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>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode *inode,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode,
62 static void ext4_invalidatepage(struct page *page, unsigned long offset);
65 * Test whether an inode is a fast symlink.
67 static int ext4_inode_is_fast_symlink(struct inode *inode)
69 int ea_blocks = EXT4_I(inode)->i_file_acl ?
70 (inode->i_sb->s_blocksize >> 9) : 0;
72 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
76 * Work out how many blocks we need to proceed with the next chunk of a
77 * truncate transaction.
79 static unsigned long blocks_for_truncate(struct inode *inode)
83 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
85 /* Give ourselves just enough room to cope with inodes in which
86 * i_blocks is corrupt: we've seen disk corruptions in the past
87 * which resulted in random data in an inode which looked enough
88 * like a regular file for ext4 to try to delete it. Things
89 * will go a bit crazy if that happens, but at least we should
90 * try not to panic the whole kernel. */
94 /* But we need to bound the transaction so we don't overflow the
96 if (needed > EXT4_MAX_TRANS_DATA)
97 needed = EXT4_MAX_TRANS_DATA;
99 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
103 * Truncate transactions can be complex and absolutely huge. So we need to
104 * be able to restart the transaction at a conventient checkpoint to make
105 * sure we don't overflow the journal.
107 * start_transaction gets us a new handle for a truncate transaction,
108 * and extend_transaction tries to extend the existing one a bit. If
109 * extend fails, we need to propagate the failure up and restart the
110 * transaction in the top-level truncate loop. --sct
112 static handle_t *start_transaction(struct inode *inode)
116 result = ext4_journal_start(inode, blocks_for_truncate(inode));
120 ext4_std_error(inode->i_sb, PTR_ERR(result));
125 * Try to extend this transaction for the purposes of truncation.
127 * Returns 0 if we managed to create more room. If we can't create more
128 * room, and the transaction must be restarted we return 1.
130 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
132 if (!ext4_handle_valid(handle))
134 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
136 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
142 * Restart the transaction associated with *handle. This does a commit,
143 * so before we call here everything must be consistently dirtied against
146 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
152 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
153 * moment, get_block can be called only for blocks inside i_size since
154 * page cache has been already dropped and writes are blocked by
155 * i_mutex. So we can safely drop the i_data_sem here.
157 BUG_ON(EXT4_JOURNAL(inode) == NULL);
158 jbd_debug(2, "restarting handle %p\n", handle);
159 up_write(&EXT4_I(inode)->i_data_sem);
160 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
161 down_write(&EXT4_I(inode)->i_data_sem);
162 ext4_discard_preallocations(inode);
168 * Called at the last iput() if i_nlink is zero.
170 void ext4_delete_inode(struct inode *inode)
175 if (!is_bad_inode(inode))
176 dquot_initialize(inode);
178 if (ext4_should_order_data(inode))
179 ext4_begin_ordered_truncate(inode, 0);
180 truncate_inode_pages(&inode->i_data, 0);
182 if (is_bad_inode(inode))
185 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
186 if (IS_ERR(handle)) {
187 ext4_std_error(inode->i_sb, PTR_ERR(handle));
189 * If we're going to skip the normal cleanup, we still need to
190 * make sure that the in-core orphan linked list is properly
193 ext4_orphan_del(NULL, inode);
198 ext4_handle_sync(handle);
200 err = ext4_mark_inode_dirty(handle, inode);
202 ext4_warning(inode->i_sb,
203 "couldn't mark inode dirty (err %d)", err);
207 ext4_truncate(inode);
210 * ext4_ext_truncate() doesn't reserve any slop when it
211 * restarts journal transactions; therefore there may not be
212 * enough credits left in the handle to remove the inode from
213 * the orphan list and set the dtime field.
215 if (!ext4_handle_has_enough_credits(handle, 3)) {
216 err = ext4_journal_extend(handle, 3);
218 err = ext4_journal_restart(handle, 3);
220 ext4_warning(inode->i_sb,
221 "couldn't extend journal (err %d)", err);
223 ext4_journal_stop(handle);
229 * Kill off the orphan record which ext4_truncate created.
230 * AKPM: I think this can be inside the above `if'.
231 * Note that ext4_orphan_del() has to be able to cope with the
232 * deletion of a non-existent orphan - this is because we don't
233 * know if ext4_truncate() actually created an orphan record.
234 * (Well, we could do this if we need to, but heck - it works)
236 ext4_orphan_del(handle, inode);
237 EXT4_I(inode)->i_dtime = get_seconds();
240 * One subtle ordering requirement: if anything has gone wrong
241 * (transaction abort, IO errors, whatever), then we can still
242 * do these next steps (the fs will already have been marked as
243 * having errors), but we can't free the inode if the mark_dirty
246 if (ext4_mark_inode_dirty(handle, inode))
247 /* If that failed, just do the required in-core inode clear. */
250 ext4_free_inode(handle, inode);
251 ext4_journal_stop(handle);
254 clear_inode(inode); /* We must guarantee clearing of inode... */
260 struct buffer_head *bh;
263 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
265 p->key = *(p->p = v);
270 * ext4_block_to_path - parse the block number into array of offsets
271 * @inode: inode in question (we are only interested in its superblock)
272 * @i_block: block number to be parsed
273 * @offsets: array to store the offsets in
274 * @boundary: set this non-zero if the referred-to block is likely to be
275 * followed (on disk) by an indirect block.
277 * To store the locations of file's data ext4 uses a data structure common
278 * for UNIX filesystems - tree of pointers anchored in the inode, with
279 * data blocks at leaves and indirect blocks in intermediate nodes.
280 * This function translates the block number into path in that tree -
281 * return value is the path length and @offsets[n] is the offset of
282 * pointer to (n+1)th node in the nth one. If @block is out of range
283 * (negative or too large) warning is printed and zero returned.
285 * Note: function doesn't find node addresses, so no IO is needed. All
286 * we need to know is the capacity of indirect blocks (taken from the
291 * Portability note: the last comparison (check that we fit into triple
292 * indirect block) is spelled differently, because otherwise on an
293 * architecture with 32-bit longs and 8Kb pages we might get into trouble
294 * if our filesystem had 8Kb blocks. We might use long long, but that would
295 * kill us on x86. Oh, well, at least the sign propagation does not matter -
296 * i_block would have to be negative in the very beginning, so we would not
300 static int ext4_block_to_path(struct inode *inode,
302 ext4_lblk_t offsets[4], int *boundary)
304 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
305 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
306 const long direct_blocks = EXT4_NDIR_BLOCKS,
307 indirect_blocks = ptrs,
308 double_blocks = (1 << (ptrs_bits * 2));
312 if (i_block < direct_blocks) {
313 offsets[n++] = i_block;
314 final = direct_blocks;
315 } else if ((i_block -= direct_blocks) < indirect_blocks) {
316 offsets[n++] = EXT4_IND_BLOCK;
317 offsets[n++] = i_block;
319 } else if ((i_block -= indirect_blocks) < double_blocks) {
320 offsets[n++] = EXT4_DIND_BLOCK;
321 offsets[n++] = i_block >> ptrs_bits;
322 offsets[n++] = i_block & (ptrs - 1);
324 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
325 offsets[n++] = EXT4_TIND_BLOCK;
326 offsets[n++] = i_block >> (ptrs_bits * 2);
327 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
328 offsets[n++] = i_block & (ptrs - 1);
331 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
332 i_block + direct_blocks +
333 indirect_blocks + double_blocks, inode->i_ino);
336 *boundary = final - 1 - (i_block & (ptrs - 1));
340 static int __ext4_check_blockref(const char *function, struct inode *inode,
341 __le32 *p, unsigned int max)
346 while (bref < p+max) {
347 blk = le32_to_cpu(*bref++);
349 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
351 __ext4_error(inode->i_sb, function,
352 "invalid block reference %u "
353 "in inode #%lu", blk, inode->i_ino);
361 #define ext4_check_indirect_blockref(inode, bh) \
362 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
363 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
365 #define ext4_check_inode_blockref(inode) \
366 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
370 * ext4_get_branch - read the chain of indirect blocks leading to data
371 * @inode: inode in question
372 * @depth: depth of the chain (1 - direct pointer, etc.)
373 * @offsets: offsets of pointers in inode/indirect blocks
374 * @chain: place to store the result
375 * @err: here we store the error value
377 * Function fills the array of triples <key, p, bh> and returns %NULL
378 * if everything went OK or the pointer to the last filled triple
379 * (incomplete one) otherwise. Upon the return chain[i].key contains
380 * the number of (i+1)-th block in the chain (as it is stored in memory,
381 * i.e. little-endian 32-bit), chain[i].p contains the address of that
382 * number (it points into struct inode for i==0 and into the bh->b_data
383 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
384 * block for i>0 and NULL for i==0. In other words, it holds the block
385 * numbers of the chain, addresses they were taken from (and where we can
386 * verify that chain did not change) and buffer_heads hosting these
389 * Function stops when it stumbles upon zero pointer (absent block)
390 * (pointer to last triple returned, *@err == 0)
391 * or when it gets an IO error reading an indirect block
392 * (ditto, *@err == -EIO)
393 * or when it reads all @depth-1 indirect blocks successfully and finds
394 * the whole chain, all way to the data (returns %NULL, *err == 0).
396 * Need to be called with
397 * down_read(&EXT4_I(inode)->i_data_sem)
399 static Indirect *ext4_get_branch(struct inode *inode, int depth,
400 ext4_lblk_t *offsets,
401 Indirect chain[4], int *err)
403 struct super_block *sb = inode->i_sb;
405 struct buffer_head *bh;
408 /* i_data is not going away, no lock needed */
409 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
413 bh = sb_getblk(sb, le32_to_cpu(p->key));
417 if (!bh_uptodate_or_lock(bh)) {
418 if (bh_submit_read(bh) < 0) {
422 /* validate block references */
423 if (ext4_check_indirect_blockref(inode, bh)) {
429 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
443 * ext4_find_near - find a place for allocation with sufficient locality
445 * @ind: descriptor of indirect block.
447 * This function returns the preferred place for block allocation.
448 * It is used when heuristic for sequential allocation fails.
450 * + if there is a block to the left of our position - allocate near it.
451 * + if pointer will live in indirect block - allocate near that block.
452 * + if pointer will live in inode - allocate in the same
455 * In the latter case we colour the starting block by the callers PID to
456 * prevent it from clashing with concurrent allocations for a different inode
457 * in the same block group. The PID is used here so that functionally related
458 * files will be close-by on-disk.
460 * Caller must make sure that @ind is valid and will stay that way.
462 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
464 struct ext4_inode_info *ei = EXT4_I(inode);
465 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
467 ext4_fsblk_t bg_start;
468 ext4_fsblk_t last_block;
469 ext4_grpblk_t colour;
470 ext4_group_t block_group;
471 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
473 /* Try to find previous block */
474 for (p = ind->p - 1; p >= start; p--) {
476 return le32_to_cpu(*p);
479 /* No such thing, so let's try location of indirect block */
481 return ind->bh->b_blocknr;
484 * It is going to be referred to from the inode itself? OK, just put it
485 * into the same cylinder group then.
487 block_group = ei->i_block_group;
488 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
489 block_group &= ~(flex_size-1);
490 if (S_ISREG(inode->i_mode))
493 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
494 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
497 * If we are doing delayed allocation, we don't need take
498 * colour into account.
500 if (test_opt(inode->i_sb, DELALLOC))
503 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
504 colour = (current->pid % 16) *
505 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
507 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
508 return bg_start + colour;
512 * ext4_find_goal - find a preferred place for allocation.
514 * @block: block we want
515 * @partial: pointer to the last triple within a chain
517 * Normally this function find the preferred place for block allocation,
519 * Because this is only used for non-extent files, we limit the block nr
522 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
528 * XXX need to get goal block from mballoc's data structures
531 goal = ext4_find_near(inode, partial);
532 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
537 * ext4_blks_to_allocate: Look up the block map and count the number
538 * of direct blocks need to be allocated for the given branch.
540 * @branch: chain of indirect blocks
541 * @k: number of blocks need for indirect blocks
542 * @blks: number of data blocks to be mapped.
543 * @blocks_to_boundary: the offset in the indirect block
545 * return the total number of blocks to be allocate, including the
546 * direct and indirect blocks.
548 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
549 int blocks_to_boundary)
551 unsigned int count = 0;
554 * Simple case, [t,d]Indirect block(s) has not allocated yet
555 * then it's clear blocks on that path have not allocated
558 /* right now we don't handle cross boundary allocation */
559 if (blks < blocks_to_boundary + 1)
562 count += blocks_to_boundary + 1;
567 while (count < blks && count <= blocks_to_boundary &&
568 le32_to_cpu(*(branch[0].p + count)) == 0) {
575 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
576 * @indirect_blks: the number of blocks need to allocate for indirect
579 * @new_blocks: on return it will store the new block numbers for
580 * the indirect blocks(if needed) and the first direct block,
581 * @blks: on return it will store the total number of allocated
584 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
585 ext4_lblk_t iblock, ext4_fsblk_t goal,
586 int indirect_blks, int blks,
587 ext4_fsblk_t new_blocks[4], int *err)
589 struct ext4_allocation_request ar;
591 unsigned long count = 0, blk_allocated = 0;
593 ext4_fsblk_t current_block = 0;
597 * Here we try to allocate the requested multiple blocks at once,
598 * on a best-effort basis.
599 * To build a branch, we should allocate blocks for
600 * the indirect blocks(if not allocated yet), and at least
601 * the first direct block of this branch. That's the
602 * minimum number of blocks need to allocate(required)
604 /* first we try to allocate the indirect blocks */
605 target = indirect_blks;
608 /* allocating blocks for indirect blocks and direct blocks */
609 current_block = ext4_new_meta_blocks(handle, inode,
614 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
615 EXT4_ERROR_INODE(inode,
616 "current_block %llu + count %lu > %d!",
617 current_block, count,
618 EXT4_MAX_BLOCK_FILE_PHYS);
624 /* allocate blocks for indirect blocks */
625 while (index < indirect_blks && count) {
626 new_blocks[index++] = current_block++;
631 * save the new block number
632 * for the first direct block
634 new_blocks[index] = current_block;
635 printk(KERN_INFO "%s returned more blocks than "
636 "requested\n", __func__);
642 target = blks - count ;
643 blk_allocated = count;
646 /* Now allocate data blocks */
647 memset(&ar, 0, sizeof(ar));
652 if (S_ISREG(inode->i_mode))
653 /* enable in-core preallocation only for regular files */
654 ar.flags = EXT4_MB_HINT_DATA;
656 current_block = ext4_mb_new_blocks(handle, &ar, err);
657 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
658 EXT4_ERROR_INODE(inode,
659 "current_block %llu + ar.len %d > %d!",
660 current_block, ar.len,
661 EXT4_MAX_BLOCK_FILE_PHYS);
666 if (*err && (target == blks)) {
668 * if the allocation failed and we didn't allocate
674 if (target == blks) {
676 * save the new block number
677 * for the first direct block
679 new_blocks[index] = current_block;
681 blk_allocated += ar.len;
684 /* total number of blocks allocated for direct blocks */
689 for (i = 0; i < index; i++)
690 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
695 * ext4_alloc_branch - allocate and set up a chain of blocks.
697 * @indirect_blks: number of allocated indirect blocks
698 * @blks: number of allocated direct blocks
699 * @offsets: offsets (in the blocks) to store the pointers to next.
700 * @branch: place to store the chain in.
702 * This function allocates blocks, zeroes out all but the last one,
703 * links them into chain and (if we are synchronous) writes them to disk.
704 * In other words, it prepares a branch that can be spliced onto the
705 * inode. It stores the information about that chain in the branch[], in
706 * the same format as ext4_get_branch() would do. We are calling it after
707 * we had read the existing part of chain and partial points to the last
708 * triple of that (one with zero ->key). Upon the exit we have the same
709 * picture as after the successful ext4_get_block(), except that in one
710 * place chain is disconnected - *branch->p is still zero (we did not
711 * set the last link), but branch->key contains the number that should
712 * be placed into *branch->p to fill that gap.
714 * If allocation fails we free all blocks we've allocated (and forget
715 * their buffer_heads) and return the error value the from failed
716 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
717 * as described above and return 0.
719 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
720 ext4_lblk_t iblock, int indirect_blks,
721 int *blks, ext4_fsblk_t goal,
722 ext4_lblk_t *offsets, Indirect *branch)
724 int blocksize = inode->i_sb->s_blocksize;
727 struct buffer_head *bh;
729 ext4_fsblk_t new_blocks[4];
730 ext4_fsblk_t current_block;
732 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
733 *blks, new_blocks, &err);
737 branch[0].key = cpu_to_le32(new_blocks[0]);
739 * metadata blocks and data blocks are allocated.
741 for (n = 1; n <= indirect_blks; n++) {
743 * Get buffer_head for parent block, zero it out
744 * and set the pointer to new one, then send
747 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
750 BUFFER_TRACE(bh, "call get_create_access");
751 err = ext4_journal_get_create_access(handle, bh);
753 /* Don't brelse(bh) here; it's done in
754 * ext4_journal_forget() below */
759 memset(bh->b_data, 0, blocksize);
760 branch[n].p = (__le32 *) bh->b_data + offsets[n];
761 branch[n].key = cpu_to_le32(new_blocks[n]);
762 *branch[n].p = branch[n].key;
763 if (n == indirect_blks) {
764 current_block = new_blocks[n];
766 * End of chain, update the last new metablock of
767 * the chain to point to the new allocated
768 * data blocks numbers
770 for (i = 1; i < num; i++)
771 *(branch[n].p + i) = cpu_to_le32(++current_block);
773 BUFFER_TRACE(bh, "marking uptodate");
774 set_buffer_uptodate(bh);
777 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
778 err = ext4_handle_dirty_metadata(handle, inode, bh);
785 /* Allocation failed, free what we already allocated */
786 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
787 for (i = 1; i <= n ; i++) {
789 * branch[i].bh is newly allocated, so there is no
790 * need to revoke the block, which is why we don't
791 * need to set EXT4_FREE_BLOCKS_METADATA.
793 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
794 EXT4_FREE_BLOCKS_FORGET);
796 for (i = n+1; i < indirect_blks; i++)
797 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
799 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
805 * ext4_splice_branch - splice the allocated branch onto inode.
807 * @block: (logical) number of block we are adding
808 * @chain: chain of indirect blocks (with a missing link - see
810 * @where: location of missing link
811 * @num: number of indirect blocks we are adding
812 * @blks: number of direct blocks we are adding
814 * This function fills the missing link and does all housekeeping needed in
815 * inode (->i_blocks, etc.). In case of success we end up with the full
816 * chain to new block and return 0.
818 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
819 ext4_lblk_t block, Indirect *where, int num,
824 ext4_fsblk_t current_block;
827 * If we're splicing into a [td]indirect block (as opposed to the
828 * inode) then we need to get write access to the [td]indirect block
832 BUFFER_TRACE(where->bh, "get_write_access");
833 err = ext4_journal_get_write_access(handle, where->bh);
839 *where->p = where->key;
842 * Update the host buffer_head or inode to point to more just allocated
843 * direct blocks blocks
845 if (num == 0 && blks > 1) {
846 current_block = le32_to_cpu(where->key) + 1;
847 for (i = 1; i < blks; i++)
848 *(where->p + i) = cpu_to_le32(current_block++);
851 /* We are done with atomic stuff, now do the rest of housekeeping */
852 /* had we spliced it onto indirect block? */
855 * If we spliced it onto an indirect block, we haven't
856 * altered the inode. Note however that if it is being spliced
857 * onto an indirect block at the very end of the file (the
858 * file is growing) then we *will* alter the inode to reflect
859 * the new i_size. But that is not done here - it is done in
860 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
862 jbd_debug(5, "splicing indirect only\n");
863 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
864 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
869 * OK, we spliced it into the inode itself on a direct block.
871 ext4_mark_inode_dirty(handle, inode);
872 jbd_debug(5, "splicing direct\n");
877 for (i = 1; i <= num; i++) {
879 * branch[i].bh is newly allocated, so there is no
880 * need to revoke the block, which is why we don't
881 * need to set EXT4_FREE_BLOCKS_METADATA.
883 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
884 EXT4_FREE_BLOCKS_FORGET);
886 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
893 * The ext4_ind_map_blocks() function handles non-extents inodes
894 * (i.e., using the traditional indirect/double-indirect i_blocks
895 * scheme) for ext4_map_blocks().
897 * Allocation strategy is simple: if we have to allocate something, we will
898 * have to go the whole way to leaf. So let's do it before attaching anything
899 * to tree, set linkage between the newborn blocks, write them if sync is
900 * required, recheck the path, free and repeat if check fails, otherwise
901 * set the last missing link (that will protect us from any truncate-generated
902 * removals - all blocks on the path are immune now) and possibly force the
903 * write on the parent block.
904 * That has a nice additional property: no special recovery from the failed
905 * allocations is needed - we simply release blocks and do not touch anything
906 * reachable from inode.
908 * `handle' can be NULL if create == 0.
910 * return > 0, # of blocks mapped or allocated.
911 * return = 0, if plain lookup failed.
912 * return < 0, error case.
914 * The ext4_ind_get_blocks() function should be called with
915 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
916 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
917 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
920 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
921 struct ext4_map_blocks *map,
925 ext4_lblk_t offsets[4];
930 int blocks_to_boundary = 0;
933 ext4_fsblk_t first_block = 0;
935 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
936 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
937 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
938 &blocks_to_boundary);
943 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
945 /* Simplest case - block found, no allocation needed */
947 first_block = le32_to_cpu(chain[depth - 1].key);
950 while (count < map->m_len && count <= blocks_to_boundary) {
953 blk = le32_to_cpu(*(chain[depth-1].p + count));
955 if (blk == first_block + count)
963 /* Next simple case - plain lookup or failed read of indirect block */
964 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
968 * Okay, we need to do block allocation.
970 goal = ext4_find_goal(inode, map->m_lblk, partial);
972 /* the number of blocks need to allocate for [d,t]indirect blocks */
973 indirect_blks = (chain + depth) - partial - 1;
976 * Next look up the indirect map to count the totoal number of
977 * direct blocks to allocate for this branch.
979 count = ext4_blks_to_allocate(partial, indirect_blks,
980 map->m_len, blocks_to_boundary);
982 * Block out ext4_truncate while we alter the tree
984 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
986 offsets + (partial - chain), partial);
989 * The ext4_splice_branch call will free and forget any buffers
990 * on the new chain if there is a failure, but that risks using
991 * up transaction credits, especially for bitmaps where the
992 * credits cannot be returned. Can we handle this somehow? We
993 * may need to return -EAGAIN upwards in the worst case. --sct
996 err = ext4_splice_branch(handle, inode, map->m_lblk,
997 partial, indirect_blks, count);
1001 map->m_flags |= EXT4_MAP_NEW;
1003 ext4_update_inode_fsync_trans(handle, inode, 1);
1005 map->m_flags |= EXT4_MAP_MAPPED;
1006 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1008 if (count > blocks_to_boundary)
1009 map->m_flags |= EXT4_MAP_BOUNDARY;
1011 /* Clean up and exit */
1012 partial = chain + depth - 1; /* the whole chain */
1014 while (partial > chain) {
1015 BUFFER_TRACE(partial->bh, "call brelse");
1016 brelse(partial->bh);
1024 qsize_t *ext4_get_reserved_space(struct inode *inode)
1026 return &EXT4_I(inode)->i_reserved_quota;
1031 * Calculate the number of metadata blocks need to reserve
1032 * to allocate a new block at @lblocks for non extent file based file
1034 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1037 struct ext4_inode_info *ei = EXT4_I(inode);
1038 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1041 if (lblock < EXT4_NDIR_BLOCKS)
1044 lblock -= EXT4_NDIR_BLOCKS;
1046 if (ei->i_da_metadata_calc_len &&
1047 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1048 ei->i_da_metadata_calc_len++;
1051 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1052 ei->i_da_metadata_calc_len = 1;
1053 blk_bits = order_base_2(lblock);
1054 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1058 * Calculate the number of metadata blocks need to reserve
1059 * to allocate a block located at @lblock
1061 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1063 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1064 return ext4_ext_calc_metadata_amount(inode, lblock);
1066 return ext4_indirect_calc_metadata_amount(inode, lblock);
1070 * Called with i_data_sem down, which is important since we can call
1071 * ext4_discard_preallocations() from here.
1073 void ext4_da_update_reserve_space(struct inode *inode,
1074 int used, int quota_claim)
1076 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1077 struct ext4_inode_info *ei = EXT4_I(inode);
1079 spin_lock(&ei->i_block_reservation_lock);
1080 trace_ext4_da_update_reserve_space(inode, used);
1081 if (unlikely(used > ei->i_reserved_data_blocks)) {
1082 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1083 "with only %d reserved data blocks\n",
1084 __func__, inode->i_ino, used,
1085 ei->i_reserved_data_blocks);
1087 used = ei->i_reserved_data_blocks;
1090 /* Update per-inode reservations */
1091 ei->i_reserved_data_blocks -= used;
1092 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1093 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1094 used + ei->i_allocated_meta_blocks);
1095 ei->i_allocated_meta_blocks = 0;
1097 if (ei->i_reserved_data_blocks == 0) {
1099 * We can release all of the reserved metadata blocks
1100 * only when we have written all of the delayed
1101 * allocation blocks.
1103 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1104 ei->i_reserved_meta_blocks);
1105 ei->i_reserved_meta_blocks = 0;
1106 ei->i_da_metadata_calc_len = 0;
1108 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1110 /* Update quota subsystem for data blocks */
1112 dquot_claim_block(inode, used);
1115 * We did fallocate with an offset that is already delayed
1116 * allocated. So on delayed allocated writeback we should
1117 * not re-claim the quota for fallocated blocks.
1119 dquot_release_reservation_block(inode, used);
1123 * If we have done all the pending block allocations and if
1124 * there aren't any writers on the inode, we can discard the
1125 * inode's preallocations.
1127 if ((ei->i_reserved_data_blocks == 0) &&
1128 (atomic_read(&inode->i_writecount) == 0))
1129 ext4_discard_preallocations(inode);
1132 static int check_block_validity(struct inode *inode, const char *msg,
1133 sector_t logical, sector_t phys, int len)
1135 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1136 __ext4_error(inode->i_sb, msg,
1137 "inode #%lu logical block %llu mapped to %llu "
1138 "(size %d)", inode->i_ino,
1139 (unsigned long long) logical,
1140 (unsigned long long) phys, len);
1147 * Return the number of contiguous dirty pages in a given inode
1148 * starting at page frame idx.
1150 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1151 unsigned int max_pages)
1153 struct address_space *mapping = inode->i_mapping;
1155 struct pagevec pvec;
1157 int i, nr_pages, done = 0;
1161 pagevec_init(&pvec, 0);
1164 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1165 PAGECACHE_TAG_DIRTY,
1166 (pgoff_t)PAGEVEC_SIZE);
1169 for (i = 0; i < nr_pages; i++) {
1170 struct page *page = pvec.pages[i];
1171 struct buffer_head *bh, *head;
1174 if (unlikely(page->mapping != mapping) ||
1176 PageWriteback(page) ||
1177 page->index != idx) {
1182 if (page_has_buffers(page)) {
1183 bh = head = page_buffers(page);
1185 if (!buffer_delay(bh) &&
1186 !buffer_unwritten(bh))
1188 bh = bh->b_this_page;
1189 } while (!done && (bh != head));
1196 if (num >= max_pages)
1199 pagevec_release(&pvec);
1205 * The ext4_map_blocks() function tries to look up the requested blocks,
1206 * and returns if the blocks are already mapped.
1208 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1209 * and store the allocated blocks in the result buffer head and mark it
1212 * If file type is extents based, it will call ext4_ext_map_blocks(),
1213 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1216 * On success, it returns the number of blocks being mapped or allocate.
1217 * if create==0 and the blocks are pre-allocated and uninitialized block,
1218 * the result buffer head is unmapped. If the create ==1, it will make sure
1219 * the buffer head is mapped.
1221 * It returns 0 if plain look up failed (blocks have not been allocated), in
1222 * that casem, buffer head is unmapped
1224 * It returns the error in case of allocation failure.
1226 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1227 struct ext4_map_blocks *map, int flags)
1232 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1233 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1234 (unsigned long) map->m_lblk);
1236 * Try to see if we can get the block without requesting a new
1237 * file system block.
1239 down_read((&EXT4_I(inode)->i_data_sem));
1240 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1241 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1243 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1245 up_read((&EXT4_I(inode)->i_data_sem));
1247 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1248 int ret = check_block_validity(inode, "file system corruption",
1249 map->m_lblk, map->m_pblk, retval);
1254 /* If it is only a block(s) look up */
1255 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1259 * Returns if the blocks have already allocated
1261 * Note that if blocks have been preallocated
1262 * ext4_ext_get_block() returns th create = 0
1263 * with buffer head unmapped.
1265 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1269 * When we call get_blocks without the create flag, the
1270 * BH_Unwritten flag could have gotten set if the blocks
1271 * requested were part of a uninitialized extent. We need to
1272 * clear this flag now that we are committed to convert all or
1273 * part of the uninitialized extent to be an initialized
1274 * extent. This is because we need to avoid the combination
1275 * of BH_Unwritten and BH_Mapped flags being simultaneously
1276 * set on the buffer_head.
1278 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1281 * New blocks allocate and/or writing to uninitialized extent
1282 * will possibly result in updating i_data, so we take
1283 * the write lock of i_data_sem, and call get_blocks()
1284 * with create == 1 flag.
1286 down_write((&EXT4_I(inode)->i_data_sem));
1289 * if the caller is from delayed allocation writeout path
1290 * we have already reserved fs blocks for allocation
1291 * let the underlying get_block() function know to
1292 * avoid double accounting
1294 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1295 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1297 * We need to check for EXT4 here because migrate
1298 * could have changed the inode type in between
1300 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1301 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1303 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1305 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1307 * We allocated new blocks which will result in
1308 * i_data's format changing. Force the migrate
1309 * to fail by clearing migrate flags
1311 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1315 * Update reserved blocks/metadata blocks after successful
1316 * block allocation which had been deferred till now. We don't
1317 * support fallocate for non extent files. So we can update
1318 * reserve space here.
1321 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1322 ext4_da_update_reserve_space(inode, retval, 1);
1324 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1325 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1327 up_write((&EXT4_I(inode)->i_data_sem));
1328 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1329 int ret = check_block_validity(inode, "file system "
1330 "corruption after allocation",
1331 map->m_lblk, map->m_pblk,
1339 /* Maximum number of blocks we map for direct IO at once. */
1340 #define DIO_MAX_BLOCKS 4096
1342 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1343 struct buffer_head *bh, int flags)
1345 handle_t *handle = ext4_journal_current_handle();
1346 struct ext4_map_blocks map;
1347 int ret = 0, started = 0;
1350 map.m_lblk = iblock;
1351 map.m_len = bh->b_size >> inode->i_blkbits;
1353 if (flags && !handle) {
1354 /* Direct IO write... */
1355 if (map.m_len > DIO_MAX_BLOCKS)
1356 map.m_len = DIO_MAX_BLOCKS;
1357 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1358 handle = ext4_journal_start(inode, dio_credits);
1359 if (IS_ERR(handle)) {
1360 ret = PTR_ERR(handle);
1366 ret = ext4_map_blocks(handle, inode, &map, flags);
1368 map_bh(bh, inode->i_sb, map.m_pblk);
1369 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1370 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1374 ext4_journal_stop(handle);
1378 int ext4_get_block(struct inode *inode, sector_t iblock,
1379 struct buffer_head *bh, int create)
1381 return _ext4_get_block(inode, iblock, bh,
1382 create ? EXT4_GET_BLOCKS_CREATE : 0);
1386 * `handle' can be NULL if create is zero
1388 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1389 ext4_lblk_t block, int create, int *errp)
1391 struct ext4_map_blocks map;
1392 struct buffer_head *bh;
1395 J_ASSERT(handle != NULL || create == 0);
1399 err = ext4_map_blocks(handle, inode, &map,
1400 create ? EXT4_GET_BLOCKS_CREATE : 0);
1408 bh = sb_getblk(inode->i_sb, map.m_pblk);
1413 if (map.m_flags & EXT4_MAP_NEW) {
1414 J_ASSERT(create != 0);
1415 J_ASSERT(handle != NULL);
1418 * Now that we do not always journal data, we should
1419 * keep in mind whether this should always journal the
1420 * new buffer as metadata. For now, regular file
1421 * writes use ext4_get_block instead, so it's not a
1425 BUFFER_TRACE(bh, "call get_create_access");
1426 fatal = ext4_journal_get_create_access(handle, bh);
1427 if (!fatal && !buffer_uptodate(bh)) {
1428 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1429 set_buffer_uptodate(bh);
1432 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1433 err = ext4_handle_dirty_metadata(handle, inode, bh);
1437 BUFFER_TRACE(bh, "not a new buffer");
1447 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1448 ext4_lblk_t block, int create, int *err)
1450 struct buffer_head *bh;
1452 bh = ext4_getblk(handle, inode, block, create, err);
1455 if (buffer_uptodate(bh))
1457 ll_rw_block(READ_META, 1, &bh);
1459 if (buffer_uptodate(bh))
1466 static int walk_page_buffers(handle_t *handle,
1467 struct buffer_head *head,
1471 int (*fn)(handle_t *handle,
1472 struct buffer_head *bh))
1474 struct buffer_head *bh;
1475 unsigned block_start, block_end;
1476 unsigned blocksize = head->b_size;
1478 struct buffer_head *next;
1480 for (bh = head, block_start = 0;
1481 ret == 0 && (bh != head || !block_start);
1482 block_start = block_end, bh = next) {
1483 next = bh->b_this_page;
1484 block_end = block_start + blocksize;
1485 if (block_end <= from || block_start >= to) {
1486 if (partial && !buffer_uptodate(bh))
1490 err = (*fn)(handle, bh);
1498 * To preserve ordering, it is essential that the hole instantiation and
1499 * the data write be encapsulated in a single transaction. We cannot
1500 * close off a transaction and start a new one between the ext4_get_block()
1501 * and the commit_write(). So doing the jbd2_journal_start at the start of
1502 * prepare_write() is the right place.
1504 * Also, this function can nest inside ext4_writepage() ->
1505 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1506 * has generated enough buffer credits to do the whole page. So we won't
1507 * block on the journal in that case, which is good, because the caller may
1510 * By accident, ext4 can be reentered when a transaction is open via
1511 * quota file writes. If we were to commit the transaction while thus
1512 * reentered, there can be a deadlock - we would be holding a quota
1513 * lock, and the commit would never complete if another thread had a
1514 * transaction open and was blocking on the quota lock - a ranking
1517 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1518 * will _not_ run commit under these circumstances because handle->h_ref
1519 * is elevated. We'll still have enough credits for the tiny quotafile
1522 static int do_journal_get_write_access(handle_t *handle,
1523 struct buffer_head *bh)
1525 if (!buffer_mapped(bh) || buffer_freed(bh))
1527 return ext4_journal_get_write_access(handle, bh);
1531 * Truncate blocks that were not used by write. We have to truncate the
1532 * pagecache as well so that corresponding buffers get properly unmapped.
1534 static void ext4_truncate_failed_write(struct inode *inode)
1536 truncate_inode_pages(inode->i_mapping, inode->i_size);
1537 ext4_truncate(inode);
1540 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1541 struct buffer_head *bh_result, int create);
1542 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1543 loff_t pos, unsigned len, unsigned flags,
1544 struct page **pagep, void **fsdata)
1546 struct inode *inode = mapping->host;
1547 int ret, needed_blocks;
1554 trace_ext4_write_begin(inode, pos, len, flags);
1556 * Reserve one block more for addition to orphan list in case
1557 * we allocate blocks but write fails for some reason
1559 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1560 index = pos >> PAGE_CACHE_SHIFT;
1561 from = pos & (PAGE_CACHE_SIZE - 1);
1565 handle = ext4_journal_start(inode, needed_blocks);
1566 if (IS_ERR(handle)) {
1567 ret = PTR_ERR(handle);
1571 /* We cannot recurse into the filesystem as the transaction is already
1573 flags |= AOP_FLAG_NOFS;
1575 page = grab_cache_page_write_begin(mapping, index, flags);
1577 ext4_journal_stop(handle);
1583 if (ext4_should_dioread_nolock(inode))
1584 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1585 fsdata, ext4_get_block_write);
1587 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1588 fsdata, ext4_get_block);
1590 if (!ret && ext4_should_journal_data(inode)) {
1591 ret = walk_page_buffers(handle, page_buffers(page),
1592 from, to, NULL, do_journal_get_write_access);
1597 page_cache_release(page);
1599 * block_write_begin may have instantiated a few blocks
1600 * outside i_size. Trim these off again. Don't need
1601 * i_size_read because we hold i_mutex.
1603 * Add inode to orphan list in case we crash before
1606 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1607 ext4_orphan_add(handle, inode);
1609 ext4_journal_stop(handle);
1610 if (pos + len > inode->i_size) {
1611 ext4_truncate_failed_write(inode);
1613 * If truncate failed early the inode might
1614 * still be on the orphan list; we need to
1615 * make sure the inode is removed from the
1616 * orphan list in that case.
1619 ext4_orphan_del(NULL, inode);
1623 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1629 /* For write_end() in data=journal mode */
1630 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1632 if (!buffer_mapped(bh) || buffer_freed(bh))
1634 set_buffer_uptodate(bh);
1635 return ext4_handle_dirty_metadata(handle, NULL, bh);
1638 static int ext4_generic_write_end(struct file *file,
1639 struct address_space *mapping,
1640 loff_t pos, unsigned len, unsigned copied,
1641 struct page *page, void *fsdata)
1643 int i_size_changed = 0;
1644 struct inode *inode = mapping->host;
1645 handle_t *handle = ext4_journal_current_handle();
1647 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1650 * No need to use i_size_read() here, the i_size
1651 * cannot change under us because we hold i_mutex.
1653 * But it's important to update i_size while still holding page lock:
1654 * page writeout could otherwise come in and zero beyond i_size.
1656 if (pos + copied > inode->i_size) {
1657 i_size_write(inode, pos + copied);
1661 if (pos + copied > EXT4_I(inode)->i_disksize) {
1662 /* We need to mark inode dirty even if
1663 * new_i_size is less that inode->i_size
1664 * bu greater than i_disksize.(hint delalloc)
1666 ext4_update_i_disksize(inode, (pos + copied));
1670 page_cache_release(page);
1673 * Don't mark the inode dirty under page lock. First, it unnecessarily
1674 * makes the holding time of page lock longer. Second, it forces lock
1675 * ordering of page lock and transaction start for journaling
1679 ext4_mark_inode_dirty(handle, inode);
1685 * We need to pick up the new inode size which generic_commit_write gave us
1686 * `file' can be NULL - eg, when called from page_symlink().
1688 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1689 * buffers are managed internally.
1691 static int ext4_ordered_write_end(struct file *file,
1692 struct address_space *mapping,
1693 loff_t pos, unsigned len, unsigned copied,
1694 struct page *page, void *fsdata)
1696 handle_t *handle = ext4_journal_current_handle();
1697 struct inode *inode = mapping->host;
1700 trace_ext4_ordered_write_end(inode, pos, len, copied);
1701 ret = ext4_jbd2_file_inode(handle, inode);
1704 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1707 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1708 /* if we have allocated more blocks and copied
1709 * less. We will have blocks allocated outside
1710 * inode->i_size. So truncate them
1712 ext4_orphan_add(handle, inode);
1716 ret2 = ext4_journal_stop(handle);
1720 if (pos + len > inode->i_size) {
1721 ext4_truncate_failed_write(inode);
1723 * If truncate failed early the inode might still be
1724 * on the orphan list; we need to make sure the inode
1725 * is removed from the orphan list in that case.
1728 ext4_orphan_del(NULL, inode);
1732 return ret ? ret : copied;
1735 static int ext4_writeback_write_end(struct file *file,
1736 struct address_space *mapping,
1737 loff_t pos, unsigned len, unsigned copied,
1738 struct page *page, void *fsdata)
1740 handle_t *handle = ext4_journal_current_handle();
1741 struct inode *inode = mapping->host;
1744 trace_ext4_writeback_write_end(inode, pos, len, copied);
1745 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1748 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1749 /* if we have allocated more blocks and copied
1750 * less. We will have blocks allocated outside
1751 * inode->i_size. So truncate them
1753 ext4_orphan_add(handle, inode);
1758 ret2 = ext4_journal_stop(handle);
1762 if (pos + len > inode->i_size) {
1763 ext4_truncate_failed_write(inode);
1765 * If truncate failed early the inode might still be
1766 * on the orphan list; we need to make sure the inode
1767 * is removed from the orphan list in that case.
1770 ext4_orphan_del(NULL, inode);
1773 return ret ? ret : copied;
1776 static int ext4_journalled_write_end(struct file *file,
1777 struct address_space *mapping,
1778 loff_t pos, unsigned len, unsigned copied,
1779 struct page *page, void *fsdata)
1781 handle_t *handle = ext4_journal_current_handle();
1782 struct inode *inode = mapping->host;
1788 trace_ext4_journalled_write_end(inode, pos, len, copied);
1789 from = pos & (PAGE_CACHE_SIZE - 1);
1793 if (!PageUptodate(page))
1795 page_zero_new_buffers(page, from+copied, to);
1798 ret = walk_page_buffers(handle, page_buffers(page), from,
1799 to, &partial, write_end_fn);
1801 SetPageUptodate(page);
1802 new_i_size = pos + copied;
1803 if (new_i_size > inode->i_size)
1804 i_size_write(inode, pos+copied);
1805 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1806 if (new_i_size > EXT4_I(inode)->i_disksize) {
1807 ext4_update_i_disksize(inode, new_i_size);
1808 ret2 = ext4_mark_inode_dirty(handle, inode);
1814 page_cache_release(page);
1815 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1816 /* if we have allocated more blocks and copied
1817 * less. We will have blocks allocated outside
1818 * inode->i_size. So truncate them
1820 ext4_orphan_add(handle, inode);
1822 ret2 = ext4_journal_stop(handle);
1825 if (pos + len > inode->i_size) {
1826 ext4_truncate_failed_write(inode);
1828 * If truncate failed early the inode might still be
1829 * on the orphan list; we need to make sure the inode
1830 * is removed from the orphan list in that case.
1833 ext4_orphan_del(NULL, inode);
1836 return ret ? ret : copied;
1840 * Reserve a single block located at lblock
1842 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1845 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1846 struct ext4_inode_info *ei = EXT4_I(inode);
1847 unsigned long md_needed;
1851 * recalculate the amount of metadata blocks to reserve
1852 * in order to allocate nrblocks
1853 * worse case is one extent per block
1856 spin_lock(&ei->i_block_reservation_lock);
1857 md_needed = ext4_calc_metadata_amount(inode, lblock);
1858 trace_ext4_da_reserve_space(inode, md_needed);
1859 spin_unlock(&ei->i_block_reservation_lock);
1862 * We will charge metadata quota at writeout time; this saves
1863 * us from metadata over-estimation, though we may go over by
1864 * a small amount in the end. Here we just reserve for data.
1866 ret = dquot_reserve_block(inode, 1);
1870 * We do still charge estimated metadata to the sb though;
1871 * we cannot afford to run out of free blocks.
1873 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1874 dquot_release_reservation_block(inode, 1);
1875 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1881 spin_lock(&ei->i_block_reservation_lock);
1882 ei->i_reserved_data_blocks++;
1883 ei->i_reserved_meta_blocks += md_needed;
1884 spin_unlock(&ei->i_block_reservation_lock);
1886 return 0; /* success */
1889 static void ext4_da_release_space(struct inode *inode, int to_free)
1891 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1892 struct ext4_inode_info *ei = EXT4_I(inode);
1895 return; /* Nothing to release, exit */
1897 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1899 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1901 * if there aren't enough reserved blocks, then the
1902 * counter is messed up somewhere. Since this
1903 * function is called from invalidate page, it's
1904 * harmless to return without any action.
1906 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1907 "ino %lu, to_free %d with only %d reserved "
1908 "data blocks\n", inode->i_ino, to_free,
1909 ei->i_reserved_data_blocks);
1911 to_free = ei->i_reserved_data_blocks;
1913 ei->i_reserved_data_blocks -= to_free;
1915 if (ei->i_reserved_data_blocks == 0) {
1917 * We can release all of the reserved metadata blocks
1918 * only when we have written all of the delayed
1919 * allocation blocks.
1921 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1922 ei->i_reserved_meta_blocks);
1923 ei->i_reserved_meta_blocks = 0;
1924 ei->i_da_metadata_calc_len = 0;
1927 /* update fs dirty data blocks counter */
1928 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1930 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1932 dquot_release_reservation_block(inode, to_free);
1935 static void ext4_da_page_release_reservation(struct page *page,
1936 unsigned long offset)
1939 struct buffer_head *head, *bh;
1940 unsigned int curr_off = 0;
1942 head = page_buffers(page);
1945 unsigned int next_off = curr_off + bh->b_size;
1947 if ((offset <= curr_off) && (buffer_delay(bh))) {
1949 clear_buffer_delay(bh);
1951 curr_off = next_off;
1952 } while ((bh = bh->b_this_page) != head);
1953 ext4_da_release_space(page->mapping->host, to_release);
1957 * Delayed allocation stuff
1961 * mpage_da_submit_io - walks through extent of pages and try to write
1962 * them with writepage() call back
1964 * @mpd->inode: inode
1965 * @mpd->first_page: first page of the extent
1966 * @mpd->next_page: page after the last page of the extent
1968 * By the time mpage_da_submit_io() is called we expect all blocks
1969 * to be allocated. this may be wrong if allocation failed.
1971 * As pages are already locked by write_cache_pages(), we can't use it
1973 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1976 struct pagevec pvec;
1977 unsigned long index, end;
1978 int ret = 0, err, nr_pages, i;
1979 struct inode *inode = mpd->inode;
1980 struct address_space *mapping = inode->i_mapping;
1982 BUG_ON(mpd->next_page <= mpd->first_page);
1984 * We need to start from the first_page to the next_page - 1
1985 * to make sure we also write the mapped dirty buffer_heads.
1986 * If we look at mpd->b_blocknr we would only be looking
1987 * at the currently mapped buffer_heads.
1989 index = mpd->first_page;
1990 end = mpd->next_page - 1;
1992 pagevec_init(&pvec, 0);
1993 while (index <= end) {
1994 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1997 for (i = 0; i < nr_pages; i++) {
1998 struct page *page = pvec.pages[i];
2000 index = page->index;
2005 BUG_ON(!PageLocked(page));
2006 BUG_ON(PageWriteback(page));
2008 pages_skipped = mpd->wbc->pages_skipped;
2009 err = mapping->a_ops->writepage(page, mpd->wbc);
2010 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2012 * have successfully written the page
2013 * without skipping the same
2015 mpd->pages_written++;
2017 * In error case, we have to continue because
2018 * remaining pages are still locked
2019 * XXX: unlock and re-dirty them?
2024 pagevec_release(&pvec);
2030 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2032 * the function goes through all passed space and put actual disk
2033 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2035 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2036 struct ext4_map_blocks *map)
2038 struct inode *inode = mpd->inode;
2039 struct address_space *mapping = inode->i_mapping;
2040 int blocks = map->m_len;
2041 sector_t pblock = map->m_pblk, cur_logical;
2042 struct buffer_head *head, *bh;
2044 struct pagevec pvec;
2047 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2048 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2049 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2051 pagevec_init(&pvec, 0);
2053 while (index <= end) {
2054 /* XXX: optimize tail */
2055 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2058 for (i = 0; i < nr_pages; i++) {
2059 struct page *page = pvec.pages[i];
2061 index = page->index;
2066 BUG_ON(!PageLocked(page));
2067 BUG_ON(PageWriteback(page));
2068 BUG_ON(!page_has_buffers(page));
2070 bh = page_buffers(page);
2073 /* skip blocks out of the range */
2075 if (cur_logical >= map->m_lblk)
2078 } while ((bh = bh->b_this_page) != head);
2081 if (cur_logical >= map->m_lblk + blocks)
2084 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2086 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2088 if (buffer_delay(bh)) {
2089 clear_buffer_delay(bh);
2090 bh->b_blocknr = pblock;
2093 * unwritten already should have
2094 * blocknr assigned. Verify that
2096 clear_buffer_unwritten(bh);
2097 BUG_ON(bh->b_blocknr != pblock);
2100 } else if (buffer_mapped(bh))
2101 BUG_ON(bh->b_blocknr != pblock);
2103 if (map->m_flags & EXT4_MAP_UNINIT)
2104 set_buffer_uninit(bh);
2107 } while ((bh = bh->b_this_page) != head);
2109 pagevec_release(&pvec);
2114 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2115 sector_t logical, long blk_cnt)
2119 struct pagevec pvec;
2120 struct inode *inode = mpd->inode;
2121 struct address_space *mapping = inode->i_mapping;
2123 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2124 end = (logical + blk_cnt - 1) >>
2125 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2126 while (index <= end) {
2127 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2130 for (i = 0; i < nr_pages; i++) {
2131 struct page *page = pvec.pages[i];
2132 if (page->index > end)
2134 BUG_ON(!PageLocked(page));
2135 BUG_ON(PageWriteback(page));
2136 block_invalidatepage(page, 0);
2137 ClearPageUptodate(page);
2140 index = pvec.pages[nr_pages - 1]->index + 1;
2141 pagevec_release(&pvec);
2146 static void ext4_print_free_blocks(struct inode *inode)
2148 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2149 printk(KERN_CRIT "Total free blocks count %lld\n",
2150 ext4_count_free_blocks(inode->i_sb));
2151 printk(KERN_CRIT "Free/Dirty block details\n");
2152 printk(KERN_CRIT "free_blocks=%lld\n",
2153 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2154 printk(KERN_CRIT "dirty_blocks=%lld\n",
2155 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2156 printk(KERN_CRIT "Block reservation details\n");
2157 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2158 EXT4_I(inode)->i_reserved_data_blocks);
2159 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2160 EXT4_I(inode)->i_reserved_meta_blocks);
2165 * mpage_da_map_blocks - go through given space
2167 * @mpd - bh describing space
2169 * The function skips space we know is already mapped to disk blocks.
2172 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2174 int err, blks, get_blocks_flags;
2175 struct ext4_map_blocks map;
2176 sector_t next = mpd->b_blocknr;
2177 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2178 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2179 handle_t *handle = NULL;
2182 * We consider only non-mapped and non-allocated blocks
2184 if ((mpd->b_state & (1 << BH_Mapped)) &&
2185 !(mpd->b_state & (1 << BH_Delay)) &&
2186 !(mpd->b_state & (1 << BH_Unwritten)))
2190 * If we didn't accumulate anything to write simply return
2195 handle = ext4_journal_current_handle();
2199 * Call ext4_get_blocks() to allocate any delayed allocation
2200 * blocks, or to convert an uninitialized extent to be
2201 * initialized (in the case where we have written into
2202 * one or more preallocated blocks).
2204 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2205 * indicate that we are on the delayed allocation path. This
2206 * affects functions in many different parts of the allocation
2207 * call path. This flag exists primarily because we don't
2208 * want to change *many* call functions, so ext4_get_blocks()
2209 * will set the magic i_delalloc_reserved_flag once the
2210 * inode's allocation semaphore is taken.
2212 * If the blocks in questions were delalloc blocks, set
2213 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2214 * variables are updated after the blocks have been allocated.
2217 map.m_len = max_blocks;
2218 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2219 if (ext4_should_dioread_nolock(mpd->inode))
2220 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2221 if (mpd->b_state & (1 << BH_Delay))
2222 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2224 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2228 * If get block returns with error we simply
2229 * return. Later writepage will redirty the page and
2230 * writepages will find the dirty page again
2235 if (err == -ENOSPC &&
2236 ext4_count_free_blocks(mpd->inode->i_sb)) {
2242 * get block failure will cause us to loop in
2243 * writepages, because a_ops->writepage won't be able
2244 * to make progress. The page will be redirtied by
2245 * writepage and writepages will again try to write
2248 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2249 "delayed block allocation failed for inode %lu at "
2250 "logical offset %llu with max blocks %zd with "
2251 "error %d", mpd->inode->i_ino,
2252 (unsigned long long) next,
2253 mpd->b_size >> mpd->inode->i_blkbits, err);
2254 printk(KERN_CRIT "This should not happen!! "
2255 "Data will be lost\n");
2256 if (err == -ENOSPC) {
2257 ext4_print_free_blocks(mpd->inode);
2259 /* invalidate all the pages */
2260 ext4_da_block_invalidatepages(mpd, next,
2261 mpd->b_size >> mpd->inode->i_blkbits);
2266 if (map.m_flags & EXT4_MAP_NEW) {
2267 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2270 for (i = 0; i < map.m_len; i++)
2271 unmap_underlying_metadata(bdev, map.m_pblk + i);
2275 * If blocks are delayed marked, we need to
2276 * put actual blocknr and drop delayed bit
2278 if ((mpd->b_state & (1 << BH_Delay)) ||
2279 (mpd->b_state & (1 << BH_Unwritten)))
2280 mpage_put_bnr_to_bhs(mpd, &map);
2282 if (ext4_should_order_data(mpd->inode)) {
2283 err = ext4_jbd2_file_inode(handle, mpd->inode);
2289 * Update on-disk size along with block allocation.
2291 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2292 if (disksize > i_size_read(mpd->inode))
2293 disksize = i_size_read(mpd->inode);
2294 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2295 ext4_update_i_disksize(mpd->inode, disksize);
2296 return ext4_mark_inode_dirty(handle, mpd->inode);
2302 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2303 (1 << BH_Delay) | (1 << BH_Unwritten))
2306 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2308 * @mpd->lbh - extent of blocks
2309 * @logical - logical number of the block in the file
2310 * @bh - bh of the block (used to access block's state)
2312 * the function is used to collect contig. blocks in same state
2314 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2315 sector_t logical, size_t b_size,
2316 unsigned long b_state)
2319 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2322 * XXX Don't go larger than mballoc is willing to allocate
2323 * This is a stopgap solution. We eventually need to fold
2324 * mpage_da_submit_io() into this function and then call
2325 * ext4_get_blocks() multiple times in a loop
2327 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2330 /* check if thereserved journal credits might overflow */
2331 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2332 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2334 * With non-extent format we are limited by the journal
2335 * credit available. Total credit needed to insert
2336 * nrblocks contiguous blocks is dependent on the
2337 * nrblocks. So limit nrblocks.
2340 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2341 EXT4_MAX_TRANS_DATA) {
2343 * Adding the new buffer_head would make it cross the
2344 * allowed limit for which we have journal credit
2345 * reserved. So limit the new bh->b_size
2347 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2348 mpd->inode->i_blkbits;
2349 /* we will do mpage_da_submit_io in the next loop */
2353 * First block in the extent
2355 if (mpd->b_size == 0) {
2356 mpd->b_blocknr = logical;
2357 mpd->b_size = b_size;
2358 mpd->b_state = b_state & BH_FLAGS;
2362 next = mpd->b_blocknr + nrblocks;
2364 * Can we merge the block to our big extent?
2366 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2367 mpd->b_size += b_size;
2373 * We couldn't merge the block to our extent, so we
2374 * need to flush current extent and start new one
2376 if (mpage_da_map_blocks(mpd) == 0)
2377 mpage_da_submit_io(mpd);
2382 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2384 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2388 * __mpage_da_writepage - finds extent of pages and blocks
2390 * @page: page to consider
2391 * @wbc: not used, we just follow rules
2394 * The function finds extents of pages and scan them for all blocks.
2396 static int __mpage_da_writepage(struct page *page,
2397 struct writeback_control *wbc, void *data)
2399 struct mpage_da_data *mpd = data;
2400 struct inode *inode = mpd->inode;
2401 struct buffer_head *bh, *head;
2405 * Can we merge this page to current extent?
2407 if (mpd->next_page != page->index) {
2409 * Nope, we can't. So, we map non-allocated blocks
2410 * and start IO on them using writepage()
2412 if (mpd->next_page != mpd->first_page) {
2413 if (mpage_da_map_blocks(mpd) == 0)
2414 mpage_da_submit_io(mpd);
2416 * skip rest of the page in the page_vec
2419 redirty_page_for_writepage(wbc, page);
2421 return MPAGE_DA_EXTENT_TAIL;
2425 * Start next extent of pages ...
2427 mpd->first_page = page->index;
2437 mpd->next_page = page->index + 1;
2438 logical = (sector_t) page->index <<
2439 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2441 if (!page_has_buffers(page)) {
2442 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2443 (1 << BH_Dirty) | (1 << BH_Uptodate));
2445 return MPAGE_DA_EXTENT_TAIL;
2448 * Page with regular buffer heads, just add all dirty ones
2450 head = page_buffers(page);
2453 BUG_ON(buffer_locked(bh));
2455 * We need to try to allocate
2456 * unmapped blocks in the same page.
2457 * Otherwise we won't make progress
2458 * with the page in ext4_writepage
2460 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2461 mpage_add_bh_to_extent(mpd, logical,
2465 return MPAGE_DA_EXTENT_TAIL;
2466 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2468 * mapped dirty buffer. We need to update
2469 * the b_state because we look at
2470 * b_state in mpage_da_map_blocks. We don't
2471 * update b_size because if we find an
2472 * unmapped buffer_head later we need to
2473 * use the b_state flag of that buffer_head.
2475 if (mpd->b_size == 0)
2476 mpd->b_state = bh->b_state & BH_FLAGS;
2479 } while ((bh = bh->b_this_page) != head);
2486 * This is a special get_blocks_t callback which is used by
2487 * ext4_da_write_begin(). It will either return mapped block or
2488 * reserve space for a single block.
2490 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2491 * We also have b_blocknr = -1 and b_bdev initialized properly
2493 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2494 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2495 * initialized properly.
2497 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2498 struct buffer_head *bh, int create)
2500 struct ext4_map_blocks map;
2502 sector_t invalid_block = ~((sector_t) 0xffff);
2504 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2507 BUG_ON(create == 0);
2508 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2510 map.m_lblk = iblock;
2514 * first, we need to know whether the block is allocated already
2515 * preallocated blocks are unmapped but should treated
2516 * the same as allocated blocks.
2518 ret = ext4_map_blocks(NULL, inode, &map, 0);
2522 if (buffer_delay(bh))
2523 return 0; /* Not sure this could or should happen */
2525 * XXX: __block_prepare_write() unmaps passed block,
2528 ret = ext4_da_reserve_space(inode, iblock);
2530 /* not enough space to reserve */
2533 map_bh(bh, inode->i_sb, invalid_block);
2535 set_buffer_delay(bh);
2539 map_bh(bh, inode->i_sb, map.m_pblk);
2540 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2542 if (buffer_unwritten(bh)) {
2543 /* A delayed write to unwritten bh should be marked
2544 * new and mapped. Mapped ensures that we don't do
2545 * get_block multiple times when we write to the same
2546 * offset and new ensures that we do proper zero out
2547 * for partial write.
2550 set_buffer_mapped(bh);
2556 * This function is used as a standard get_block_t calback function
2557 * when there is no desire to allocate any blocks. It is used as a
2558 * callback function for block_prepare_write(), nobh_writepage(), and
2559 * block_write_full_page(). These functions should only try to map a
2560 * single block at a time.
2562 * Since this function doesn't do block allocations even if the caller
2563 * requests it by passing in create=1, it is critically important that
2564 * any caller checks to make sure that any buffer heads are returned
2565 * by this function are either all already mapped or marked for
2566 * delayed allocation before calling nobh_writepage() or
2567 * block_write_full_page(). Otherwise, b_blocknr could be left
2568 * unitialized, and the page write functions will be taken by
2571 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2572 struct buffer_head *bh_result, int create)
2574 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2575 return _ext4_get_block(inode, iblock, bh_result, 0);
2578 static int bget_one(handle_t *handle, struct buffer_head *bh)
2584 static int bput_one(handle_t *handle, struct buffer_head *bh)
2590 static int __ext4_journalled_writepage(struct page *page,
2593 struct address_space *mapping = page->mapping;
2594 struct inode *inode = mapping->host;
2595 struct buffer_head *page_bufs;
2596 handle_t *handle = NULL;
2600 page_bufs = page_buffers(page);
2602 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2603 /* As soon as we unlock the page, it can go away, but we have
2604 * references to buffers so we are safe */
2607 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2608 if (IS_ERR(handle)) {
2609 ret = PTR_ERR(handle);
2613 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2614 do_journal_get_write_access);
2616 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2620 err = ext4_journal_stop(handle);
2624 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2625 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2630 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2631 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2634 * Note that we don't need to start a transaction unless we're journaling data
2635 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2636 * need to file the inode to the transaction's list in ordered mode because if
2637 * we are writing back data added by write(), the inode is already there and if
2638 * we are writing back data modified via mmap(), noone guarantees in which
2639 * transaction the data will hit the disk. In case we are journaling data, we
2640 * cannot start transaction directly because transaction start ranks above page
2641 * lock so we have to do some magic.
2643 * This function can get called via...
2644 * - ext4_da_writepages after taking page lock (have journal handle)
2645 * - journal_submit_inode_data_buffers (no journal handle)
2646 * - shrink_page_list via pdflush (no journal handle)
2647 * - grab_page_cache when doing write_begin (have journal handle)
2649 * We don't do any block allocation in this function. If we have page with
2650 * multiple blocks we need to write those buffer_heads that are mapped. This
2651 * is important for mmaped based write. So if we do with blocksize 1K
2652 * truncate(f, 1024);
2653 * a = mmap(f, 0, 4096);
2655 * truncate(f, 4096);
2656 * we have in the page first buffer_head mapped via page_mkwrite call back
2657 * but other bufer_heads would be unmapped but dirty(dirty done via the
2658 * do_wp_page). So writepage should write the first block. If we modify
2659 * the mmap area beyond 1024 we will again get a page_fault and the
2660 * page_mkwrite callback will do the block allocation and mark the
2661 * buffer_heads mapped.
2663 * We redirty the page if we have any buffer_heads that is either delay or
2664 * unwritten in the page.
2666 * We can get recursively called as show below.
2668 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2671 * But since we don't do any block allocation we should not deadlock.
2672 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2674 static int ext4_writepage(struct page *page,
2675 struct writeback_control *wbc)
2680 struct buffer_head *page_bufs = NULL;
2681 struct inode *inode = page->mapping->host;
2683 trace_ext4_writepage(inode, page);
2684 size = i_size_read(inode);
2685 if (page->index == size >> PAGE_CACHE_SHIFT)
2686 len = size & ~PAGE_CACHE_MASK;
2688 len = PAGE_CACHE_SIZE;
2690 if (page_has_buffers(page)) {
2691 page_bufs = page_buffers(page);
2692 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2693 ext4_bh_delay_or_unwritten)) {
2695 * We don't want to do block allocation
2696 * So redirty the page and return
2697 * We may reach here when we do a journal commit
2698 * via journal_submit_inode_data_buffers.
2699 * If we don't have mapping block we just ignore
2700 * them. We can also reach here via shrink_page_list
2702 redirty_page_for_writepage(wbc, page);
2708 * The test for page_has_buffers() is subtle:
2709 * We know the page is dirty but it lost buffers. That means
2710 * that at some moment in time after write_begin()/write_end()
2711 * has been called all buffers have been clean and thus they
2712 * must have been written at least once. So they are all
2713 * mapped and we can happily proceed with mapping them
2714 * and writing the page.
2716 * Try to initialize the buffer_heads and check whether
2717 * all are mapped and non delay. We don't want to
2718 * do block allocation here.
2720 ret = block_prepare_write(page, 0, len,
2721 noalloc_get_block_write);
2723 page_bufs = page_buffers(page);
2724 /* check whether all are mapped and non delay */
2725 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2726 ext4_bh_delay_or_unwritten)) {
2727 redirty_page_for_writepage(wbc, page);
2733 * We can't do block allocation here
2734 * so just redity the page and unlock
2737 redirty_page_for_writepage(wbc, page);
2741 /* now mark the buffer_heads as dirty and uptodate */
2742 block_commit_write(page, 0, len);
2745 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2747 * It's mmapped pagecache. Add buffers and journal it. There
2748 * doesn't seem much point in redirtying the page here.
2750 ClearPageChecked(page);
2751 return __ext4_journalled_writepage(page, len);
2754 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2755 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2756 else if (page_bufs && buffer_uninit(page_bufs)) {
2757 ext4_set_bh_endio(page_bufs, inode);
2758 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2759 wbc, ext4_end_io_buffer_write);
2761 ret = block_write_full_page(page, noalloc_get_block_write,
2768 * This is called via ext4_da_writepages() to
2769 * calulate the total number of credits to reserve to fit
2770 * a single extent allocation into a single transaction,
2771 * ext4_da_writpeages() will loop calling this before
2772 * the block allocation.
2775 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2777 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2780 * With non-extent format the journal credit needed to
2781 * insert nrblocks contiguous block is dependent on
2782 * number of contiguous block. So we will limit
2783 * number of contiguous block to a sane value
2785 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) &&
2786 (max_blocks > EXT4_MAX_TRANS_DATA))
2787 max_blocks = EXT4_MAX_TRANS_DATA;
2789 return ext4_chunk_trans_blocks(inode, max_blocks);
2793 * write_cache_pages_da - walk the list of dirty pages of the given
2794 * address space and call the callback function (which usually writes
2797 * This is a forked version of write_cache_pages(). Differences:
2798 * Range cyclic is ignored.
2799 * no_nrwrite_index_update is always presumed true
2801 static int write_cache_pages_da(struct address_space *mapping,
2802 struct writeback_control *wbc,
2803 struct mpage_da_data *mpd)
2807 struct pagevec pvec;
2810 pgoff_t end; /* Inclusive */
2811 long nr_to_write = wbc->nr_to_write;
2813 pagevec_init(&pvec, 0);
2814 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2815 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2817 while (!done && (index <= end)) {
2820 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2821 PAGECACHE_TAG_DIRTY,
2822 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2826 for (i = 0; i < nr_pages; i++) {
2827 struct page *page = pvec.pages[i];
2830 * At this point, the page may be truncated or
2831 * invalidated (changing page->mapping to NULL), or
2832 * even swizzled back from swapper_space to tmpfs file
2833 * mapping. However, page->index will not change
2834 * because we have a reference on the page.
2836 if (page->index > end) {
2844 * Page truncated or invalidated. We can freely skip it
2845 * then, even for data integrity operations: the page
2846 * has disappeared concurrently, so there could be no
2847 * real expectation of this data interity operation
2848 * even if there is now a new, dirty page at the same
2849 * pagecache address.
2851 if (unlikely(page->mapping != mapping)) {
2857 if (!PageDirty(page)) {
2858 /* someone wrote it for us */
2859 goto continue_unlock;
2862 if (PageWriteback(page)) {
2863 if (wbc->sync_mode != WB_SYNC_NONE)
2864 wait_on_page_writeback(page);
2866 goto continue_unlock;
2869 BUG_ON(PageWriteback(page));
2870 if (!clear_page_dirty_for_io(page))
2871 goto continue_unlock;
2873 ret = __mpage_da_writepage(page, wbc, mpd);
2874 if (unlikely(ret)) {
2875 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2884 if (nr_to_write > 0) {
2886 if (nr_to_write == 0 &&
2887 wbc->sync_mode == WB_SYNC_NONE) {
2889 * We stop writing back only if we are
2890 * not doing integrity sync. In case of
2891 * integrity sync we have to keep going
2892 * because someone may be concurrently
2893 * dirtying pages, and we might have
2894 * synced a lot of newly appeared dirty
2895 * pages, but have not synced all of the
2903 pagevec_release(&pvec);
2910 static int ext4_da_writepages(struct address_space *mapping,
2911 struct writeback_control *wbc)
2914 int range_whole = 0;
2915 handle_t *handle = NULL;
2916 struct mpage_da_data mpd;
2917 struct inode *inode = mapping->host;
2918 int pages_written = 0;
2920 unsigned int max_pages;
2921 int range_cyclic, cycled = 1, io_done = 0;
2922 int needed_blocks, ret = 0;
2923 long desired_nr_to_write, nr_to_writebump = 0;
2924 loff_t range_start = wbc->range_start;
2925 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2927 trace_ext4_da_writepages(inode, wbc);
2930 * No pages to write? This is mainly a kludge to avoid starting
2931 * a transaction for special inodes like journal inode on last iput()
2932 * because that could violate lock ordering on umount
2934 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2938 * If the filesystem has aborted, it is read-only, so return
2939 * right away instead of dumping stack traces later on that
2940 * will obscure the real source of the problem. We test
2941 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2942 * the latter could be true if the filesystem is mounted
2943 * read-only, and in that case, ext4_da_writepages should
2944 * *never* be called, so if that ever happens, we would want
2947 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2950 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2953 range_cyclic = wbc->range_cyclic;
2954 if (wbc->range_cyclic) {
2955 index = mapping->writeback_index;
2958 wbc->range_start = index << PAGE_CACHE_SHIFT;
2959 wbc->range_end = LLONG_MAX;
2960 wbc->range_cyclic = 0;
2962 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2965 * This works around two forms of stupidity. The first is in
2966 * the writeback code, which caps the maximum number of pages
2967 * written to be 1024 pages. This is wrong on multiple
2968 * levels; different architectues have a different page size,
2969 * which changes the maximum amount of data which gets
2970 * written. Secondly, 4 megabytes is way too small. XFS
2971 * forces this value to be 16 megabytes by multiplying
2972 * nr_to_write parameter by four, and then relies on its
2973 * allocator to allocate larger extents to make them
2974 * contiguous. Unfortunately this brings us to the second
2975 * stupidity, which is that ext4's mballoc code only allocates
2976 * at most 2048 blocks. So we force contiguous writes up to
2977 * the number of dirty blocks in the inode, or
2978 * sbi->max_writeback_mb_bump whichever is smaller.
2980 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2981 if (!range_cyclic && range_whole)
2982 desired_nr_to_write = wbc->nr_to_write * 8;
2984 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2986 if (desired_nr_to_write > max_pages)
2987 desired_nr_to_write = max_pages;
2989 if (wbc->nr_to_write < desired_nr_to_write) {
2990 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2991 wbc->nr_to_write = desired_nr_to_write;
2995 mpd.inode = mapping->host;
2997 pages_skipped = wbc->pages_skipped;
3000 while (!ret && wbc->nr_to_write > 0) {
3003 * we insert one extent at a time. So we need
3004 * credit needed for single extent allocation.
3005 * journalled mode is currently not supported
3008 BUG_ON(ext4_should_journal_data(inode));
3009 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3011 /* start a new transaction*/
3012 handle = ext4_journal_start(inode, needed_blocks);
3013 if (IS_ERR(handle)) {
3014 ret = PTR_ERR(handle);
3015 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3016 "%ld pages, ino %lu; err %d", __func__,
3017 wbc->nr_to_write, inode->i_ino, ret);
3018 goto out_writepages;
3022 * Now call __mpage_da_writepage to find the next
3023 * contiguous region of logical blocks that need
3024 * blocks to be allocated by ext4. We don't actually
3025 * submit the blocks for I/O here, even though
3026 * write_cache_pages thinks it will, and will set the
3027 * pages as clean for write before calling
3028 * __mpage_da_writepage().
3036 mpd.pages_written = 0;
3038 ret = write_cache_pages_da(mapping, wbc, &mpd);
3040 * If we have a contiguous extent of pages and we
3041 * haven't done the I/O yet, map the blocks and submit
3044 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3045 if (mpage_da_map_blocks(&mpd) == 0)
3046 mpage_da_submit_io(&mpd);
3048 ret = MPAGE_DA_EXTENT_TAIL;
3050 trace_ext4_da_write_pages(inode, &mpd);
3051 wbc->nr_to_write -= mpd.pages_written;
3053 ext4_journal_stop(handle);
3055 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3056 /* commit the transaction which would
3057 * free blocks released in the transaction
3060 jbd2_journal_force_commit_nested(sbi->s_journal);
3061 wbc->pages_skipped = pages_skipped;
3063 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3065 * got one extent now try with
3068 pages_written += mpd.pages_written;
3069 wbc->pages_skipped = pages_skipped;
3072 } else if (wbc->nr_to_write)
3074 * There is no more writeout needed
3075 * or we requested for a noblocking writeout
3076 * and we found the device congested
3080 if (!io_done && !cycled) {
3083 wbc->range_start = index << PAGE_CACHE_SHIFT;
3084 wbc->range_end = mapping->writeback_index - 1;
3087 if (pages_skipped != wbc->pages_skipped)
3088 ext4_msg(inode->i_sb, KERN_CRIT,
3089 "This should not happen leaving %s "
3090 "with nr_to_write = %ld ret = %d",
3091 __func__, wbc->nr_to_write, ret);
3094 index += pages_written;
3095 wbc->range_cyclic = range_cyclic;
3096 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3098 * set the writeback_index so that range_cyclic
3099 * mode will write it back later
3101 mapping->writeback_index = index;
3104 wbc->nr_to_write -= nr_to_writebump;
3105 wbc->range_start = range_start;
3106 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3110 #define FALL_BACK_TO_NONDELALLOC 1
3111 static int ext4_nonda_switch(struct super_block *sb)
3113 s64 free_blocks, dirty_blocks;
3114 struct ext4_sb_info *sbi = EXT4_SB(sb);
3117 * switch to non delalloc mode if we are running low
3118 * on free block. The free block accounting via percpu
3119 * counters can get slightly wrong with percpu_counter_batch getting
3120 * accumulated on each CPU without updating global counters
3121 * Delalloc need an accurate free block accounting. So switch
3122 * to non delalloc when we are near to error range.
3124 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3125 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3126 if (2 * free_blocks < 3 * dirty_blocks ||
3127 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3129 * free block count is less than 150% of dirty blocks
3130 * or free blocks is less than watermark
3135 * Even if we don't switch but are nearing capacity,
3136 * start pushing delalloc when 1/2 of free blocks are dirty.
3138 if (free_blocks < 2 * dirty_blocks)
3139 writeback_inodes_sb_if_idle(sb);
3144 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3145 loff_t pos, unsigned len, unsigned flags,
3146 struct page **pagep, void **fsdata)
3148 int ret, retries = 0;
3152 struct inode *inode = mapping->host;
3155 index = pos >> PAGE_CACHE_SHIFT;
3156 from = pos & (PAGE_CACHE_SIZE - 1);
3159 if (ext4_nonda_switch(inode->i_sb)) {
3160 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3161 return ext4_write_begin(file, mapping, pos,
3162 len, flags, pagep, fsdata);
3164 *fsdata = (void *)0;
3165 trace_ext4_da_write_begin(inode, pos, len, flags);
3168 * With delayed allocation, we don't log the i_disksize update
3169 * if there is delayed block allocation. But we still need
3170 * to journalling the i_disksize update if writes to the end
3171 * of file which has an already mapped buffer.
3173 handle = ext4_journal_start(inode, 1);
3174 if (IS_ERR(handle)) {
3175 ret = PTR_ERR(handle);
3178 /* We cannot recurse into the filesystem as the transaction is already
3180 flags |= AOP_FLAG_NOFS;
3182 page = grab_cache_page_write_begin(mapping, index, flags);
3184 ext4_journal_stop(handle);
3190 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3191 ext4_da_get_block_prep);
3194 ext4_journal_stop(handle);
3195 page_cache_release(page);
3197 * block_write_begin may have instantiated a few blocks
3198 * outside i_size. Trim these off again. Don't need
3199 * i_size_read because we hold i_mutex.
3201 if (pos + len > inode->i_size)
3202 ext4_truncate_failed_write(inode);
3205 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3212 * Check if we should update i_disksize
3213 * when write to the end of file but not require block allocation
3215 static int ext4_da_should_update_i_disksize(struct page *page,
3216 unsigned long offset)
3218 struct buffer_head *bh;
3219 struct inode *inode = page->mapping->host;
3223 bh = page_buffers(page);
3224 idx = offset >> inode->i_blkbits;
3226 for (i = 0; i < idx; i++)
3227 bh = bh->b_this_page;
3229 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3234 static int ext4_da_write_end(struct file *file,
3235 struct address_space *mapping,
3236 loff_t pos, unsigned len, unsigned copied,
3237 struct page *page, void *fsdata)
3239 struct inode *inode = mapping->host;
3241 handle_t *handle = ext4_journal_current_handle();
3243 unsigned long start, end;
3244 int write_mode = (int)(unsigned long)fsdata;
3246 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3247 if (ext4_should_order_data(inode)) {
3248 return ext4_ordered_write_end(file, mapping, pos,
3249 len, copied, page, fsdata);
3250 } else if (ext4_should_writeback_data(inode)) {
3251 return ext4_writeback_write_end(file, mapping, pos,
3252 len, copied, page, fsdata);
3258 trace_ext4_da_write_end(inode, pos, len, copied);
3259 start = pos & (PAGE_CACHE_SIZE - 1);
3260 end = start + copied - 1;
3263 * generic_write_end() will run mark_inode_dirty() if i_size
3264 * changes. So let's piggyback the i_disksize mark_inode_dirty
3268 new_i_size = pos + copied;
3269 if (new_i_size > EXT4_I(inode)->i_disksize) {
3270 if (ext4_da_should_update_i_disksize(page, end)) {
3271 down_write(&EXT4_I(inode)->i_data_sem);
3272 if (new_i_size > EXT4_I(inode)->i_disksize) {
3274 * Updating i_disksize when extending file
3275 * without needing block allocation
3277 if (ext4_should_order_data(inode))
3278 ret = ext4_jbd2_file_inode(handle,
3281 EXT4_I(inode)->i_disksize = new_i_size;
3283 up_write(&EXT4_I(inode)->i_data_sem);
3284 /* We need to mark inode dirty even if
3285 * new_i_size is less that inode->i_size
3286 * bu greater than i_disksize.(hint delalloc)
3288 ext4_mark_inode_dirty(handle, inode);
3291 ret2 = generic_write_end(file, mapping, pos, len, copied,
3296 ret2 = ext4_journal_stop(handle);
3300 return ret ? ret : copied;
3303 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3306 * Drop reserved blocks
3308 BUG_ON(!PageLocked(page));
3309 if (!page_has_buffers(page))
3312 ext4_da_page_release_reservation(page, offset);
3315 ext4_invalidatepage(page, offset);
3321 * Force all delayed allocation blocks to be allocated for a given inode.
3323 int ext4_alloc_da_blocks(struct inode *inode)
3325 trace_ext4_alloc_da_blocks(inode);
3327 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3328 !EXT4_I(inode)->i_reserved_meta_blocks)
3332 * We do something simple for now. The filemap_flush() will
3333 * also start triggering a write of the data blocks, which is
3334 * not strictly speaking necessary (and for users of
3335 * laptop_mode, not even desirable). However, to do otherwise
3336 * would require replicating code paths in:
3338 * ext4_da_writepages() ->
3339 * write_cache_pages() ---> (via passed in callback function)
3340 * __mpage_da_writepage() -->
3341 * mpage_add_bh_to_extent()
3342 * mpage_da_map_blocks()
3344 * The problem is that write_cache_pages(), located in
3345 * mm/page-writeback.c, marks pages clean in preparation for
3346 * doing I/O, which is not desirable if we're not planning on
3349 * We could call write_cache_pages(), and then redirty all of
3350 * the pages by calling redirty_page_for_writeback() but that
3351 * would be ugly in the extreme. So instead we would need to
3352 * replicate parts of the code in the above functions,
3353 * simplifying them becuase we wouldn't actually intend to
3354 * write out the pages, but rather only collect contiguous
3355 * logical block extents, call the multi-block allocator, and
3356 * then update the buffer heads with the block allocations.
3358 * For now, though, we'll cheat by calling filemap_flush(),
3359 * which will map the blocks, and start the I/O, but not
3360 * actually wait for the I/O to complete.
3362 return filemap_flush(inode->i_mapping);
3366 * bmap() is special. It gets used by applications such as lilo and by
3367 * the swapper to find the on-disk block of a specific piece of data.
3369 * Naturally, this is dangerous if the block concerned is still in the
3370 * journal. If somebody makes a swapfile on an ext4 data-journaling
3371 * filesystem and enables swap, then they may get a nasty shock when the
3372 * data getting swapped to that swapfile suddenly gets overwritten by
3373 * the original zero's written out previously to the journal and
3374 * awaiting writeback in the kernel's buffer cache.
3376 * So, if we see any bmap calls here on a modified, data-journaled file,
3377 * take extra steps to flush any blocks which might be in the cache.
3379 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3381 struct inode *inode = mapping->host;
3385 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3386 test_opt(inode->i_sb, DELALLOC)) {
3388 * With delalloc we want to sync the file
3389 * so that we can make sure we allocate
3392 filemap_write_and_wait(mapping);
3395 if (EXT4_JOURNAL(inode) &&
3396 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3398 * This is a REALLY heavyweight approach, but the use of
3399 * bmap on dirty files is expected to be extremely rare:
3400 * only if we run lilo or swapon on a freshly made file
3401 * do we expect this to happen.
3403 * (bmap requires CAP_SYS_RAWIO so this does not
3404 * represent an unprivileged user DOS attack --- we'd be
3405 * in trouble if mortal users could trigger this path at
3408 * NB. EXT4_STATE_JDATA is not set on files other than
3409 * regular files. If somebody wants to bmap a directory
3410 * or symlink and gets confused because the buffer
3411 * hasn't yet been flushed to disk, they deserve
3412 * everything they get.
3415 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3416 journal = EXT4_JOURNAL(inode);
3417 jbd2_journal_lock_updates(journal);
3418 err = jbd2_journal_flush(journal);
3419 jbd2_journal_unlock_updates(journal);
3425 return generic_block_bmap(mapping, block, ext4_get_block);
3428 static int ext4_readpage(struct file *file, struct page *page)
3430 return mpage_readpage(page, ext4_get_block);
3434 ext4_readpages(struct file *file, struct address_space *mapping,
3435 struct list_head *pages, unsigned nr_pages)
3437 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3440 static void ext4_free_io_end(ext4_io_end_t *io)
3449 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3451 struct buffer_head *head, *bh;
3452 unsigned int curr_off = 0;
3454 if (!page_has_buffers(page))
3456 head = bh = page_buffers(page);
3458 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3460 ext4_free_io_end(bh->b_private);
3461 bh->b_private = NULL;
3462 bh->b_end_io = NULL;
3464 curr_off = curr_off + bh->b_size;
3465 bh = bh->b_this_page;
3466 } while (bh != head);
3469 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3471 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3474 * free any io_end structure allocated for buffers to be discarded
3476 if (ext4_should_dioread_nolock(page->mapping->host))
3477 ext4_invalidatepage_free_endio(page, offset);
3479 * If it's a full truncate we just forget about the pending dirtying
3482 ClearPageChecked(page);
3485 jbd2_journal_invalidatepage(journal, page, offset);
3487 block_invalidatepage(page, offset);
3490 static int ext4_releasepage(struct page *page, gfp_t wait)
3492 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3494 WARN_ON(PageChecked(page));
3495 if (!page_has_buffers(page))
3498 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3500 return try_to_free_buffers(page);
3504 * O_DIRECT for ext3 (or indirect map) based files
3506 * If the O_DIRECT write will extend the file then add this inode to the
3507 * orphan list. So recovery will truncate it back to the original size
3508 * if the machine crashes during the write.
3510 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3511 * crashes then stale disk data _may_ be exposed inside the file. But current
3512 * VFS code falls back into buffered path in that case so we are safe.
3514 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3515 const struct iovec *iov, loff_t offset,
3516 unsigned long nr_segs)
3518 struct file *file = iocb->ki_filp;
3519 struct inode *inode = file->f_mapping->host;
3520 struct ext4_inode_info *ei = EXT4_I(inode);
3524 size_t count = iov_length(iov, nr_segs);
3528 loff_t final_size = offset + count;
3530 if (final_size > inode->i_size) {
3531 /* Credits for sb + inode write */
3532 handle = ext4_journal_start(inode, 2);
3533 if (IS_ERR(handle)) {
3534 ret = PTR_ERR(handle);
3537 ret = ext4_orphan_add(handle, inode);
3539 ext4_journal_stop(handle);
3543 ei->i_disksize = inode->i_size;
3544 ext4_journal_stop(handle);
3549 if (rw == READ && ext4_should_dioread_nolock(inode))
3550 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3551 inode->i_sb->s_bdev, iov,
3553 ext4_get_block, NULL);
3555 ret = blockdev_direct_IO(rw, iocb, inode,
3556 inode->i_sb->s_bdev, iov,
3558 ext4_get_block, NULL);
3559 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3565 /* Credits for sb + inode write */
3566 handle = ext4_journal_start(inode, 2);
3567 if (IS_ERR(handle)) {
3568 /* This is really bad luck. We've written the data
3569 * but cannot extend i_size. Bail out and pretend
3570 * the write failed... */
3571 ret = PTR_ERR(handle);
3573 ext4_orphan_del(NULL, inode);
3578 ext4_orphan_del(handle, inode);
3580 loff_t end = offset + ret;
3581 if (end > inode->i_size) {
3582 ei->i_disksize = end;
3583 i_size_write(inode, end);
3585 * We're going to return a positive `ret'
3586 * here due to non-zero-length I/O, so there's
3587 * no way of reporting error returns from
3588 * ext4_mark_inode_dirty() to userspace. So
3591 ext4_mark_inode_dirty(handle, inode);
3594 err = ext4_journal_stop(handle);
3603 * ext4_get_block used when preparing for a DIO write or buffer write.
3604 * We allocate an uinitialized extent if blocks haven't been allocated.
3605 * The extent will be converted to initialized after the IO is complete.
3607 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3608 struct buffer_head *bh_result, int create)
3610 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3611 inode->i_ino, create);
3612 return _ext4_get_block(inode, iblock, bh_result,
3613 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3616 static void dump_completed_IO(struct inode * inode)
3619 struct list_head *cur, *before, *after;
3620 ext4_io_end_t *io, *io0, *io1;
3621 unsigned long flags;
3623 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3624 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3628 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3629 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3630 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3633 io0 = container_of(before, ext4_io_end_t, list);
3635 io1 = container_of(after, ext4_io_end_t, list);
3637 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3638 io, inode->i_ino, io0, io1);
3640 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3645 * check a range of space and convert unwritten extents to written.
3647 static int ext4_end_io_nolock(ext4_io_end_t *io)
3649 struct inode *inode = io->inode;
3650 loff_t offset = io->offset;
3651 ssize_t size = io->size;
3654 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3655 "list->prev 0x%p\n",
3656 io, inode->i_ino, io->list.next, io->list.prev);
3658 if (list_empty(&io->list))
3661 if (io->flag != EXT4_IO_UNWRITTEN)
3664 ret = ext4_convert_unwritten_extents(inode, offset, size);
3666 printk(KERN_EMERG "%s: failed to convert unwritten"
3667 "extents to written extents, error is %d"
3668 " io is still on inode %lu aio dio list\n",
3669 __func__, ret, inode->i_ino);
3673 /* clear the DIO AIO unwritten flag */
3679 * work on completed aio dio IO, to convert unwritten extents to extents
3681 static void ext4_end_io_work(struct work_struct *work)
3683 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3684 struct inode *inode = io->inode;
3685 struct ext4_inode_info *ei = EXT4_I(inode);
3686 unsigned long flags;
3689 mutex_lock(&inode->i_mutex);
3690 ret = ext4_end_io_nolock(io);
3692 mutex_unlock(&inode->i_mutex);
3696 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3697 if (!list_empty(&io->list))
3698 list_del_init(&io->list);
3699 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3700 mutex_unlock(&inode->i_mutex);
3701 ext4_free_io_end(io);
3705 * This function is called from ext4_sync_file().
3707 * When IO is completed, the work to convert unwritten extents to
3708 * written is queued on workqueue but may not get immediately
3709 * scheduled. When fsync is called, we need to ensure the
3710 * conversion is complete before fsync returns.
3711 * The inode keeps track of a list of pending/completed IO that
3712 * might needs to do the conversion. This function walks through
3713 * the list and convert the related unwritten extents for completed IO
3715 * The function return the number of pending IOs on success.
3717 int flush_completed_IO(struct inode *inode)
3720 struct ext4_inode_info *ei = EXT4_I(inode);
3721 unsigned long flags;
3725 if (list_empty(&ei->i_completed_io_list))
3728 dump_completed_IO(inode);
3729 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3730 while (!list_empty(&ei->i_completed_io_list)){
3731 io = list_entry(ei->i_completed_io_list.next,
3732 ext4_io_end_t, list);
3734 * Calling ext4_end_io_nolock() to convert completed
3737 * When ext4_sync_file() is called, run_queue() may already
3738 * about to flush the work corresponding to this io structure.
3739 * It will be upset if it founds the io structure related
3740 * to the work-to-be schedule is freed.
3742 * Thus we need to keep the io structure still valid here after
3743 * convertion finished. The io structure has a flag to
3744 * avoid double converting from both fsync and background work
3747 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3748 ret = ext4_end_io_nolock(io);
3749 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3753 list_del_init(&io->list);
3755 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3756 return (ret2 < 0) ? ret2 : 0;
3759 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3761 ext4_io_end_t *io = NULL;
3763 io = kmalloc(sizeof(*io), flags);
3772 INIT_WORK(&io->work, ext4_end_io_work);
3773 INIT_LIST_HEAD(&io->list);
3779 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3780 ssize_t size, void *private)
3782 ext4_io_end_t *io_end = iocb->private;
3783 struct workqueue_struct *wq;
3784 unsigned long flags;
3785 struct ext4_inode_info *ei;
3787 /* if not async direct IO or dio with 0 bytes write, just return */
3788 if (!io_end || !size)
3791 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3792 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3793 iocb->private, io_end->inode->i_ino, iocb, offset,
3796 /* if not aio dio with unwritten extents, just free io and return */
3797 if (io_end->flag != EXT4_IO_UNWRITTEN){
3798 ext4_free_io_end(io_end);
3799 iocb->private = NULL;
3803 io_end->offset = offset;
3804 io_end->size = size;
3805 io_end->flag = EXT4_IO_UNWRITTEN;
3806 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3808 /* queue the work to convert unwritten extents to written */
3809 queue_work(wq, &io_end->work);
3811 /* Add the io_end to per-inode completed aio dio list*/
3812 ei = EXT4_I(io_end->inode);
3813 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3814 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3815 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3816 iocb->private = NULL;
3819 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3821 ext4_io_end_t *io_end = bh->b_private;
3822 struct workqueue_struct *wq;
3823 struct inode *inode;
3824 unsigned long flags;
3826 if (!test_clear_buffer_uninit(bh) || !io_end)
3829 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3830 printk("sb umounted, discard end_io request for inode %lu\n",
3831 io_end->inode->i_ino);
3832 ext4_free_io_end(io_end);
3836 io_end->flag = EXT4_IO_UNWRITTEN;
3837 inode = io_end->inode;
3839 /* Add the io_end to per-inode completed io list*/
3840 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3841 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3842 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3844 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3845 /* queue the work to convert unwritten extents to written */
3846 queue_work(wq, &io_end->work);
3848 bh->b_private = NULL;
3849 bh->b_end_io = NULL;
3850 clear_buffer_uninit(bh);
3851 end_buffer_async_write(bh, uptodate);
3854 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3856 ext4_io_end_t *io_end;
3857 struct page *page = bh->b_page;
3858 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3859 size_t size = bh->b_size;
3862 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3864 if (printk_ratelimit())
3865 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3869 io_end->offset = offset;
3870 io_end->size = size;
3872 * We need to hold a reference to the page to make sure it
3873 * doesn't get evicted before ext4_end_io_work() has a chance
3874 * to convert the extent from written to unwritten.
3876 io_end->page = page;
3877 get_page(io_end->page);
3879 bh->b_private = io_end;
3880 bh->b_end_io = ext4_end_io_buffer_write;
3885 * For ext4 extent files, ext4 will do direct-io write to holes,
3886 * preallocated extents, and those write extend the file, no need to
3887 * fall back to buffered IO.
3889 * For holes, we fallocate those blocks, mark them as unintialized
3890 * If those blocks were preallocated, we mark sure they are splited, but
3891 * still keep the range to write as unintialized.
3893 * The unwrritten extents will be converted to written when DIO is completed.
3894 * For async direct IO, since the IO may still pending when return, we
3895 * set up an end_io call back function, which will do the convertion
3896 * when async direct IO completed.
3898 * If the O_DIRECT write will extend the file then add this inode to the
3899 * orphan list. So recovery will truncate it back to the original size
3900 * if the machine crashes during the write.
3903 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3904 const struct iovec *iov, loff_t offset,
3905 unsigned long nr_segs)
3907 struct file *file = iocb->ki_filp;
3908 struct inode *inode = file->f_mapping->host;
3910 size_t count = iov_length(iov, nr_segs);
3912 loff_t final_size = offset + count;
3913 if (rw == WRITE && final_size <= inode->i_size) {
3915 * We could direct write to holes and fallocate.
3917 * Allocated blocks to fill the hole are marked as uninitialized
3918 * to prevent paralel buffered read to expose the stale data
3919 * before DIO complete the data IO.
3921 * As to previously fallocated extents, ext4 get_block
3922 * will just simply mark the buffer mapped but still
3923 * keep the extents uninitialized.
3925 * for non AIO case, we will convert those unwritten extents
3926 * to written after return back from blockdev_direct_IO.
3928 * for async DIO, the conversion needs to be defered when
3929 * the IO is completed. The ext4 end_io callback function
3930 * will be called to take care of the conversion work.
3931 * Here for async case, we allocate an io_end structure to
3934 iocb->private = NULL;
3935 EXT4_I(inode)->cur_aio_dio = NULL;
3936 if (!is_sync_kiocb(iocb)) {
3937 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3941 * we save the io structure for current async
3942 * direct IO, so that later ext4_get_blocks()
3943 * could flag the io structure whether there
3944 * is a unwritten extents needs to be converted
3945 * when IO is completed.
3947 EXT4_I(inode)->cur_aio_dio = iocb->private;
3950 ret = blockdev_direct_IO(rw, iocb, inode,
3951 inode->i_sb->s_bdev, iov,
3953 ext4_get_block_write,
3956 EXT4_I(inode)->cur_aio_dio = NULL;
3958 * The io_end structure takes a reference to the inode,
3959 * that structure needs to be destroyed and the
3960 * reference to the inode need to be dropped, when IO is
3961 * complete, even with 0 byte write, or failed.
3963 * In the successful AIO DIO case, the io_end structure will be
3964 * desctroyed and the reference to the inode will be dropped
3965 * after the end_io call back function is called.
3967 * In the case there is 0 byte write, or error case, since
3968 * VFS direct IO won't invoke the end_io call back function,
3969 * we need to free the end_io structure here.
3971 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3972 ext4_free_io_end(iocb->private);
3973 iocb->private = NULL;
3974 } else if (ret > 0 && ext4_test_inode_state(inode,
3975 EXT4_STATE_DIO_UNWRITTEN)) {
3978 * for non AIO case, since the IO is already
3979 * completed, we could do the convertion right here
3981 err = ext4_convert_unwritten_extents(inode,
3985 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3990 /* for write the the end of file case, we fall back to old way */
3991 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3994 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3995 const struct iovec *iov, loff_t offset,
3996 unsigned long nr_segs)
3998 struct file *file = iocb->ki_filp;
3999 struct inode *inode = file->f_mapping->host;
4001 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
4002 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4004 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4008 * Pages can be marked dirty completely asynchronously from ext4's journalling
4009 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4010 * much here because ->set_page_dirty is called under VFS locks. The page is
4011 * not necessarily locked.
4013 * We cannot just dirty the page and leave attached buffers clean, because the
4014 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4015 * or jbddirty because all the journalling code will explode.
4017 * So what we do is to mark the page "pending dirty" and next time writepage
4018 * is called, propagate that into the buffers appropriately.
4020 static int ext4_journalled_set_page_dirty(struct page *page)
4022 SetPageChecked(page);
4023 return __set_page_dirty_nobuffers(page);
4026 static const struct address_space_operations ext4_ordered_aops = {
4027 .readpage = ext4_readpage,
4028 .readpages = ext4_readpages,
4029 .writepage = ext4_writepage,
4030 .sync_page = block_sync_page,
4031 .write_begin = ext4_write_begin,
4032 .write_end = ext4_ordered_write_end,
4034 .invalidatepage = ext4_invalidatepage,
4035 .releasepage = ext4_releasepage,
4036 .direct_IO = ext4_direct_IO,
4037 .migratepage = buffer_migrate_page,
4038 .is_partially_uptodate = block_is_partially_uptodate,
4039 .error_remove_page = generic_error_remove_page,
4042 static const struct address_space_operations ext4_writeback_aops = {
4043 .readpage = ext4_readpage,
4044 .readpages = ext4_readpages,
4045 .writepage = ext4_writepage,
4046 .sync_page = block_sync_page,
4047 .write_begin = ext4_write_begin,
4048 .write_end = ext4_writeback_write_end,
4050 .invalidatepage = ext4_invalidatepage,
4051 .releasepage = ext4_releasepage,
4052 .direct_IO = ext4_direct_IO,
4053 .migratepage = buffer_migrate_page,
4054 .is_partially_uptodate = block_is_partially_uptodate,
4055 .error_remove_page = generic_error_remove_page,
4058 static const struct address_space_operations ext4_journalled_aops = {
4059 .readpage = ext4_readpage,
4060 .readpages = ext4_readpages,
4061 .writepage = ext4_writepage,
4062 .sync_page = block_sync_page,
4063 .write_begin = ext4_write_begin,
4064 .write_end = ext4_journalled_write_end,
4065 .set_page_dirty = ext4_journalled_set_page_dirty,
4067 .invalidatepage = ext4_invalidatepage,
4068 .releasepage = ext4_releasepage,
4069 .is_partially_uptodate = block_is_partially_uptodate,
4070 .error_remove_page = generic_error_remove_page,
4073 static const struct address_space_operations ext4_da_aops = {
4074 .readpage = ext4_readpage,
4075 .readpages = ext4_readpages,
4076 .writepage = ext4_writepage,
4077 .writepages = ext4_da_writepages,
4078 .sync_page = block_sync_page,
4079 .write_begin = ext4_da_write_begin,
4080 .write_end = ext4_da_write_end,
4082 .invalidatepage = ext4_da_invalidatepage,
4083 .releasepage = ext4_releasepage,
4084 .direct_IO = ext4_direct_IO,
4085 .migratepage = buffer_migrate_page,
4086 .is_partially_uptodate = block_is_partially_uptodate,
4087 .error_remove_page = generic_error_remove_page,
4090 void ext4_set_aops(struct inode *inode)
4092 if (ext4_should_order_data(inode) &&
4093 test_opt(inode->i_sb, DELALLOC))
4094 inode->i_mapping->a_ops = &ext4_da_aops;
4095 else if (ext4_should_order_data(inode))
4096 inode->i_mapping->a_ops = &ext4_ordered_aops;
4097 else if (ext4_should_writeback_data(inode) &&
4098 test_opt(inode->i_sb, DELALLOC))
4099 inode->i_mapping->a_ops = &ext4_da_aops;
4100 else if (ext4_should_writeback_data(inode))
4101 inode->i_mapping->a_ops = &ext4_writeback_aops;
4103 inode->i_mapping->a_ops = &ext4_journalled_aops;
4107 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4108 * up to the end of the block which corresponds to `from'.
4109 * This required during truncate. We need to physically zero the tail end
4110 * of that block so it doesn't yield old data if the file is later grown.
4112 int ext4_block_truncate_page(handle_t *handle,
4113 struct address_space *mapping, loff_t from)
4115 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4116 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4117 unsigned blocksize, length, pos;
4119 struct inode *inode = mapping->host;
4120 struct buffer_head *bh;
4124 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4125 mapping_gfp_mask(mapping) & ~__GFP_FS);
4129 blocksize = inode->i_sb->s_blocksize;
4130 length = blocksize - (offset & (blocksize - 1));
4131 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4134 * For "nobh" option, we can only work if we don't need to
4135 * read-in the page - otherwise we create buffers to do the IO.
4137 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
4138 ext4_should_writeback_data(inode) && PageUptodate(page)) {
4139 zero_user(page, offset, length);
4140 set_page_dirty(page);
4144 if (!page_has_buffers(page))
4145 create_empty_buffers(page, blocksize, 0);
4147 /* Find the buffer that contains "offset" */
4148 bh = page_buffers(page);
4150 while (offset >= pos) {
4151 bh = bh->b_this_page;
4157 if (buffer_freed(bh)) {
4158 BUFFER_TRACE(bh, "freed: skip");
4162 if (!buffer_mapped(bh)) {
4163 BUFFER_TRACE(bh, "unmapped");
4164 ext4_get_block(inode, iblock, bh, 0);
4165 /* unmapped? It's a hole - nothing to do */
4166 if (!buffer_mapped(bh)) {
4167 BUFFER_TRACE(bh, "still unmapped");
4172 /* Ok, it's mapped. Make sure it's up-to-date */
4173 if (PageUptodate(page))
4174 set_buffer_uptodate(bh);
4176 if (!buffer_uptodate(bh)) {
4178 ll_rw_block(READ, 1, &bh);
4180 /* Uhhuh. Read error. Complain and punt. */
4181 if (!buffer_uptodate(bh))
4185 if (ext4_should_journal_data(inode)) {
4186 BUFFER_TRACE(bh, "get write access");
4187 err = ext4_journal_get_write_access(handle, bh);
4192 zero_user(page, offset, length);
4194 BUFFER_TRACE(bh, "zeroed end of block");
4197 if (ext4_should_journal_data(inode)) {
4198 err = ext4_handle_dirty_metadata(handle, inode, bh);
4200 if (ext4_should_order_data(inode))
4201 err = ext4_jbd2_file_inode(handle, inode);
4202 mark_buffer_dirty(bh);
4207 page_cache_release(page);
4212 * Probably it should be a library function... search for first non-zero word
4213 * or memcmp with zero_page, whatever is better for particular architecture.
4216 static inline int all_zeroes(__le32 *p, __le32 *q)
4225 * ext4_find_shared - find the indirect blocks for partial truncation.
4226 * @inode: inode in question
4227 * @depth: depth of the affected branch
4228 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4229 * @chain: place to store the pointers to partial indirect blocks
4230 * @top: place to the (detached) top of branch
4232 * This is a helper function used by ext4_truncate().
4234 * When we do truncate() we may have to clean the ends of several
4235 * indirect blocks but leave the blocks themselves alive. Block is
4236 * partially truncated if some data below the new i_size is refered
4237 * from it (and it is on the path to the first completely truncated
4238 * data block, indeed). We have to free the top of that path along
4239 * with everything to the right of the path. Since no allocation
4240 * past the truncation point is possible until ext4_truncate()
4241 * finishes, we may safely do the latter, but top of branch may
4242 * require special attention - pageout below the truncation point
4243 * might try to populate it.
4245 * We atomically detach the top of branch from the tree, store the
4246 * block number of its root in *@top, pointers to buffer_heads of
4247 * partially truncated blocks - in @chain[].bh and pointers to
4248 * their last elements that should not be removed - in
4249 * @chain[].p. Return value is the pointer to last filled element
4252 * The work left to caller to do the actual freeing of subtrees:
4253 * a) free the subtree starting from *@top
4254 * b) free the subtrees whose roots are stored in
4255 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4256 * c) free the subtrees growing from the inode past the @chain[0].
4257 * (no partially truncated stuff there). */
4259 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4260 ext4_lblk_t offsets[4], Indirect chain[4],
4263 Indirect *partial, *p;
4267 /* Make k index the deepest non-null offset + 1 */
4268 for (k = depth; k > 1 && !offsets[k-1]; k--)
4270 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4271 /* Writer: pointers */
4273 partial = chain + k-1;
4275 * If the branch acquired continuation since we've looked at it -
4276 * fine, it should all survive and (new) top doesn't belong to us.
4278 if (!partial->key && *partial->p)
4281 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4284 * OK, we've found the last block that must survive. The rest of our
4285 * branch should be detached before unlocking. However, if that rest
4286 * of branch is all ours and does not grow immediately from the inode
4287 * it's easier to cheat and just decrement partial->p.
4289 if (p == chain + k - 1 && p > chain) {
4293 /* Nope, don't do this in ext4. Must leave the tree intact */
4300 while (partial > p) {
4301 brelse(partial->bh);
4309 * Zero a number of block pointers in either an inode or an indirect block.
4310 * If we restart the transaction we must again get write access to the
4311 * indirect block for further modification.
4313 * We release `count' blocks on disk, but (last - first) may be greater
4314 * than `count' because there can be holes in there.
4316 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4317 struct buffer_head *bh,
4318 ext4_fsblk_t block_to_free,
4319 unsigned long count, __le32 *first,
4323 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4325 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4326 flags |= EXT4_FREE_BLOCKS_METADATA;
4328 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4330 ext4_error(inode->i_sb, "inode #%lu: "
4331 "attempt to clear blocks %llu len %lu, invalid",
4332 inode->i_ino, (unsigned long long) block_to_free,
4337 if (try_to_extend_transaction(handle, inode)) {
4339 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4340 ext4_handle_dirty_metadata(handle, inode, bh);
4342 ext4_mark_inode_dirty(handle, inode);
4343 ext4_truncate_restart_trans(handle, inode,
4344 blocks_for_truncate(inode));
4346 BUFFER_TRACE(bh, "retaking write access");
4347 ext4_journal_get_write_access(handle, bh);
4351 for (p = first; p < last; p++)
4354 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4359 * ext4_free_data - free a list of data blocks
4360 * @handle: handle for this transaction
4361 * @inode: inode we are dealing with
4362 * @this_bh: indirect buffer_head which contains *@first and *@last
4363 * @first: array of block numbers
4364 * @last: points immediately past the end of array
4366 * We are freeing all blocks refered from that array (numbers are stored as
4367 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4369 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4370 * blocks are contiguous then releasing them at one time will only affect one
4371 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4372 * actually use a lot of journal space.
4374 * @this_bh will be %NULL if @first and @last point into the inode's direct
4377 static void ext4_free_data(handle_t *handle, struct inode *inode,
4378 struct buffer_head *this_bh,
4379 __le32 *first, __le32 *last)
4381 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4382 unsigned long count = 0; /* Number of blocks in the run */
4383 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4386 ext4_fsblk_t nr; /* Current block # */
4387 __le32 *p; /* Pointer into inode/ind
4388 for current block */
4391 if (this_bh) { /* For indirect block */
4392 BUFFER_TRACE(this_bh, "get_write_access");
4393 err = ext4_journal_get_write_access(handle, this_bh);
4394 /* Important: if we can't update the indirect pointers
4395 * to the blocks, we can't free them. */
4400 for (p = first; p < last; p++) {
4401 nr = le32_to_cpu(*p);
4403 /* accumulate blocks to free if they're contiguous */
4406 block_to_free_p = p;
4408 } else if (nr == block_to_free + count) {
4411 if (ext4_clear_blocks(handle, inode, this_bh,
4412 block_to_free, count,
4413 block_to_free_p, p))
4416 block_to_free_p = p;
4423 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4424 count, block_to_free_p, p);
4427 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4430 * The buffer head should have an attached journal head at this
4431 * point. However, if the data is corrupted and an indirect
4432 * block pointed to itself, it would have been detached when
4433 * the block was cleared. Check for this instead of OOPSing.
4435 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4436 ext4_handle_dirty_metadata(handle, inode, this_bh);
4438 ext4_error(inode->i_sb,
4439 "circular indirect block detected, "
4440 "inode=%lu, block=%llu",
4442 (unsigned long long) this_bh->b_blocknr);
4447 * ext4_free_branches - free an array of branches
4448 * @handle: JBD handle for this transaction
4449 * @inode: inode we are dealing with
4450 * @parent_bh: the buffer_head which contains *@first and *@last
4451 * @first: array of block numbers
4452 * @last: pointer immediately past the end of array
4453 * @depth: depth of the branches to free
4455 * We are freeing all blocks refered from these branches (numbers are
4456 * stored as little-endian 32-bit) and updating @inode->i_blocks
4459 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4460 struct buffer_head *parent_bh,
4461 __le32 *first, __le32 *last, int depth)
4466 if (ext4_handle_is_aborted(handle))
4470 struct buffer_head *bh;
4471 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4473 while (--p >= first) {
4474 nr = le32_to_cpu(*p);
4476 continue; /* A hole */
4478 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4480 ext4_error(inode->i_sb,
4481 "indirect mapped block in inode "
4482 "#%lu invalid (level %d, blk #%lu)",
4483 inode->i_ino, depth,
4484 (unsigned long) nr);
4488 /* Go read the buffer for the next level down */
4489 bh = sb_bread(inode->i_sb, nr);
4492 * A read failure? Report error and clear slot
4496 ext4_error(inode->i_sb,
4497 "Read failure, inode=%lu, block=%llu",
4502 /* This zaps the entire block. Bottom up. */
4503 BUFFER_TRACE(bh, "free child branches");
4504 ext4_free_branches(handle, inode, bh,
4505 (__le32 *) bh->b_data,
4506 (__le32 *) bh->b_data + addr_per_block,
4510 * We've probably journalled the indirect block several
4511 * times during the truncate. But it's no longer
4512 * needed and we now drop it from the transaction via
4513 * jbd2_journal_revoke().
4515 * That's easy if it's exclusively part of this
4516 * transaction. But if it's part of the committing
4517 * transaction then jbd2_journal_forget() will simply
4518 * brelse() it. That means that if the underlying
4519 * block is reallocated in ext4_get_block(),
4520 * unmap_underlying_metadata() will find this block
4521 * and will try to get rid of it. damn, damn.
4523 * If this block has already been committed to the
4524 * journal, a revoke record will be written. And
4525 * revoke records must be emitted *before* clearing
4526 * this block's bit in the bitmaps.
4528 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4531 * Everything below this this pointer has been
4532 * released. Now let this top-of-subtree go.
4534 * We want the freeing of this indirect block to be
4535 * atomic in the journal with the updating of the
4536 * bitmap block which owns it. So make some room in
4539 * We zero the parent pointer *after* freeing its
4540 * pointee in the bitmaps, so if extend_transaction()
4541 * for some reason fails to put the bitmap changes and
4542 * the release into the same transaction, recovery
4543 * will merely complain about releasing a free block,
4544 * rather than leaking blocks.
4546 if (ext4_handle_is_aborted(handle))
4548 if (try_to_extend_transaction(handle, inode)) {
4549 ext4_mark_inode_dirty(handle, inode);
4550 ext4_truncate_restart_trans(handle, inode,
4551 blocks_for_truncate(inode));
4554 ext4_free_blocks(handle, inode, 0, nr, 1,
4555 EXT4_FREE_BLOCKS_METADATA);
4559 * The block which we have just freed is
4560 * pointed to by an indirect block: journal it
4562 BUFFER_TRACE(parent_bh, "get_write_access");
4563 if (!ext4_journal_get_write_access(handle,
4566 BUFFER_TRACE(parent_bh,
4567 "call ext4_handle_dirty_metadata");
4568 ext4_handle_dirty_metadata(handle,
4575 /* We have reached the bottom of the tree. */
4576 BUFFER_TRACE(parent_bh, "free data blocks");
4577 ext4_free_data(handle, inode, parent_bh, first, last);
4581 int ext4_can_truncate(struct inode *inode)
4583 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4585 if (S_ISREG(inode->i_mode))
4587 if (S_ISDIR(inode->i_mode))
4589 if (S_ISLNK(inode->i_mode))
4590 return !ext4_inode_is_fast_symlink(inode);
4597 * We block out ext4_get_block() block instantiations across the entire
4598 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4599 * simultaneously on behalf of the same inode.
4601 * As we work through the truncate and commmit bits of it to the journal there
4602 * is one core, guiding principle: the file's tree must always be consistent on
4603 * disk. We must be able to restart the truncate after a crash.
4605 * The file's tree may be transiently inconsistent in memory (although it
4606 * probably isn't), but whenever we close off and commit a journal transaction,
4607 * the contents of (the filesystem + the journal) must be consistent and
4608 * restartable. It's pretty simple, really: bottom up, right to left (although
4609 * left-to-right works OK too).
4611 * Note that at recovery time, journal replay occurs *before* the restart of
4612 * truncate against the orphan inode list.
4614 * The committed inode has the new, desired i_size (which is the same as
4615 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4616 * that this inode's truncate did not complete and it will again call
4617 * ext4_truncate() to have another go. So there will be instantiated blocks
4618 * to the right of the truncation point in a crashed ext4 filesystem. But
4619 * that's fine - as long as they are linked from the inode, the post-crash
4620 * ext4_truncate() run will find them and release them.
4622 void ext4_truncate(struct inode *inode)
4625 struct ext4_inode_info *ei = EXT4_I(inode);
4626 __le32 *i_data = ei->i_data;
4627 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4628 struct address_space *mapping = inode->i_mapping;
4629 ext4_lblk_t offsets[4];
4634 ext4_lblk_t last_block;
4635 unsigned blocksize = inode->i_sb->s_blocksize;
4637 if (!ext4_can_truncate(inode))
4640 EXT4_I(inode)->i_flags &= ~EXT4_EOFBLOCKS_FL;
4642 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4643 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4645 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4646 ext4_ext_truncate(inode);
4650 handle = start_transaction(inode);
4652 return; /* AKPM: return what? */
4654 last_block = (inode->i_size + blocksize-1)
4655 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4657 if (inode->i_size & (blocksize - 1))
4658 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4661 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4663 goto out_stop; /* error */
4666 * OK. This truncate is going to happen. We add the inode to the
4667 * orphan list, so that if this truncate spans multiple transactions,
4668 * and we crash, we will resume the truncate when the filesystem
4669 * recovers. It also marks the inode dirty, to catch the new size.
4671 * Implication: the file must always be in a sane, consistent
4672 * truncatable state while each transaction commits.
4674 if (ext4_orphan_add(handle, inode))
4678 * From here we block out all ext4_get_block() callers who want to
4679 * modify the block allocation tree.
4681 down_write(&ei->i_data_sem);
4683 ext4_discard_preallocations(inode);
4686 * The orphan list entry will now protect us from any crash which
4687 * occurs before the truncate completes, so it is now safe to propagate
4688 * the new, shorter inode size (held for now in i_size) into the
4689 * on-disk inode. We do this via i_disksize, which is the value which
4690 * ext4 *really* writes onto the disk inode.
4692 ei->i_disksize = inode->i_size;
4694 if (n == 1) { /* direct blocks */
4695 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4696 i_data + EXT4_NDIR_BLOCKS);
4700 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4701 /* Kill the top of shared branch (not detached) */
4703 if (partial == chain) {
4704 /* Shared branch grows from the inode */
4705 ext4_free_branches(handle, inode, NULL,
4706 &nr, &nr+1, (chain+n-1) - partial);
4709 * We mark the inode dirty prior to restart,
4710 * and prior to stop. No need for it here.
4713 /* Shared branch grows from an indirect block */
4714 BUFFER_TRACE(partial->bh, "get_write_access");
4715 ext4_free_branches(handle, inode, partial->bh,
4717 partial->p+1, (chain+n-1) - partial);
4720 /* Clear the ends of indirect blocks on the shared branch */
4721 while (partial > chain) {
4722 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4723 (__le32*)partial->bh->b_data+addr_per_block,
4724 (chain+n-1) - partial);
4725 BUFFER_TRACE(partial->bh, "call brelse");
4726 brelse(partial->bh);
4730 /* Kill the remaining (whole) subtrees */
4731 switch (offsets[0]) {
4733 nr = i_data[EXT4_IND_BLOCK];
4735 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4736 i_data[EXT4_IND_BLOCK] = 0;
4738 case EXT4_IND_BLOCK:
4739 nr = i_data[EXT4_DIND_BLOCK];
4741 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4742 i_data[EXT4_DIND_BLOCK] = 0;
4744 case EXT4_DIND_BLOCK:
4745 nr = i_data[EXT4_TIND_BLOCK];
4747 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4748 i_data[EXT4_TIND_BLOCK] = 0;
4750 case EXT4_TIND_BLOCK:
4754 up_write(&ei->i_data_sem);
4755 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4756 ext4_mark_inode_dirty(handle, inode);
4759 * In a multi-transaction truncate, we only make the final transaction
4763 ext4_handle_sync(handle);
4766 * If this was a simple ftruncate(), and the file will remain alive
4767 * then we need to clear up the orphan record which we created above.
4768 * However, if this was a real unlink then we were called by
4769 * ext4_delete_inode(), and we allow that function to clean up the
4770 * orphan info for us.
4773 ext4_orphan_del(handle, inode);
4775 ext4_journal_stop(handle);
4779 * ext4_get_inode_loc returns with an extra refcount against the inode's
4780 * underlying buffer_head on success. If 'in_mem' is true, we have all
4781 * data in memory that is needed to recreate the on-disk version of this
4784 static int __ext4_get_inode_loc(struct inode *inode,
4785 struct ext4_iloc *iloc, int in_mem)
4787 struct ext4_group_desc *gdp;
4788 struct buffer_head *bh;
4789 struct super_block *sb = inode->i_sb;
4791 int inodes_per_block, inode_offset;
4794 if (!ext4_valid_inum(sb, inode->i_ino))
4797 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4798 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4803 * Figure out the offset within the block group inode table
4805 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4806 inode_offset = ((inode->i_ino - 1) %
4807 EXT4_INODES_PER_GROUP(sb));
4808 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4809 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4811 bh = sb_getblk(sb, block);
4813 ext4_error(sb, "unable to read inode block - "
4814 "inode=%lu, block=%llu", inode->i_ino, block);
4817 if (!buffer_uptodate(bh)) {
4821 * If the buffer has the write error flag, we have failed
4822 * to write out another inode in the same block. In this
4823 * case, we don't have to read the block because we may
4824 * read the old inode data successfully.
4826 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4827 set_buffer_uptodate(bh);
4829 if (buffer_uptodate(bh)) {
4830 /* someone brought it uptodate while we waited */
4836 * If we have all information of the inode in memory and this
4837 * is the only valid inode in the block, we need not read the
4841 struct buffer_head *bitmap_bh;
4844 start = inode_offset & ~(inodes_per_block - 1);
4846 /* Is the inode bitmap in cache? */
4847 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4852 * If the inode bitmap isn't in cache then the
4853 * optimisation may end up performing two reads instead
4854 * of one, so skip it.
4856 if (!buffer_uptodate(bitmap_bh)) {
4860 for (i = start; i < start + inodes_per_block; i++) {
4861 if (i == inode_offset)
4863 if (ext4_test_bit(i, bitmap_bh->b_data))
4867 if (i == start + inodes_per_block) {
4868 /* all other inodes are free, so skip I/O */
4869 memset(bh->b_data, 0, bh->b_size);
4870 set_buffer_uptodate(bh);
4878 * If we need to do any I/O, try to pre-readahead extra
4879 * blocks from the inode table.
4881 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4882 ext4_fsblk_t b, end, table;
4885 table = ext4_inode_table(sb, gdp);
4886 /* s_inode_readahead_blks is always a power of 2 */
4887 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4890 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4891 num = EXT4_INODES_PER_GROUP(sb);
4892 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4893 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4894 num -= ext4_itable_unused_count(sb, gdp);
4895 table += num / inodes_per_block;
4899 sb_breadahead(sb, b++);
4903 * There are other valid inodes in the buffer, this inode
4904 * has in-inode xattrs, or we don't have this inode in memory.
4905 * Read the block from disk.
4908 bh->b_end_io = end_buffer_read_sync;
4909 submit_bh(READ_META, bh);
4911 if (!buffer_uptodate(bh)) {
4912 ext4_error(sb, "unable to read inode block - inode=%lu,"
4913 " block=%llu", inode->i_ino, block);
4923 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4925 /* We have all inode data except xattrs in memory here. */
4926 return __ext4_get_inode_loc(inode, iloc,
4927 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4930 void ext4_set_inode_flags(struct inode *inode)
4932 unsigned int flags = EXT4_I(inode)->i_flags;
4934 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4935 if (flags & EXT4_SYNC_FL)
4936 inode->i_flags |= S_SYNC;
4937 if (flags & EXT4_APPEND_FL)
4938 inode->i_flags |= S_APPEND;
4939 if (flags & EXT4_IMMUTABLE_FL)
4940 inode->i_flags |= S_IMMUTABLE;
4941 if (flags & EXT4_NOATIME_FL)
4942 inode->i_flags |= S_NOATIME;
4943 if (flags & EXT4_DIRSYNC_FL)
4944 inode->i_flags |= S_DIRSYNC;
4947 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4948 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4950 unsigned int flags = ei->vfs_inode.i_flags;
4952 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4953 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4955 ei->i_flags |= EXT4_SYNC_FL;
4956 if (flags & S_APPEND)
4957 ei->i_flags |= EXT4_APPEND_FL;
4958 if (flags & S_IMMUTABLE)
4959 ei->i_flags |= EXT4_IMMUTABLE_FL;
4960 if (flags & S_NOATIME)
4961 ei->i_flags |= EXT4_NOATIME_FL;
4962 if (flags & S_DIRSYNC)
4963 ei->i_flags |= EXT4_DIRSYNC_FL;
4966 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4967 struct ext4_inode_info *ei)
4970 struct inode *inode = &(ei->vfs_inode);
4971 struct super_block *sb = inode->i_sb;
4973 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4974 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4975 /* we are using combined 48 bit field */
4976 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4977 le32_to_cpu(raw_inode->i_blocks_lo);
4978 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4979 /* i_blocks represent file system block size */
4980 return i_blocks << (inode->i_blkbits - 9);
4985 return le32_to_cpu(raw_inode->i_blocks_lo);
4989 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4991 struct ext4_iloc iloc;
4992 struct ext4_inode *raw_inode;
4993 struct ext4_inode_info *ei;
4994 struct inode *inode;
4995 journal_t *journal = EXT4_SB(sb)->s_journal;
4999 inode = iget_locked(sb, ino);
5001 return ERR_PTR(-ENOMEM);
5002 if (!(inode->i_state & I_NEW))
5008 ret = __ext4_get_inode_loc(inode, &iloc, 0);
5011 raw_inode = ext4_raw_inode(&iloc);
5012 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5013 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5014 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5015 if (!(test_opt(inode->i_sb, NO_UID32))) {
5016 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5017 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5019 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5021 ei->i_state_flags = 0;
5022 ei->i_dir_start_lookup = 0;
5023 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5024 /* We now have enough fields to check if the inode was active or not.
5025 * This is needed because nfsd might try to access dead inodes
5026 * the test is that same one that e2fsck uses
5027 * NeilBrown 1999oct15
5029 if (inode->i_nlink == 0) {
5030 if (inode->i_mode == 0 ||
5031 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5032 /* this inode is deleted */
5036 /* The only unlinked inodes we let through here have
5037 * valid i_mode and are being read by the orphan
5038 * recovery code: that's fine, we're about to complete
5039 * the process of deleting those. */
5041 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5042 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5043 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5044 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5046 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5047 inode->i_size = ext4_isize(raw_inode);
5048 ei->i_disksize = inode->i_size;
5050 ei->i_reserved_quota = 0;
5052 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5053 ei->i_block_group = iloc.block_group;
5054 ei->i_last_alloc_group = ~0;
5056 * NOTE! The in-memory inode i_data array is in little-endian order
5057 * even on big-endian machines: we do NOT byteswap the block numbers!
5059 for (block = 0; block < EXT4_N_BLOCKS; block++)
5060 ei->i_data[block] = raw_inode->i_block[block];
5061 INIT_LIST_HEAD(&ei->i_orphan);
5064 * Set transaction id's of transactions that have to be committed
5065 * to finish f[data]sync. We set them to currently running transaction
5066 * as we cannot be sure that the inode or some of its metadata isn't
5067 * part of the transaction - the inode could have been reclaimed and
5068 * now it is reread from disk.
5071 transaction_t *transaction;
5074 spin_lock(&journal->j_state_lock);
5075 if (journal->j_running_transaction)
5076 transaction = journal->j_running_transaction;
5078 transaction = journal->j_committing_transaction;
5080 tid = transaction->t_tid;
5082 tid = journal->j_commit_sequence;
5083 spin_unlock(&journal->j_state_lock);
5084 ei->i_sync_tid = tid;
5085 ei->i_datasync_tid = tid;
5088 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5089 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5090 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5091 EXT4_INODE_SIZE(inode->i_sb)) {
5095 if (ei->i_extra_isize == 0) {
5096 /* The extra space is currently unused. Use it. */
5097 ei->i_extra_isize = sizeof(struct ext4_inode) -
5098 EXT4_GOOD_OLD_INODE_SIZE;
5100 __le32 *magic = (void *)raw_inode +
5101 EXT4_GOOD_OLD_INODE_SIZE +
5103 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5104 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5107 ei->i_extra_isize = 0;
5109 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5110 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5111 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5112 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5114 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5115 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5116 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5118 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5122 if (ei->i_file_acl &&
5123 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5124 ext4_error(sb, "bad extended attribute block %llu inode #%lu",
5125 ei->i_file_acl, inode->i_ino);
5128 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
5129 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5130 (S_ISLNK(inode->i_mode) &&
5131 !ext4_inode_is_fast_symlink(inode)))
5132 /* Validate extent which is part of inode */
5133 ret = ext4_ext_check_inode(inode);
5134 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5135 (S_ISLNK(inode->i_mode) &&
5136 !ext4_inode_is_fast_symlink(inode))) {
5137 /* Validate block references which are part of inode */
5138 ret = ext4_check_inode_blockref(inode);
5143 if (S_ISREG(inode->i_mode)) {
5144 inode->i_op = &ext4_file_inode_operations;
5145 inode->i_fop = &ext4_file_operations;
5146 ext4_set_aops(inode);
5147 } else if (S_ISDIR(inode->i_mode)) {
5148 inode->i_op = &ext4_dir_inode_operations;
5149 inode->i_fop = &ext4_dir_operations;
5150 } else if (S_ISLNK(inode->i_mode)) {
5151 if (ext4_inode_is_fast_symlink(inode)) {
5152 inode->i_op = &ext4_fast_symlink_inode_operations;
5153 nd_terminate_link(ei->i_data, inode->i_size,
5154 sizeof(ei->i_data) - 1);
5156 inode->i_op = &ext4_symlink_inode_operations;
5157 ext4_set_aops(inode);
5159 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5160 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5161 inode->i_op = &ext4_special_inode_operations;
5162 if (raw_inode->i_block[0])
5163 init_special_inode(inode, inode->i_mode,
5164 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5166 init_special_inode(inode, inode->i_mode,
5167 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5170 ext4_error(inode->i_sb, "bogus i_mode (%o) for inode=%lu",
5171 inode->i_mode, inode->i_ino);
5175 ext4_set_inode_flags(inode);
5176 unlock_new_inode(inode);
5182 return ERR_PTR(ret);
5185 static int ext4_inode_blocks_set(handle_t *handle,
5186 struct ext4_inode *raw_inode,
5187 struct ext4_inode_info *ei)
5189 struct inode *inode = &(ei->vfs_inode);
5190 u64 i_blocks = inode->i_blocks;
5191 struct super_block *sb = inode->i_sb;
5193 if (i_blocks <= ~0U) {
5195 * i_blocks can be represnted in a 32 bit variable
5196 * as multiple of 512 bytes
5198 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5199 raw_inode->i_blocks_high = 0;
5200 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5203 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5206 if (i_blocks <= 0xffffffffffffULL) {
5208 * i_blocks can be represented in a 48 bit variable
5209 * as multiple of 512 bytes
5211 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5212 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5213 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
5215 ei->i_flags |= EXT4_HUGE_FILE_FL;
5216 /* i_block is stored in file system block size */
5217 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5218 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5219 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5225 * Post the struct inode info into an on-disk inode location in the
5226 * buffer-cache. This gobbles the caller's reference to the
5227 * buffer_head in the inode location struct.
5229 * The caller must have write access to iloc->bh.
5231 static int ext4_do_update_inode(handle_t *handle,
5232 struct inode *inode,
5233 struct ext4_iloc *iloc)
5235 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5236 struct ext4_inode_info *ei = EXT4_I(inode);
5237 struct buffer_head *bh = iloc->bh;
5238 int err = 0, rc, block;
5240 /* For fields not not tracking in the in-memory inode,
5241 * initialise them to zero for new inodes. */
5242 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5243 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5245 ext4_get_inode_flags(ei);
5246 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5247 if (!(test_opt(inode->i_sb, NO_UID32))) {
5248 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5249 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5251 * Fix up interoperability with old kernels. Otherwise, old inodes get
5252 * re-used with the upper 16 bits of the uid/gid intact
5255 raw_inode->i_uid_high =
5256 cpu_to_le16(high_16_bits(inode->i_uid));
5257 raw_inode->i_gid_high =
5258 cpu_to_le16(high_16_bits(inode->i_gid));
5260 raw_inode->i_uid_high = 0;
5261 raw_inode->i_gid_high = 0;
5264 raw_inode->i_uid_low =
5265 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5266 raw_inode->i_gid_low =
5267 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5268 raw_inode->i_uid_high = 0;
5269 raw_inode->i_gid_high = 0;
5271 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5273 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5274 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5275 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5276 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5278 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5280 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5281 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5282 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5283 cpu_to_le32(EXT4_OS_HURD))
5284 raw_inode->i_file_acl_high =
5285 cpu_to_le16(ei->i_file_acl >> 32);
5286 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5287 ext4_isize_set(raw_inode, ei->i_disksize);
5288 if (ei->i_disksize > 0x7fffffffULL) {
5289 struct super_block *sb = inode->i_sb;
5290 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5291 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5292 EXT4_SB(sb)->s_es->s_rev_level ==
5293 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5294 /* If this is the first large file
5295 * created, add a flag to the superblock.
5297 err = ext4_journal_get_write_access(handle,
5298 EXT4_SB(sb)->s_sbh);
5301 ext4_update_dynamic_rev(sb);
5302 EXT4_SET_RO_COMPAT_FEATURE(sb,
5303 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5305 ext4_handle_sync(handle);
5306 err = ext4_handle_dirty_metadata(handle, NULL,
5307 EXT4_SB(sb)->s_sbh);
5310 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5311 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5312 if (old_valid_dev(inode->i_rdev)) {
5313 raw_inode->i_block[0] =
5314 cpu_to_le32(old_encode_dev(inode->i_rdev));
5315 raw_inode->i_block[1] = 0;
5317 raw_inode->i_block[0] = 0;
5318 raw_inode->i_block[1] =
5319 cpu_to_le32(new_encode_dev(inode->i_rdev));
5320 raw_inode->i_block[2] = 0;
5323 for (block = 0; block < EXT4_N_BLOCKS; block++)
5324 raw_inode->i_block[block] = ei->i_data[block];
5326 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5327 if (ei->i_extra_isize) {
5328 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5329 raw_inode->i_version_hi =
5330 cpu_to_le32(inode->i_version >> 32);
5331 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5334 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5335 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5338 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5340 ext4_update_inode_fsync_trans(handle, inode, 0);
5343 ext4_std_error(inode->i_sb, err);
5348 * ext4_write_inode()
5350 * We are called from a few places:
5352 * - Within generic_file_write() for O_SYNC files.
5353 * Here, there will be no transaction running. We wait for any running
5354 * trasnaction to commit.
5356 * - Within sys_sync(), kupdate and such.
5357 * We wait on commit, if tol to.
5359 * - Within prune_icache() (PF_MEMALLOC == true)
5360 * Here we simply return. We can't afford to block kswapd on the
5363 * In all cases it is actually safe for us to return without doing anything,
5364 * because the inode has been copied into a raw inode buffer in
5365 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5368 * Note that we are absolutely dependent upon all inode dirtiers doing the
5369 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5370 * which we are interested.
5372 * It would be a bug for them to not do this. The code:
5374 * mark_inode_dirty(inode)
5376 * inode->i_size = expr;
5378 * is in error because a kswapd-driven write_inode() could occur while
5379 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5380 * will no longer be on the superblock's dirty inode list.
5382 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5386 if (current->flags & PF_MEMALLOC)
5389 if (EXT4_SB(inode->i_sb)->s_journal) {
5390 if (ext4_journal_current_handle()) {
5391 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5396 if (wbc->sync_mode != WB_SYNC_ALL)
5399 err = ext4_force_commit(inode->i_sb);
5401 struct ext4_iloc iloc;
5403 err = __ext4_get_inode_loc(inode, &iloc, 0);
5406 if (wbc->sync_mode == WB_SYNC_ALL)
5407 sync_dirty_buffer(iloc.bh);
5408 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5409 ext4_error(inode->i_sb, "IO error syncing inode, "
5410 "inode=%lu, block=%llu", inode->i_ino,
5411 (unsigned long long)iloc.bh->b_blocknr);
5422 * Called from notify_change.
5424 * We want to trap VFS attempts to truncate the file as soon as
5425 * possible. In particular, we want to make sure that when the VFS
5426 * shrinks i_size, we put the inode on the orphan list and modify
5427 * i_disksize immediately, so that during the subsequent flushing of
5428 * dirty pages and freeing of disk blocks, we can guarantee that any
5429 * commit will leave the blocks being flushed in an unused state on
5430 * disk. (On recovery, the inode will get truncated and the blocks will
5431 * be freed, so we have a strong guarantee that no future commit will
5432 * leave these blocks visible to the user.)
5434 * Another thing we have to assure is that if we are in ordered mode
5435 * and inode is still attached to the committing transaction, we must
5436 * we start writeout of all the dirty pages which are being truncated.
5437 * This way we are sure that all the data written in the previous
5438 * transaction are already on disk (truncate waits for pages under
5441 * Called with inode->i_mutex down.
5443 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5445 struct inode *inode = dentry->d_inode;
5447 const unsigned int ia_valid = attr->ia_valid;
5449 error = inode_change_ok(inode, attr);
5453 if (ia_valid & ATTR_SIZE)
5454 dquot_initialize(inode);
5455 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5456 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5459 /* (user+group)*(old+new) structure, inode write (sb,
5460 * inode block, ? - but truncate inode update has it) */
5461 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5462 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5463 if (IS_ERR(handle)) {
5464 error = PTR_ERR(handle);
5467 error = dquot_transfer(inode, attr);
5469 ext4_journal_stop(handle);
5472 /* Update corresponding info in inode so that everything is in
5473 * one transaction */
5474 if (attr->ia_valid & ATTR_UID)
5475 inode->i_uid = attr->ia_uid;
5476 if (attr->ia_valid & ATTR_GID)
5477 inode->i_gid = attr->ia_gid;
5478 error = ext4_mark_inode_dirty(handle, inode);
5479 ext4_journal_stop(handle);
5482 if (attr->ia_valid & ATTR_SIZE) {
5483 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5484 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5486 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5493 if (S_ISREG(inode->i_mode) &&
5494 attr->ia_valid & ATTR_SIZE &&
5495 (attr->ia_size < inode->i_size ||
5496 (EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))) {
5499 handle = ext4_journal_start(inode, 3);
5500 if (IS_ERR(handle)) {
5501 error = PTR_ERR(handle);
5505 error = ext4_orphan_add(handle, inode);
5506 EXT4_I(inode)->i_disksize = attr->ia_size;
5507 rc = ext4_mark_inode_dirty(handle, inode);
5510 ext4_journal_stop(handle);
5512 if (ext4_should_order_data(inode)) {
5513 error = ext4_begin_ordered_truncate(inode,
5516 /* Do as much error cleanup as possible */
5517 handle = ext4_journal_start(inode, 3);
5518 if (IS_ERR(handle)) {
5519 ext4_orphan_del(NULL, inode);
5522 ext4_orphan_del(handle, inode);
5523 ext4_journal_stop(handle);
5527 /* ext4_truncate will clear the flag */
5528 if ((EXT4_I(inode)->i_flags & EXT4_EOFBLOCKS_FL))
5529 ext4_truncate(inode);
5532 rc = inode_setattr(inode, attr);
5534 /* If inode_setattr's call to ext4_truncate failed to get a
5535 * transaction handle at all, we need to clean up the in-core
5536 * orphan list manually. */
5538 ext4_orphan_del(NULL, inode);
5540 if (!rc && (ia_valid & ATTR_MODE))
5541 rc = ext4_acl_chmod(inode);
5544 ext4_std_error(inode->i_sb, error);
5550 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5553 struct inode *inode;
5554 unsigned long delalloc_blocks;
5556 inode = dentry->d_inode;
5557 generic_fillattr(inode, stat);
5560 * We can't update i_blocks if the block allocation is delayed
5561 * otherwise in the case of system crash before the real block
5562 * allocation is done, we will have i_blocks inconsistent with
5563 * on-disk file blocks.
5564 * We always keep i_blocks updated together with real
5565 * allocation. But to not confuse with user, stat
5566 * will return the blocks that include the delayed allocation
5567 * blocks for this file.
5569 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5570 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5571 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5573 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5577 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5582 /* if nrblocks are contiguous */
5585 * With N contiguous data blocks, it need at most
5586 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5587 * 2 dindirect blocks
5590 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5591 return indirects + 3;
5594 * if nrblocks are not contiguous, worse case, each block touch
5595 * a indirect block, and each indirect block touch a double indirect
5596 * block, plus a triple indirect block
5598 indirects = nrblocks * 2 + 1;
5602 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5604 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5605 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5606 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5610 * Account for index blocks, block groups bitmaps and block group
5611 * descriptor blocks if modify datablocks and index blocks
5612 * worse case, the indexs blocks spread over different block groups
5614 * If datablocks are discontiguous, they are possible to spread over
5615 * different block groups too. If they are contiuguous, with flexbg,
5616 * they could still across block group boundary.
5618 * Also account for superblock, inode, quota and xattr blocks
5620 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5622 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5628 * How many index blocks need to touch to modify nrblocks?
5629 * The "Chunk" flag indicating whether the nrblocks is
5630 * physically contiguous on disk
5632 * For Direct IO and fallocate, they calls get_block to allocate
5633 * one single extent at a time, so they could set the "Chunk" flag
5635 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5640 * Now let's see how many group bitmaps and group descriptors need
5650 if (groups > ngroups)
5652 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5653 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5655 /* bitmaps and block group descriptor blocks */
5656 ret += groups + gdpblocks;
5658 /* Blocks for super block, inode, quota and xattr blocks */
5659 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5665 * Calulate the total number of credits to reserve to fit
5666 * the modification of a single pages into a single transaction,
5667 * which may include multiple chunks of block allocations.
5669 * This could be called via ext4_write_begin()
5671 * We need to consider the worse case, when
5672 * one new block per extent.
5674 int ext4_writepage_trans_blocks(struct inode *inode)
5676 int bpp = ext4_journal_blocks_per_page(inode);
5679 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5681 /* Account for data blocks for journalled mode */
5682 if (ext4_should_journal_data(inode))
5688 * Calculate the journal credits for a chunk of data modification.
5690 * This is called from DIO, fallocate or whoever calling
5691 * ext4_get_blocks() to map/allocate a chunk of contiguous disk blocks.
5693 * journal buffers for data blocks are not included here, as DIO
5694 * and fallocate do no need to journal data buffers.
5696 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5698 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5702 * The caller must have previously called ext4_reserve_inode_write().
5703 * Give this, we know that the caller already has write access to iloc->bh.
5705 int ext4_mark_iloc_dirty(handle_t *handle,
5706 struct inode *inode, struct ext4_iloc *iloc)
5710 if (test_opt(inode->i_sb, I_VERSION))
5711 inode_inc_iversion(inode);
5713 /* the do_update_inode consumes one bh->b_count */
5716 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5717 err = ext4_do_update_inode(handle, inode, iloc);
5723 * On success, We end up with an outstanding reference count against
5724 * iloc->bh. This _must_ be cleaned up later.
5728 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5729 struct ext4_iloc *iloc)
5733 err = ext4_get_inode_loc(inode, iloc);
5735 BUFFER_TRACE(iloc->bh, "get_write_access");
5736 err = ext4_journal_get_write_access(handle, iloc->bh);
5742 ext4_std_error(inode->i_sb, err);
5747 * Expand an inode by new_extra_isize bytes.
5748 * Returns 0 on success or negative error number on failure.
5750 static int ext4_expand_extra_isize(struct inode *inode,
5751 unsigned int new_extra_isize,
5752 struct ext4_iloc iloc,
5755 struct ext4_inode *raw_inode;
5756 struct ext4_xattr_ibody_header *header;
5757 struct ext4_xattr_entry *entry;
5759 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5762 raw_inode = ext4_raw_inode(&iloc);
5764 header = IHDR(inode, raw_inode);
5765 entry = IFIRST(header);
5767 /* No extended attributes present */
5768 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5769 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5770 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5772 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5776 /* try to expand with EAs present */
5777 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5782 * What we do here is to mark the in-core inode as clean with respect to inode
5783 * dirtiness (it may still be data-dirty).
5784 * This means that the in-core inode may be reaped by prune_icache
5785 * without having to perform any I/O. This is a very good thing,
5786 * because *any* task may call prune_icache - even ones which
5787 * have a transaction open against a different journal.
5789 * Is this cheating? Not really. Sure, we haven't written the
5790 * inode out, but prune_icache isn't a user-visible syncing function.
5791 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5792 * we start and wait on commits.
5794 * Is this efficient/effective? Well, we're being nice to the system
5795 * by cleaning up our inodes proactively so they can be reaped
5796 * without I/O. But we are potentially leaving up to five seconds'
5797 * worth of inodes floating about which prune_icache wants us to
5798 * write out. One way to fix that would be to get prune_icache()
5799 * to do a write_super() to free up some memory. It has the desired
5802 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5804 struct ext4_iloc iloc;
5805 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5806 static unsigned int mnt_count;
5810 err = ext4_reserve_inode_write(handle, inode, &iloc);
5811 if (ext4_handle_valid(handle) &&
5812 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5813 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5815 * We need extra buffer credits since we may write into EA block
5816 * with this same handle. If journal_extend fails, then it will
5817 * only result in a minor loss of functionality for that inode.
5818 * If this is felt to be critical, then e2fsck should be run to
5819 * force a large enough s_min_extra_isize.
5821 if ((jbd2_journal_extend(handle,
5822 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5823 ret = ext4_expand_extra_isize(inode,
5824 sbi->s_want_extra_isize,
5827 ext4_set_inode_state(inode,
5828 EXT4_STATE_NO_EXPAND);
5830 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5831 ext4_warning(inode->i_sb,
5832 "Unable to expand inode %lu. Delete"
5833 " some EAs or run e2fsck.",
5836 le16_to_cpu(sbi->s_es->s_mnt_count);
5842 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5847 * ext4_dirty_inode() is called from __mark_inode_dirty()
5849 * We're really interested in the case where a file is being extended.
5850 * i_size has been changed by generic_commit_write() and we thus need
5851 * to include the updated inode in the current transaction.
5853 * Also, dquot_alloc_block() will always dirty the inode when blocks
5854 * are allocated to the file.
5856 * If the inode is marked synchronous, we don't honour that here - doing
5857 * so would cause a commit on atime updates, which we don't bother doing.
5858 * We handle synchronous inodes at the highest possible level.
5860 void ext4_dirty_inode(struct inode *inode)
5864 handle = ext4_journal_start(inode, 2);
5868 ext4_mark_inode_dirty(handle, inode);
5870 ext4_journal_stop(handle);
5877 * Bind an inode's backing buffer_head into this transaction, to prevent
5878 * it from being flushed to disk early. Unlike
5879 * ext4_reserve_inode_write, this leaves behind no bh reference and
5880 * returns no iloc structure, so the caller needs to repeat the iloc
5881 * lookup to mark the inode dirty later.
5883 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5885 struct ext4_iloc iloc;
5889 err = ext4_get_inode_loc(inode, &iloc);
5891 BUFFER_TRACE(iloc.bh, "get_write_access");
5892 err = jbd2_journal_get_write_access(handle, iloc.bh);
5894 err = ext4_handle_dirty_metadata(handle,
5900 ext4_std_error(inode->i_sb, err);
5905 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5912 * We have to be very careful here: changing a data block's
5913 * journaling status dynamically is dangerous. If we write a
5914 * data block to the journal, change the status and then delete
5915 * that block, we risk forgetting to revoke the old log record
5916 * from the journal and so a subsequent replay can corrupt data.
5917 * So, first we make sure that the journal is empty and that
5918 * nobody is changing anything.
5921 journal = EXT4_JOURNAL(inode);
5924 if (is_journal_aborted(journal))
5927 jbd2_journal_lock_updates(journal);
5928 jbd2_journal_flush(journal);
5931 * OK, there are no updates running now, and all cached data is
5932 * synced to disk. We are now in a completely consistent state
5933 * which doesn't have anything in the journal, and we know that
5934 * no filesystem updates are running, so it is safe to modify
5935 * the inode's in-core data-journaling state flag now.
5939 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5941 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5942 ext4_set_aops(inode);
5944 jbd2_journal_unlock_updates(journal);
5946 /* Finally we can mark the inode as dirty. */
5948 handle = ext4_journal_start(inode, 1);
5950 return PTR_ERR(handle);
5952 err = ext4_mark_inode_dirty(handle, inode);
5953 ext4_handle_sync(handle);
5954 ext4_journal_stop(handle);
5955 ext4_std_error(inode->i_sb, err);
5960 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5962 return !buffer_mapped(bh);
5965 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5967 struct page *page = vmf->page;
5972 struct file *file = vma->vm_file;
5973 struct inode *inode = file->f_path.dentry->d_inode;
5974 struct address_space *mapping = inode->i_mapping;
5977 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5978 * get i_mutex because we are already holding mmap_sem.
5980 down_read(&inode->i_alloc_sem);
5981 size = i_size_read(inode);
5982 if (page->mapping != mapping || size <= page_offset(page)
5983 || !PageUptodate(page)) {
5984 /* page got truncated from under us? */
5988 if (PageMappedToDisk(page))
5991 if (page->index == size >> PAGE_CACHE_SHIFT)
5992 len = size & ~PAGE_CACHE_MASK;
5994 len = PAGE_CACHE_SIZE;
5998 * return if we have all the buffers mapped. This avoid
5999 * the need to call write_begin/write_end which does a
6000 * journal_start/journal_stop which can block and take
6003 if (page_has_buffers(page)) {
6004 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
6005 ext4_bh_unmapped)) {
6012 * OK, we need to fill the hole... Do write_begin write_end
6013 * to do block allocation/reservation.We are not holding
6014 * inode.i__mutex here. That allow * parallel write_begin,
6015 * write_end call. lock_page prevent this from happening
6016 * on the same page though
6018 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6019 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6022 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6023 len, len, page, fsdata);
6029 ret = VM_FAULT_SIGBUS;
6030 up_read(&inode->i_alloc_sem);