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);
224 ext4_orphan_del(NULL, inode);
230 * Kill off the orphan record which ext4_truncate created.
231 * AKPM: I think this can be inside the above `if'.
232 * Note that ext4_orphan_del() has to be able to cope with the
233 * deletion of a non-existent orphan - this is because we don't
234 * know if ext4_truncate() actually created an orphan record.
235 * (Well, we could do this if we need to, but heck - it works)
237 ext4_orphan_del(handle, inode);
238 EXT4_I(inode)->i_dtime = get_seconds();
241 * One subtle ordering requirement: if anything has gone wrong
242 * (transaction abort, IO errors, whatever), then we can still
243 * do these next steps (the fs will already have been marked as
244 * having errors), but we can't free the inode if the mark_dirty
247 if (ext4_mark_inode_dirty(handle, inode))
248 /* If that failed, just do the required in-core inode clear. */
251 ext4_free_inode(handle, inode);
252 ext4_journal_stop(handle);
255 clear_inode(inode); /* We must guarantee clearing of inode... */
261 struct buffer_head *bh;
264 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
266 p->key = *(p->p = v);
271 * ext4_block_to_path - parse the block number into array of offsets
272 * @inode: inode in question (we are only interested in its superblock)
273 * @i_block: block number to be parsed
274 * @offsets: array to store the offsets in
275 * @boundary: set this non-zero if the referred-to block is likely to be
276 * followed (on disk) by an indirect block.
278 * To store the locations of file's data ext4 uses a data structure common
279 * for UNIX filesystems - tree of pointers anchored in the inode, with
280 * data blocks at leaves and indirect blocks in intermediate nodes.
281 * This function translates the block number into path in that tree -
282 * return value is the path length and @offsets[n] is the offset of
283 * pointer to (n+1)th node in the nth one. If @block is out of range
284 * (negative or too large) warning is printed and zero returned.
286 * Note: function doesn't find node addresses, so no IO is needed. All
287 * we need to know is the capacity of indirect blocks (taken from the
292 * Portability note: the last comparison (check that we fit into triple
293 * indirect block) is spelled differently, because otherwise on an
294 * architecture with 32-bit longs and 8Kb pages we might get into trouble
295 * if our filesystem had 8Kb blocks. We might use long long, but that would
296 * kill us on x86. Oh, well, at least the sign propagation does not matter -
297 * i_block would have to be negative in the very beginning, so we would not
301 static int ext4_block_to_path(struct inode *inode,
303 ext4_lblk_t offsets[4], int *boundary)
305 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
306 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
307 const long direct_blocks = EXT4_NDIR_BLOCKS,
308 indirect_blocks = ptrs,
309 double_blocks = (1 << (ptrs_bits * 2));
313 if (i_block < direct_blocks) {
314 offsets[n++] = i_block;
315 final = direct_blocks;
316 } else if ((i_block -= direct_blocks) < indirect_blocks) {
317 offsets[n++] = EXT4_IND_BLOCK;
318 offsets[n++] = i_block;
320 } else if ((i_block -= indirect_blocks) < double_blocks) {
321 offsets[n++] = EXT4_DIND_BLOCK;
322 offsets[n++] = i_block >> ptrs_bits;
323 offsets[n++] = i_block & (ptrs - 1);
325 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
326 offsets[n++] = EXT4_TIND_BLOCK;
327 offsets[n++] = i_block >> (ptrs_bits * 2);
328 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
329 offsets[n++] = i_block & (ptrs - 1);
332 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
333 i_block + direct_blocks +
334 indirect_blocks + double_blocks, inode->i_ino);
337 *boundary = final - 1 - (i_block & (ptrs - 1));
341 static int __ext4_check_blockref(const char *function, unsigned int line,
343 __le32 *p, unsigned int max)
345 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
349 while (bref < p+max) {
350 blk = le32_to_cpu(*bref++);
352 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
354 es->s_last_error_block = cpu_to_le64(blk);
355 ext4_error_inode(inode, function, line, blk,
364 #define ext4_check_indirect_blockref(inode, bh) \
365 __ext4_check_blockref(__func__, __LINE__, inode, \
366 (__le32 *)(bh)->b_data, \
367 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
369 #define ext4_check_inode_blockref(inode) \
370 __ext4_check_blockref(__func__, __LINE__, inode, \
371 EXT4_I(inode)->i_data, \
375 * ext4_get_branch - read the chain of indirect blocks leading to data
376 * @inode: inode in question
377 * @depth: depth of the chain (1 - direct pointer, etc.)
378 * @offsets: offsets of pointers in inode/indirect blocks
379 * @chain: place to store the result
380 * @err: here we store the error value
382 * Function fills the array of triples <key, p, bh> and returns %NULL
383 * if everything went OK or the pointer to the last filled triple
384 * (incomplete one) otherwise. Upon the return chain[i].key contains
385 * the number of (i+1)-th block in the chain (as it is stored in memory,
386 * i.e. little-endian 32-bit), chain[i].p contains the address of that
387 * number (it points into struct inode for i==0 and into the bh->b_data
388 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
389 * block for i>0 and NULL for i==0. In other words, it holds the block
390 * numbers of the chain, addresses they were taken from (and where we can
391 * verify that chain did not change) and buffer_heads hosting these
394 * Function stops when it stumbles upon zero pointer (absent block)
395 * (pointer to last triple returned, *@err == 0)
396 * or when it gets an IO error reading an indirect block
397 * (ditto, *@err == -EIO)
398 * or when it reads all @depth-1 indirect blocks successfully and finds
399 * the whole chain, all way to the data (returns %NULL, *err == 0).
401 * Need to be called with
402 * down_read(&EXT4_I(inode)->i_data_sem)
404 static Indirect *ext4_get_branch(struct inode *inode, int depth,
405 ext4_lblk_t *offsets,
406 Indirect chain[4], int *err)
408 struct super_block *sb = inode->i_sb;
410 struct buffer_head *bh;
413 /* i_data is not going away, no lock needed */
414 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
418 bh = sb_getblk(sb, le32_to_cpu(p->key));
422 if (!bh_uptodate_or_lock(bh)) {
423 if (bh_submit_read(bh) < 0) {
427 /* validate block references */
428 if (ext4_check_indirect_blockref(inode, bh)) {
434 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
448 * ext4_find_near - find a place for allocation with sufficient locality
450 * @ind: descriptor of indirect block.
452 * This function returns the preferred place for block allocation.
453 * It is used when heuristic for sequential allocation fails.
455 * + if there is a block to the left of our position - allocate near it.
456 * + if pointer will live in indirect block - allocate near that block.
457 * + if pointer will live in inode - allocate in the same
460 * In the latter case we colour the starting block by the callers PID to
461 * prevent it from clashing with concurrent allocations for a different inode
462 * in the same block group. The PID is used here so that functionally related
463 * files will be close-by on-disk.
465 * Caller must make sure that @ind is valid and will stay that way.
467 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
469 struct ext4_inode_info *ei = EXT4_I(inode);
470 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
472 ext4_fsblk_t bg_start;
473 ext4_fsblk_t last_block;
474 ext4_grpblk_t colour;
475 ext4_group_t block_group;
476 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
478 /* Try to find previous block */
479 for (p = ind->p - 1; p >= start; p--) {
481 return le32_to_cpu(*p);
484 /* No such thing, so let's try location of indirect block */
486 return ind->bh->b_blocknr;
489 * It is going to be referred to from the inode itself? OK, just put it
490 * into the same cylinder group then.
492 block_group = ei->i_block_group;
493 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
494 block_group &= ~(flex_size-1);
495 if (S_ISREG(inode->i_mode))
498 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
499 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
502 * If we are doing delayed allocation, we don't need take
503 * colour into account.
505 if (test_opt(inode->i_sb, DELALLOC))
508 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
509 colour = (current->pid % 16) *
510 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
512 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
513 return bg_start + colour;
517 * ext4_find_goal - find a preferred place for allocation.
519 * @block: block we want
520 * @partial: pointer to the last triple within a chain
522 * Normally this function find the preferred place for block allocation,
524 * Because this is only used for non-extent files, we limit the block nr
527 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
533 * XXX need to get goal block from mballoc's data structures
536 goal = ext4_find_near(inode, partial);
537 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
542 * ext4_blks_to_allocate: Look up the block map and count the number
543 * of direct blocks need to be allocated for the given branch.
545 * @branch: chain of indirect blocks
546 * @k: number of blocks need for indirect blocks
547 * @blks: number of data blocks to be mapped.
548 * @blocks_to_boundary: the offset in the indirect block
550 * return the total number of blocks to be allocate, including the
551 * direct and indirect blocks.
553 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
554 int blocks_to_boundary)
556 unsigned int count = 0;
559 * Simple case, [t,d]Indirect block(s) has not allocated yet
560 * then it's clear blocks on that path have not allocated
563 /* right now we don't handle cross boundary allocation */
564 if (blks < blocks_to_boundary + 1)
567 count += blocks_to_boundary + 1;
572 while (count < blks && count <= blocks_to_boundary &&
573 le32_to_cpu(*(branch[0].p + count)) == 0) {
580 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
581 * @indirect_blks: the number of blocks need to allocate for indirect
584 * @new_blocks: on return it will store the new block numbers for
585 * the indirect blocks(if needed) and the first direct block,
586 * @blks: on return it will store the total number of allocated
589 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
590 ext4_lblk_t iblock, ext4_fsblk_t goal,
591 int indirect_blks, int blks,
592 ext4_fsblk_t new_blocks[4], int *err)
594 struct ext4_allocation_request ar;
596 unsigned long count = 0, blk_allocated = 0;
598 ext4_fsblk_t current_block = 0;
602 * Here we try to allocate the requested multiple blocks at once,
603 * on a best-effort basis.
604 * To build a branch, we should allocate blocks for
605 * the indirect blocks(if not allocated yet), and at least
606 * the first direct block of this branch. That's the
607 * minimum number of blocks need to allocate(required)
609 /* first we try to allocate the indirect blocks */
610 target = indirect_blks;
613 /* allocating blocks for indirect blocks and direct blocks */
614 current_block = ext4_new_meta_blocks(handle, inode,
619 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
620 EXT4_ERROR_INODE(inode,
621 "current_block %llu + count %lu > %d!",
622 current_block, count,
623 EXT4_MAX_BLOCK_FILE_PHYS);
629 /* allocate blocks for indirect blocks */
630 while (index < indirect_blks && count) {
631 new_blocks[index++] = current_block++;
636 * save the new block number
637 * for the first direct block
639 new_blocks[index] = current_block;
640 printk(KERN_INFO "%s returned more blocks than "
641 "requested\n", __func__);
647 target = blks - count ;
648 blk_allocated = count;
651 /* Now allocate data blocks */
652 memset(&ar, 0, sizeof(ar));
657 if (S_ISREG(inode->i_mode))
658 /* enable in-core preallocation only for regular files */
659 ar.flags = EXT4_MB_HINT_DATA;
661 current_block = ext4_mb_new_blocks(handle, &ar, err);
662 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
663 EXT4_ERROR_INODE(inode,
664 "current_block %llu + ar.len %d > %d!",
665 current_block, ar.len,
666 EXT4_MAX_BLOCK_FILE_PHYS);
671 if (*err && (target == blks)) {
673 * if the allocation failed and we didn't allocate
679 if (target == blks) {
681 * save the new block number
682 * for the first direct block
684 new_blocks[index] = current_block;
686 blk_allocated += ar.len;
689 /* total number of blocks allocated for direct blocks */
694 for (i = 0; i < index; i++)
695 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
700 * ext4_alloc_branch - allocate and set up a chain of blocks.
702 * @indirect_blks: number of allocated indirect blocks
703 * @blks: number of allocated direct blocks
704 * @offsets: offsets (in the blocks) to store the pointers to next.
705 * @branch: place to store the chain in.
707 * This function allocates blocks, zeroes out all but the last one,
708 * links them into chain and (if we are synchronous) writes them to disk.
709 * In other words, it prepares a branch that can be spliced onto the
710 * inode. It stores the information about that chain in the branch[], in
711 * the same format as ext4_get_branch() would do. We are calling it after
712 * we had read the existing part of chain and partial points to the last
713 * triple of that (one with zero ->key). Upon the exit we have the same
714 * picture as after the successful ext4_get_block(), except that in one
715 * place chain is disconnected - *branch->p is still zero (we did not
716 * set the last link), but branch->key contains the number that should
717 * be placed into *branch->p to fill that gap.
719 * If allocation fails we free all blocks we've allocated (and forget
720 * their buffer_heads) and return the error value the from failed
721 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
722 * as described above and return 0.
724 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
725 ext4_lblk_t iblock, int indirect_blks,
726 int *blks, ext4_fsblk_t goal,
727 ext4_lblk_t *offsets, Indirect *branch)
729 int blocksize = inode->i_sb->s_blocksize;
732 struct buffer_head *bh;
734 ext4_fsblk_t new_blocks[4];
735 ext4_fsblk_t current_block;
737 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
738 *blks, new_blocks, &err);
742 branch[0].key = cpu_to_le32(new_blocks[0]);
744 * metadata blocks and data blocks are allocated.
746 for (n = 1; n <= indirect_blks; n++) {
748 * Get buffer_head for parent block, zero it out
749 * and set the pointer to new one, then send
752 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
755 BUFFER_TRACE(bh, "call get_create_access");
756 err = ext4_journal_get_create_access(handle, bh);
758 /* Don't brelse(bh) here; it's done in
759 * ext4_journal_forget() below */
764 memset(bh->b_data, 0, blocksize);
765 branch[n].p = (__le32 *) bh->b_data + offsets[n];
766 branch[n].key = cpu_to_le32(new_blocks[n]);
767 *branch[n].p = branch[n].key;
768 if (n == indirect_blks) {
769 current_block = new_blocks[n];
771 * End of chain, update the last new metablock of
772 * the chain to point to the new allocated
773 * data blocks numbers
775 for (i = 1; i < num; i++)
776 *(branch[n].p + i) = cpu_to_le32(++current_block);
778 BUFFER_TRACE(bh, "marking uptodate");
779 set_buffer_uptodate(bh);
782 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
783 err = ext4_handle_dirty_metadata(handle, inode, bh);
790 /* Allocation failed, free what we already allocated */
791 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
792 for (i = 1; i <= n ; i++) {
794 * branch[i].bh is newly allocated, so there is no
795 * need to revoke the block, which is why we don't
796 * need to set EXT4_FREE_BLOCKS_METADATA.
798 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
799 EXT4_FREE_BLOCKS_FORGET);
801 for (i = n+1; i < indirect_blks; i++)
802 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
804 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
810 * ext4_splice_branch - splice the allocated branch onto inode.
812 * @block: (logical) number of block we are adding
813 * @chain: chain of indirect blocks (with a missing link - see
815 * @where: location of missing link
816 * @num: number of indirect blocks we are adding
817 * @blks: number of direct blocks we are adding
819 * This function fills the missing link and does all housekeeping needed in
820 * inode (->i_blocks, etc.). In case of success we end up with the full
821 * chain to new block and return 0.
823 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
824 ext4_lblk_t block, Indirect *where, int num,
829 ext4_fsblk_t current_block;
832 * If we're splicing into a [td]indirect block (as opposed to the
833 * inode) then we need to get write access to the [td]indirect block
837 BUFFER_TRACE(where->bh, "get_write_access");
838 err = ext4_journal_get_write_access(handle, where->bh);
844 *where->p = where->key;
847 * Update the host buffer_head or inode to point to more just allocated
848 * direct blocks blocks
850 if (num == 0 && blks > 1) {
851 current_block = le32_to_cpu(where->key) + 1;
852 for (i = 1; i < blks; i++)
853 *(where->p + i) = cpu_to_le32(current_block++);
856 /* We are done with atomic stuff, now do the rest of housekeeping */
857 /* had we spliced it onto indirect block? */
860 * If we spliced it onto an indirect block, we haven't
861 * altered the inode. Note however that if it is being spliced
862 * onto an indirect block at the very end of the file (the
863 * file is growing) then we *will* alter the inode to reflect
864 * the new i_size. But that is not done here - it is done in
865 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
867 jbd_debug(5, "splicing indirect only\n");
868 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
869 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
874 * OK, we spliced it into the inode itself on a direct block.
876 ext4_mark_inode_dirty(handle, inode);
877 jbd_debug(5, "splicing direct\n");
882 for (i = 1; i <= num; i++) {
884 * branch[i].bh is newly allocated, so there is no
885 * need to revoke the block, which is why we don't
886 * need to set EXT4_FREE_BLOCKS_METADATA.
888 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
889 EXT4_FREE_BLOCKS_FORGET);
891 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
898 * The ext4_ind_map_blocks() function handles non-extents inodes
899 * (i.e., using the traditional indirect/double-indirect i_blocks
900 * scheme) for ext4_map_blocks().
902 * Allocation strategy is simple: if we have to allocate something, we will
903 * have to go the whole way to leaf. So let's do it before attaching anything
904 * to tree, set linkage between the newborn blocks, write them if sync is
905 * required, recheck the path, free and repeat if check fails, otherwise
906 * set the last missing link (that will protect us from any truncate-generated
907 * removals - all blocks on the path are immune now) and possibly force the
908 * write on the parent block.
909 * That has a nice additional property: no special recovery from the failed
910 * allocations is needed - we simply release blocks and do not touch anything
911 * reachable from inode.
913 * `handle' can be NULL if create == 0.
915 * return > 0, # of blocks mapped or allocated.
916 * return = 0, if plain lookup failed.
917 * return < 0, error case.
919 * The ext4_ind_get_blocks() function should be called with
920 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
921 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
922 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
925 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
926 struct ext4_map_blocks *map,
930 ext4_lblk_t offsets[4];
935 int blocks_to_boundary = 0;
938 ext4_fsblk_t first_block = 0;
940 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
941 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
942 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
943 &blocks_to_boundary);
948 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
950 /* Simplest case - block found, no allocation needed */
952 first_block = le32_to_cpu(chain[depth - 1].key);
955 while (count < map->m_len && count <= blocks_to_boundary) {
958 blk = le32_to_cpu(*(chain[depth-1].p + count));
960 if (blk == first_block + count)
968 /* Next simple case - plain lookup or failed read of indirect block */
969 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
973 * Okay, we need to do block allocation.
975 goal = ext4_find_goal(inode, map->m_lblk, partial);
977 /* the number of blocks need to allocate for [d,t]indirect blocks */
978 indirect_blks = (chain + depth) - partial - 1;
981 * Next look up the indirect map to count the totoal number of
982 * direct blocks to allocate for this branch.
984 count = ext4_blks_to_allocate(partial, indirect_blks,
985 map->m_len, blocks_to_boundary);
987 * Block out ext4_truncate while we alter the tree
989 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
991 offsets + (partial - chain), partial);
994 * The ext4_splice_branch call will free and forget any buffers
995 * on the new chain if there is a failure, but that risks using
996 * up transaction credits, especially for bitmaps where the
997 * credits cannot be returned. Can we handle this somehow? We
998 * may need to return -EAGAIN upwards in the worst case. --sct
1001 err = ext4_splice_branch(handle, inode, map->m_lblk,
1002 partial, indirect_blks, count);
1006 map->m_flags |= EXT4_MAP_NEW;
1008 ext4_update_inode_fsync_trans(handle, inode, 1);
1010 map->m_flags |= EXT4_MAP_MAPPED;
1011 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1013 if (count > blocks_to_boundary)
1014 map->m_flags |= EXT4_MAP_BOUNDARY;
1016 /* Clean up and exit */
1017 partial = chain + depth - 1; /* the whole chain */
1019 while (partial > chain) {
1020 BUFFER_TRACE(partial->bh, "call brelse");
1021 brelse(partial->bh);
1029 qsize_t *ext4_get_reserved_space(struct inode *inode)
1031 return &EXT4_I(inode)->i_reserved_quota;
1036 * Calculate the number of metadata blocks need to reserve
1037 * to allocate a new block at @lblocks for non extent file based file
1039 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1042 struct ext4_inode_info *ei = EXT4_I(inode);
1043 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1046 if (lblock < EXT4_NDIR_BLOCKS)
1049 lblock -= EXT4_NDIR_BLOCKS;
1051 if (ei->i_da_metadata_calc_len &&
1052 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1053 ei->i_da_metadata_calc_len++;
1056 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1057 ei->i_da_metadata_calc_len = 1;
1058 blk_bits = order_base_2(lblock);
1059 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1063 * Calculate the number of metadata blocks need to reserve
1064 * to allocate a block located at @lblock
1066 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1068 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1069 return ext4_ext_calc_metadata_amount(inode, lblock);
1071 return ext4_indirect_calc_metadata_amount(inode, lblock);
1075 * Called with i_data_sem down, which is important since we can call
1076 * ext4_discard_preallocations() from here.
1078 void ext4_da_update_reserve_space(struct inode *inode,
1079 int used, int quota_claim)
1081 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1082 struct ext4_inode_info *ei = EXT4_I(inode);
1084 spin_lock(&ei->i_block_reservation_lock);
1085 trace_ext4_da_update_reserve_space(inode, used);
1086 if (unlikely(used > ei->i_reserved_data_blocks)) {
1087 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1088 "with only %d reserved data blocks\n",
1089 __func__, inode->i_ino, used,
1090 ei->i_reserved_data_blocks);
1092 used = ei->i_reserved_data_blocks;
1095 /* Update per-inode reservations */
1096 ei->i_reserved_data_blocks -= used;
1097 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1098 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1099 used + ei->i_allocated_meta_blocks);
1100 ei->i_allocated_meta_blocks = 0;
1102 if (ei->i_reserved_data_blocks == 0) {
1104 * We can release all of the reserved metadata blocks
1105 * only when we have written all of the delayed
1106 * allocation blocks.
1108 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1109 ei->i_reserved_meta_blocks);
1110 ei->i_reserved_meta_blocks = 0;
1111 ei->i_da_metadata_calc_len = 0;
1113 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1115 /* Update quota subsystem for data blocks */
1117 dquot_claim_block(inode, used);
1120 * We did fallocate with an offset that is already delayed
1121 * allocated. So on delayed allocated writeback we should
1122 * not re-claim the quota for fallocated blocks.
1124 dquot_release_reservation_block(inode, used);
1128 * If we have done all the pending block allocations and if
1129 * there aren't any writers on the inode, we can discard the
1130 * inode's preallocations.
1132 if ((ei->i_reserved_data_blocks == 0) &&
1133 (atomic_read(&inode->i_writecount) == 0))
1134 ext4_discard_preallocations(inode);
1137 static int __check_block_validity(struct inode *inode, const char *func,
1139 struct ext4_map_blocks *map)
1141 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1143 ext4_error_inode(inode, func, line, map->m_pblk,
1144 "lblock %lu mapped to illegal pblock "
1145 "(length %d)", (unsigned long) map->m_lblk,
1152 #define check_block_validity(inode, map) \
1153 __check_block_validity((inode), __func__, __LINE__, (map))
1156 * Return the number of contiguous dirty pages in a given inode
1157 * starting at page frame idx.
1159 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1160 unsigned int max_pages)
1162 struct address_space *mapping = inode->i_mapping;
1164 struct pagevec pvec;
1166 int i, nr_pages, done = 0;
1170 pagevec_init(&pvec, 0);
1173 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1174 PAGECACHE_TAG_DIRTY,
1175 (pgoff_t)PAGEVEC_SIZE);
1178 for (i = 0; i < nr_pages; i++) {
1179 struct page *page = pvec.pages[i];
1180 struct buffer_head *bh, *head;
1183 if (unlikely(page->mapping != mapping) ||
1185 PageWriteback(page) ||
1186 page->index != idx) {
1191 if (page_has_buffers(page)) {
1192 bh = head = page_buffers(page);
1194 if (!buffer_delay(bh) &&
1195 !buffer_unwritten(bh))
1197 bh = bh->b_this_page;
1198 } while (!done && (bh != head));
1205 if (num >= max_pages)
1208 pagevec_release(&pvec);
1214 * The ext4_map_blocks() function tries to look up the requested blocks,
1215 * and returns if the blocks are already mapped.
1217 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1218 * and store the allocated blocks in the result buffer head and mark it
1221 * If file type is extents based, it will call ext4_ext_map_blocks(),
1222 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1225 * On success, it returns the number of blocks being mapped or allocate.
1226 * if create==0 and the blocks are pre-allocated and uninitialized block,
1227 * the result buffer head is unmapped. If the create ==1, it will make sure
1228 * the buffer head is mapped.
1230 * It returns 0 if plain look up failed (blocks have not been allocated), in
1231 * that casem, buffer head is unmapped
1233 * It returns the error in case of allocation failure.
1235 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1236 struct ext4_map_blocks *map, int flags)
1241 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1242 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1243 (unsigned long) map->m_lblk);
1245 * Try to see if we can get the block without requesting a new
1246 * file system block.
1248 down_read((&EXT4_I(inode)->i_data_sem));
1249 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1250 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1252 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1254 up_read((&EXT4_I(inode)->i_data_sem));
1256 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1257 int ret = check_block_validity(inode, map);
1262 /* If it is only a block(s) look up */
1263 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1267 * Returns if the blocks have already allocated
1269 * Note that if blocks have been preallocated
1270 * ext4_ext_get_block() returns th create = 0
1271 * with buffer head unmapped.
1273 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1277 * When we call get_blocks without the create flag, the
1278 * BH_Unwritten flag could have gotten set if the blocks
1279 * requested were part of a uninitialized extent. We need to
1280 * clear this flag now that we are committed to convert all or
1281 * part of the uninitialized extent to be an initialized
1282 * extent. This is because we need to avoid the combination
1283 * of BH_Unwritten and BH_Mapped flags being simultaneously
1284 * set on the buffer_head.
1286 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1289 * New blocks allocate and/or writing to uninitialized extent
1290 * will possibly result in updating i_data, so we take
1291 * the write lock of i_data_sem, and call get_blocks()
1292 * with create == 1 flag.
1294 down_write((&EXT4_I(inode)->i_data_sem));
1297 * if the caller is from delayed allocation writeout path
1298 * we have already reserved fs blocks for allocation
1299 * let the underlying get_block() function know to
1300 * avoid double accounting
1302 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1303 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1305 * We need to check for EXT4 here because migrate
1306 * could have changed the inode type in between
1308 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1309 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1311 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1313 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1315 * We allocated new blocks which will result in
1316 * i_data's format changing. Force the migrate
1317 * to fail by clearing migrate flags
1319 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1323 * Update reserved blocks/metadata blocks after successful
1324 * block allocation which had been deferred till now. We don't
1325 * support fallocate for non extent files. So we can update
1326 * reserve space here.
1329 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1330 ext4_da_update_reserve_space(inode, retval, 1);
1332 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1333 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1335 up_write((&EXT4_I(inode)->i_data_sem));
1336 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1337 int ret = check_block_validity(inode, map);
1344 /* Maximum number of blocks we map for direct IO at once. */
1345 #define DIO_MAX_BLOCKS 4096
1347 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1348 struct buffer_head *bh, int flags)
1350 handle_t *handle = ext4_journal_current_handle();
1351 struct ext4_map_blocks map;
1352 int ret = 0, started = 0;
1355 map.m_lblk = iblock;
1356 map.m_len = bh->b_size >> inode->i_blkbits;
1358 if (flags && !handle) {
1359 /* Direct IO write... */
1360 if (map.m_len > DIO_MAX_BLOCKS)
1361 map.m_len = DIO_MAX_BLOCKS;
1362 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1363 handle = ext4_journal_start(inode, dio_credits);
1364 if (IS_ERR(handle)) {
1365 ret = PTR_ERR(handle);
1371 ret = ext4_map_blocks(handle, inode, &map, flags);
1373 map_bh(bh, inode->i_sb, map.m_pblk);
1374 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1375 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1379 ext4_journal_stop(handle);
1383 int ext4_get_block(struct inode *inode, sector_t iblock,
1384 struct buffer_head *bh, int create)
1386 return _ext4_get_block(inode, iblock, bh,
1387 create ? EXT4_GET_BLOCKS_CREATE : 0);
1391 * `handle' can be NULL if create is zero
1393 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1394 ext4_lblk_t block, int create, int *errp)
1396 struct ext4_map_blocks map;
1397 struct buffer_head *bh;
1400 J_ASSERT(handle != NULL || create == 0);
1404 err = ext4_map_blocks(handle, inode, &map,
1405 create ? EXT4_GET_BLOCKS_CREATE : 0);
1413 bh = sb_getblk(inode->i_sb, map.m_pblk);
1418 if (map.m_flags & EXT4_MAP_NEW) {
1419 J_ASSERT(create != 0);
1420 J_ASSERT(handle != NULL);
1423 * Now that we do not always journal data, we should
1424 * keep in mind whether this should always journal the
1425 * new buffer as metadata. For now, regular file
1426 * writes use ext4_get_block instead, so it's not a
1430 BUFFER_TRACE(bh, "call get_create_access");
1431 fatal = ext4_journal_get_create_access(handle, bh);
1432 if (!fatal && !buffer_uptodate(bh)) {
1433 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1434 set_buffer_uptodate(bh);
1437 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1438 err = ext4_handle_dirty_metadata(handle, inode, bh);
1442 BUFFER_TRACE(bh, "not a new buffer");
1452 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1453 ext4_lblk_t block, int create, int *err)
1455 struct buffer_head *bh;
1457 bh = ext4_getblk(handle, inode, block, create, err);
1460 if (buffer_uptodate(bh))
1462 ll_rw_block(READ_META, 1, &bh);
1464 if (buffer_uptodate(bh))
1471 static int walk_page_buffers(handle_t *handle,
1472 struct buffer_head *head,
1476 int (*fn)(handle_t *handle,
1477 struct buffer_head *bh))
1479 struct buffer_head *bh;
1480 unsigned block_start, block_end;
1481 unsigned blocksize = head->b_size;
1483 struct buffer_head *next;
1485 for (bh = head, block_start = 0;
1486 ret == 0 && (bh != head || !block_start);
1487 block_start = block_end, bh = next) {
1488 next = bh->b_this_page;
1489 block_end = block_start + blocksize;
1490 if (block_end <= from || block_start >= to) {
1491 if (partial && !buffer_uptodate(bh))
1495 err = (*fn)(handle, bh);
1503 * To preserve ordering, it is essential that the hole instantiation and
1504 * the data write be encapsulated in a single transaction. We cannot
1505 * close off a transaction and start a new one between the ext4_get_block()
1506 * and the commit_write(). So doing the jbd2_journal_start at the start of
1507 * prepare_write() is the right place.
1509 * Also, this function can nest inside ext4_writepage() ->
1510 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1511 * has generated enough buffer credits to do the whole page. So we won't
1512 * block on the journal in that case, which is good, because the caller may
1515 * By accident, ext4 can be reentered when a transaction is open via
1516 * quota file writes. If we were to commit the transaction while thus
1517 * reentered, there can be a deadlock - we would be holding a quota
1518 * lock, and the commit would never complete if another thread had a
1519 * transaction open and was blocking on the quota lock - a ranking
1522 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1523 * will _not_ run commit under these circumstances because handle->h_ref
1524 * is elevated. We'll still have enough credits for the tiny quotafile
1527 static int do_journal_get_write_access(handle_t *handle,
1528 struct buffer_head *bh)
1530 if (!buffer_mapped(bh) || buffer_freed(bh))
1532 return ext4_journal_get_write_access(handle, bh);
1536 * Truncate blocks that were not used by write. We have to truncate the
1537 * pagecache as well so that corresponding buffers get properly unmapped.
1539 static void ext4_truncate_failed_write(struct inode *inode)
1541 truncate_inode_pages(inode->i_mapping, inode->i_size);
1542 ext4_truncate(inode);
1545 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1546 struct buffer_head *bh_result, int create);
1547 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1548 loff_t pos, unsigned len, unsigned flags,
1549 struct page **pagep, void **fsdata)
1551 struct inode *inode = mapping->host;
1552 int ret, needed_blocks;
1559 trace_ext4_write_begin(inode, pos, len, flags);
1561 * Reserve one block more for addition to orphan list in case
1562 * we allocate blocks but write fails for some reason
1564 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1565 index = pos >> PAGE_CACHE_SHIFT;
1566 from = pos & (PAGE_CACHE_SIZE - 1);
1570 handle = ext4_journal_start(inode, needed_blocks);
1571 if (IS_ERR(handle)) {
1572 ret = PTR_ERR(handle);
1576 /* We cannot recurse into the filesystem as the transaction is already
1578 flags |= AOP_FLAG_NOFS;
1580 page = grab_cache_page_write_begin(mapping, index, flags);
1582 ext4_journal_stop(handle);
1588 if (ext4_should_dioread_nolock(inode))
1589 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1590 fsdata, ext4_get_block_write);
1592 ret = block_write_begin(file, mapping, pos, len, flags, pagep,
1593 fsdata, ext4_get_block);
1595 if (!ret && ext4_should_journal_data(inode)) {
1596 ret = walk_page_buffers(handle, page_buffers(page),
1597 from, to, NULL, do_journal_get_write_access);
1602 page_cache_release(page);
1604 * block_write_begin may have instantiated a few blocks
1605 * outside i_size. Trim these off again. Don't need
1606 * i_size_read because we hold i_mutex.
1608 * Add inode to orphan list in case we crash before
1611 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1612 ext4_orphan_add(handle, inode);
1614 ext4_journal_stop(handle);
1615 if (pos + len > inode->i_size) {
1616 ext4_truncate_failed_write(inode);
1618 * If truncate failed early the inode might
1619 * still be on the orphan list; we need to
1620 * make sure the inode is removed from the
1621 * orphan list in that case.
1624 ext4_orphan_del(NULL, inode);
1628 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1634 /* For write_end() in data=journal mode */
1635 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1637 if (!buffer_mapped(bh) || buffer_freed(bh))
1639 set_buffer_uptodate(bh);
1640 return ext4_handle_dirty_metadata(handle, NULL, bh);
1643 static int ext4_generic_write_end(struct file *file,
1644 struct address_space *mapping,
1645 loff_t pos, unsigned len, unsigned copied,
1646 struct page *page, void *fsdata)
1648 int i_size_changed = 0;
1649 struct inode *inode = mapping->host;
1650 handle_t *handle = ext4_journal_current_handle();
1652 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1655 * No need to use i_size_read() here, the i_size
1656 * cannot change under us because we hold i_mutex.
1658 * But it's important to update i_size while still holding page lock:
1659 * page writeout could otherwise come in and zero beyond i_size.
1661 if (pos + copied > inode->i_size) {
1662 i_size_write(inode, pos + copied);
1666 if (pos + copied > EXT4_I(inode)->i_disksize) {
1667 /* We need to mark inode dirty even if
1668 * new_i_size is less that inode->i_size
1669 * bu greater than i_disksize.(hint delalloc)
1671 ext4_update_i_disksize(inode, (pos + copied));
1675 page_cache_release(page);
1678 * Don't mark the inode dirty under page lock. First, it unnecessarily
1679 * makes the holding time of page lock longer. Second, it forces lock
1680 * ordering of page lock and transaction start for journaling
1684 ext4_mark_inode_dirty(handle, inode);
1690 * We need to pick up the new inode size which generic_commit_write gave us
1691 * `file' can be NULL - eg, when called from page_symlink().
1693 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1694 * buffers are managed internally.
1696 static int ext4_ordered_write_end(struct file *file,
1697 struct address_space *mapping,
1698 loff_t pos, unsigned len, unsigned copied,
1699 struct page *page, void *fsdata)
1701 handle_t *handle = ext4_journal_current_handle();
1702 struct inode *inode = mapping->host;
1705 trace_ext4_ordered_write_end(inode, pos, len, copied);
1706 ret = ext4_jbd2_file_inode(handle, inode);
1709 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1712 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1713 /* if we have allocated more blocks and copied
1714 * less. We will have blocks allocated outside
1715 * inode->i_size. So truncate them
1717 ext4_orphan_add(handle, inode);
1721 ret2 = ext4_journal_stop(handle);
1725 if (pos + len > inode->i_size) {
1726 ext4_truncate_failed_write(inode);
1728 * If truncate failed early the inode might still be
1729 * on the orphan list; we need to make sure the inode
1730 * is removed from the orphan list in that case.
1733 ext4_orphan_del(NULL, inode);
1737 return ret ? ret : copied;
1740 static int ext4_writeback_write_end(struct file *file,
1741 struct address_space *mapping,
1742 loff_t pos, unsigned len, unsigned copied,
1743 struct page *page, void *fsdata)
1745 handle_t *handle = ext4_journal_current_handle();
1746 struct inode *inode = mapping->host;
1749 trace_ext4_writeback_write_end(inode, pos, len, copied);
1750 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1753 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1754 /* if we have allocated more blocks and copied
1755 * less. We will have blocks allocated outside
1756 * inode->i_size. So truncate them
1758 ext4_orphan_add(handle, inode);
1763 ret2 = ext4_journal_stop(handle);
1767 if (pos + len > inode->i_size) {
1768 ext4_truncate_failed_write(inode);
1770 * If truncate failed early the inode might still be
1771 * on the orphan list; we need to make sure the inode
1772 * is removed from the orphan list in that case.
1775 ext4_orphan_del(NULL, inode);
1778 return ret ? ret : copied;
1781 static int ext4_journalled_write_end(struct file *file,
1782 struct address_space *mapping,
1783 loff_t pos, unsigned len, unsigned copied,
1784 struct page *page, void *fsdata)
1786 handle_t *handle = ext4_journal_current_handle();
1787 struct inode *inode = mapping->host;
1793 trace_ext4_journalled_write_end(inode, pos, len, copied);
1794 from = pos & (PAGE_CACHE_SIZE - 1);
1798 if (!PageUptodate(page))
1800 page_zero_new_buffers(page, from+copied, to);
1803 ret = walk_page_buffers(handle, page_buffers(page), from,
1804 to, &partial, write_end_fn);
1806 SetPageUptodate(page);
1807 new_i_size = pos + copied;
1808 if (new_i_size > inode->i_size)
1809 i_size_write(inode, pos+copied);
1810 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1811 if (new_i_size > EXT4_I(inode)->i_disksize) {
1812 ext4_update_i_disksize(inode, new_i_size);
1813 ret2 = ext4_mark_inode_dirty(handle, inode);
1819 page_cache_release(page);
1820 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1821 /* if we have allocated more blocks and copied
1822 * less. We will have blocks allocated outside
1823 * inode->i_size. So truncate them
1825 ext4_orphan_add(handle, inode);
1827 ret2 = ext4_journal_stop(handle);
1830 if (pos + len > inode->i_size) {
1831 ext4_truncate_failed_write(inode);
1833 * If truncate failed early the inode might still be
1834 * on the orphan list; we need to make sure the inode
1835 * is removed from the orphan list in that case.
1838 ext4_orphan_del(NULL, inode);
1841 return ret ? ret : copied;
1845 * Reserve a single block located at lblock
1847 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1850 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1851 struct ext4_inode_info *ei = EXT4_I(inode);
1852 unsigned long md_needed;
1856 * recalculate the amount of metadata blocks to reserve
1857 * in order to allocate nrblocks
1858 * worse case is one extent per block
1861 spin_lock(&ei->i_block_reservation_lock);
1862 md_needed = ext4_calc_metadata_amount(inode, lblock);
1863 trace_ext4_da_reserve_space(inode, md_needed);
1864 spin_unlock(&ei->i_block_reservation_lock);
1867 * We will charge metadata quota at writeout time; this saves
1868 * us from metadata over-estimation, though we may go over by
1869 * a small amount in the end. Here we just reserve for data.
1871 ret = dquot_reserve_block(inode, 1);
1875 * We do still charge estimated metadata to the sb though;
1876 * we cannot afford to run out of free blocks.
1878 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1879 dquot_release_reservation_block(inode, 1);
1880 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1886 spin_lock(&ei->i_block_reservation_lock);
1887 ei->i_reserved_data_blocks++;
1888 ei->i_reserved_meta_blocks += md_needed;
1889 spin_unlock(&ei->i_block_reservation_lock);
1891 return 0; /* success */
1894 static void ext4_da_release_space(struct inode *inode, int to_free)
1896 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1897 struct ext4_inode_info *ei = EXT4_I(inode);
1900 return; /* Nothing to release, exit */
1902 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1904 trace_ext4_da_release_space(inode, to_free);
1905 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1907 * if there aren't enough reserved blocks, then the
1908 * counter is messed up somewhere. Since this
1909 * function is called from invalidate page, it's
1910 * harmless to return without any action.
1912 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1913 "ino %lu, to_free %d with only %d reserved "
1914 "data blocks\n", inode->i_ino, to_free,
1915 ei->i_reserved_data_blocks);
1917 to_free = ei->i_reserved_data_blocks;
1919 ei->i_reserved_data_blocks -= to_free;
1921 if (ei->i_reserved_data_blocks == 0) {
1923 * We can release all of the reserved metadata blocks
1924 * only when we have written all of the delayed
1925 * allocation blocks.
1927 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1928 ei->i_reserved_meta_blocks);
1929 ei->i_reserved_meta_blocks = 0;
1930 ei->i_da_metadata_calc_len = 0;
1933 /* update fs dirty data blocks counter */
1934 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1936 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1938 dquot_release_reservation_block(inode, to_free);
1941 static void ext4_da_page_release_reservation(struct page *page,
1942 unsigned long offset)
1945 struct buffer_head *head, *bh;
1946 unsigned int curr_off = 0;
1948 head = page_buffers(page);
1951 unsigned int next_off = curr_off + bh->b_size;
1953 if ((offset <= curr_off) && (buffer_delay(bh))) {
1955 clear_buffer_delay(bh);
1957 curr_off = next_off;
1958 } while ((bh = bh->b_this_page) != head);
1959 ext4_da_release_space(page->mapping->host, to_release);
1963 * Delayed allocation stuff
1967 * mpage_da_submit_io - walks through extent of pages and try to write
1968 * them with writepage() call back
1970 * @mpd->inode: inode
1971 * @mpd->first_page: first page of the extent
1972 * @mpd->next_page: page after the last page of the extent
1974 * By the time mpage_da_submit_io() is called we expect all blocks
1975 * to be allocated. this may be wrong if allocation failed.
1977 * As pages are already locked by write_cache_pages(), we can't use it
1979 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1982 struct pagevec pvec;
1983 unsigned long index, end;
1984 int ret = 0, err, nr_pages, i;
1985 struct inode *inode = mpd->inode;
1986 struct address_space *mapping = inode->i_mapping;
1988 BUG_ON(mpd->next_page <= mpd->first_page);
1990 * We need to start from the first_page to the next_page - 1
1991 * to make sure we also write the mapped dirty buffer_heads.
1992 * If we look at mpd->b_blocknr we would only be looking
1993 * at the currently mapped buffer_heads.
1995 index = mpd->first_page;
1996 end = mpd->next_page - 1;
1998 pagevec_init(&pvec, 0);
1999 while (index <= end) {
2000 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2003 for (i = 0; i < nr_pages; i++) {
2004 struct page *page = pvec.pages[i];
2006 index = page->index;
2011 BUG_ON(!PageLocked(page));
2012 BUG_ON(PageWriteback(page));
2014 pages_skipped = mpd->wbc->pages_skipped;
2015 err = mapping->a_ops->writepage(page, mpd->wbc);
2016 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2018 * have successfully written the page
2019 * without skipping the same
2021 mpd->pages_written++;
2023 * In error case, we have to continue because
2024 * remaining pages are still locked
2025 * XXX: unlock and re-dirty them?
2030 pagevec_release(&pvec);
2036 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2038 * the function goes through all passed space and put actual disk
2039 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2041 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2042 struct ext4_map_blocks *map)
2044 struct inode *inode = mpd->inode;
2045 struct address_space *mapping = inode->i_mapping;
2046 int blocks = map->m_len;
2047 sector_t pblock = map->m_pblk, cur_logical;
2048 struct buffer_head *head, *bh;
2050 struct pagevec pvec;
2053 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2054 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2055 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2057 pagevec_init(&pvec, 0);
2059 while (index <= end) {
2060 /* XXX: optimize tail */
2061 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2064 for (i = 0; i < nr_pages; i++) {
2065 struct page *page = pvec.pages[i];
2067 index = page->index;
2072 BUG_ON(!PageLocked(page));
2073 BUG_ON(PageWriteback(page));
2074 BUG_ON(!page_has_buffers(page));
2076 bh = page_buffers(page);
2079 /* skip blocks out of the range */
2081 if (cur_logical >= map->m_lblk)
2084 } while ((bh = bh->b_this_page) != head);
2087 if (cur_logical >= map->m_lblk + blocks)
2090 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2092 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2094 if (buffer_delay(bh)) {
2095 clear_buffer_delay(bh);
2096 bh->b_blocknr = pblock;
2099 * unwritten already should have
2100 * blocknr assigned. Verify that
2102 clear_buffer_unwritten(bh);
2103 BUG_ON(bh->b_blocknr != pblock);
2106 } else if (buffer_mapped(bh))
2107 BUG_ON(bh->b_blocknr != pblock);
2109 if (map->m_flags & EXT4_MAP_UNINIT)
2110 set_buffer_uninit(bh);
2113 } while ((bh = bh->b_this_page) != head);
2115 pagevec_release(&pvec);
2120 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2121 sector_t logical, long blk_cnt)
2125 struct pagevec pvec;
2126 struct inode *inode = mpd->inode;
2127 struct address_space *mapping = inode->i_mapping;
2129 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2130 end = (logical + blk_cnt - 1) >>
2131 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2132 while (index <= end) {
2133 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2136 for (i = 0; i < nr_pages; i++) {
2137 struct page *page = pvec.pages[i];
2138 if (page->index > end)
2140 BUG_ON(!PageLocked(page));
2141 BUG_ON(PageWriteback(page));
2142 block_invalidatepage(page, 0);
2143 ClearPageUptodate(page);
2146 index = pvec.pages[nr_pages - 1]->index + 1;
2147 pagevec_release(&pvec);
2152 static void ext4_print_free_blocks(struct inode *inode)
2154 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2155 printk(KERN_CRIT "Total free blocks count %lld\n",
2156 ext4_count_free_blocks(inode->i_sb));
2157 printk(KERN_CRIT "Free/Dirty block details\n");
2158 printk(KERN_CRIT "free_blocks=%lld\n",
2159 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2160 printk(KERN_CRIT "dirty_blocks=%lld\n",
2161 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2162 printk(KERN_CRIT "Block reservation details\n");
2163 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2164 EXT4_I(inode)->i_reserved_data_blocks);
2165 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2166 EXT4_I(inode)->i_reserved_meta_blocks);
2171 * mpage_da_map_blocks - go through given space
2173 * @mpd - bh describing space
2175 * The function skips space we know is already mapped to disk blocks.
2178 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2180 int err, blks, get_blocks_flags;
2181 struct ext4_map_blocks map;
2182 sector_t next = mpd->b_blocknr;
2183 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2184 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2185 handle_t *handle = NULL;
2188 * We consider only non-mapped and non-allocated blocks
2190 if ((mpd->b_state & (1 << BH_Mapped)) &&
2191 !(mpd->b_state & (1 << BH_Delay)) &&
2192 !(mpd->b_state & (1 << BH_Unwritten)))
2196 * If we didn't accumulate anything to write simply return
2201 handle = ext4_journal_current_handle();
2205 * Call ext4_map_blocks() to allocate any delayed allocation
2206 * blocks, or to convert an uninitialized extent to be
2207 * initialized (in the case where we have written into
2208 * one or more preallocated blocks).
2210 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2211 * indicate that we are on the delayed allocation path. This
2212 * affects functions in many different parts of the allocation
2213 * call path. This flag exists primarily because we don't
2214 * want to change *many* call functions, so ext4_map_blocks()
2215 * will set the magic i_delalloc_reserved_flag once the
2216 * inode's allocation semaphore is taken.
2218 * If the blocks in questions were delalloc blocks, set
2219 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2220 * variables are updated after the blocks have been allocated.
2223 map.m_len = max_blocks;
2224 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2225 if (ext4_should_dioread_nolock(mpd->inode))
2226 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2227 if (mpd->b_state & (1 << BH_Delay))
2228 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2230 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2232 struct super_block *sb = mpd->inode->i_sb;
2236 * If get block returns with error we simply
2237 * return. Later writepage will redirty the page and
2238 * writepages will find the dirty page again
2243 if (err == -ENOSPC &&
2244 ext4_count_free_blocks(sb)) {
2250 * get block failure will cause us to loop in
2251 * writepages, because a_ops->writepage won't be able
2252 * to make progress. The page will be redirtied by
2253 * writepage and writepages will again try to write
2256 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2257 ext4_msg(sb, KERN_CRIT,
2258 "delayed block allocation failed for inode %lu "
2259 "at logical offset %llu with max blocks %zd "
2260 "with error %d", mpd->inode->i_ino,
2261 (unsigned long long) next,
2262 mpd->b_size >> mpd->inode->i_blkbits, err);
2263 ext4_msg(sb, KERN_CRIT,
2264 "This should not happen!! Data will be lost\n");
2266 ext4_print_free_blocks(mpd->inode);
2268 /* invalidate all the pages */
2269 ext4_da_block_invalidatepages(mpd, next,
2270 mpd->b_size >> mpd->inode->i_blkbits);
2275 if (map.m_flags & EXT4_MAP_NEW) {
2276 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2279 for (i = 0; i < map.m_len; i++)
2280 unmap_underlying_metadata(bdev, map.m_pblk + i);
2284 * If blocks are delayed marked, we need to
2285 * put actual blocknr and drop delayed bit
2287 if ((mpd->b_state & (1 << BH_Delay)) ||
2288 (mpd->b_state & (1 << BH_Unwritten)))
2289 mpage_put_bnr_to_bhs(mpd, &map);
2291 if (ext4_should_order_data(mpd->inode)) {
2292 err = ext4_jbd2_file_inode(handle, mpd->inode);
2298 * Update on-disk size along with block allocation.
2300 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2301 if (disksize > i_size_read(mpd->inode))
2302 disksize = i_size_read(mpd->inode);
2303 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2304 ext4_update_i_disksize(mpd->inode, disksize);
2305 return ext4_mark_inode_dirty(handle, mpd->inode);
2311 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2312 (1 << BH_Delay) | (1 << BH_Unwritten))
2315 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2317 * @mpd->lbh - extent of blocks
2318 * @logical - logical number of the block in the file
2319 * @bh - bh of the block (used to access block's state)
2321 * the function is used to collect contig. blocks in same state
2323 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2324 sector_t logical, size_t b_size,
2325 unsigned long b_state)
2328 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2331 * XXX Don't go larger than mballoc is willing to allocate
2332 * This is a stopgap solution. We eventually need to fold
2333 * mpage_da_submit_io() into this function and then call
2334 * ext4_map_blocks() multiple times in a loop
2336 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2339 /* check if thereserved journal credits might overflow */
2340 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2341 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2343 * With non-extent format we are limited by the journal
2344 * credit available. Total credit needed to insert
2345 * nrblocks contiguous blocks is dependent on the
2346 * nrblocks. So limit nrblocks.
2349 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2350 EXT4_MAX_TRANS_DATA) {
2352 * Adding the new buffer_head would make it cross the
2353 * allowed limit for which we have journal credit
2354 * reserved. So limit the new bh->b_size
2356 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2357 mpd->inode->i_blkbits;
2358 /* we will do mpage_da_submit_io in the next loop */
2362 * First block in the extent
2364 if (mpd->b_size == 0) {
2365 mpd->b_blocknr = logical;
2366 mpd->b_size = b_size;
2367 mpd->b_state = b_state & BH_FLAGS;
2371 next = mpd->b_blocknr + nrblocks;
2373 * Can we merge the block to our big extent?
2375 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2376 mpd->b_size += b_size;
2382 * We couldn't merge the block to our extent, so we
2383 * need to flush current extent and start new one
2385 if (mpage_da_map_blocks(mpd) == 0)
2386 mpage_da_submit_io(mpd);
2391 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2393 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2397 * __mpage_da_writepage - finds extent of pages and blocks
2399 * @page: page to consider
2400 * @wbc: not used, we just follow rules
2403 * The function finds extents of pages and scan them for all blocks.
2405 static int __mpage_da_writepage(struct page *page,
2406 struct writeback_control *wbc, void *data)
2408 struct mpage_da_data *mpd = data;
2409 struct inode *inode = mpd->inode;
2410 struct buffer_head *bh, *head;
2414 * Can we merge this page to current extent?
2416 if (mpd->next_page != page->index) {
2418 * Nope, we can't. So, we map non-allocated blocks
2419 * and start IO on them using writepage()
2421 if (mpd->next_page != mpd->first_page) {
2422 if (mpage_da_map_blocks(mpd) == 0)
2423 mpage_da_submit_io(mpd);
2425 * skip rest of the page in the page_vec
2428 redirty_page_for_writepage(wbc, page);
2430 return MPAGE_DA_EXTENT_TAIL;
2434 * Start next extent of pages ...
2436 mpd->first_page = page->index;
2446 mpd->next_page = page->index + 1;
2447 logical = (sector_t) page->index <<
2448 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2450 if (!page_has_buffers(page)) {
2451 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2452 (1 << BH_Dirty) | (1 << BH_Uptodate));
2454 return MPAGE_DA_EXTENT_TAIL;
2457 * Page with regular buffer heads, just add all dirty ones
2459 head = page_buffers(page);
2462 BUG_ON(buffer_locked(bh));
2464 * We need to try to allocate
2465 * unmapped blocks in the same page.
2466 * Otherwise we won't make progress
2467 * with the page in ext4_writepage
2469 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2470 mpage_add_bh_to_extent(mpd, logical,
2474 return MPAGE_DA_EXTENT_TAIL;
2475 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2477 * mapped dirty buffer. We need to update
2478 * the b_state because we look at
2479 * b_state in mpage_da_map_blocks. We don't
2480 * update b_size because if we find an
2481 * unmapped buffer_head later we need to
2482 * use the b_state flag of that buffer_head.
2484 if (mpd->b_size == 0)
2485 mpd->b_state = bh->b_state & BH_FLAGS;
2488 } while ((bh = bh->b_this_page) != head);
2495 * This is a special get_blocks_t callback which is used by
2496 * ext4_da_write_begin(). It will either return mapped block or
2497 * reserve space for a single block.
2499 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2500 * We also have b_blocknr = -1 and b_bdev initialized properly
2502 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2503 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2504 * initialized properly.
2506 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2507 struct buffer_head *bh, int create)
2509 struct ext4_map_blocks map;
2511 sector_t invalid_block = ~((sector_t) 0xffff);
2513 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2516 BUG_ON(create == 0);
2517 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2519 map.m_lblk = iblock;
2523 * first, we need to know whether the block is allocated already
2524 * preallocated blocks are unmapped but should treated
2525 * the same as allocated blocks.
2527 ret = ext4_map_blocks(NULL, inode, &map, 0);
2531 if (buffer_delay(bh))
2532 return 0; /* Not sure this could or should happen */
2534 * XXX: __block_prepare_write() unmaps passed block,
2537 ret = ext4_da_reserve_space(inode, iblock);
2539 /* not enough space to reserve */
2542 map_bh(bh, inode->i_sb, invalid_block);
2544 set_buffer_delay(bh);
2548 map_bh(bh, inode->i_sb, map.m_pblk);
2549 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2551 if (buffer_unwritten(bh)) {
2552 /* A delayed write to unwritten bh should be marked
2553 * new and mapped. Mapped ensures that we don't do
2554 * get_block multiple times when we write to the same
2555 * offset and new ensures that we do proper zero out
2556 * for partial write.
2559 set_buffer_mapped(bh);
2565 * This function is used as a standard get_block_t calback function
2566 * when there is no desire to allocate any blocks. It is used as a
2567 * callback function for block_prepare_write() and block_write_full_page().
2568 * These functions should only try to map a single block at a time.
2570 * Since this function doesn't do block allocations even if the caller
2571 * requests it by passing in create=1, it is critically important that
2572 * any caller checks to make sure that any buffer heads are returned
2573 * by this function are either all already mapped or marked for
2574 * delayed allocation before calling block_write_full_page(). Otherwise,
2575 * b_blocknr could be left unitialized, and the page write functions will
2576 * be taken by surprise.
2578 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2579 struct buffer_head *bh_result, int create)
2581 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2582 return _ext4_get_block(inode, iblock, bh_result, 0);
2585 static int bget_one(handle_t *handle, struct buffer_head *bh)
2591 static int bput_one(handle_t *handle, struct buffer_head *bh)
2597 static int __ext4_journalled_writepage(struct page *page,
2600 struct address_space *mapping = page->mapping;
2601 struct inode *inode = mapping->host;
2602 struct buffer_head *page_bufs;
2603 handle_t *handle = NULL;
2607 page_bufs = page_buffers(page);
2609 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2610 /* As soon as we unlock the page, it can go away, but we have
2611 * references to buffers so we are safe */
2614 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2615 if (IS_ERR(handle)) {
2616 ret = PTR_ERR(handle);
2620 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2621 do_journal_get_write_access);
2623 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2627 err = ext4_journal_stop(handle);
2631 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2632 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2637 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2638 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2641 * Note that we don't need to start a transaction unless we're journaling data
2642 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2643 * need to file the inode to the transaction's list in ordered mode because if
2644 * we are writing back data added by write(), the inode is already there and if
2645 * we are writing back data modified via mmap(), noone guarantees in which
2646 * transaction the data will hit the disk. In case we are journaling data, we
2647 * cannot start transaction directly because transaction start ranks above page
2648 * lock so we have to do some magic.
2650 * This function can get called via...
2651 * - ext4_da_writepages after taking page lock (have journal handle)
2652 * - journal_submit_inode_data_buffers (no journal handle)
2653 * - shrink_page_list via pdflush (no journal handle)
2654 * - grab_page_cache when doing write_begin (have journal handle)
2656 * We don't do any block allocation in this function. If we have page with
2657 * multiple blocks we need to write those buffer_heads that are mapped. This
2658 * is important for mmaped based write. So if we do with blocksize 1K
2659 * truncate(f, 1024);
2660 * a = mmap(f, 0, 4096);
2662 * truncate(f, 4096);
2663 * we have in the page first buffer_head mapped via page_mkwrite call back
2664 * but other bufer_heads would be unmapped but dirty(dirty done via the
2665 * do_wp_page). So writepage should write the first block. If we modify
2666 * the mmap area beyond 1024 we will again get a page_fault and the
2667 * page_mkwrite callback will do the block allocation and mark the
2668 * buffer_heads mapped.
2670 * We redirty the page if we have any buffer_heads that is either delay or
2671 * unwritten in the page.
2673 * We can get recursively called as show below.
2675 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2678 * But since we don't do any block allocation we should not deadlock.
2679 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2681 static int ext4_writepage(struct page *page,
2682 struct writeback_control *wbc)
2687 struct buffer_head *page_bufs = NULL;
2688 struct inode *inode = page->mapping->host;
2690 trace_ext4_writepage(inode, page);
2691 size = i_size_read(inode);
2692 if (page->index == size >> PAGE_CACHE_SHIFT)
2693 len = size & ~PAGE_CACHE_MASK;
2695 len = PAGE_CACHE_SIZE;
2697 if (page_has_buffers(page)) {
2698 page_bufs = page_buffers(page);
2699 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2700 ext4_bh_delay_or_unwritten)) {
2702 * We don't want to do block allocation
2703 * So redirty the page and return
2704 * We may reach here when we do a journal commit
2705 * via journal_submit_inode_data_buffers.
2706 * If we don't have mapping block we just ignore
2707 * them. We can also reach here via shrink_page_list
2709 redirty_page_for_writepage(wbc, page);
2715 * The test for page_has_buffers() is subtle:
2716 * We know the page is dirty but it lost buffers. That means
2717 * that at some moment in time after write_begin()/write_end()
2718 * has been called all buffers have been clean and thus they
2719 * must have been written at least once. So they are all
2720 * mapped and we can happily proceed with mapping them
2721 * and writing the page.
2723 * Try to initialize the buffer_heads and check whether
2724 * all are mapped and non delay. We don't want to
2725 * do block allocation here.
2727 ret = block_prepare_write(page, 0, len,
2728 noalloc_get_block_write);
2730 page_bufs = page_buffers(page);
2731 /* check whether all are mapped and non delay */
2732 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2733 ext4_bh_delay_or_unwritten)) {
2734 redirty_page_for_writepage(wbc, page);
2740 * We can't do block allocation here
2741 * so just redity the page and unlock
2744 redirty_page_for_writepage(wbc, page);
2748 /* now mark the buffer_heads as dirty and uptodate */
2749 block_commit_write(page, 0, len);
2752 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2754 * It's mmapped pagecache. Add buffers and journal it. There
2755 * doesn't seem much point in redirtying the page here.
2757 ClearPageChecked(page);
2758 return __ext4_journalled_writepage(page, len);
2761 if (page_bufs && buffer_uninit(page_bufs)) {
2762 ext4_set_bh_endio(page_bufs, inode);
2763 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2764 wbc, ext4_end_io_buffer_write);
2766 ret = block_write_full_page(page, noalloc_get_block_write,
2773 * This is called via ext4_da_writepages() to
2774 * calulate the total number of credits to reserve to fit
2775 * a single extent allocation into a single transaction,
2776 * ext4_da_writpeages() will loop calling this before
2777 * the block allocation.
2780 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2782 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2785 * With non-extent format the journal credit needed to
2786 * insert nrblocks contiguous block is dependent on
2787 * number of contiguous block. So we will limit
2788 * number of contiguous block to a sane value
2790 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2791 (max_blocks > EXT4_MAX_TRANS_DATA))
2792 max_blocks = EXT4_MAX_TRANS_DATA;
2794 return ext4_chunk_trans_blocks(inode, max_blocks);
2798 * write_cache_pages_da - walk the list of dirty pages of the given
2799 * address space and call the callback function (which usually writes
2802 * This is a forked version of write_cache_pages(). Differences:
2803 * Range cyclic is ignored.
2804 * no_nrwrite_index_update is always presumed true
2806 static int write_cache_pages_da(struct address_space *mapping,
2807 struct writeback_control *wbc,
2808 struct mpage_da_data *mpd)
2812 struct pagevec pvec;
2815 pgoff_t end; /* Inclusive */
2816 long nr_to_write = wbc->nr_to_write;
2818 pagevec_init(&pvec, 0);
2819 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2820 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2822 while (!done && (index <= end)) {
2825 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2826 PAGECACHE_TAG_DIRTY,
2827 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2831 for (i = 0; i < nr_pages; i++) {
2832 struct page *page = pvec.pages[i];
2835 * At this point, the page may be truncated or
2836 * invalidated (changing page->mapping to NULL), or
2837 * even swizzled back from swapper_space to tmpfs file
2838 * mapping. However, page->index will not change
2839 * because we have a reference on the page.
2841 if (page->index > end) {
2849 * Page truncated or invalidated. We can freely skip it
2850 * then, even for data integrity operations: the page
2851 * has disappeared concurrently, so there could be no
2852 * real expectation of this data interity operation
2853 * even if there is now a new, dirty page at the same
2854 * pagecache address.
2856 if (unlikely(page->mapping != mapping)) {
2862 if (!PageDirty(page)) {
2863 /* someone wrote it for us */
2864 goto continue_unlock;
2867 if (PageWriteback(page)) {
2868 if (wbc->sync_mode != WB_SYNC_NONE)
2869 wait_on_page_writeback(page);
2871 goto continue_unlock;
2874 BUG_ON(PageWriteback(page));
2875 if (!clear_page_dirty_for_io(page))
2876 goto continue_unlock;
2878 ret = __mpage_da_writepage(page, wbc, mpd);
2879 if (unlikely(ret)) {
2880 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2889 if (nr_to_write > 0) {
2891 if (nr_to_write == 0 &&
2892 wbc->sync_mode == WB_SYNC_NONE) {
2894 * We stop writing back only if we are
2895 * not doing integrity sync. In case of
2896 * integrity sync we have to keep going
2897 * because someone may be concurrently
2898 * dirtying pages, and we might have
2899 * synced a lot of newly appeared dirty
2900 * pages, but have not synced all of the
2908 pagevec_release(&pvec);
2915 static int ext4_da_writepages(struct address_space *mapping,
2916 struct writeback_control *wbc)
2919 int range_whole = 0;
2920 handle_t *handle = NULL;
2921 struct mpage_da_data mpd;
2922 struct inode *inode = mapping->host;
2923 int pages_written = 0;
2925 unsigned int max_pages;
2926 int range_cyclic, cycled = 1, io_done = 0;
2927 int needed_blocks, ret = 0;
2928 long desired_nr_to_write, nr_to_writebump = 0;
2929 loff_t range_start = wbc->range_start;
2930 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2932 trace_ext4_da_writepages(inode, wbc);
2935 * No pages to write? This is mainly a kludge to avoid starting
2936 * a transaction for special inodes like journal inode on last iput()
2937 * because that could violate lock ordering on umount
2939 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2943 * If the filesystem has aborted, it is read-only, so return
2944 * right away instead of dumping stack traces later on that
2945 * will obscure the real source of the problem. We test
2946 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2947 * the latter could be true if the filesystem is mounted
2948 * read-only, and in that case, ext4_da_writepages should
2949 * *never* be called, so if that ever happens, we would want
2952 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2955 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2958 range_cyclic = wbc->range_cyclic;
2959 if (wbc->range_cyclic) {
2960 index = mapping->writeback_index;
2963 wbc->range_start = index << PAGE_CACHE_SHIFT;
2964 wbc->range_end = LLONG_MAX;
2965 wbc->range_cyclic = 0;
2967 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2970 * This works around two forms of stupidity. The first is in
2971 * the writeback code, which caps the maximum number of pages
2972 * written to be 1024 pages. This is wrong on multiple
2973 * levels; different architectues have a different page size,
2974 * which changes the maximum amount of data which gets
2975 * written. Secondly, 4 megabytes is way too small. XFS
2976 * forces this value to be 16 megabytes by multiplying
2977 * nr_to_write parameter by four, and then relies on its
2978 * allocator to allocate larger extents to make them
2979 * contiguous. Unfortunately this brings us to the second
2980 * stupidity, which is that ext4's mballoc code only allocates
2981 * at most 2048 blocks. So we force contiguous writes up to
2982 * the number of dirty blocks in the inode, or
2983 * sbi->max_writeback_mb_bump whichever is smaller.
2985 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2986 if (!range_cyclic && range_whole)
2987 desired_nr_to_write = wbc->nr_to_write * 8;
2989 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2991 if (desired_nr_to_write > max_pages)
2992 desired_nr_to_write = max_pages;
2994 if (wbc->nr_to_write < desired_nr_to_write) {
2995 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2996 wbc->nr_to_write = desired_nr_to_write;
3000 mpd.inode = mapping->host;
3002 pages_skipped = wbc->pages_skipped;
3005 while (!ret && wbc->nr_to_write > 0) {
3008 * we insert one extent at a time. So we need
3009 * credit needed for single extent allocation.
3010 * journalled mode is currently not supported
3013 BUG_ON(ext4_should_journal_data(inode));
3014 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3016 /* start a new transaction*/
3017 handle = ext4_journal_start(inode, needed_blocks);
3018 if (IS_ERR(handle)) {
3019 ret = PTR_ERR(handle);
3020 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3021 "%ld pages, ino %lu; err %d", __func__,
3022 wbc->nr_to_write, inode->i_ino, ret);
3023 goto out_writepages;
3027 * Now call __mpage_da_writepage to find the next
3028 * contiguous region of logical blocks that need
3029 * blocks to be allocated by ext4. We don't actually
3030 * submit the blocks for I/O here, even though
3031 * write_cache_pages thinks it will, and will set the
3032 * pages as clean for write before calling
3033 * __mpage_da_writepage().
3041 mpd.pages_written = 0;
3043 ret = write_cache_pages_da(mapping, wbc, &mpd);
3045 * If we have a contiguous extent of pages and we
3046 * haven't done the I/O yet, map the blocks and submit
3049 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3050 if (mpage_da_map_blocks(&mpd) == 0)
3051 mpage_da_submit_io(&mpd);
3053 ret = MPAGE_DA_EXTENT_TAIL;
3055 trace_ext4_da_write_pages(inode, &mpd);
3056 wbc->nr_to_write -= mpd.pages_written;
3058 ext4_journal_stop(handle);
3060 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3061 /* commit the transaction which would
3062 * free blocks released in the transaction
3065 jbd2_journal_force_commit_nested(sbi->s_journal);
3066 wbc->pages_skipped = pages_skipped;
3068 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3070 * got one extent now try with
3073 pages_written += mpd.pages_written;
3074 wbc->pages_skipped = pages_skipped;
3077 } else if (wbc->nr_to_write)
3079 * There is no more writeout needed
3080 * or we requested for a noblocking writeout
3081 * and we found the device congested
3085 if (!io_done && !cycled) {
3088 wbc->range_start = index << PAGE_CACHE_SHIFT;
3089 wbc->range_end = mapping->writeback_index - 1;
3092 if (pages_skipped != wbc->pages_skipped)
3093 ext4_msg(inode->i_sb, KERN_CRIT,
3094 "This should not happen leaving %s "
3095 "with nr_to_write = %ld ret = %d",
3096 __func__, wbc->nr_to_write, ret);
3099 index += pages_written;
3100 wbc->range_cyclic = range_cyclic;
3101 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3103 * set the writeback_index so that range_cyclic
3104 * mode will write it back later
3106 mapping->writeback_index = index;
3109 wbc->nr_to_write -= nr_to_writebump;
3110 wbc->range_start = range_start;
3111 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3115 #define FALL_BACK_TO_NONDELALLOC 1
3116 static int ext4_nonda_switch(struct super_block *sb)
3118 s64 free_blocks, dirty_blocks;
3119 struct ext4_sb_info *sbi = EXT4_SB(sb);
3122 * switch to non delalloc mode if we are running low
3123 * on free block. The free block accounting via percpu
3124 * counters can get slightly wrong with percpu_counter_batch getting
3125 * accumulated on each CPU without updating global counters
3126 * Delalloc need an accurate free block accounting. So switch
3127 * to non delalloc when we are near to error range.
3129 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3130 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3131 if (2 * free_blocks < 3 * dirty_blocks ||
3132 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3134 * free block count is less than 150% of dirty blocks
3135 * or free blocks is less than watermark
3140 * Even if we don't switch but are nearing capacity,
3141 * start pushing delalloc when 1/2 of free blocks are dirty.
3143 if (free_blocks < 2 * dirty_blocks)
3144 writeback_inodes_sb_if_idle(sb);
3149 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3150 loff_t pos, unsigned len, unsigned flags,
3151 struct page **pagep, void **fsdata)
3153 int ret, retries = 0;
3156 struct inode *inode = mapping->host;
3159 index = pos >> PAGE_CACHE_SHIFT;
3161 if (ext4_nonda_switch(inode->i_sb)) {
3162 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3163 return ext4_write_begin(file, mapping, pos,
3164 len, flags, pagep, fsdata);
3166 *fsdata = (void *)0;
3167 trace_ext4_da_write_begin(inode, pos, len, flags);
3170 * With delayed allocation, we don't log the i_disksize update
3171 * if there is delayed block allocation. But we still need
3172 * to journalling the i_disksize update if writes to the end
3173 * of file which has an already mapped buffer.
3175 handle = ext4_journal_start(inode, 1);
3176 if (IS_ERR(handle)) {
3177 ret = PTR_ERR(handle);
3180 /* We cannot recurse into the filesystem as the transaction is already
3182 flags |= AOP_FLAG_NOFS;
3184 page = grab_cache_page_write_begin(mapping, index, flags);
3186 ext4_journal_stop(handle);
3192 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3193 ext4_da_get_block_prep);
3196 ext4_journal_stop(handle);
3197 page_cache_release(page);
3199 * block_write_begin may have instantiated a few blocks
3200 * outside i_size. Trim these off again. Don't need
3201 * i_size_read because we hold i_mutex.
3203 if (pos + len > inode->i_size)
3204 ext4_truncate_failed_write(inode);
3207 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3214 * Check if we should update i_disksize
3215 * when write to the end of file but not require block allocation
3217 static int ext4_da_should_update_i_disksize(struct page *page,
3218 unsigned long offset)
3220 struct buffer_head *bh;
3221 struct inode *inode = page->mapping->host;
3225 bh = page_buffers(page);
3226 idx = offset >> inode->i_blkbits;
3228 for (i = 0; i < idx; i++)
3229 bh = bh->b_this_page;
3231 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3236 static int ext4_da_write_end(struct file *file,
3237 struct address_space *mapping,
3238 loff_t pos, unsigned len, unsigned copied,
3239 struct page *page, void *fsdata)
3241 struct inode *inode = mapping->host;
3243 handle_t *handle = ext4_journal_current_handle();
3245 unsigned long start, end;
3246 int write_mode = (int)(unsigned long)fsdata;
3248 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3249 if (ext4_should_order_data(inode)) {
3250 return ext4_ordered_write_end(file, mapping, pos,
3251 len, copied, page, fsdata);
3252 } else if (ext4_should_writeback_data(inode)) {
3253 return ext4_writeback_write_end(file, mapping, pos,
3254 len, copied, page, fsdata);
3260 trace_ext4_da_write_end(inode, pos, len, copied);
3261 start = pos & (PAGE_CACHE_SIZE - 1);
3262 end = start + copied - 1;
3265 * generic_write_end() will run mark_inode_dirty() if i_size
3266 * changes. So let's piggyback the i_disksize mark_inode_dirty
3270 new_i_size = pos + copied;
3271 if (new_i_size > EXT4_I(inode)->i_disksize) {
3272 if (ext4_da_should_update_i_disksize(page, end)) {
3273 down_write(&EXT4_I(inode)->i_data_sem);
3274 if (new_i_size > EXT4_I(inode)->i_disksize) {
3276 * Updating i_disksize when extending file
3277 * without needing block allocation
3279 if (ext4_should_order_data(inode))
3280 ret = ext4_jbd2_file_inode(handle,
3283 EXT4_I(inode)->i_disksize = new_i_size;
3285 up_write(&EXT4_I(inode)->i_data_sem);
3286 /* We need to mark inode dirty even if
3287 * new_i_size is less that inode->i_size
3288 * bu greater than i_disksize.(hint delalloc)
3290 ext4_mark_inode_dirty(handle, inode);
3293 ret2 = generic_write_end(file, mapping, pos, len, copied,
3298 ret2 = ext4_journal_stop(handle);
3302 return ret ? ret : copied;
3305 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3308 * Drop reserved blocks
3310 BUG_ON(!PageLocked(page));
3311 if (!page_has_buffers(page))
3314 ext4_da_page_release_reservation(page, offset);
3317 ext4_invalidatepage(page, offset);
3323 * Force all delayed allocation blocks to be allocated for a given inode.
3325 int ext4_alloc_da_blocks(struct inode *inode)
3327 trace_ext4_alloc_da_blocks(inode);
3329 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3330 !EXT4_I(inode)->i_reserved_meta_blocks)
3334 * We do something simple for now. The filemap_flush() will
3335 * also start triggering a write of the data blocks, which is
3336 * not strictly speaking necessary (and for users of
3337 * laptop_mode, not even desirable). However, to do otherwise
3338 * would require replicating code paths in:
3340 * ext4_da_writepages() ->
3341 * write_cache_pages() ---> (via passed in callback function)
3342 * __mpage_da_writepage() -->
3343 * mpage_add_bh_to_extent()
3344 * mpage_da_map_blocks()
3346 * The problem is that write_cache_pages(), located in
3347 * mm/page-writeback.c, marks pages clean in preparation for
3348 * doing I/O, which is not desirable if we're not planning on
3351 * We could call write_cache_pages(), and then redirty all of
3352 * the pages by calling redirty_page_for_writeback() but that
3353 * would be ugly in the extreme. So instead we would need to
3354 * replicate parts of the code in the above functions,
3355 * simplifying them becuase we wouldn't actually intend to
3356 * write out the pages, but rather only collect contiguous
3357 * logical block extents, call the multi-block allocator, and
3358 * then update the buffer heads with the block allocations.
3360 * For now, though, we'll cheat by calling filemap_flush(),
3361 * which will map the blocks, and start the I/O, but not
3362 * actually wait for the I/O to complete.
3364 return filemap_flush(inode->i_mapping);
3368 * bmap() is special. It gets used by applications such as lilo and by
3369 * the swapper to find the on-disk block of a specific piece of data.
3371 * Naturally, this is dangerous if the block concerned is still in the
3372 * journal. If somebody makes a swapfile on an ext4 data-journaling
3373 * filesystem and enables swap, then they may get a nasty shock when the
3374 * data getting swapped to that swapfile suddenly gets overwritten by
3375 * the original zero's written out previously to the journal and
3376 * awaiting writeback in the kernel's buffer cache.
3378 * So, if we see any bmap calls here on a modified, data-journaled file,
3379 * take extra steps to flush any blocks which might be in the cache.
3381 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3383 struct inode *inode = mapping->host;
3387 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3388 test_opt(inode->i_sb, DELALLOC)) {
3390 * With delalloc we want to sync the file
3391 * so that we can make sure we allocate
3394 filemap_write_and_wait(mapping);
3397 if (EXT4_JOURNAL(inode) &&
3398 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3400 * This is a REALLY heavyweight approach, but the use of
3401 * bmap on dirty files is expected to be extremely rare:
3402 * only if we run lilo or swapon on a freshly made file
3403 * do we expect this to happen.
3405 * (bmap requires CAP_SYS_RAWIO so this does not
3406 * represent an unprivileged user DOS attack --- we'd be
3407 * in trouble if mortal users could trigger this path at
3410 * NB. EXT4_STATE_JDATA is not set on files other than
3411 * regular files. If somebody wants to bmap a directory
3412 * or symlink and gets confused because the buffer
3413 * hasn't yet been flushed to disk, they deserve
3414 * everything they get.
3417 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3418 journal = EXT4_JOURNAL(inode);
3419 jbd2_journal_lock_updates(journal);
3420 err = jbd2_journal_flush(journal);
3421 jbd2_journal_unlock_updates(journal);
3427 return generic_block_bmap(mapping, block, ext4_get_block);
3430 static int ext4_readpage(struct file *file, struct page *page)
3432 return mpage_readpage(page, ext4_get_block);
3436 ext4_readpages(struct file *file, struct address_space *mapping,
3437 struct list_head *pages, unsigned nr_pages)
3439 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3442 static void ext4_free_io_end(ext4_io_end_t *io)
3451 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3453 struct buffer_head *head, *bh;
3454 unsigned int curr_off = 0;
3456 if (!page_has_buffers(page))
3458 head = bh = page_buffers(page);
3460 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3462 ext4_free_io_end(bh->b_private);
3463 bh->b_private = NULL;
3464 bh->b_end_io = NULL;
3466 curr_off = curr_off + bh->b_size;
3467 bh = bh->b_this_page;
3468 } while (bh != head);
3471 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3473 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3476 * free any io_end structure allocated for buffers to be discarded
3478 if (ext4_should_dioread_nolock(page->mapping->host))
3479 ext4_invalidatepage_free_endio(page, offset);
3481 * If it's a full truncate we just forget about the pending dirtying
3484 ClearPageChecked(page);
3487 jbd2_journal_invalidatepage(journal, page, offset);
3489 block_invalidatepage(page, offset);
3492 static int ext4_releasepage(struct page *page, gfp_t wait)
3494 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3496 WARN_ON(PageChecked(page));
3497 if (!page_has_buffers(page))
3500 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3502 return try_to_free_buffers(page);
3506 * O_DIRECT for ext3 (or indirect map) based files
3508 * If the O_DIRECT write will extend the file then add this inode to the
3509 * orphan list. So recovery will truncate it back to the original size
3510 * if the machine crashes during the write.
3512 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3513 * crashes then stale disk data _may_ be exposed inside the file. But current
3514 * VFS code falls back into buffered path in that case so we are safe.
3516 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3517 const struct iovec *iov, loff_t offset,
3518 unsigned long nr_segs)
3520 struct file *file = iocb->ki_filp;
3521 struct inode *inode = file->f_mapping->host;
3522 struct ext4_inode_info *ei = EXT4_I(inode);
3526 size_t count = iov_length(iov, nr_segs);
3530 loff_t final_size = offset + count;
3532 if (final_size > inode->i_size) {
3533 /* Credits for sb + inode write */
3534 handle = ext4_journal_start(inode, 2);
3535 if (IS_ERR(handle)) {
3536 ret = PTR_ERR(handle);
3539 ret = ext4_orphan_add(handle, inode);
3541 ext4_journal_stop(handle);
3545 ei->i_disksize = inode->i_size;
3546 ext4_journal_stop(handle);
3551 if (rw == READ && ext4_should_dioread_nolock(inode))
3552 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
3553 inode->i_sb->s_bdev, iov,
3555 ext4_get_block, NULL);
3557 ret = blockdev_direct_IO(rw, iocb, inode,
3558 inode->i_sb->s_bdev, iov,
3560 ext4_get_block, NULL);
3561 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3567 /* Credits for sb + inode write */
3568 handle = ext4_journal_start(inode, 2);
3569 if (IS_ERR(handle)) {
3570 /* This is really bad luck. We've written the data
3571 * but cannot extend i_size. Bail out and pretend
3572 * the write failed... */
3573 ret = PTR_ERR(handle);
3575 ext4_orphan_del(NULL, inode);
3580 ext4_orphan_del(handle, inode);
3582 loff_t end = offset + ret;
3583 if (end > inode->i_size) {
3584 ei->i_disksize = end;
3585 i_size_write(inode, end);
3587 * We're going to return a positive `ret'
3588 * here due to non-zero-length I/O, so there's
3589 * no way of reporting error returns from
3590 * ext4_mark_inode_dirty() to userspace. So
3593 ext4_mark_inode_dirty(handle, inode);
3596 err = ext4_journal_stop(handle);
3605 * ext4_get_block used when preparing for a DIO write or buffer write.
3606 * We allocate an uinitialized extent if blocks haven't been allocated.
3607 * The extent will be converted to initialized after the IO is complete.
3609 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3610 struct buffer_head *bh_result, int create)
3612 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3613 inode->i_ino, create);
3614 return _ext4_get_block(inode, iblock, bh_result,
3615 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3618 static void dump_completed_IO(struct inode * inode)
3621 struct list_head *cur, *before, *after;
3622 ext4_io_end_t *io, *io0, *io1;
3623 unsigned long flags;
3625 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3626 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3630 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3631 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3632 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3635 io0 = container_of(before, ext4_io_end_t, list);
3637 io1 = container_of(after, ext4_io_end_t, list);
3639 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3640 io, inode->i_ino, io0, io1);
3642 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3647 * check a range of space and convert unwritten extents to written.
3649 static int ext4_end_io_nolock(ext4_io_end_t *io)
3651 struct inode *inode = io->inode;
3652 loff_t offset = io->offset;
3653 ssize_t size = io->size;
3656 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3657 "list->prev 0x%p\n",
3658 io, inode->i_ino, io->list.next, io->list.prev);
3660 if (list_empty(&io->list))
3663 if (io->flag != EXT4_IO_UNWRITTEN)
3666 ret = ext4_convert_unwritten_extents(inode, offset, size);
3668 printk(KERN_EMERG "%s: failed to convert unwritten"
3669 "extents to written extents, error is %d"
3670 " io is still on inode %lu aio dio list\n",
3671 __func__, ret, inode->i_ino);
3676 aio_complete(io->iocb, io->result, 0);
3677 /* clear the DIO AIO unwritten flag */
3683 * work on completed aio dio IO, to convert unwritten extents to extents
3685 static void ext4_end_io_work(struct work_struct *work)
3687 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3688 struct inode *inode = io->inode;
3689 struct ext4_inode_info *ei = EXT4_I(inode);
3690 unsigned long flags;
3693 mutex_lock(&inode->i_mutex);
3694 ret = ext4_end_io_nolock(io);
3696 mutex_unlock(&inode->i_mutex);
3700 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3701 if (!list_empty(&io->list))
3702 list_del_init(&io->list);
3703 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3704 mutex_unlock(&inode->i_mutex);
3705 ext4_free_io_end(io);
3709 * This function is called from ext4_sync_file().
3711 * When IO is completed, the work to convert unwritten extents to
3712 * written is queued on workqueue but may not get immediately
3713 * scheduled. When fsync is called, we need to ensure the
3714 * conversion is complete before fsync returns.
3715 * The inode keeps track of a list of pending/completed IO that
3716 * might needs to do the conversion. This function walks through
3717 * the list and convert the related unwritten extents for completed IO
3719 * The function return the number of pending IOs on success.
3721 int flush_completed_IO(struct inode *inode)
3724 struct ext4_inode_info *ei = EXT4_I(inode);
3725 unsigned long flags;
3729 if (list_empty(&ei->i_completed_io_list))
3732 dump_completed_IO(inode);
3733 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3734 while (!list_empty(&ei->i_completed_io_list)){
3735 io = list_entry(ei->i_completed_io_list.next,
3736 ext4_io_end_t, list);
3738 * Calling ext4_end_io_nolock() to convert completed
3741 * When ext4_sync_file() is called, run_queue() may already
3742 * about to flush the work corresponding to this io structure.
3743 * It will be upset if it founds the io structure related
3744 * to the work-to-be schedule is freed.
3746 * Thus we need to keep the io structure still valid here after
3747 * convertion finished. The io structure has a flag to
3748 * avoid double converting from both fsync and background work
3751 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3752 ret = ext4_end_io_nolock(io);
3753 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3757 list_del_init(&io->list);
3759 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3760 return (ret2 < 0) ? ret2 : 0;
3763 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3765 ext4_io_end_t *io = NULL;
3767 io = kmalloc(sizeof(*io), flags);
3778 INIT_WORK(&io->work, ext4_end_io_work);
3779 INIT_LIST_HEAD(&io->list);
3785 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3786 ssize_t size, void *private, int ret,
3789 ext4_io_end_t *io_end = iocb->private;
3790 struct workqueue_struct *wq;
3791 unsigned long flags;
3792 struct ext4_inode_info *ei;
3794 /* if not async direct IO or dio with 0 bytes write, just return */
3795 if (!io_end || !size)
3798 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3799 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3800 iocb->private, io_end->inode->i_ino, iocb, offset,
3803 /* if not aio dio with unwritten extents, just free io and return */
3804 if (io_end->flag != EXT4_IO_UNWRITTEN){
3805 ext4_free_io_end(io_end);
3806 iocb->private = NULL;
3809 aio_complete(iocb, ret, 0);
3813 io_end->offset = offset;
3814 io_end->size = size;
3816 io_end->iocb = iocb;
3817 io_end->result = ret;
3819 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3821 /* queue the work to convert unwritten extents to written */
3822 queue_work(wq, &io_end->work);
3824 /* Add the io_end to per-inode completed aio dio list*/
3825 ei = EXT4_I(io_end->inode);
3826 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3827 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3828 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3829 iocb->private = NULL;
3832 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3834 ext4_io_end_t *io_end = bh->b_private;
3835 struct workqueue_struct *wq;
3836 struct inode *inode;
3837 unsigned long flags;
3839 if (!test_clear_buffer_uninit(bh) || !io_end)
3842 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3843 printk("sb umounted, discard end_io request for inode %lu\n",
3844 io_end->inode->i_ino);
3845 ext4_free_io_end(io_end);
3849 io_end->flag = EXT4_IO_UNWRITTEN;
3850 inode = io_end->inode;
3852 /* Add the io_end to per-inode completed io list*/
3853 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3854 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3855 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3857 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3858 /* queue the work to convert unwritten extents to written */
3859 queue_work(wq, &io_end->work);
3861 bh->b_private = NULL;
3862 bh->b_end_io = NULL;
3863 clear_buffer_uninit(bh);
3864 end_buffer_async_write(bh, uptodate);
3867 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3869 ext4_io_end_t *io_end;
3870 struct page *page = bh->b_page;
3871 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3872 size_t size = bh->b_size;
3875 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3877 if (printk_ratelimit())
3878 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3882 io_end->offset = offset;
3883 io_end->size = size;
3885 * We need to hold a reference to the page to make sure it
3886 * doesn't get evicted before ext4_end_io_work() has a chance
3887 * to convert the extent from written to unwritten.
3889 io_end->page = page;
3890 get_page(io_end->page);
3892 bh->b_private = io_end;
3893 bh->b_end_io = ext4_end_io_buffer_write;
3898 * For ext4 extent files, ext4 will do direct-io write to holes,
3899 * preallocated extents, and those write extend the file, no need to
3900 * fall back to buffered IO.
3902 * For holes, we fallocate those blocks, mark them as unintialized
3903 * If those blocks were preallocated, we mark sure they are splited, but
3904 * still keep the range to write as unintialized.
3906 * The unwrritten extents will be converted to written when DIO is completed.
3907 * For async direct IO, since the IO may still pending when return, we
3908 * set up an end_io call back function, which will do the convertion
3909 * when async direct IO completed.
3911 * If the O_DIRECT write will extend the file then add this inode to the
3912 * orphan list. So recovery will truncate it back to the original size
3913 * if the machine crashes during the write.
3916 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3917 const struct iovec *iov, loff_t offset,
3918 unsigned long nr_segs)
3920 struct file *file = iocb->ki_filp;
3921 struct inode *inode = file->f_mapping->host;
3923 size_t count = iov_length(iov, nr_segs);
3925 loff_t final_size = offset + count;
3926 if (rw == WRITE && final_size <= inode->i_size) {
3928 * We could direct write to holes and fallocate.
3930 * Allocated blocks to fill the hole are marked as uninitialized
3931 * to prevent paralel buffered read to expose the stale data
3932 * before DIO complete the data IO.
3934 * As to previously fallocated extents, ext4 get_block
3935 * will just simply mark the buffer mapped but still
3936 * keep the extents uninitialized.
3938 * for non AIO case, we will convert those unwritten extents
3939 * to written after return back from blockdev_direct_IO.
3941 * for async DIO, the conversion needs to be defered when
3942 * the IO is completed. The ext4 end_io callback function
3943 * will be called to take care of the conversion work.
3944 * Here for async case, we allocate an io_end structure to
3947 iocb->private = NULL;
3948 EXT4_I(inode)->cur_aio_dio = NULL;
3949 if (!is_sync_kiocb(iocb)) {
3950 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3954 * we save the io structure for current async
3955 * direct IO, so that later ext4_map_blocks()
3956 * could flag the io structure whether there
3957 * is a unwritten extents needs to be converted
3958 * when IO is completed.
3960 EXT4_I(inode)->cur_aio_dio = iocb->private;
3963 ret = blockdev_direct_IO(rw, iocb, inode,
3964 inode->i_sb->s_bdev, iov,
3966 ext4_get_block_write,
3969 EXT4_I(inode)->cur_aio_dio = NULL;
3971 * The io_end structure takes a reference to the inode,
3972 * that structure needs to be destroyed and the
3973 * reference to the inode need to be dropped, when IO is
3974 * complete, even with 0 byte write, or failed.
3976 * In the successful AIO DIO case, the io_end structure will be
3977 * desctroyed and the reference to the inode will be dropped
3978 * after the end_io call back function is called.
3980 * In the case there is 0 byte write, or error case, since
3981 * VFS direct IO won't invoke the end_io call back function,
3982 * we need to free the end_io structure here.
3984 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3985 ext4_free_io_end(iocb->private);
3986 iocb->private = NULL;
3987 } else if (ret > 0 && ext4_test_inode_state(inode,
3988 EXT4_STATE_DIO_UNWRITTEN)) {
3991 * for non AIO case, since the IO is already
3992 * completed, we could do the convertion right here
3994 err = ext4_convert_unwritten_extents(inode,
3998 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4003 /* for write the the end of file case, we fall back to old way */
4004 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4007 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
4008 const struct iovec *iov, loff_t offset,
4009 unsigned long nr_segs)
4011 struct file *file = iocb->ki_filp;
4012 struct inode *inode = file->f_mapping->host;
4014 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4015 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4017 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4021 * Pages can be marked dirty completely asynchronously from ext4's journalling
4022 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4023 * much here because ->set_page_dirty is called under VFS locks. The page is
4024 * not necessarily locked.
4026 * We cannot just dirty the page and leave attached buffers clean, because the
4027 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4028 * or jbddirty because all the journalling code will explode.
4030 * So what we do is to mark the page "pending dirty" and next time writepage
4031 * is called, propagate that into the buffers appropriately.
4033 static int ext4_journalled_set_page_dirty(struct page *page)
4035 SetPageChecked(page);
4036 return __set_page_dirty_nobuffers(page);
4039 static const struct address_space_operations ext4_ordered_aops = {
4040 .readpage = ext4_readpage,
4041 .readpages = ext4_readpages,
4042 .writepage = ext4_writepage,
4043 .sync_page = block_sync_page,
4044 .write_begin = ext4_write_begin,
4045 .write_end = ext4_ordered_write_end,
4047 .invalidatepage = ext4_invalidatepage,
4048 .releasepage = ext4_releasepage,
4049 .direct_IO = ext4_direct_IO,
4050 .migratepage = buffer_migrate_page,
4051 .is_partially_uptodate = block_is_partially_uptodate,
4052 .error_remove_page = generic_error_remove_page,
4055 static const struct address_space_operations ext4_writeback_aops = {
4056 .readpage = ext4_readpage,
4057 .readpages = ext4_readpages,
4058 .writepage = ext4_writepage,
4059 .sync_page = block_sync_page,
4060 .write_begin = ext4_write_begin,
4061 .write_end = ext4_writeback_write_end,
4063 .invalidatepage = ext4_invalidatepage,
4064 .releasepage = ext4_releasepage,
4065 .direct_IO = ext4_direct_IO,
4066 .migratepage = buffer_migrate_page,
4067 .is_partially_uptodate = block_is_partially_uptodate,
4068 .error_remove_page = generic_error_remove_page,
4071 static const struct address_space_operations ext4_journalled_aops = {
4072 .readpage = ext4_readpage,
4073 .readpages = ext4_readpages,
4074 .writepage = ext4_writepage,
4075 .sync_page = block_sync_page,
4076 .write_begin = ext4_write_begin,
4077 .write_end = ext4_journalled_write_end,
4078 .set_page_dirty = ext4_journalled_set_page_dirty,
4080 .invalidatepage = ext4_invalidatepage,
4081 .releasepage = ext4_releasepage,
4082 .is_partially_uptodate = block_is_partially_uptodate,
4083 .error_remove_page = generic_error_remove_page,
4086 static const struct address_space_operations ext4_da_aops = {
4087 .readpage = ext4_readpage,
4088 .readpages = ext4_readpages,
4089 .writepage = ext4_writepage,
4090 .writepages = ext4_da_writepages,
4091 .sync_page = block_sync_page,
4092 .write_begin = ext4_da_write_begin,
4093 .write_end = ext4_da_write_end,
4095 .invalidatepage = ext4_da_invalidatepage,
4096 .releasepage = ext4_releasepage,
4097 .direct_IO = ext4_direct_IO,
4098 .migratepage = buffer_migrate_page,
4099 .is_partially_uptodate = block_is_partially_uptodate,
4100 .error_remove_page = generic_error_remove_page,
4103 void ext4_set_aops(struct inode *inode)
4105 if (ext4_should_order_data(inode) &&
4106 test_opt(inode->i_sb, DELALLOC))
4107 inode->i_mapping->a_ops = &ext4_da_aops;
4108 else if (ext4_should_order_data(inode))
4109 inode->i_mapping->a_ops = &ext4_ordered_aops;
4110 else if (ext4_should_writeback_data(inode) &&
4111 test_opt(inode->i_sb, DELALLOC))
4112 inode->i_mapping->a_ops = &ext4_da_aops;
4113 else if (ext4_should_writeback_data(inode))
4114 inode->i_mapping->a_ops = &ext4_writeback_aops;
4116 inode->i_mapping->a_ops = &ext4_journalled_aops;
4120 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4121 * up to the end of the block which corresponds to `from'.
4122 * This required during truncate. We need to physically zero the tail end
4123 * of that block so it doesn't yield old data if the file is later grown.
4125 int ext4_block_truncate_page(handle_t *handle,
4126 struct address_space *mapping, loff_t from)
4128 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4129 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4130 unsigned blocksize, length, pos;
4132 struct inode *inode = mapping->host;
4133 struct buffer_head *bh;
4137 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4138 mapping_gfp_mask(mapping) & ~__GFP_FS);
4142 blocksize = inode->i_sb->s_blocksize;
4143 length = blocksize - (offset & (blocksize - 1));
4144 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4146 if (!page_has_buffers(page))
4147 create_empty_buffers(page, blocksize, 0);
4149 /* Find the buffer that contains "offset" */
4150 bh = page_buffers(page);
4152 while (offset >= pos) {
4153 bh = bh->b_this_page;
4159 if (buffer_freed(bh)) {
4160 BUFFER_TRACE(bh, "freed: skip");
4164 if (!buffer_mapped(bh)) {
4165 BUFFER_TRACE(bh, "unmapped");
4166 ext4_get_block(inode, iblock, bh, 0);
4167 /* unmapped? It's a hole - nothing to do */
4168 if (!buffer_mapped(bh)) {
4169 BUFFER_TRACE(bh, "still unmapped");
4174 /* Ok, it's mapped. Make sure it's up-to-date */
4175 if (PageUptodate(page))
4176 set_buffer_uptodate(bh);
4178 if (!buffer_uptodate(bh)) {
4180 ll_rw_block(READ, 1, &bh);
4182 /* Uhhuh. Read error. Complain and punt. */
4183 if (!buffer_uptodate(bh))
4187 if (ext4_should_journal_data(inode)) {
4188 BUFFER_TRACE(bh, "get write access");
4189 err = ext4_journal_get_write_access(handle, bh);
4194 zero_user(page, offset, length);
4196 BUFFER_TRACE(bh, "zeroed end of block");
4199 if (ext4_should_journal_data(inode)) {
4200 err = ext4_handle_dirty_metadata(handle, inode, bh);
4202 if (ext4_should_order_data(inode))
4203 err = ext4_jbd2_file_inode(handle, inode);
4204 mark_buffer_dirty(bh);
4209 page_cache_release(page);
4214 * Probably it should be a library function... search for first non-zero word
4215 * or memcmp with zero_page, whatever is better for particular architecture.
4218 static inline int all_zeroes(__le32 *p, __le32 *q)
4227 * ext4_find_shared - find the indirect blocks for partial truncation.
4228 * @inode: inode in question
4229 * @depth: depth of the affected branch
4230 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4231 * @chain: place to store the pointers to partial indirect blocks
4232 * @top: place to the (detached) top of branch
4234 * This is a helper function used by ext4_truncate().
4236 * When we do truncate() we may have to clean the ends of several
4237 * indirect blocks but leave the blocks themselves alive. Block is
4238 * partially truncated if some data below the new i_size is refered
4239 * from it (and it is on the path to the first completely truncated
4240 * data block, indeed). We have to free the top of that path along
4241 * with everything to the right of the path. Since no allocation
4242 * past the truncation point is possible until ext4_truncate()
4243 * finishes, we may safely do the latter, but top of branch may
4244 * require special attention - pageout below the truncation point
4245 * might try to populate it.
4247 * We atomically detach the top of branch from the tree, store the
4248 * block number of its root in *@top, pointers to buffer_heads of
4249 * partially truncated blocks - in @chain[].bh and pointers to
4250 * their last elements that should not be removed - in
4251 * @chain[].p. Return value is the pointer to last filled element
4254 * The work left to caller to do the actual freeing of subtrees:
4255 * a) free the subtree starting from *@top
4256 * b) free the subtrees whose roots are stored in
4257 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4258 * c) free the subtrees growing from the inode past the @chain[0].
4259 * (no partially truncated stuff there). */
4261 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4262 ext4_lblk_t offsets[4], Indirect chain[4],
4265 Indirect *partial, *p;
4269 /* Make k index the deepest non-null offset + 1 */
4270 for (k = depth; k > 1 && !offsets[k-1]; k--)
4272 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4273 /* Writer: pointers */
4275 partial = chain + k-1;
4277 * If the branch acquired continuation since we've looked at it -
4278 * fine, it should all survive and (new) top doesn't belong to us.
4280 if (!partial->key && *partial->p)
4283 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4286 * OK, we've found the last block that must survive. The rest of our
4287 * branch should be detached before unlocking. However, if that rest
4288 * of branch is all ours and does not grow immediately from the inode
4289 * it's easier to cheat and just decrement partial->p.
4291 if (p == chain + k - 1 && p > chain) {
4295 /* Nope, don't do this in ext4. Must leave the tree intact */
4302 while (partial > p) {
4303 brelse(partial->bh);
4311 * Zero a number of block pointers in either an inode or an indirect block.
4312 * If we restart the transaction we must again get write access to the
4313 * indirect block for further modification.
4315 * We release `count' blocks on disk, but (last - first) may be greater
4316 * than `count' because there can be holes in there.
4318 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4319 struct buffer_head *bh,
4320 ext4_fsblk_t block_to_free,
4321 unsigned long count, __le32 *first,
4325 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4327 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4328 flags |= EXT4_FREE_BLOCKS_METADATA;
4330 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4332 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4333 "blocks %llu len %lu",
4334 (unsigned long long) block_to_free, count);
4338 if (try_to_extend_transaction(handle, inode)) {
4340 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4341 ext4_handle_dirty_metadata(handle, inode, bh);
4343 ext4_mark_inode_dirty(handle, inode);
4344 ext4_truncate_restart_trans(handle, inode,
4345 blocks_for_truncate(inode));
4347 BUFFER_TRACE(bh, "retaking write access");
4348 ext4_journal_get_write_access(handle, bh);
4352 for (p = first; p < last; p++)
4355 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4360 * ext4_free_data - free a list of data blocks
4361 * @handle: handle for this transaction
4362 * @inode: inode we are dealing with
4363 * @this_bh: indirect buffer_head which contains *@first and *@last
4364 * @first: array of block numbers
4365 * @last: points immediately past the end of array
4367 * We are freeing all blocks refered from that array (numbers are stored as
4368 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4370 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4371 * blocks are contiguous then releasing them at one time will only affect one
4372 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4373 * actually use a lot of journal space.
4375 * @this_bh will be %NULL if @first and @last point into the inode's direct
4378 static void ext4_free_data(handle_t *handle, struct inode *inode,
4379 struct buffer_head *this_bh,
4380 __le32 *first, __le32 *last)
4382 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4383 unsigned long count = 0; /* Number of blocks in the run */
4384 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4387 ext4_fsblk_t nr; /* Current block # */
4388 __le32 *p; /* Pointer into inode/ind
4389 for current block */
4392 if (this_bh) { /* For indirect block */
4393 BUFFER_TRACE(this_bh, "get_write_access");
4394 err = ext4_journal_get_write_access(handle, this_bh);
4395 /* Important: if we can't update the indirect pointers
4396 * to the blocks, we can't free them. */
4401 for (p = first; p < last; p++) {
4402 nr = le32_to_cpu(*p);
4404 /* accumulate blocks to free if they're contiguous */
4407 block_to_free_p = p;
4409 } else if (nr == block_to_free + count) {
4412 if (ext4_clear_blocks(handle, inode, this_bh,
4413 block_to_free, count,
4414 block_to_free_p, p))
4417 block_to_free_p = p;
4424 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4425 count, block_to_free_p, p);
4428 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4431 * The buffer head should have an attached journal head at this
4432 * point. However, if the data is corrupted and an indirect
4433 * block pointed to itself, it would have been detached when
4434 * the block was cleared. Check for this instead of OOPSing.
4436 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4437 ext4_handle_dirty_metadata(handle, inode, this_bh);
4439 EXT4_ERROR_INODE(inode,
4440 "circular indirect block detected at "
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(inode,
4481 "invalid indirect mapped "
4482 "block %lu (level %d)",
4483 (unsigned long) nr, depth);
4487 /* Go read the buffer for the next level down */
4488 bh = sb_bread(inode->i_sb, nr);
4491 * A read failure? Report error and clear slot
4495 EXT4_ERROR_INODE_BLOCK(inode, nr,
4500 /* This zaps the entire block. Bottom up. */
4501 BUFFER_TRACE(bh, "free child branches");
4502 ext4_free_branches(handle, inode, bh,
4503 (__le32 *) bh->b_data,
4504 (__le32 *) bh->b_data + addr_per_block,
4508 * Everything below this this pointer has been
4509 * released. Now let this top-of-subtree go.
4511 * We want the freeing of this indirect block to be
4512 * atomic in the journal with the updating of the
4513 * bitmap block which owns it. So make some room in
4516 * We zero the parent pointer *after* freeing its
4517 * pointee in the bitmaps, so if extend_transaction()
4518 * for some reason fails to put the bitmap changes and
4519 * the release into the same transaction, recovery
4520 * will merely complain about releasing a free block,
4521 * rather than leaking blocks.
4523 if (ext4_handle_is_aborted(handle))
4525 if (try_to_extend_transaction(handle, inode)) {
4526 ext4_mark_inode_dirty(handle, inode);
4527 ext4_truncate_restart_trans(handle, inode,
4528 blocks_for_truncate(inode));
4532 * The forget flag here is critical because if
4533 * we are journaling (and not doing data
4534 * journaling), we have to make sure a revoke
4535 * record is written to prevent the journal
4536 * replay from overwriting the (former)
4537 * indirect block if it gets reallocated as a
4538 * data block. This must happen in the same
4539 * transaction where the data blocks are
4542 ext4_free_blocks(handle, inode, 0, nr, 1,
4543 EXT4_FREE_BLOCKS_METADATA|
4544 EXT4_FREE_BLOCKS_FORGET);
4548 * The block which we have just freed is
4549 * pointed to by an indirect block: journal it
4551 BUFFER_TRACE(parent_bh, "get_write_access");
4552 if (!ext4_journal_get_write_access(handle,
4555 BUFFER_TRACE(parent_bh,
4556 "call ext4_handle_dirty_metadata");
4557 ext4_handle_dirty_metadata(handle,
4564 /* We have reached the bottom of the tree. */
4565 BUFFER_TRACE(parent_bh, "free data blocks");
4566 ext4_free_data(handle, inode, parent_bh, first, last);
4570 int ext4_can_truncate(struct inode *inode)
4572 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4574 if (S_ISREG(inode->i_mode))
4576 if (S_ISDIR(inode->i_mode))
4578 if (S_ISLNK(inode->i_mode))
4579 return !ext4_inode_is_fast_symlink(inode);
4586 * We block out ext4_get_block() block instantiations across the entire
4587 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4588 * simultaneously on behalf of the same inode.
4590 * As we work through the truncate and commmit bits of it to the journal there
4591 * is one core, guiding principle: the file's tree must always be consistent on
4592 * disk. We must be able to restart the truncate after a crash.
4594 * The file's tree may be transiently inconsistent in memory (although it
4595 * probably isn't), but whenever we close off and commit a journal transaction,
4596 * the contents of (the filesystem + the journal) must be consistent and
4597 * restartable. It's pretty simple, really: bottom up, right to left (although
4598 * left-to-right works OK too).
4600 * Note that at recovery time, journal replay occurs *before* the restart of
4601 * truncate against the orphan inode list.
4603 * The committed inode has the new, desired i_size (which is the same as
4604 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4605 * that this inode's truncate did not complete and it will again call
4606 * ext4_truncate() to have another go. So there will be instantiated blocks
4607 * to the right of the truncation point in a crashed ext4 filesystem. But
4608 * that's fine - as long as they are linked from the inode, the post-crash
4609 * ext4_truncate() run will find them and release them.
4611 void ext4_truncate(struct inode *inode)
4614 struct ext4_inode_info *ei = EXT4_I(inode);
4615 __le32 *i_data = ei->i_data;
4616 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4617 struct address_space *mapping = inode->i_mapping;
4618 ext4_lblk_t offsets[4];
4623 ext4_lblk_t last_block;
4624 unsigned blocksize = inode->i_sb->s_blocksize;
4626 if (!ext4_can_truncate(inode))
4629 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4631 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4632 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4634 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4635 ext4_ext_truncate(inode);
4639 handle = start_transaction(inode);
4641 return; /* AKPM: return what? */
4643 last_block = (inode->i_size + blocksize-1)
4644 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4646 if (inode->i_size & (blocksize - 1))
4647 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4650 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4652 goto out_stop; /* error */
4655 * OK. This truncate is going to happen. We add the inode to the
4656 * orphan list, so that if this truncate spans multiple transactions,
4657 * and we crash, we will resume the truncate when the filesystem
4658 * recovers. It also marks the inode dirty, to catch the new size.
4660 * Implication: the file must always be in a sane, consistent
4661 * truncatable state while each transaction commits.
4663 if (ext4_orphan_add(handle, inode))
4667 * From here we block out all ext4_get_block() callers who want to
4668 * modify the block allocation tree.
4670 down_write(&ei->i_data_sem);
4672 ext4_discard_preallocations(inode);
4675 * The orphan list entry will now protect us from any crash which
4676 * occurs before the truncate completes, so it is now safe to propagate
4677 * the new, shorter inode size (held for now in i_size) into the
4678 * on-disk inode. We do this via i_disksize, which is the value which
4679 * ext4 *really* writes onto the disk inode.
4681 ei->i_disksize = inode->i_size;
4683 if (n == 1) { /* direct blocks */
4684 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4685 i_data + EXT4_NDIR_BLOCKS);
4689 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4690 /* Kill the top of shared branch (not detached) */
4692 if (partial == chain) {
4693 /* Shared branch grows from the inode */
4694 ext4_free_branches(handle, inode, NULL,
4695 &nr, &nr+1, (chain+n-1) - partial);
4698 * We mark the inode dirty prior to restart,
4699 * and prior to stop. No need for it here.
4702 /* Shared branch grows from an indirect block */
4703 BUFFER_TRACE(partial->bh, "get_write_access");
4704 ext4_free_branches(handle, inode, partial->bh,
4706 partial->p+1, (chain+n-1) - partial);
4709 /* Clear the ends of indirect blocks on the shared branch */
4710 while (partial > chain) {
4711 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4712 (__le32*)partial->bh->b_data+addr_per_block,
4713 (chain+n-1) - partial);
4714 BUFFER_TRACE(partial->bh, "call brelse");
4715 brelse(partial->bh);
4719 /* Kill the remaining (whole) subtrees */
4720 switch (offsets[0]) {
4722 nr = i_data[EXT4_IND_BLOCK];
4724 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4725 i_data[EXT4_IND_BLOCK] = 0;
4727 case EXT4_IND_BLOCK:
4728 nr = i_data[EXT4_DIND_BLOCK];
4730 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4731 i_data[EXT4_DIND_BLOCK] = 0;
4733 case EXT4_DIND_BLOCK:
4734 nr = i_data[EXT4_TIND_BLOCK];
4736 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4737 i_data[EXT4_TIND_BLOCK] = 0;
4739 case EXT4_TIND_BLOCK:
4743 up_write(&ei->i_data_sem);
4744 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4745 ext4_mark_inode_dirty(handle, inode);
4748 * In a multi-transaction truncate, we only make the final transaction
4752 ext4_handle_sync(handle);
4755 * If this was a simple ftruncate(), and the file will remain alive
4756 * then we need to clear up the orphan record which we created above.
4757 * However, if this was a real unlink then we were called by
4758 * ext4_delete_inode(), and we allow that function to clean up the
4759 * orphan info for us.
4762 ext4_orphan_del(handle, inode);
4764 ext4_journal_stop(handle);
4768 * ext4_get_inode_loc returns with an extra refcount against the inode's
4769 * underlying buffer_head on success. If 'in_mem' is true, we have all
4770 * data in memory that is needed to recreate the on-disk version of this
4773 static int __ext4_get_inode_loc(struct inode *inode,
4774 struct ext4_iloc *iloc, int in_mem)
4776 struct ext4_group_desc *gdp;
4777 struct buffer_head *bh;
4778 struct super_block *sb = inode->i_sb;
4780 int inodes_per_block, inode_offset;
4783 if (!ext4_valid_inum(sb, inode->i_ino))
4786 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4787 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4792 * Figure out the offset within the block group inode table
4794 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4795 inode_offset = ((inode->i_ino - 1) %
4796 EXT4_INODES_PER_GROUP(sb));
4797 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4798 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4800 bh = sb_getblk(sb, block);
4802 EXT4_ERROR_INODE_BLOCK(inode, block,
4803 "unable to read itable block");
4806 if (!buffer_uptodate(bh)) {
4810 * If the buffer has the write error flag, we have failed
4811 * to write out another inode in the same block. In this
4812 * case, we don't have to read the block because we may
4813 * read the old inode data successfully.
4815 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4816 set_buffer_uptodate(bh);
4818 if (buffer_uptodate(bh)) {
4819 /* someone brought it uptodate while we waited */
4825 * If we have all information of the inode in memory and this
4826 * is the only valid inode in the block, we need not read the
4830 struct buffer_head *bitmap_bh;
4833 start = inode_offset & ~(inodes_per_block - 1);
4835 /* Is the inode bitmap in cache? */
4836 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4841 * If the inode bitmap isn't in cache then the
4842 * optimisation may end up performing two reads instead
4843 * of one, so skip it.
4845 if (!buffer_uptodate(bitmap_bh)) {
4849 for (i = start; i < start + inodes_per_block; i++) {
4850 if (i == inode_offset)
4852 if (ext4_test_bit(i, bitmap_bh->b_data))
4856 if (i == start + inodes_per_block) {
4857 /* all other inodes are free, so skip I/O */
4858 memset(bh->b_data, 0, bh->b_size);
4859 set_buffer_uptodate(bh);
4867 * If we need to do any I/O, try to pre-readahead extra
4868 * blocks from the inode table.
4870 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4871 ext4_fsblk_t b, end, table;
4874 table = ext4_inode_table(sb, gdp);
4875 /* s_inode_readahead_blks is always a power of 2 */
4876 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4879 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4880 num = EXT4_INODES_PER_GROUP(sb);
4881 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4882 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4883 num -= ext4_itable_unused_count(sb, gdp);
4884 table += num / inodes_per_block;
4888 sb_breadahead(sb, b++);
4892 * There are other valid inodes in the buffer, this inode
4893 * has in-inode xattrs, or we don't have this inode in memory.
4894 * Read the block from disk.
4897 bh->b_end_io = end_buffer_read_sync;
4898 submit_bh(READ_META, bh);
4900 if (!buffer_uptodate(bh)) {
4901 EXT4_ERROR_INODE_BLOCK(inode, block,
4902 "unable to read itable block");
4912 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4914 /* We have all inode data except xattrs in memory here. */
4915 return __ext4_get_inode_loc(inode, iloc,
4916 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4919 void ext4_set_inode_flags(struct inode *inode)
4921 unsigned int flags = EXT4_I(inode)->i_flags;
4923 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4924 if (flags & EXT4_SYNC_FL)
4925 inode->i_flags |= S_SYNC;
4926 if (flags & EXT4_APPEND_FL)
4927 inode->i_flags |= S_APPEND;
4928 if (flags & EXT4_IMMUTABLE_FL)
4929 inode->i_flags |= S_IMMUTABLE;
4930 if (flags & EXT4_NOATIME_FL)
4931 inode->i_flags |= S_NOATIME;
4932 if (flags & EXT4_DIRSYNC_FL)
4933 inode->i_flags |= S_DIRSYNC;
4936 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4937 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4939 unsigned int vfs_fl;
4940 unsigned long old_fl, new_fl;
4943 vfs_fl = ei->vfs_inode.i_flags;
4944 old_fl = ei->i_flags;
4945 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4946 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4948 if (vfs_fl & S_SYNC)
4949 new_fl |= EXT4_SYNC_FL;
4950 if (vfs_fl & S_APPEND)
4951 new_fl |= EXT4_APPEND_FL;
4952 if (vfs_fl & S_IMMUTABLE)
4953 new_fl |= EXT4_IMMUTABLE_FL;
4954 if (vfs_fl & S_NOATIME)
4955 new_fl |= EXT4_NOATIME_FL;
4956 if (vfs_fl & S_DIRSYNC)
4957 new_fl |= EXT4_DIRSYNC_FL;
4958 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4961 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4962 struct ext4_inode_info *ei)
4965 struct inode *inode = &(ei->vfs_inode);
4966 struct super_block *sb = inode->i_sb;
4968 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4969 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4970 /* we are using combined 48 bit field */
4971 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4972 le32_to_cpu(raw_inode->i_blocks_lo);
4973 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4974 /* i_blocks represent file system block size */
4975 return i_blocks << (inode->i_blkbits - 9);
4980 return le32_to_cpu(raw_inode->i_blocks_lo);
4984 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4986 struct ext4_iloc iloc;
4987 struct ext4_inode *raw_inode;
4988 struct ext4_inode_info *ei;
4989 struct inode *inode;
4990 journal_t *journal = EXT4_SB(sb)->s_journal;
4994 inode = iget_locked(sb, ino);
4996 return ERR_PTR(-ENOMEM);
4997 if (!(inode->i_state & I_NEW))
5003 ret = __ext4_get_inode_loc(inode, &iloc, 0);
5006 raw_inode = ext4_raw_inode(&iloc);
5007 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5008 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5009 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5010 if (!(test_opt(inode->i_sb, NO_UID32))) {
5011 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5012 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5014 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5016 ei->i_state_flags = 0;
5017 ei->i_dir_start_lookup = 0;
5018 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5019 /* We now have enough fields to check if the inode was active or not.
5020 * This is needed because nfsd might try to access dead inodes
5021 * the test is that same one that e2fsck uses
5022 * NeilBrown 1999oct15
5024 if (inode->i_nlink == 0) {
5025 if (inode->i_mode == 0 ||
5026 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5027 /* this inode is deleted */
5031 /* The only unlinked inodes we let through here have
5032 * valid i_mode and are being read by the orphan
5033 * recovery code: that's fine, we're about to complete
5034 * the process of deleting those. */
5036 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5037 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5038 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5039 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5041 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5042 inode->i_size = ext4_isize(raw_inode);
5043 ei->i_disksize = inode->i_size;
5045 ei->i_reserved_quota = 0;
5047 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5048 ei->i_block_group = iloc.block_group;
5049 ei->i_last_alloc_group = ~0;
5051 * NOTE! The in-memory inode i_data array is in little-endian order
5052 * even on big-endian machines: we do NOT byteswap the block numbers!
5054 for (block = 0; block < EXT4_N_BLOCKS; block++)
5055 ei->i_data[block] = raw_inode->i_block[block];
5056 INIT_LIST_HEAD(&ei->i_orphan);
5059 * Set transaction id's of transactions that have to be committed
5060 * to finish f[data]sync. We set them to currently running transaction
5061 * as we cannot be sure that the inode or some of its metadata isn't
5062 * part of the transaction - the inode could have been reclaimed and
5063 * now it is reread from disk.
5066 transaction_t *transaction;
5069 read_lock(&journal->j_state_lock);
5070 if (journal->j_running_transaction)
5071 transaction = journal->j_running_transaction;
5073 transaction = journal->j_committing_transaction;
5075 tid = transaction->t_tid;
5077 tid = journal->j_commit_sequence;
5078 read_unlock(&journal->j_state_lock);
5079 ei->i_sync_tid = tid;
5080 ei->i_datasync_tid = tid;
5083 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5084 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5085 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5086 EXT4_INODE_SIZE(inode->i_sb)) {
5090 if (ei->i_extra_isize == 0) {
5091 /* The extra space is currently unused. Use it. */
5092 ei->i_extra_isize = sizeof(struct ext4_inode) -
5093 EXT4_GOOD_OLD_INODE_SIZE;
5095 __le32 *magic = (void *)raw_inode +
5096 EXT4_GOOD_OLD_INODE_SIZE +
5098 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5099 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5102 ei->i_extra_isize = 0;
5104 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5105 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5106 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5107 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5109 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5110 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5111 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5113 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5117 if (ei->i_file_acl &&
5118 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5119 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5123 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5124 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5125 (S_ISLNK(inode->i_mode) &&
5126 !ext4_inode_is_fast_symlink(inode)))
5127 /* Validate extent which is part of inode */
5128 ret = ext4_ext_check_inode(inode);
5129 } else 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 block references which are part of inode */
5133 ret = ext4_check_inode_blockref(inode);
5138 if (S_ISREG(inode->i_mode)) {
5139 inode->i_op = &ext4_file_inode_operations;
5140 inode->i_fop = &ext4_file_operations;
5141 ext4_set_aops(inode);
5142 } else if (S_ISDIR(inode->i_mode)) {
5143 inode->i_op = &ext4_dir_inode_operations;
5144 inode->i_fop = &ext4_dir_operations;
5145 } else if (S_ISLNK(inode->i_mode)) {
5146 if (ext4_inode_is_fast_symlink(inode)) {
5147 inode->i_op = &ext4_fast_symlink_inode_operations;
5148 nd_terminate_link(ei->i_data, inode->i_size,
5149 sizeof(ei->i_data) - 1);
5151 inode->i_op = &ext4_symlink_inode_operations;
5152 ext4_set_aops(inode);
5154 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5155 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5156 inode->i_op = &ext4_special_inode_operations;
5157 if (raw_inode->i_block[0])
5158 init_special_inode(inode, inode->i_mode,
5159 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5161 init_special_inode(inode, inode->i_mode,
5162 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5165 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5169 ext4_set_inode_flags(inode);
5170 unlock_new_inode(inode);
5176 return ERR_PTR(ret);
5179 static int ext4_inode_blocks_set(handle_t *handle,
5180 struct ext4_inode *raw_inode,
5181 struct ext4_inode_info *ei)
5183 struct inode *inode = &(ei->vfs_inode);
5184 u64 i_blocks = inode->i_blocks;
5185 struct super_block *sb = inode->i_sb;
5187 if (i_blocks <= ~0U) {
5189 * i_blocks can be represnted in a 32 bit variable
5190 * as multiple of 512 bytes
5192 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5193 raw_inode->i_blocks_high = 0;
5194 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5197 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5200 if (i_blocks <= 0xffffffffffffULL) {
5202 * i_blocks can be represented in a 48 bit variable
5203 * as multiple of 512 bytes
5205 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5206 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5207 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5209 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5210 /* i_block is stored in file system block size */
5211 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5212 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5213 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5219 * Post the struct inode info into an on-disk inode location in the
5220 * buffer-cache. This gobbles the caller's reference to the
5221 * buffer_head in the inode location struct.
5223 * The caller must have write access to iloc->bh.
5225 static int ext4_do_update_inode(handle_t *handle,
5226 struct inode *inode,
5227 struct ext4_iloc *iloc)
5229 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5230 struct ext4_inode_info *ei = EXT4_I(inode);
5231 struct buffer_head *bh = iloc->bh;
5232 int err = 0, rc, block;
5234 /* For fields not not tracking in the in-memory inode,
5235 * initialise them to zero for new inodes. */
5236 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5237 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5239 ext4_get_inode_flags(ei);
5240 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5241 if (!(test_opt(inode->i_sb, NO_UID32))) {
5242 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5243 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5245 * Fix up interoperability with old kernels. Otherwise, old inodes get
5246 * re-used with the upper 16 bits of the uid/gid intact
5249 raw_inode->i_uid_high =
5250 cpu_to_le16(high_16_bits(inode->i_uid));
5251 raw_inode->i_gid_high =
5252 cpu_to_le16(high_16_bits(inode->i_gid));
5254 raw_inode->i_uid_high = 0;
5255 raw_inode->i_gid_high = 0;
5258 raw_inode->i_uid_low =
5259 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5260 raw_inode->i_gid_low =
5261 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5262 raw_inode->i_uid_high = 0;
5263 raw_inode->i_gid_high = 0;
5265 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5267 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5268 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5269 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5270 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5272 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5274 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5275 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5276 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5277 cpu_to_le32(EXT4_OS_HURD))
5278 raw_inode->i_file_acl_high =
5279 cpu_to_le16(ei->i_file_acl >> 32);
5280 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5281 ext4_isize_set(raw_inode, ei->i_disksize);
5282 if (ei->i_disksize > 0x7fffffffULL) {
5283 struct super_block *sb = inode->i_sb;
5284 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5285 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5286 EXT4_SB(sb)->s_es->s_rev_level ==
5287 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5288 /* If this is the first large file
5289 * created, add a flag to the superblock.
5291 err = ext4_journal_get_write_access(handle,
5292 EXT4_SB(sb)->s_sbh);
5295 ext4_update_dynamic_rev(sb);
5296 EXT4_SET_RO_COMPAT_FEATURE(sb,
5297 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5299 ext4_handle_sync(handle);
5300 err = ext4_handle_dirty_metadata(handle, NULL,
5301 EXT4_SB(sb)->s_sbh);
5304 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5305 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5306 if (old_valid_dev(inode->i_rdev)) {
5307 raw_inode->i_block[0] =
5308 cpu_to_le32(old_encode_dev(inode->i_rdev));
5309 raw_inode->i_block[1] = 0;
5311 raw_inode->i_block[0] = 0;
5312 raw_inode->i_block[1] =
5313 cpu_to_le32(new_encode_dev(inode->i_rdev));
5314 raw_inode->i_block[2] = 0;
5317 for (block = 0; block < EXT4_N_BLOCKS; block++)
5318 raw_inode->i_block[block] = ei->i_data[block];
5320 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5321 if (ei->i_extra_isize) {
5322 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5323 raw_inode->i_version_hi =
5324 cpu_to_le32(inode->i_version >> 32);
5325 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5328 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5329 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5332 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5334 ext4_update_inode_fsync_trans(handle, inode, 0);
5337 ext4_std_error(inode->i_sb, err);
5342 * ext4_write_inode()
5344 * We are called from a few places:
5346 * - Within generic_file_write() for O_SYNC files.
5347 * Here, there will be no transaction running. We wait for any running
5348 * trasnaction to commit.
5350 * - Within sys_sync(), kupdate and such.
5351 * We wait on commit, if tol to.
5353 * - Within prune_icache() (PF_MEMALLOC == true)
5354 * Here we simply return. We can't afford to block kswapd on the
5357 * In all cases it is actually safe for us to return without doing anything,
5358 * because the inode has been copied into a raw inode buffer in
5359 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5362 * Note that we are absolutely dependent upon all inode dirtiers doing the
5363 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5364 * which we are interested.
5366 * It would be a bug for them to not do this. The code:
5368 * mark_inode_dirty(inode)
5370 * inode->i_size = expr;
5372 * is in error because a kswapd-driven write_inode() could occur while
5373 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5374 * will no longer be on the superblock's dirty inode list.
5376 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5380 if (current->flags & PF_MEMALLOC)
5383 if (EXT4_SB(inode->i_sb)->s_journal) {
5384 if (ext4_journal_current_handle()) {
5385 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5390 if (wbc->sync_mode != WB_SYNC_ALL)
5393 err = ext4_force_commit(inode->i_sb);
5395 struct ext4_iloc iloc;
5397 err = __ext4_get_inode_loc(inode, &iloc, 0);
5400 if (wbc->sync_mode == WB_SYNC_ALL)
5401 sync_dirty_buffer(iloc.bh);
5402 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5403 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5404 "IO error syncing inode");
5415 * Called from notify_change.
5417 * We want to trap VFS attempts to truncate the file as soon as
5418 * possible. In particular, we want to make sure that when the VFS
5419 * shrinks i_size, we put the inode on the orphan list and modify
5420 * i_disksize immediately, so that during the subsequent flushing of
5421 * dirty pages and freeing of disk blocks, we can guarantee that any
5422 * commit will leave the blocks being flushed in an unused state on
5423 * disk. (On recovery, the inode will get truncated and the blocks will
5424 * be freed, so we have a strong guarantee that no future commit will
5425 * leave these blocks visible to the user.)
5427 * Another thing we have to assure is that if we are in ordered mode
5428 * and inode is still attached to the committing transaction, we must
5429 * we start writeout of all the dirty pages which are being truncated.
5430 * This way we are sure that all the data written in the previous
5431 * transaction are already on disk (truncate waits for pages under
5434 * Called with inode->i_mutex down.
5436 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5438 struct inode *inode = dentry->d_inode;
5440 const unsigned int ia_valid = attr->ia_valid;
5442 error = inode_change_ok(inode, attr);
5446 if (is_quota_modification(inode, attr))
5447 dquot_initialize(inode);
5448 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5449 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5452 /* (user+group)*(old+new) structure, inode write (sb,
5453 * inode block, ? - but truncate inode update has it) */
5454 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5455 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5456 if (IS_ERR(handle)) {
5457 error = PTR_ERR(handle);
5460 error = dquot_transfer(inode, attr);
5462 ext4_journal_stop(handle);
5465 /* Update corresponding info in inode so that everything is in
5466 * one transaction */
5467 if (attr->ia_valid & ATTR_UID)
5468 inode->i_uid = attr->ia_uid;
5469 if (attr->ia_valid & ATTR_GID)
5470 inode->i_gid = attr->ia_gid;
5471 error = ext4_mark_inode_dirty(handle, inode);
5472 ext4_journal_stop(handle);
5475 if (attr->ia_valid & ATTR_SIZE) {
5476 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5477 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5479 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5484 if (S_ISREG(inode->i_mode) &&
5485 attr->ia_valid & ATTR_SIZE &&
5486 (attr->ia_size < inode->i_size ||
5487 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5490 handle = ext4_journal_start(inode, 3);
5491 if (IS_ERR(handle)) {
5492 error = PTR_ERR(handle);
5496 error = ext4_orphan_add(handle, inode);
5497 EXT4_I(inode)->i_disksize = attr->ia_size;
5498 rc = ext4_mark_inode_dirty(handle, inode);
5501 ext4_journal_stop(handle);
5503 if (ext4_should_order_data(inode)) {
5504 error = ext4_begin_ordered_truncate(inode,
5507 /* Do as much error cleanup as possible */
5508 handle = ext4_journal_start(inode, 3);
5509 if (IS_ERR(handle)) {
5510 ext4_orphan_del(NULL, inode);
5513 ext4_orphan_del(handle, inode);
5514 ext4_journal_stop(handle);
5518 /* ext4_truncate will clear the flag */
5519 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5520 ext4_truncate(inode);
5523 rc = inode_setattr(inode, attr);
5525 /* If inode_setattr's call to ext4_truncate failed to get a
5526 * transaction handle at all, we need to clean up the in-core
5527 * orphan list manually. */
5529 ext4_orphan_del(NULL, inode);
5531 if (!rc && (ia_valid & ATTR_MODE))
5532 rc = ext4_acl_chmod(inode);
5535 ext4_std_error(inode->i_sb, error);
5541 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5544 struct inode *inode;
5545 unsigned long delalloc_blocks;
5547 inode = dentry->d_inode;
5548 generic_fillattr(inode, stat);
5551 * We can't update i_blocks if the block allocation is delayed
5552 * otherwise in the case of system crash before the real block
5553 * allocation is done, we will have i_blocks inconsistent with
5554 * on-disk file blocks.
5555 * We always keep i_blocks updated together with real
5556 * allocation. But to not confuse with user, stat
5557 * will return the blocks that include the delayed allocation
5558 * blocks for this file.
5560 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5561 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5562 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5564 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5568 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5573 /* if nrblocks are contiguous */
5576 * With N contiguous data blocks, it need at most
5577 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5578 * 2 dindirect blocks
5581 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5582 return indirects + 3;
5585 * if nrblocks are not contiguous, worse case, each block touch
5586 * a indirect block, and each indirect block touch a double indirect
5587 * block, plus a triple indirect block
5589 indirects = nrblocks * 2 + 1;
5593 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5595 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5596 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5597 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5601 * Account for index blocks, block groups bitmaps and block group
5602 * descriptor blocks if modify datablocks and index blocks
5603 * worse case, the indexs blocks spread over different block groups
5605 * If datablocks are discontiguous, they are possible to spread over
5606 * different block groups too. If they are contiuguous, with flexbg,
5607 * they could still across block group boundary.
5609 * Also account for superblock, inode, quota and xattr blocks
5611 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5613 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5619 * How many index blocks need to touch to modify nrblocks?
5620 * The "Chunk" flag indicating whether the nrblocks is
5621 * physically contiguous on disk
5623 * For Direct IO and fallocate, they calls get_block to allocate
5624 * one single extent at a time, so they could set the "Chunk" flag
5626 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5631 * Now let's see how many group bitmaps and group descriptors need
5641 if (groups > ngroups)
5643 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5644 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5646 /* bitmaps and block group descriptor blocks */
5647 ret += groups + gdpblocks;
5649 /* Blocks for super block, inode, quota and xattr blocks */
5650 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5656 * Calulate the total number of credits to reserve to fit
5657 * the modification of a single pages into a single transaction,
5658 * which may include multiple chunks of block allocations.
5660 * This could be called via ext4_write_begin()
5662 * We need to consider the worse case, when
5663 * one new block per extent.
5665 int ext4_writepage_trans_blocks(struct inode *inode)
5667 int bpp = ext4_journal_blocks_per_page(inode);
5670 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5672 /* Account for data blocks for journalled mode */
5673 if (ext4_should_journal_data(inode))
5679 * Calculate the journal credits for a chunk of data modification.
5681 * This is called from DIO, fallocate or whoever calling
5682 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5684 * journal buffers for data blocks are not included here, as DIO
5685 * and fallocate do no need to journal data buffers.
5687 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5689 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5693 * The caller must have previously called ext4_reserve_inode_write().
5694 * Give this, we know that the caller already has write access to iloc->bh.
5696 int ext4_mark_iloc_dirty(handle_t *handle,
5697 struct inode *inode, struct ext4_iloc *iloc)
5701 if (test_opt(inode->i_sb, I_VERSION))
5702 inode_inc_iversion(inode);
5704 /* the do_update_inode consumes one bh->b_count */
5707 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5708 err = ext4_do_update_inode(handle, inode, iloc);
5714 * On success, We end up with an outstanding reference count against
5715 * iloc->bh. This _must_ be cleaned up later.
5719 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5720 struct ext4_iloc *iloc)
5724 err = ext4_get_inode_loc(inode, iloc);
5726 BUFFER_TRACE(iloc->bh, "get_write_access");
5727 err = ext4_journal_get_write_access(handle, iloc->bh);
5733 ext4_std_error(inode->i_sb, err);
5738 * Expand an inode by new_extra_isize bytes.
5739 * Returns 0 on success or negative error number on failure.
5741 static int ext4_expand_extra_isize(struct inode *inode,
5742 unsigned int new_extra_isize,
5743 struct ext4_iloc iloc,
5746 struct ext4_inode *raw_inode;
5747 struct ext4_xattr_ibody_header *header;
5749 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5752 raw_inode = ext4_raw_inode(&iloc);
5754 header = IHDR(inode, raw_inode);
5756 /* No extended attributes present */
5757 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5758 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5759 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5761 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5765 /* try to expand with EAs present */
5766 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5771 * What we do here is to mark the in-core inode as clean with respect to inode
5772 * dirtiness (it may still be data-dirty).
5773 * This means that the in-core inode may be reaped by prune_icache
5774 * without having to perform any I/O. This is a very good thing,
5775 * because *any* task may call prune_icache - even ones which
5776 * have a transaction open against a different journal.
5778 * Is this cheating? Not really. Sure, we haven't written the
5779 * inode out, but prune_icache isn't a user-visible syncing function.
5780 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5781 * we start and wait on commits.
5783 * Is this efficient/effective? Well, we're being nice to the system
5784 * by cleaning up our inodes proactively so they can be reaped
5785 * without I/O. But we are potentially leaving up to five seconds'
5786 * worth of inodes floating about which prune_icache wants us to
5787 * write out. One way to fix that would be to get prune_icache()
5788 * to do a write_super() to free up some memory. It has the desired
5791 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5793 struct ext4_iloc iloc;
5794 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5795 static unsigned int mnt_count;
5799 err = ext4_reserve_inode_write(handle, inode, &iloc);
5800 if (ext4_handle_valid(handle) &&
5801 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5802 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5804 * We need extra buffer credits since we may write into EA block
5805 * with this same handle. If journal_extend fails, then it will
5806 * only result in a minor loss of functionality for that inode.
5807 * If this is felt to be critical, then e2fsck should be run to
5808 * force a large enough s_min_extra_isize.
5810 if ((jbd2_journal_extend(handle,
5811 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5812 ret = ext4_expand_extra_isize(inode,
5813 sbi->s_want_extra_isize,
5816 ext4_set_inode_state(inode,
5817 EXT4_STATE_NO_EXPAND);
5819 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5820 ext4_warning(inode->i_sb,
5821 "Unable to expand inode %lu. Delete"
5822 " some EAs or run e2fsck.",
5825 le16_to_cpu(sbi->s_es->s_mnt_count);
5831 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5836 * ext4_dirty_inode() is called from __mark_inode_dirty()
5838 * We're really interested in the case where a file is being extended.
5839 * i_size has been changed by generic_commit_write() and we thus need
5840 * to include the updated inode in the current transaction.
5842 * Also, dquot_alloc_block() will always dirty the inode when blocks
5843 * are allocated to the file.
5845 * If the inode is marked synchronous, we don't honour that here - doing
5846 * so would cause a commit on atime updates, which we don't bother doing.
5847 * We handle synchronous inodes at the highest possible level.
5849 void ext4_dirty_inode(struct inode *inode)
5853 handle = ext4_journal_start(inode, 2);
5857 ext4_mark_inode_dirty(handle, inode);
5859 ext4_journal_stop(handle);
5866 * Bind an inode's backing buffer_head into this transaction, to prevent
5867 * it from being flushed to disk early. Unlike
5868 * ext4_reserve_inode_write, this leaves behind no bh reference and
5869 * returns no iloc structure, so the caller needs to repeat the iloc
5870 * lookup to mark the inode dirty later.
5872 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5874 struct ext4_iloc iloc;
5878 err = ext4_get_inode_loc(inode, &iloc);
5880 BUFFER_TRACE(iloc.bh, "get_write_access");
5881 err = jbd2_journal_get_write_access(handle, iloc.bh);
5883 err = ext4_handle_dirty_metadata(handle,
5889 ext4_std_error(inode->i_sb, err);
5894 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5901 * We have to be very careful here: changing a data block's
5902 * journaling status dynamically is dangerous. If we write a
5903 * data block to the journal, change the status and then delete
5904 * that block, we risk forgetting to revoke the old log record
5905 * from the journal and so a subsequent replay can corrupt data.
5906 * So, first we make sure that the journal is empty and that
5907 * nobody is changing anything.
5910 journal = EXT4_JOURNAL(inode);
5913 if (is_journal_aborted(journal))
5916 jbd2_journal_lock_updates(journal);
5917 jbd2_journal_flush(journal);
5920 * OK, there are no updates running now, and all cached data is
5921 * synced to disk. We are now in a completely consistent state
5922 * which doesn't have anything in the journal, and we know that
5923 * no filesystem updates are running, so it is safe to modify
5924 * the inode's in-core data-journaling state flag now.
5928 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5930 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5931 ext4_set_aops(inode);
5933 jbd2_journal_unlock_updates(journal);
5935 /* Finally we can mark the inode as dirty. */
5937 handle = ext4_journal_start(inode, 1);
5939 return PTR_ERR(handle);
5941 err = ext4_mark_inode_dirty(handle, inode);
5942 ext4_handle_sync(handle);
5943 ext4_journal_stop(handle);
5944 ext4_std_error(inode->i_sb, err);
5949 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5951 return !buffer_mapped(bh);
5954 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5956 struct page *page = vmf->page;
5961 struct file *file = vma->vm_file;
5962 struct inode *inode = file->f_path.dentry->d_inode;
5963 struct address_space *mapping = inode->i_mapping;
5966 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5967 * get i_mutex because we are already holding mmap_sem.
5969 down_read(&inode->i_alloc_sem);
5970 size = i_size_read(inode);
5971 if (page->mapping != mapping || size <= page_offset(page)
5972 || !PageUptodate(page)) {
5973 /* page got truncated from under us? */
5977 if (PageMappedToDisk(page))
5980 if (page->index == size >> PAGE_CACHE_SHIFT)
5981 len = size & ~PAGE_CACHE_MASK;
5983 len = PAGE_CACHE_SIZE;
5987 * return if we have all the buffers mapped. This avoid
5988 * the need to call write_begin/write_end which does a
5989 * journal_start/journal_stop which can block and take
5992 if (page_has_buffers(page)) {
5993 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5994 ext4_bh_unmapped)) {
6001 * OK, we need to fill the hole... Do write_begin write_end
6002 * to do block allocation/reservation.We are not holding
6003 * inode.i__mutex here. That allow * parallel write_begin,
6004 * write_end call. lock_page prevent this from happening
6005 * on the same page though
6007 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6008 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6011 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6012 len, len, page, fsdata);
6018 ret = VM_FAULT_SIGBUS;
6019 up_read(&inode->i_alloc_sem);