inode->i_mapping->a_ops = &ext4_writeback_aops;
3453 inode->i_mapping->a_ops = &ext4_journalled_aops;
3457 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3458 * up to the end of the block which corresponds to `from'.
3459 * This required during truncate. We need to physically zero the tail end
3460 * of that block so it doesn't yield old data if the file is later grown.
3462 int ext4_block_truncate_page(handle_t *handle,
3463 struct address_space *mapping, loff_t from)
3465 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3466 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3467 unsigned blocksize, length, pos;
3469 struct inode *inode = mapping->host;
3470 struct buffer_head *bh;
3474 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3475 mapping_gfp_mask(mapping) & ~__GFP_FS);
3479 blocksize = inode->i_sb->s_blocksize;
3480 length = blocksize - (offset & (blocksize - 1));
3481 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3484 * For "nobh" option, we can only work if we don't need to
3485 * read-in the page - otherwise we create buffers to do the IO.
3487 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3488 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3489 zero_user(page, offset, length);
3490 set_page_dirty(page);
3494 if (!page_has_buffers(page))
3495 create_empty_buffers(page, blocksize, 0);
3497 /* Find the buffer that contains "offset" */
3498 bh = page_buffers(page);
3500 while (offset >= pos) {
3501 bh = bh->b_this_page;
3507 if (buffer_freed(bh)) {
3508 BUFFER_TRACE(bh, "freed: skip");
3512 if (!buffer_mapped(bh)) {
3513 BUFFER_TRACE(bh, "unmapped");
3514 ext4_get_block(inode, iblock, bh, 0);
3515 /* unmapped? It's a hole - nothing to do */
3516 if (!buffer_mapped(bh)) {
3517 BUFFER_TRACE(bh, "still unmapped");
3522 /* Ok, it's mapped. Make sure it's up-to-date */
3523 if (PageUptodate(page))
3524 set_buffer_uptodate(bh);
3526 if (!buffer_uptodate(bh)) {
3528 ll_rw_block(READ, 1, &bh);
3530 /* Uhhuh. Read error. Complain and punt. */
3531 if (!buffer_uptodate(bh))
3535 if (ext4_should_journal_data(inode)) {
3536 BUFFER_TRACE(bh, "get write access");
3537 err = ext4_journal_get_write_access(handle, bh);
3542 zero_user(page, offset, length);
3544 BUFFER_TRACE(bh, "zeroed end of block");
3547 if (ext4_should_journal_data(inode)) {
3548 err = ext4_handle_dirty_metadata(handle, inode, bh);
3550 if (ext4_should_order_data(inode))
3551 err = ext4_jbd2_file_inode(handle, inode);
3552 mark_buffer_dirty(bh);
3557 page_cache_release(page);
3562 * Probably it should be a library function... search for first non-zero word
3563 * or memcmp with zero_page, whatever is better for particular architecture.
3566 static inline int all_zeroes(__le32 *p, __le32 *q)
3575 * ext4_find_shared - find the indirect blocks for partial truncation.
3576 * @inode: inode in question
3577 * @depth: depth of the affected branch
3578 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3579 * @chain: place to store the pointers to partial indirect blocks
3580 * @top: place to the (detached) top of branch
3582 * This is a helper function used by ext4_truncate().
3584 * When we do truncate() we may have to clean the ends of several
3585 * indirect blocks but leave the blocks themselves alive. Block is
3586 * partially truncated if some data below the new i_size is refered
3587 * from it (and it is on the path to the first completely truncated
3588 * data block, indeed). We have to free the top of that path along
3589 * with everything to the right of the path. Since no allocation
3590 * past the truncation point is possible until ext4_truncate()
3591 * finishes, we may safely do the latter, but top of branch may
3592 * require special attention - pageout below the truncation point
3593 * might try to populate it.
3595 * We atomically detach the top of branch from the tree, store the
3596 * block number of its root in *@top, pointers to buffer_heads of
3597 * partially truncated blocks - in @chain[].bh and pointers to
3598 * their last elements that should not be removed - in
3599 * @chain[].p. Return value is the pointer to last filled element
3602 * The work left to caller to do the actual freeing of subtrees:
3603 * a) free the subtree starting from *@top
3604 * b) free the subtrees whose roots are stored in
3605 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3606 * c) free the subtrees growing from the inode past the @chain[0].
3607 * (no partially truncated stuff there). */
3609 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3610 ext4_lblk_t offsets[4], Indirect chain[4],
3613 Indirect *partial, *p;
3617 /* Make k index the deepest non-null offest + 1 */
3618 for (k = depth; k > 1 && !offsets[k-1]; k--)
3620 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3621 /* Writer: pointers */
3623 partial = chain + k-1;
3625 * If the branch acquired continuation since we've looked at it -
3626 * fine, it should all survive and (new) top doesn't belong to us.
3628 if (!partial->key && *partial->p)
3631 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3634 * OK, we've found the last block that must survive. The rest of our
3635 * branch should be detached before unlocking. However, if that rest
3636 * of branch is all ours and does not grow immediately from the inode
3637 * it's easier to cheat and just decrement partial->p.
3639 if (p == chain + k - 1 && p > chain) {
3643 /* Nope, don't do this in ext4. Must leave the tree intact */
3650 while (partial > p) {
3651 brelse(partial->bh);
3659 * Zero a number of block pointers in either an inode or an indirect block.
3660 * If we restart the transaction we must again get write access to the
3661 * indirect block for further modification.
3663 * We release `count' blocks on disk, but (last - first) may be greater
3664 * than `count' because there can be holes in there.
3666 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3667 struct buffer_head *bh,
3668 ext4_fsblk_t block_to_free,
3669 unsigned long count, __le32 *first,
3673 if (try_to_extend_transaction(handle, inode)) {
3675 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3676 ext4_handle_dirty_metadata(handle, inode, bh);
3678 ext4_mark_inode_dirty(handle, inode);
3679 ext4_truncate_restart_trans(handle, inode,
3680 blocks_for_truncate(inode));
3682 BUFFER_TRACE(bh, "retaking write access");
3683 ext4_journal_get_write_access(handle, bh);
3688 * Any buffers which are on the journal will be in memory. We
3689 * find them on the hash table so jbd2_journal_revoke() will
3690 * run jbd2_journal_forget() on them. We've already detached
3691 * each block from the file, so bforget() in
3692 * jbd2_journal_forget() should be safe.
3694 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3696 for (p = first; p < last; p++) {
3697 u32 nr = le32_to_cpu(*p);
3699 struct buffer_head *tbh;
3702 tbh = sb_find_get_block(inode->i_sb, nr);
3703 ext4_forget(handle, 0, inode, tbh, nr);
3707 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3711 * ext4_free_data - free a list of data blocks
3712 * @handle: handle for this transaction
3713 * @inode: inode we are dealing with
3714 * @this_bh: indirect buffer_head which contains *@first and *@last
3715 * @first: array of block numbers
3716 * @last: points immediately past the end of array
3718 * We are freeing all blocks refered from that array (numbers are stored as
3719 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3721 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3722 * blocks are contiguous then releasing them at one time will only affect one
3723 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3724 * actually use a lot of journal space.
3726 * @this_bh will be %NULL if @first and @last point into the inode's direct
3729 static void ext4_free_data(handle_t *handle, struct inode *inode,
3730 struct buffer_head *this_bh,
3731 __le32 *first, __le32 *last)
3733 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3734 unsigned long count = 0; /* Number of blocks in the run */
3735 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3738 ext4_fsblk_t nr; /* Current block # */
3739 __le32 *p; /* Pointer into inode/ind
3740 for current block */
3743 if (this_bh) { /* For indirect block */
3744 BUFFER_TRACE(this_bh, "get_write_access");
3745 err = ext4_journal_get_write_access(handle, this_bh);
3746 /* Important: if we can't update the indirect pointers
3747 * to the blocks, we can't free them. */
3752 for (p = first; p < last; p++) {
3753 nr = le32_to_cpu(*p);
3755 /* accumulate blocks to free if they're contiguous */
3758 block_to_free_p = p;
3760 } else if (nr == block_to_free + count) {
3763 ext4_clear_blocks(handle, inode, this_bh,
3765 count, block_to_free_p, p);
3767 block_to_free_p = p;
3774 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3775 count, block_to_free_p, p);
3778 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3781 * The buffer head should have an attached journal head at this
3782 * point. However, if the data is corrupted and an indirect
3783 * block pointed to itself, it would have been detached when
3784 * the block was cleared. Check for this instead of OOPSing.
3786 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3787 ext4_handle_dirty_metadata(handle, inode, this_bh);
3789 ext4_error(inode->i_sb, __func__,
3790 "circular indirect block detected, "
3791 "inode=%lu, block=%llu",
3793 (unsigned long long) this_bh->b_blocknr);
3798 * ext4_free_branches - free an array of branches
3799 * @handle: JBD handle for this transaction
3800 * @inode: inode we are dealing with
3801 * @parent_bh: the buffer_head which contains *@first and *@last
3802 * @first: array of block numbers
3803 * @last: pointer immediately past the end of array
3804 * @depth: depth of the branches to free
3806 * We are freeing all blocks refered from these branches (numbers are
3807 * stored as little-endian 32-bit) and updating @inode->i_blocks
3810 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3811 struct buffer_head *parent_bh,
3812 __le32 *first, __le32 *last, int depth)
3817 if (ext4_handle_is_aborted(handle))
3821 struct buffer_head *bh;
3822 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3824 while (--p >= first) {
3825 nr = le32_to_cpu(*p);
3827 continue; /* A hole */
3829 /* Go read the buffer for the next level down */
3830 bh = sb_bread(inode->i_sb, nr);
3833 * A read failure? Report error and clear slot
3837 ext4_error(inode->i_sb, "ext4_free_branches",
3838 "Read failure, inode=%lu, block=%llu",
3843 /* This zaps the entire block. Bottom up. */
3844 BUFFER_TRACE(bh, "free child branches");
3845 ext4_free_branches(handle, inode, bh,
3846 (__le32 *) bh->b_data,
3847 (__le32 *) bh->b_data + addr_per_block,
3851 * We've probably journalled the indirect block several
3852 * times during the truncate. But it's no longer
3853 * needed and we now drop it from the transaction via
3854 * jbd2_journal_revoke().
3856 * That's easy if it's exclusively part of this
3857 * transaction. But if it's part of the committing
3858 * transaction then jbd2_journal_forget() will simply
3859 * brelse() it. That means that if the underlying
3860 * block is reallocated in ext4_get_block(),
3861 * unmap_underlying_metadata() will find this block
3862 * and will try to get rid of it. damn, damn.
3864 * If this block has already been committed to the
3865 * journal, a revoke record will be written. And
3866 * revoke records must be emitted *before* clearing
3867 * this block's bit in the bitmaps.
3869 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3872 * Everything below this this pointer has been
3873 * released. Now let this top-of-subtree go.
3875 * We want the freeing of this indirect block to be
3876 * atomic in the journal with the updating of the
3877 * bitmap block which owns it. So make some room in
3880 * We zero the parent pointer *after* freeing its
3881 * pointee in the bitmaps, so if extend_transaction()
3882 * for some reason fails to put the bitmap changes and
3883 * the release into the same transaction, recovery
3884 * will merely complain about releasing a free block,
3885 * rather than leaking blocks.
3887 if (ext4_handle_is_aborted(handle))
3889 if (try_to_extend_transaction(handle, inode)) {
3890 ext4_mark_inode_dirty(handle, inode);
3891 ext4_truncate_restart_trans(handle, inode,
3892 blocks_for_truncate(inode));
3895 ext4_free_blocks(handle, inode, nr, 1, 1);
3899 * The block which we have just freed is
3900 * pointed to by an indirect block: journal it
3902 BUFFER_TRACE(parent_bh, "get_write_access");
3903 if (!ext4_journal_get_write_access(handle,
3906 BUFFER_TRACE(parent_bh,
3907 "call ext4_handle_dirty_metadata");
3908 ext4_handle_dirty_metadata(handle,
3915 /* We have reached the bottom of the tree. */
3916 BUFFER_TRACE(parent_bh, "free data blocks");
3917 ext4_free_data(handle, inode, parent_bh, first, last);
3921 int ext4_can_truncate(struct inode *inode)
3923 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3925 if (S_ISREG(inode->i_mode))
3927 if (S_ISDIR(inode->i_mode))
3929 if (S_ISLNK(inode->i_mode))
3930 return !ext4_inode_is_fast_symlink(inode);
3937 * We block out ext4_get_block() block instantiations across the entire
3938 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3939 * simultaneously on behalf of the same inode.
3941 * As we work through the truncate and commmit bits of it to the journal there
3942 * is one core, guiding principle: the file's tree must always be consistent on
3943 * disk. We must be able to restart the truncate after a crash.
3945 * The file's tree may be transiently inconsistent in memory (although it
3946 * probably isn't), but whenever we close off and commit a journal transaction,
3947 * the contents of (the filesystem + the journal) must be consistent and
3948 * restartable. It's pretty simple, really: bottom up, right to left (although
3949 * left-to-right works OK too).
3951 * Note that at recovery time, journal replay occurs *before* the restart of
3952 * truncate against the orphan inode list.
3954 * The committed inode has the new, desired i_size (which is the same as
3955 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3956 * that this inode's truncate did not complete and it will again call
3957 * ext4_truncate() to have another go. So there will be instantiated blocks
3958 * to the right of the truncation point in a crashed ext4 filesystem. But
3959 * that's fine - as long as they are linked from the inode, the post-crash
3960 * ext4_truncate() run will find them and release them.
3962 void ext4_truncate(struct inode *inode)
3965 struct ext4_inode_info *ei = EXT4_I(inode);
3966 __le32 *i_data = ei->i_data;
3967 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3968 struct address_space *mapping = inode->i_mapping;
3969 ext4_lblk_t offsets[4];
3974 ext4_lblk_t last_block;
3975 unsigned blocksize = inode->i_sb->s_blocksize;
3977 if (!ext4_can_truncate(inode))
3980 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3981 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3983 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3984 ext4_ext_truncate(inode);
3988 handle = start_transaction(inode);
3990 return; /* AKPM: return what? */
3992 last_block = (inode->i_size + blocksize-1)
3993 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3995 if (inode->i_size & (blocksize - 1))
3996 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3999 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4001 goto out_stop; /* error */
4004 * OK. This truncate is going to happen. We add the inode to the
4005 * orphan list, so that if this truncate spans multiple transactions,
4006 * and we crash, we will resume the truncate when the filesystem
4007 * recovers. It also marks the inode dirty, to catch the new size.
4009 * Implication: the file must always be in a sane, consistent
4010 * truncatable state while each transaction commits.
4012 if (ext4_orphan_add(handle, inode))
4016 * From here we block out all ext4_get_block() callers who want to
4017 * modify the block allocation tree.
4019 down_write(&ei->i_data_sem);
4021 ext4_discard_preallocations(inode);
4024 * The orphan list entry will now protect us from any crash which
4025 * occurs before the truncate completes, so it is now safe to propagate
4026 * the new, shorter inode size (held for now in i_size) into the
4027 * on-disk inode. We do this via i_disksize, which is the value which
4028 * ext4 *really* writes onto the disk inode.
4030 ei->i_disksize = inode->i_size;
4032 if (n == 1) { /* direct blocks */
4033 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4034 i_data + EXT4_NDIR_BLOCKS);
4038 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4039 /* Kill the top of shared branch (not detached) */
4041 if (partial == chain) {
4042 /* Shared branch grows from the inode */
4043 ext4_free_branches(handle, inode, NULL,
4044 &nr, &nr+1, (chain+n-1) - partial);
4047 * We mark the inode dirty prior to restart,
4048 * and prior to stop. No need for it here.
4051 /* Shared branch grows from an indirect block */
4052 BUFFER_TRACE(partial->bh, "get_write_access");
4053 ext4_free_branches(handle, inode, partial->bh,
4055 partial->p+1, (chain+n-1) - partial);
4058 /* Clear the ends of indirect blocks on the shared branch */
4059 while (partial > chain) {
4060 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4061 (__le32*)partial->bh->b_data+addr_per_block,
4062 (chain+n-1) - partial);
4063 BUFFER_TRACE(partial->bh, "call brelse");
4064 brelse(partial->bh);
4068 /* Kill the remaining (whole) subtrees */
4069 switch (offsets[0]) {
4071 nr = i_data[EXT4_IND_BLOCK];
4073 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4074 i_data[EXT4_IND_BLOCK] = 0;
4076 case EXT4_IND_BLOCK:
4077 nr = i_data[EXT4_DIND_BLOCK];
4079 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4080 i_data[EXT4_DIND_BLOCK] = 0;
4082 case EXT4_DIND_BLOCK:
4083 nr = i_data[EXT4_TIND_BLOCK];
4085 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4086 i_data[EXT4_TIND_BLOCK] = 0;
4088 case EXT4_TIND_BLOCK:
4092 up_write(&ei->i_data_sem);
4093 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4094 ext4_mark_inode_dirty(handle, inode);
4097 * In a multi-transaction truncate, we only make the final transaction
4101 ext4_handle_sync(handle);
4104 * If this was a simple ftruncate(), and the file will remain alive
4105 * then we need to clear up the orphan record which we created above.
4106 * However, if this was a real unlink then we were called by
4107 * ext4_delete_inode(), and we allow that function to clean up the
4108 * orphan info for us.
4111 ext4_orphan_del(handle, inode);
4113 ext4_journal_stop(handle);
4117 * ext4_get_inode_loc returns with an extra refcount against the inode's
4118 * underlying buffer_head on success. If 'in_mem' is true, we have all
4119 * data in memory that is needed to recreate the on-disk version of this
4122 static int __ext4_get_inode_loc(struct inode *inode,
4123 struct ext4_iloc *iloc, int in_mem)
4125 struct ext4_group_desc *gdp;
4126 struct buffer_head *bh;
4127 struct super_block *sb = inode->i_sb;
4129 int inodes_per_block, inode_offset;
4132 if (!ext4_valid_inum(sb, inode->i_ino))
4135 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4136 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4141 * Figure out the offset within the block group inode table
4143 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4144 inode_offset = ((inode->i_ino - 1) %
4145 EXT4_INODES_PER_GROUP(sb));
4146 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4147 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4149 bh = sb_getblk(sb, block);
4151 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4152 "inode block - inode=%lu, block=%llu",
4153 inode->i_ino, block);
4156 if (!buffer_uptodate(bh)) {
4160 * If the buffer has the write error flag, we have failed
4161 * to write out another inode in the same block. In this
4162 * case, we don't have to read the block because we may
4163 * read the old inode data successfully.
4165 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4166 set_buffer_uptodate(bh);
4168 if (buffer_uptodate(bh)) {
4169 /* someone brought it uptodate while we waited */
4175 * If we have all information of the inode in memory and this
4176 * is the only valid inode in the block, we need not read the
4180 struct buffer_head *bitmap_bh;
4183 start = inode_offset & ~(inodes_per_block - 1);
4185 /* Is the inode bitmap in cache? */
4186 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4191 * If the inode bitmap isn't in cache then the
4192 * optimisation may end up performing two reads instead
4193 * of one, so skip it.
4195 if (!buffer_uptodate(bitmap_bh)) {
4199 for (i = start; i < start + inodes_per_block; i++) {
4200 if (i == inode_offset)
4202 if (ext4_test_bit(i, bitmap_bh->b_data))
4206 if (i == start + inodes_per_block) {
4207 /* all other inodes are free, so skip I/O */
4208 memset(bh->b_data, 0, bh->b_size);
4209 set_buffer_uptodate(bh);
4217 * If we need to do any I/O, try to pre-readahead extra
4218 * blocks from the inode table.
4220 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4221 ext4_fsblk_t b, end, table;
4224 table = ext4_inode_table(sb, gdp);
4225 /* s_inode_readahead_blks is always a power of 2 */
4226 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4229 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4230 num = EXT4_INODES_PER_GROUP(sb);
4231 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4232 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4233 num -= ext4_itable_unused_count(sb, gdp);
4234 table += num / inodes_per_block;
4238 sb_breadahead(sb, b++);
4242 * There are other valid inodes in the buffer, this inode
4243 * has in-inode xattrs, or we don't have this inode in memory.
4244 * Read the block from disk.
4247 bh->b_end_io = end_buffer_read_sync;
4248 submit_bh(READ_META, bh);
4250 if (!buffer_uptodate(bh)) {
4251 ext4_error(sb, __func__,
4252 "unable to read inode block - inode=%lu, "
4253 "block=%llu", inode->i_ino, block);
4263 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4265 /* We have all inode data except xattrs in memory here. */
4266 return __ext4_get_inode_loc(inode, iloc,
4267 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4270 void ext4_set_inode_flags(struct inode *inode)
4272 unsigned int flags = EXT4_I(inode)->i_flags;
4274 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4275 if (flags & EXT4_SYNC_FL)
4276 inode->i_flags |= S_SYNC;
4277 if (flags & EXT4_APPEND_FL)
4278 inode->i_flags |= S_APPEND;
4279 if (flags & EXT4_IMMUTABLE_FL)
4280 inode->i_flags |= S_IMMUTABLE;
4281 if (flags & EXT4_NOATIME_FL)
4282 inode->i_flags |= S_NOATIME;
4283 if (flags & EXT4_DIRSYNC_FL)
4284 inode->i_flags |= S_DIRSYNC;
4287 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4288 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4290 unsigned int flags = ei->vfs_inode.i_flags;
4292 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4293 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4295 ei->i_flags |= EXT4_SYNC_FL;
4296 if (flags & S_APPEND)
4297 ei->i_flags |= EXT4_APPEND_FL;
4298 if (flags & S_IMMUTABLE)
4299 ei->i_flags |= EXT4_IMMUTABLE_FL;
4300 if (flags & S_NOATIME)
4301 ei->i_flags |= EXT4_NOATIME_FL;
4302 if (flags & S_DIRSYNC)
4303 ei->i_flags |= EXT4_DIRSYNC_FL;
4306 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4307 struct ext4_inode_info *ei)
4310 struct inode *inode = &(ei->vfs_inode);
4311 struct super_block *sb = inode->i_sb;
4313 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4314 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4315 /* we are using combined 48 bit field */
4316 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4317 le32_to_cpu(raw_inode->i_blocks_lo);
4318 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4319 /* i_blocks represent file system block size */
4320 return i_blocks << (inode->i_blkbits - 9);
4325 return le32_to_cpu(raw_inode->i_blocks_lo);
4329 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4331 struct ext4_iloc iloc;
4332 struct ext4_inode *raw_inode;
4333 struct ext4_inode_info *ei;
4334 struct buffer_head *bh;
4335 struct inode *inode;
4339 inode = iget_locked(sb, ino);
4341 return ERR_PTR(-ENOMEM);
4342 if (!(inode->i_state & I_NEW))
4347 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4351 raw_inode = ext4_raw_inode(&iloc);
4352 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4353 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4354 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4355 if (!(test_opt(inode->i_sb, NO_UID32))) {
4356 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4357 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4359 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4362 ei->i_dir_start_lookup = 0;
4363 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4364 /* We now have enough fields to check if the inode was active or not.
4365 * This is needed because nfsd might try to access dead inodes
4366 * the test is that same one that e2fsck uses
4367 * NeilBrown 1999oct15
4369 if (inode->i_nlink == 0) {
4370 if (inode->i_mode == 0 ||
4371 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4372 /* this inode is deleted */
4377 /* The only unlinked inodes we let through here have
4378 * valid i_mode and are being read by the orphan
4379 * recovery code: that's fine, we're about to complete
4380 * the process of deleting those. */
4382 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4383 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4384 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4385 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4387 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4388 inode->i_size = ext4_isize(raw_inode);
4389 ei->i_disksize = inode->i_size;
4390 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4391 ei->i_block_group = iloc.block_group;
4392 ei->i_last_alloc_group = ~0;
4394 * NOTE! The in-memory inode i_data array is in little-endian order
4395 * even on big-endian machines: we do NOT byteswap the block numbers!
4397 for (block = 0; block < EXT4_N_BLOCKS; block++)
4398 ei->i_data[block] = raw_inode->i_block[block];
4399 INIT_LIST_HEAD(&ei->i_orphan);
4401 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4402 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4403 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4404 EXT4_INODE_SIZE(inode->i_sb)) {
4409 if (ei->i_extra_isize == 0) {
4410 /* The extra space is currently unused. Use it. */
4411 ei->i_extra_isize = sizeof(struct ext4_inode) -
4412 EXT4_GOOD_OLD_INODE_SIZE;
4414 __le32 *magic = (void *)raw_inode +
4415 EXT4_GOOD_OLD_INODE_SIZE +
4417 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4418 ei->i_state |= EXT4_STATE_XATTR;
4421 ei->i_extra_isize = 0;
4423 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4424 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4425 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4426 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4428 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4429 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4430 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4432 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4436 if (ei->i_file_acl &&
4438 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4439 EXT4_SB(sb)->s_gdb_count)) ||
4440 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4441 ext4_error(sb, __func__,
4442 "bad extended attribute block %llu in inode #%lu",
4443 ei->i_file_acl, inode->i_ino);
4446 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4447 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4448 (S_ISLNK(inode->i_mode) &&
4449 !ext4_inode_is_fast_symlink(inode)))
4450 /* Validate extent which is part of inode */
4451 ret = ext4_ext_check_inode(inode);
4452 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4453 (S_ISLNK(inode->i_mode) &&
4454 !ext4_inode_is_fast_symlink(inode))) {
4455 /* Validate block references which are part of inode */
4456 ret = ext4_check_inode_blockref(inode);
4463 if (S_ISREG(inode->i_mode)) {
4464 inode->i_op = &ext4_file_inode_operations;
4465 inode->i_fop = &ext4_file_operations;
4466 ext4_set_aops(inode);
4467 } else if (S_ISDIR(inode->i_mode)) {
4468 inode->i_op = &ext4_dir_inode_operations;
4469 inode->i_fop = &ext4_dir_operations;
4470 } else if (S_ISLNK(inode->i_mode)) {
4471 if (ext4_inode_is_fast_symlink(inode)) {
4472 inode->i_op = &ext4_fast_symlink_inode_operations;
4473 nd_terminate_link(ei->i_data, inode->i_size,
4474 sizeof(ei->i_data) - 1);
4476 inode->i_op = &ext4_symlink_inode_operations;
4477 ext4_set_aops(inode);
4479 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4480 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4481 inode->i_op = &ext4_special_inode_operations;
4482 if (raw_inode->i_block[0])
4483 init_special_inode(inode, inode->i_mode,
4484 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4486 init_special_inode(inode, inode->i_mode,
4487 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4491 ext4_error(inode->i_sb, __func__,
4492 "bogus i_mode (%o) for inode=%lu",
4493 inode->i_mode, inode->i_ino);
4497 ext4_set_inode_flags(inode);
4498 unlock_new_inode(inode);
4503 return ERR_PTR(ret);
4506 static int ext4_inode_blocks_set(handle_t *handle,
4507 struct ext4_inode *raw_inode,
4508 struct ext4_inode_info *ei)
4510 struct inode *inode = &(ei->vfs_inode);
4511 u64 i_blocks = inode->i_blocks;
4512 struct super_block *sb = inode->i_sb;
4514 if (i_blocks <= ~0U) {
4516 * i_blocks can be represnted in a 32 bit variable
4517 * as multiple of 512 bytes
4519 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4520 raw_inode->i_blocks_high = 0;
4521 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4524 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4527 if (i_blocks <= 0xffffffffffffULL) {
4529 * i_blocks can be represented in a 48 bit variable
4530 * as multiple of 512 bytes
4532 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4533 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4534 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4536 ei->i_flags |= EXT4_HUGE_FILE_FL;
4537 /* i_block is stored in file system block size */
4538 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4539 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4540 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4546 * Post the struct inode info into an on-disk inode location in the
4547 * buffer-cache. This gobbles the caller's reference to the
4548 * buffer_head in the inode location struct.
4550 * The caller must have write access to iloc->bh.
4552 static int ext4_do_update_inode(handle_t *handle,
4553 struct inode *inode,
4554 struct ext4_iloc *iloc,
4557 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4558 struct ext4_inode_info *ei = EXT4_I(inode);
4559 struct buffer_head *bh = iloc->bh;
4560 int err = 0, rc, block;
4562 /* For fields not not tracking in the in-memory inode,
4563 * initialise them to zero for new inodes. */
4564 if (ei->i_state & EXT4_STATE_NEW)
4565 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4567 ext4_get_inode_flags(ei);
4568 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4569 if (!(test_opt(inode->i_sb, NO_UID32))) {
4570 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4571 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4573 * Fix up interoperability with old kernels. Otherwise, old inodes get
4574 * re-used with the upper 16 bits of the uid/gid intact
4577 raw_inode->i_uid_high =
4578 cpu_to_le16(high_16_bits(inode->i_uid));
4579 raw_inode->i_gid_high =
4580 cpu_to_le16(high_16_bits(inode->i_gid));
4582 raw_inode->i_uid_high = 0;
4583 raw_inode->i_gid_high = 0;
4586 raw_inode->i_uid_low =
4587 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4588 raw_inode->i_gid_low =
4589 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4590 raw_inode->i_uid_high = 0;
4591 raw_inode->i_gid_high = 0;
4593 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4595 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4596 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4597 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4598 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4600 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4602 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4603 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
4604 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4605 cpu_to_le32(EXT4_OS_HURD))
4606 raw_inode->i_file_acl_high =
4607 cpu_to_le16(ei->i_file_acl >> 32);
4608 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4609 ext4_isize_set(raw_inode, ei->i_disksize);
4610 if (ei->i_disksize > 0x7fffffffULL) {
4611 struct super_block *sb = inode->i_sb;
4612 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4613 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4614 EXT4_SB(sb)->s_es->s_rev_level ==
4615 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4616 /* If this is the first large file
4617 * created, add a flag to the superblock.
4619 err = ext4_journal_get_write_access(handle,
4620 EXT4_SB(sb)->s_sbh);
4623 ext4_update_dynamic_rev(sb);
4624 EXT4_SET_RO_COMPAT_FEATURE(sb,
4625 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4627 ext4_handle_sync(handle);
4628 err = ext4_handle_dirty_metadata(handle, inode,
4629 EXT4_SB(sb)->s_sbh);
4632 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4633 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4634 if (old_valid_dev(inode->i_rdev)) {
4635 raw_inode->i_block[0] =
4636 cpu_to_le32(old_encode_dev(inode->i_rdev));
4637 raw_inode->i_block[1] = 0;
4639 raw_inode->i_block[0] = 0;
4640 raw_inode->i_block[1] =
4641 cpu_to_le32(new_encode_dev(inode->i_rdev));
4642 raw_inode->i_block[2] = 0;
4645 for (block = 0; block < EXT4_N_BLOCKS; block++)
4646 raw_inode->i_block[block] = ei->i_data[block];
4648 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4649 if (ei->i_extra_isize) {
4650 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4651 raw_inode->i_version_hi =
4652 cpu_to_le32(inode->i_version >> 32);
4653 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4657 * If we're not using a journal and we were called from
4658 * ext4_write_inode() to sync the inode (making do_sync true),
4659 * we can just use sync_dirty_buffer() directly to do our dirty
4660 * work. Testing s_journal here is a bit redundant but it's
4661 * worth it to avoid potential future trouble.
4663 if (EXT4_SB(inode->i_sb)->s_journal == NULL && do_sync) {
4664 BUFFER_TRACE(bh, "call sync_dirty_buffer");
4665 sync_dirty_buffer(bh);
4667 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4668 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4672 ei->i_state &= ~EXT4_STATE_NEW;
4676 ext4_std_error(inode->i_sb, err);
4681 * ext4_write_inode()
4683 * We are called from a few places:
4685 * - Within generic_file_write() for O_SYNC files.
4686 * Here, there will be no transaction running. We wait for any running
4687 * trasnaction to commit.
4689 * - Within sys_sync(), kupdate and such.
4690 * We wait on commit, if tol to.
4692 * - Within prune_icache() (PF_MEMALLOC == true)
4693 * Here we simply return. We can't afford to block kswapd on the
4696 * In all cases it is actually safe for us to return without doing anything,
4697 * because the inode has been copied into a raw inode buffer in
4698 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4701 * Note that we are absolutely dependent upon all inode dirtiers doing the
4702 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4703 * which we are interested.
4705 * It would be a bug for them to not do this. The code:
4707 * mark_inode_dirty(inode)
4709 * inode->i_size = expr;
4711 * is in error because a kswapd-driven write_inode() could occur while
4712 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4713 * will no longer be on the superblock's dirty inode list.
4715 int ext4_write_inode(struct inode *inode, int wait)
4719 if (current->flags & PF_MEMALLOC)
4722 if (EXT4_SB(inode->i_sb)->s_journal) {
4723 if (ext4_journal_current_handle()) {
4724 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4732 err = ext4_force_commit(inode->i_sb);
4734 struct ext4_iloc iloc;
4736 err = ext4_get_inode_loc(inode, &iloc);
4739 err = ext4_do_update_inode(EXT4_NOJOURNAL_HANDLE,
4740 inode, &iloc, wait);
4748 * Called from notify_change.
4750 * We want to trap VFS attempts to truncate the file as soon as
4751 * possible. In particular, we want to make sure that when the VFS
4752 * shrinks i_size, we put the inode on the orphan list and modify
4753 * i_disksize immediately, so that during the subsequent flushing of
4754 * dirty pages and freeing of disk blocks, we can guarantee that any
4755 * commit will leave the blocks being flushed in an unused state on
4756 * disk. (On recovery, the inode will get truncated and the blocks will
4757 * be freed, so we have a strong guarantee that no future commit will
4758 * leave these blocks visible to the user.)
4760 * Another thing we have to assure is that if we are in ordered mode
4761 * and inode is still attached to the committing transaction, we must
4762 * we start writeout of all the dirty pages which are being truncated.
4763 * This way we are sure that all the data written in the previous
4764 * transaction are already on disk (truncate waits for pages under
4767 * Called with inode->i_mutex down.
4769 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4771 struct inode *inode = dentry->d_inode;
4773 const unsigned int ia_valid = attr->ia_valid;
4775 error = inode_change_ok(inode, attr);
4779 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4780 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4783 /* (user+group)*(old+new) structure, inode write (sb,
4784 * inode block, ? - but truncate inode update has it) */
4785 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4786 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4787 if (IS_ERR(handle)) {
4788 error = PTR_ERR(handle);
4791 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4793 ext4_journal_stop(handle);
4796 /* Update corresponding info in inode so that everything is in
4797 * one transaction */
4798 if (attr->ia_valid & ATTR_UID)
4799 inode->i_uid = attr->ia_uid;
4800 if (attr->ia_valid & ATTR_GID)
4801 inode->i_gid = attr->ia_gid;
4802 error = ext4_mark_inode_dirty(handle, inode);
4803 ext4_journal_stop(handle);
4806 if (attr->ia_valid & ATTR_SIZE) {
4807 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4808 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4810 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4817 if (S_ISREG(inode->i_mode) &&
4818 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4821 handle = ext4_journal_start(inode, 3);
4822 if (IS_ERR(handle)) {
4823 error = PTR_ERR(handle);
4827 error = ext4_orphan_add(handle, inode);
4828 EXT4_I(inode)->i_disksize = attr->ia_size;
4829 rc = ext4_mark_inode_dirty(handle, inode);
4832 ext4_journal_stop(handle);
4834 if (ext4_should_order_data(inode)) {
4835 error = ext4_begin_ordered_truncate(inode,
4838 /* Do as much error cleanup as possible */
4839 handle = ext4_journal_start(inode, 3);
4840 if (IS_ERR(handle)) {
4841 ext4_orphan_del(NULL, inode);
4844 ext4_orphan_del(handle, inode);
4845 ext4_journal_stop(handle);
4851 rc = inode_setattr(inode, attr);
4853 /* If inode_setattr's call to ext4_truncate failed to get a
4854 * transaction handle at all, we need to clean up the in-core
4855 * orphan list manually. */
4857 ext4_orphan_del(NULL, inode);
4859 if (!rc && (ia_valid & ATTR_MODE))
4860 rc = ext4_acl_chmod(inode);
4863 ext4_std_error(inode->i_sb, error);
4869 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4872 struct inode *inode;
4873 unsigned long delalloc_blocks;
4875 inode = dentry->d_inode;
4876 generic_fillattr(inode, stat);
4879 * We can't update i_blocks if the block allocation is delayed
4880 * otherwise in the case of system crash before the real block
4881 * allocation is done, we will have i_blocks inconsistent with
4882 * on-disk file blocks.
4883 * We always keep i_blocks updated together with real
4884 * allocation. But to not confuse with user, stat
4885 * will return the blocks that include the delayed allocation
4886 * blocks for this file.
4888 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4889 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4890 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4892 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4896 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4901 /* if nrblocks are contiguous */
4904 * With N contiguous data blocks, it need at most
4905 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4906 * 2 dindirect blocks
4909 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4910 return indirects + 3;
4913 * if nrblocks are not contiguous, worse case, each block touch
4914 * a indirect block, and each indirect block touch a double indirect
4915 * block, plus a triple indirect block
4917 indirects = nrblocks * 2 + 1;
4921 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4923 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4924 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4925 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4929 * Account for index blocks, block groups bitmaps and block group
4930 * descriptor blocks if modify datablocks and index blocks
4931 * worse case, the indexs blocks spread over different block groups
4933 * If datablocks are discontiguous, they are possible to spread over
4934 * different block groups too. If they are contiugous, with flexbg,
4935 * they could still across block group boundary.
4937 * Also account for superblock, inode, quota and xattr blocks
4939 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4941 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4947 * How many index blocks need to touch to modify nrblocks?
4948 * The "Chunk" flag indicating whether the nrblocks is
4949 * physically contiguous on disk
4951 * For Direct IO and fallocate, they calls get_block to allocate
4952 * one single extent at a time, so they could set the "Chunk" flag
4954 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4959 * Now let's see how many group bitmaps and group descriptors need
4969 if (groups > ngroups)
4971 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4972 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4974 /* bitmaps and block group descriptor blocks */
4975 ret += groups + gdpblocks;
4977 /* Blocks for super block, inode, quota and xattr blocks */
4978 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4984 * Calulate the total number of credits to reserve to fit
4985 * the modification of a single pages into a single transaction,
4986 * which may include multiple chunks of block allocations.
4988 * This could be called via ext4_write_begin()
4990 * We need to consider the worse case, when
4991 * one new block per extent.
4993 int ext4_writepage_trans_blocks(struct inode *inode)
4995 int bpp = ext4_journal_blocks_per_page(inode);
4998 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5000 /* Account for data blocks for journalled mode */
5001 if (ext4_should_journal_data(inode))
5007 * Calculate the journal credits for a chunk of data modification.
5009 * This is called from DIO, fallocate or whoever calling
5010 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5012 * journal buffers for data blocks are not included here, as DIO
5013 * and fallocate do no need to journal data buffers.
5015 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5017 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5021 * The caller must have previously called ext4_reserve_inode_write().
5022 * Give this, we know that the caller already has write access to iloc->bh.
5024 int ext4_mark_iloc_dirty(handle_t *handle,
5025 struct inode *inode, struct ext4_iloc *iloc)
5029 if (test_opt(inode->i_sb, I_VERSION))
5030 inode_inc_iversion(inode);
5032 /* the do_update_inode consumes one bh->b_count */
5035 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5036 err = ext4_do_update_inode(handle, inode, iloc, 0);
5042 * On success, We end up with an outstanding reference count against
5043 * iloc->bh. This _must_ be cleaned up later.
5047 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5048 struct ext4_iloc *iloc)
5052 err = ext4_get_inode_loc(inode, iloc);
5054 BUFFER_TRACE(iloc->bh, "get_write_access");
5055 err = ext4_journal_get_write_access(handle, iloc->bh);
5061 ext4_std_error(inode->i_sb, err);
5066 * Expand an inode by new_extra_isize bytes.
5067 * Returns 0 on success or negative error number on failure.
5069 static int ext4_expand_extra_isize(struct inode *inode,
5070 unsigned int new_extra_isize,
5071 struct ext4_iloc iloc,
5074 struct ext4_inode *raw_inode;
5075 struct ext4_xattr_ibody_header *header;
5076 struct ext4_xattr_entry *entry;
5078 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5081 raw_inode = ext4_raw_inode(&iloc);
5083 header = IHDR(inode, raw_inode);
5084 entry = IFIRST(header);
5086 /* No extended attributes present */
5087 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5088 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5089 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5091 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5095 /* try to expand with EAs present */
5096 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5101 * What we do here is to mark the in-core inode as clean with respect to inode
5102 * dirtiness (it may still be data-dirty).
5103 * This means that the in-core inode may be reaped by prune_icache
5104 * without having to perform any I/O. This is a very good thing,
5105 * because *any* task may call prune_icache - even ones which
5106 * have a transaction open against a different journal.
5108 * Is this cheating? Not really. Sure, we haven't written the
5109 * inode out, but prune_icache isn't a user-visible syncing function.
5110 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5111 * we start and wait on commits.
5113 * Is this efficient/effective? Well, we're being nice to the system
5114 * by cleaning up our inodes proactively so they can be reaped
5115 * without I/O. But we are potentially leaving up to five seconds'
5116 * worth of inodes floating about which prune_icache wants us to
5117 * write out. One way to fix that would be to get prune_icache()
5118 * to do a write_super() to free up some memory. It has the desired
5121 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5123 struct ext4_iloc iloc;
5124 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5125 static unsigned int mnt_count;
5129 err = ext4_reserve_inode_write(handle, inode, &iloc);
5130 if (ext4_handle_valid(handle) &&
5131 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5132 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5134 * We need extra buffer credits since we may write into EA block
5135 * with this same handle. If journal_extend fails, then it will
5136 * only result in a minor loss of functionality for that inode.
5137 * If this is felt to be critical, then e2fsck should be run to
5138 * force a large enough s_min_extra_isize.
5140 if ((jbd2_journal_extend(handle,
5141 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5142 ret = ext4_expand_extra_isize(inode,
5143 sbi->s_want_extra_isize,
5146 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5148 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5149 ext4_warning(inode->i_sb, __func__,
5150 "Unable to expand inode %lu. Delete"
5151 " some EAs or run e2fsck.",
5154 le16_to_cpu(sbi->s_es->s_mnt_count);
5160 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5165 * ext4_dirty_inode() is called from __mark_inode_dirty()
5167 * We're really interested in the case where a file is being extended.
5168 * i_size has been changed by generic_commit_write() and we thus need
5169 * to include the updated inode in the current transaction.
5171 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5172 * are allocated to the file.
5174 * If the inode is marked synchronous, we don't honour that here - doing
5175 * so would cause a commit on atime updates, which we don't bother doing.
5176 * We handle synchronous inodes at the highest possible level.
5178 void ext4_dirty_inode(struct inode *inode)
5180 handle_t *current_handle = ext4_journal_current_handle();
5183 if (!ext4_handle_valid(current_handle)) {
5184 ext4_mark_inode_dirty(current_handle, inode);
5188 handle = ext4_journal_start(inode, 2);
5191 if (current_handle &&
5192 current_handle->h_transaction != handle->h_transaction) {
5193 /* This task has a transaction open against a different fs */
5194 printk(KERN_EMERG "%s: transactions do not match!\n",
5197 jbd_debug(5, "marking dirty. outer handle=%p\n",
5199 ext4_mark_inode_dirty(handle, inode);
5201 ext4_journal_stop(handle);
5208 * Bind an inode's backing buffer_head into this transaction, to prevent
5209 * it from being flushed to disk early. Unlike
5210 * ext4_reserve_inode_write, this leaves behind no bh reference and
5211 * returns no iloc structure, so the caller needs to repeat the iloc
5212 * lookup to mark the inode dirty later.
5214 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5216 struct ext4_iloc iloc;
5220 err = ext4_get_inode_loc(inode, &iloc);
5222 BUFFER_TRACE(iloc.bh, "get_write_access");
5223 err = jbd2_journal_get_write_access(handle, iloc.bh);
5225 err = ext4_handle_dirty_metadata(handle,
5231 ext4_std_error(inode->i_sb, err);
5236 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5243 * We have to be very careful here: changing a data block's
5244 * journaling status dynamically is dangerous. If we write a
5245 * data block to the journal, change the status and then delete
5246 * that block, we risk forgetting to revoke the old log record
5247 * from the journal and so a subsequent replay can corrupt data.
5248 * So, first we make sure that the journal is empty and that
5249 * nobody is changing anything.
5252 journal = EXT4_JOURNAL(inode);
5255 if (is_journal_aborted(journal))
5258 jbd2_journal_lock_updates(journal);
5259 jbd2_journal_flush(journal);
5262 * OK, there are no updates running now, and all cached data is
5263 * synced to disk. We are now in a completely consistent state
5264 * which doesn't have anything in the journal, and we know that
5265 * no filesystem updates are running, so it is safe to modify
5266 * the inode's in-core data-journaling state flag now.
5270 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5272 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5273 ext4_set_aops(inode);
5275 jbd2_journal_unlock_updates(journal);
5277 /* Finally we can mark the inode as dirty. */
5279 handle = ext4_journal_start(inode, 1);
5281 return PTR_ERR(handle);
5283 err = ext4_mark_inode_dirty(handle, inode);
5284 ext4_handle_sync(handle);
5285 ext4_journal_stop(handle);
5286 ext4_std_error(inode->i_sb, err);
5291 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5293 return !buffer_mapped(bh);
5296 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5298 struct page *page = vmf->page;
5303 struct file *file = vma->vm_file;
5304 struct inode *inode = file->f_path.dentry->d_inode;
5305 struct address_space *mapping = inode->i_mapping;
5308 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5309 * get i_mutex because we are already holding mmap_sem.
5311 down_read(&inode->i_alloc_sem);
5312 size = i_size_read(inode);
5313 if (page->mapping != mapping || size <= page_offset(page)
5314 || !PageUptodate(page)) {
5315 /* page got truncated from under us? */
5319 if (PageMappedToDisk(page))
5322 if (page->index == size >> PAGE_CACHE_SHIFT)
5323 len = size & ~PAGE_CACHE_MASK;
5325 len = PAGE_CACHE_SIZE;
5329 * return if we have all the buffers mapped. This avoid
5330 * the need to call write_begin/write_end which does a
5331 * journal_start/journal_stop which can block and take
5334 if (page_has_buffers(page)) {
5335 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5336 ext4_bh_unmapped)) {
5343 * OK, we need to fill the hole... Do write_begin write_end
5344 * to do block allocation/reservation.We are not holding
5345 * inode.i__mutex here. That allow * parallel write_begin,
5346 * write_end call. lock_page prevent this from happening
5347 * on the same page though
5349 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5350 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5353 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5354 len, len, page, fsdata);
5360 ret = VM_FAULT_SIGBUS;
5361 up_read(&inode->i_alloc_sem);