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>
42 #include "ext4_jbd2.h"
45 #include "ext4_extents.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static inline int ext4_begin_ordered_truncate(struct inode *inode,
54 return jbd2_journal_begin_ordered_truncate(
55 EXT4_SB(inode->i_sb)->s_journal,
56 &EXT4_I(inode)->jinode,
60 static void ext4_invalidatepage(struct page *page, unsigned long offset);
63 * Test whether an inode is a fast symlink.
65 static int ext4_inode_is_fast_symlink(struct inode *inode)
67 int ea_blocks = EXT4_I(inode)->i_file_acl ?
68 (inode->i_sb->s_blocksize >> 9) : 0;
70 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
74 * The ext4 forget function must perform a revoke if we are freeing data
75 * which has been journaled. Metadata (eg. indirect blocks) must be
76 * revoked in all cases.
78 * "bh" may be NULL: a metadata block may have been freed from memory
79 * but there may still be a record of it in the journal, and that record
80 * still needs to be revoked.
82 * If the handle isn't valid we're not journaling, but we still need to
83 * call into ext4_journal_revoke() to put the buffer head.
85 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
86 struct buffer_head *bh, ext4_fsblk_t blocknr)
92 BUFFER_TRACE(bh, "enter");
94 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
96 bh, is_metadata, inode->i_mode,
97 test_opt(inode->i_sb, DATA_FLAGS));
99 /* Never use the revoke function if we are doing full data
100 * journaling: there is no need to, and a V1 superblock won't
101 * support it. Otherwise, only skip the revoke on un-journaled
104 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
105 (!is_metadata && !ext4_should_journal_data(inode))) {
107 BUFFER_TRACE(bh, "call jbd2_journal_forget");
108 return ext4_journal_forget(handle, bh);
114 * data!=journal && (is_metadata || should_journal_data(inode))
116 BUFFER_TRACE(bh, "call ext4_journal_revoke");
117 err = ext4_journal_revoke(handle, blocknr, bh);
119 ext4_abort(inode->i_sb, __func__,
120 "error %d when attempting revoke", err);
121 BUFFER_TRACE(bh, "exit");
126 * Work out how many blocks we need to proceed with the next chunk of a
127 * truncate transaction.
129 static unsigned long blocks_for_truncate(struct inode *inode)
133 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
135 /* Give ourselves just enough room to cope with inodes in which
136 * i_blocks is corrupt: we've seen disk corruptions in the past
137 * which resulted in random data in an inode which looked enough
138 * like a regular file for ext4 to try to delete it. Things
139 * will go a bit crazy if that happens, but at least we should
140 * try not to panic the whole kernel. */
144 /* But we need to bound the transaction so we don't overflow the
146 if (needed > EXT4_MAX_TRANS_DATA)
147 needed = EXT4_MAX_TRANS_DATA;
149 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
153 * Truncate transactions can be complex and absolutely huge. So we need to
154 * be able to restart the transaction at a conventient checkpoint to make
155 * sure we don't overflow the journal.
157 * start_transaction gets us a new handle for a truncate transaction,
158 * and extend_transaction tries to extend the existing one a bit. If
159 * extend fails, we need to propagate the failure up and restart the
160 * transaction in the top-level truncate loop. --sct
162 static handle_t *start_transaction(struct inode *inode)
166 result = ext4_journal_start(inode, blocks_for_truncate(inode));
170 ext4_std_error(inode->i_sb, PTR_ERR(result));
175 * Try to extend this transaction for the purposes of truncation.
177 * Returns 0 if we managed to create more room. If we can't create more
178 * room, and the transaction must be restarted we return 1.
180 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
182 if (!ext4_handle_valid(handle))
184 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
186 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
192 * Restart the transaction associated with *handle. This does a commit,
193 * so before we call here everything must be consistently dirtied against
196 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
202 * Drop i_data_sem to avoid deadlock with ext4_get_blocks At this
203 * moment, get_block can be called only for blocks inside i_size since
204 * page cache has been already dropped and writes are blocked by
205 * i_mutex. So we can safely drop the i_data_sem here.
207 BUG_ON(EXT4_JOURNAL(inode) == NULL);
208 jbd_debug(2, "restarting handle %p\n", handle);
209 up_write(&EXT4_I(inode)->i_data_sem);
210 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
211 down_write(&EXT4_I(inode)->i_data_sem);
212 ext4_discard_preallocations(inode);
218 * Called at the last iput() if i_nlink is zero.
220 void ext4_delete_inode(struct inode *inode)
225 if (ext4_should_order_data(inode))
226 ext4_begin_ordered_truncate(inode, 0);
227 truncate_inode_pages(&inode->i_data, 0);
229 if (is_bad_inode(inode))
232 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
233 if (IS_ERR(handle)) {
234 ext4_std_error(inode->i_sb, PTR_ERR(handle));
236 * If we're going to skip the normal cleanup, we still need to
237 * make sure that the in-core orphan linked list is properly
240 ext4_orphan_del(NULL, inode);
245 ext4_handle_sync(handle);
247 err = ext4_mark_inode_dirty(handle, inode);
249 ext4_warning(inode->i_sb, __func__,
250 "couldn't mark inode dirty (err %d)", err);
254 ext4_truncate(inode);
257 * ext4_ext_truncate() doesn't reserve any slop when it
258 * restarts journal transactions; therefore there may not be
259 * enough credits left in the handle to remove the inode from
260 * the orphan list and set the dtime field.
262 if (!ext4_handle_has_enough_credits(handle, 3)) {
263 err = ext4_journal_extend(handle, 3);
265 err = ext4_journal_restart(handle, 3);
267 ext4_warning(inode->i_sb, __func__,
268 "couldn't extend journal (err %d)", err);
270 ext4_journal_stop(handle);
276 * Kill off the orphan record which ext4_truncate created.
277 * AKPM: I think this can be inside the above `if'.
278 * Note that ext4_orphan_del() has to be able to cope with the
279 * deletion of a non-existent orphan - this is because we don't
280 * know if ext4_truncate() actually created an orphan record.
281 * (Well, we could do this if we need to, but heck - it works)
283 ext4_orphan_del(handle, inode);
284 EXT4_I(inode)->i_dtime = get_seconds();
287 * One subtle ordering requirement: if anything has gone wrong
288 * (transaction abort, IO errors, whatever), then we can still
289 * do these next steps (the fs will already have been marked as
290 * having errors), but we can't free the inode if the mark_dirty
293 if (ext4_mark_inode_dirty(handle, inode))
294 /* If that failed, just do the required in-core inode clear. */
297 ext4_free_inode(handle, inode);
298 ext4_journal_stop(handle);
301 clear_inode(inode); /* We must guarantee clearing of inode... */
307 struct buffer_head *bh;
310 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
312 p->key = *(p->p = v);
317 * ext4_block_to_path - parse the block number into array of offsets
318 * @inode: inode in question (we are only interested in its superblock)
319 * @i_block: block number to be parsed
320 * @offsets: array to store the offsets in
321 * @boundary: set this non-zero if the referred-to block is likely to be
322 * followed (on disk) by an indirect block.
324 * To store the locations of file's data ext4 uses a data structure common
325 * for UNIX filesystems - tree of pointers anchored in the inode, with
326 * data blocks at leaves and indirect blocks in intermediate nodes.
327 * This function translates the block number into path in that tree -
328 * return value is the path length and @offsets[n] is the offset of
329 * pointer to (n+1)th node in the nth one. If @block is out of range
330 * (negative or too large) warning is printed and zero returned.
332 * Note: function doesn't find node addresses, so no IO is needed. All
333 * we need to know is the capacity of indirect blocks (taken from the
338 * Portability note: the last comparison (check that we fit into triple
339 * indirect block) is spelled differently, because otherwise on an
340 * architecture with 32-bit longs and 8Kb pages we might get into trouble
341 * if our filesystem had 8Kb blocks. We might use long long, but that would
342 * kill us on x86. Oh, well, at least the sign propagation does not matter -
343 * i_block would have to be negative in the very beginning, so we would not
347 static int ext4_block_to_path(struct inode *inode,
349 ext4_lblk_t offsets[4], int *boundary)
351 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
352 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
353 const long direct_blocks = EXT4_NDIR_BLOCKS,
354 indirect_blocks = ptrs,
355 double_blocks = (1 << (ptrs_bits * 2));
359 if (i_block < direct_blocks) {
360 offsets[n++] = i_block;
361 final = direct_blocks;
362 } else if ((i_block -= direct_blocks) < indirect_blocks) {
363 offsets[n++] = EXT4_IND_BLOCK;
364 offsets[n++] = i_block;
366 } else if ((i_block -= indirect_blocks) < double_blocks) {
367 offsets[n++] = EXT4_DIND_BLOCK;
368 offsets[n++] = i_block >> ptrs_bits;
369 offsets[n++] = i_block & (ptrs - 1);
371 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
372 offsets[n++] = EXT4_TIND_BLOCK;
373 offsets[n++] = i_block >> (ptrs_bits * 2);
374 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
375 offsets[n++] = i_block & (ptrs - 1);
378 ext4_warning(inode->i_sb, "ext4_block_to_path",
379 "block %lu > max in inode %lu",
380 i_block + direct_blocks +
381 indirect_blocks + double_blocks, inode->i_ino);
384 *boundary = final - 1 - (i_block & (ptrs - 1));
388 static int __ext4_check_blockref(const char *function, struct inode *inode,
389 __le32 *p, unsigned int max)
394 while (bref < p+max) {
395 blk = le32_to_cpu(*bref++);
397 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
399 ext4_error(inode->i_sb, function,
400 "invalid block reference %u "
401 "in inode #%lu", blk, inode->i_ino);
409 #define ext4_check_indirect_blockref(inode, bh) \
410 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
411 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
413 #define ext4_check_inode_blockref(inode) \
414 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
418 * ext4_get_branch - read the chain of indirect blocks leading to data
419 * @inode: inode in question
420 * @depth: depth of the chain (1 - direct pointer, etc.)
421 * @offsets: offsets of pointers in inode/indirect blocks
422 * @chain: place to store the result
423 * @err: here we store the error value
425 * Function fills the array of triples <key, p, bh> and returns %NULL
426 * if everything went OK or the pointer to the last filled triple
427 * (incomplete one) otherwise. Upon the return chain[i].key contains
428 * the number of (i+1)-th block in the chain (as it is stored in memory,
429 * i.e. little-endian 32-bit), chain[i].p contains the address of that
430 * number (it points into struct inode for i==0 and into the bh->b_data
431 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
432 * block for i>0 and NULL for i==0. In other words, it holds the block
433 * numbers of the chain, addresses they were taken from (and where we can
434 * verify that chain did not change) and buffer_heads hosting these
437 * Function stops when it stumbles upon zero pointer (absent block)
438 * (pointer to last triple returned, *@err == 0)
439 * or when it gets an IO error reading an indirect block
440 * (ditto, *@err == -EIO)
441 * or when it reads all @depth-1 indirect blocks successfully and finds
442 * the whole chain, all way to the data (returns %NULL, *err == 0).
444 * Need to be called with
445 * down_read(&EXT4_I(inode)->i_data_sem)
447 static Indirect *ext4_get_branch(struct inode *inode, int depth,
448 ext4_lblk_t *offsets,
449 Indirect chain[4], int *err)
451 struct super_block *sb = inode->i_sb;
453 struct buffer_head *bh;
456 /* i_data is not going away, no lock needed */
457 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
461 bh = sb_getblk(sb, le32_to_cpu(p->key));
465 if (!bh_uptodate_or_lock(bh)) {
466 if (bh_submit_read(bh) < 0) {
470 /* validate block references */
471 if (ext4_check_indirect_blockref(inode, bh)) {
477 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
491 * ext4_find_near - find a place for allocation with sufficient locality
493 * @ind: descriptor of indirect block.
495 * This function returns the preferred place for block allocation.
496 * It is used when heuristic for sequential allocation fails.
498 * + if there is a block to the left of our position - allocate near it.
499 * + if pointer will live in indirect block - allocate near that block.
500 * + if pointer will live in inode - allocate in the same
503 * In the latter case we colour the starting block by the callers PID to
504 * prevent it from clashing with concurrent allocations for a different inode
505 * in the same block group. The PID is used here so that functionally related
506 * files will be close-by on-disk.
508 * Caller must make sure that @ind is valid and will stay that way.
510 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
512 struct ext4_inode_info *ei = EXT4_I(inode);
513 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
515 ext4_fsblk_t bg_start;
516 ext4_fsblk_t last_block;
517 ext4_grpblk_t colour;
518 ext4_group_t block_group;
519 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
521 /* Try to find previous block */
522 for (p = ind->p - 1; p >= start; p--) {
524 return le32_to_cpu(*p);
527 /* No such thing, so let's try location of indirect block */
529 return ind->bh->b_blocknr;
532 * It is going to be referred to from the inode itself? OK, just put it
533 * into the same cylinder group then.
535 block_group = ei->i_block_group;
536 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
537 block_group &= ~(flex_size-1);
538 if (S_ISREG(inode->i_mode))
541 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
542 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
545 * If we are doing delayed allocation, we don't need take
546 * colour into account.
548 if (test_opt(inode->i_sb, DELALLOC))
551 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
552 colour = (current->pid % 16) *
553 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
555 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
556 return bg_start + colour;
560 * ext4_find_goal - find a preferred place for allocation.
562 * @block: block we want
563 * @partial: pointer to the last triple within a chain
565 * Normally this function find the preferred place for block allocation,
567 * Because this is only used for non-extent files, we limit the block nr
570 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
576 * XXX need to get goal block from mballoc's data structures
579 goal = ext4_find_near(inode, partial);
580 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
585 * ext4_blks_to_allocate: Look up the block map and count the number
586 * of direct blocks need to be allocated for the given branch.
588 * @branch: chain of indirect blocks
589 * @k: number of blocks need for indirect blocks
590 * @blks: number of data blocks to be mapped.
591 * @blocks_to_boundary: the offset in the indirect block
593 * return the total number of blocks to be allocate, including the
594 * direct and indirect blocks.
596 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
597 int blocks_to_boundary)
599 unsigned int count = 0;
602 * Simple case, [t,d]Indirect block(s) has not allocated yet
603 * then it's clear blocks on that path have not allocated
606 /* right now we don't handle cross boundary allocation */
607 if (blks < blocks_to_boundary + 1)
610 count += blocks_to_boundary + 1;
615 while (count < blks && count <= blocks_to_boundary &&
616 le32_to_cpu(*(branch[0].p + count)) == 0) {
623 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
624 * @indirect_blks: the number of blocks need to allocate for indirect
627 * @new_blocks: on return it will store the new block numbers for
628 * the indirect blocks(if needed) and the first direct block,
629 * @blks: on return it will store the total number of allocated
632 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
633 ext4_lblk_t iblock, ext4_fsblk_t goal,
634 int indirect_blks, int blks,
635 ext4_fsblk_t new_blocks[4], int *err)
637 struct ext4_allocation_request ar;
639 unsigned long count = 0, blk_allocated = 0;
641 ext4_fsblk_t current_block = 0;
645 * Here we try to allocate the requested multiple blocks at once,
646 * on a best-effort basis.
647 * To build a branch, we should allocate blocks for
648 * the indirect blocks(if not allocated yet), and at least
649 * the first direct block of this branch. That's the
650 * minimum number of blocks need to allocate(required)
652 /* first we try to allocate the indirect blocks */
653 target = indirect_blks;
656 /* allocating blocks for indirect blocks and direct blocks */
657 current_block = ext4_new_meta_blocks(handle, inode,
662 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
665 /* allocate blocks for indirect blocks */
666 while (index < indirect_blks && count) {
667 new_blocks[index++] = current_block++;
672 * save the new block number
673 * for the first direct block
675 new_blocks[index] = current_block;
676 printk(KERN_INFO "%s returned more blocks than "
677 "requested\n", __func__);
683 target = blks - count ;
684 blk_allocated = count;
687 /* Now allocate data blocks */
688 memset(&ar, 0, sizeof(ar));
693 if (S_ISREG(inode->i_mode))
694 /* enable in-core preallocation only for regular files */
695 ar.flags = EXT4_MB_HINT_DATA;
697 current_block = ext4_mb_new_blocks(handle, &ar, err);
698 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
700 if (*err && (target == blks)) {
702 * if the allocation failed and we didn't allocate
708 if (target == blks) {
710 * save the new block number
711 * for the first direct block
713 new_blocks[index] = current_block;
715 blk_allocated += ar.len;
718 /* total number of blocks allocated for direct blocks */
723 for (i = 0; i < index; i++)
724 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
729 * ext4_alloc_branch - allocate and set up a chain of blocks.
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @offsets: offsets (in the blocks) to store the pointers to next.
734 * @branch: place to store the chain in.
736 * This function allocates blocks, zeroes out all but the last one,
737 * links them into chain and (if we are synchronous) writes them to disk.
738 * In other words, it prepares a branch that can be spliced onto the
739 * inode. It stores the information about that chain in the branch[], in
740 * the same format as ext4_get_branch() would do. We are calling it after
741 * we had read the existing part of chain and partial points to the last
742 * triple of that (one with zero ->key). Upon the exit we have the same
743 * picture as after the successful ext4_get_block(), except that in one
744 * place chain is disconnected - *branch->p is still zero (we did not
745 * set the last link), but branch->key contains the number that should
746 * be placed into *branch->p to fill that gap.
748 * If allocation fails we free all blocks we've allocated (and forget
749 * their buffer_heads) and return the error value the from failed
750 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
751 * as described above and return 0.
753 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
754 ext4_lblk_t iblock, int indirect_blks,
755 int *blks, ext4_fsblk_t goal,
756 ext4_lblk_t *offsets, Indirect *branch)
758 int blocksize = inode->i_sb->s_blocksize;
761 struct buffer_head *bh;
763 ext4_fsblk_t new_blocks[4];
764 ext4_fsblk_t current_block;
766 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
767 *blks, new_blocks, &err);
771 branch[0].key = cpu_to_le32(new_blocks[0]);
773 * metadata blocks and data blocks are allocated.
775 for (n = 1; n <= indirect_blks; n++) {
777 * Get buffer_head for parent block, zero it out
778 * and set the pointer to new one, then send
781 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
784 BUFFER_TRACE(bh, "call get_create_access");
785 err = ext4_journal_get_create_access(handle, bh);
787 /* Don't brelse(bh) here; it's done in
788 * ext4_journal_forget() below */
793 memset(bh->b_data, 0, blocksize);
794 branch[n].p = (__le32 *) bh->b_data + offsets[n];
795 branch[n].key = cpu_to_le32(new_blocks[n]);
796 *branch[n].p = branch[n].key;
797 if (n == indirect_blks) {
798 current_block = new_blocks[n];
800 * End of chain, update the last new metablock of
801 * the chain to point to the new allocated
802 * data blocks numbers
804 for (i = 1; i < num; i++)
805 *(branch[n].p + i) = cpu_to_le32(++current_block);
807 BUFFER_TRACE(bh, "marking uptodate");
808 set_buffer_uptodate(bh);
811 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
812 err = ext4_handle_dirty_metadata(handle, inode, bh);
819 /* Allocation failed, free what we already allocated */
820 for (i = 1; i <= n ; i++) {
821 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
822 ext4_journal_forget(handle, branch[i].bh);
824 for (i = 0; i < indirect_blks; i++)
825 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
827 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
833 * ext4_splice_branch - splice the allocated branch onto inode.
835 * @block: (logical) number of block we are adding
836 * @chain: chain of indirect blocks (with a missing link - see
838 * @where: location of missing link
839 * @num: number of indirect blocks we are adding
840 * @blks: number of direct blocks we are adding
842 * This function fills the missing link and does all housekeeping needed in
843 * inode (->i_blocks, etc.). In case of success we end up with the full
844 * chain to new block and return 0.
846 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
847 ext4_lblk_t block, Indirect *where, int num,
852 ext4_fsblk_t current_block;
855 * If we're splicing into a [td]indirect block (as opposed to the
856 * inode) then we need to get write access to the [td]indirect block
860 BUFFER_TRACE(where->bh, "get_write_access");
861 err = ext4_journal_get_write_access(handle, where->bh);
867 *where->p = where->key;
870 * Update the host buffer_head or inode to point to more just allocated
871 * direct blocks blocks
873 if (num == 0 && blks > 1) {
874 current_block = le32_to_cpu(where->key) + 1;
875 for (i = 1; i < blks; i++)
876 *(where->p + i) = cpu_to_le32(current_block++);
879 /* We are done with atomic stuff, now do the rest of housekeeping */
880 /* had we spliced it onto indirect block? */
883 * If we spliced it onto an indirect block, we haven't
884 * altered the inode. Note however that if it is being spliced
885 * onto an indirect block at the very end of the file (the
886 * file is growing) then we *will* alter the inode to reflect
887 * the new i_size. But that is not done here - it is done in
888 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
890 jbd_debug(5, "splicing indirect only\n");
891 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
892 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
897 * OK, we spliced it into the inode itself on a direct block.
899 ext4_mark_inode_dirty(handle, inode);
900 jbd_debug(5, "splicing direct\n");
905 for (i = 1; i <= num; i++) {
906 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
907 ext4_journal_forget(handle, where[i].bh);
908 ext4_free_blocks(handle, inode,
909 le32_to_cpu(where[i-1].key), 1, 0);
911 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
917 * The ext4_ind_get_blocks() function handles non-extents inodes
918 * (i.e., using the traditional indirect/double-indirect i_blocks
919 * scheme) for ext4_get_blocks().
921 * Allocation strategy is simple: if we have to allocate something, we will
922 * have to go the whole way to leaf. So let's do it before attaching anything
923 * to tree, set linkage between the newborn blocks, write them if sync is
924 * required, recheck the path, free and repeat if check fails, otherwise
925 * set the last missing link (that will protect us from any truncate-generated
926 * removals - all blocks on the path are immune now) and possibly force the
927 * write on the parent block.
928 * That has a nice additional property: no special recovery from the failed
929 * allocations is needed - we simply release blocks and do not touch anything
930 * reachable from inode.
932 * `handle' can be NULL if create == 0.
934 * return > 0, # of blocks mapped or allocated.
935 * return = 0, if plain lookup failed.
936 * return < 0, error case.
938 * The ext4_ind_get_blocks() function should be called with
939 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
940 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
941 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
944 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
945 ext4_lblk_t iblock, unsigned int maxblocks,
946 struct buffer_head *bh_result,
950 ext4_lblk_t offsets[4];
955 int blocks_to_boundary = 0;
958 ext4_fsblk_t first_block = 0;
960 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
961 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
962 depth = ext4_block_to_path(inode, iblock, offsets,
963 &blocks_to_boundary);
968 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
970 /* Simplest case - block found, no allocation needed */
972 first_block = le32_to_cpu(chain[depth - 1].key);
973 clear_buffer_new(bh_result);
976 while (count < maxblocks && count <= blocks_to_boundary) {
979 blk = le32_to_cpu(*(chain[depth-1].p + count));
981 if (blk == first_block + count)
989 /* Next simple case - plain lookup or failed read of indirect block */
990 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
994 * Okay, we need to do block allocation.
996 goal = ext4_find_goal(inode, iblock, partial);
998 /* the number of blocks need to allocate for [d,t]indirect blocks */
999 indirect_blks = (chain + depth) - partial - 1;
1002 * Next look up the indirect map to count the totoal number of
1003 * direct blocks to allocate for this branch.
1005 count = ext4_blks_to_allocate(partial, indirect_blks,
1006 maxblocks, blocks_to_boundary);
1008 * Block out ext4_truncate while we alter the tree
1010 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
1012 offsets + (partial - chain), partial);
1015 * The ext4_splice_branch call will free and forget any buffers
1016 * on the new chain if there is a failure, but that risks using
1017 * up transaction credits, especially for bitmaps where the
1018 * credits cannot be returned. Can we handle this somehow? We
1019 * may need to return -EAGAIN upwards in the worst case. --sct
1022 err = ext4_splice_branch(handle, inode, iblock,
1023 partial, indirect_blks, count);
1027 set_buffer_new(bh_result);
1029 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1030 if (count > blocks_to_boundary)
1031 set_buffer_boundary(bh_result);
1033 /* Clean up and exit */
1034 partial = chain + depth - 1; /* the whole chain */
1036 while (partial > chain) {
1037 BUFFER_TRACE(partial->bh, "call brelse");
1038 brelse(partial->bh);
1041 BUFFER_TRACE(bh_result, "returned");
1046 qsize_t ext4_get_reserved_space(struct inode *inode)
1048 unsigned long long total;
1050 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1051 total = EXT4_I(inode)->i_reserved_data_blocks +
1052 EXT4_I(inode)->i_reserved_meta_blocks;
1053 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1055 return (total << inode->i_blkbits);
1058 * Calculate the number of metadata blocks need to reserve
1059 * to allocate @blocks for non extent file based file
1061 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1063 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1064 int ind_blks, dind_blks, tind_blks;
1066 /* number of new indirect blocks needed */
1067 ind_blks = (blocks + icap - 1) / icap;
1069 dind_blks = (ind_blks + icap - 1) / icap;
1073 return ind_blks + dind_blks + tind_blks;
1077 * Calculate the number of metadata blocks need to reserve
1078 * to allocate given number of blocks
1080 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1085 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1086 return ext4_ext_calc_metadata_amount(inode, blocks);
1088 return ext4_indirect_calc_metadata_amount(inode, blocks);
1091 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1093 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1094 int total, mdb, mdb_free;
1096 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1097 /* recalculate the number of metablocks still need to be reserved */
1098 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1099 mdb = ext4_calc_metadata_amount(inode, total);
1101 /* figure out how many metablocks to release */
1102 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1103 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1106 /* Account for allocated meta_blocks */
1107 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1109 /* update fs dirty blocks counter */
1110 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1111 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1112 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1115 /* update per-inode reservations */
1116 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1117 EXT4_I(inode)->i_reserved_data_blocks -= used;
1118 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1121 * free those over-booking quota for metadata blocks
1124 vfs_dq_release_reservation_block(inode, mdb_free);
1127 * If we have done all the pending block allocations and if
1128 * there aren't any writers on the inode, we can discard the
1129 * inode's preallocations.
1131 if (!total && (atomic_read(&inode->i_writecount) == 0))
1132 ext4_discard_preallocations(inode);
1135 static int check_block_validity(struct inode *inode, const char *msg,
1136 sector_t logical, sector_t phys, int len)
1138 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1139 ext4_error(inode->i_sb, msg,
1140 "inode #%lu logical block %llu mapped to %llu "
1141 "(size %d)", inode->i_ino,
1142 (unsigned long long) logical,
1143 (unsigned long long) phys, len);
1150 * Return the number of contiguous dirty pages in a given inode
1151 * starting at page frame idx.
1153 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1154 unsigned int max_pages)
1156 struct address_space *mapping = inode->i_mapping;
1158 struct pagevec pvec;
1160 int i, nr_pages, done = 0;
1164 pagevec_init(&pvec, 0);
1167 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1168 PAGECACHE_TAG_DIRTY,
1169 (pgoff_t)PAGEVEC_SIZE);
1172 for (i = 0; i < nr_pages; i++) {
1173 struct page *page = pvec.pages[i];
1174 struct buffer_head *bh, *head;
1177 if (unlikely(page->mapping != mapping) ||
1179 PageWriteback(page) ||
1180 page->index != idx) {
1185 if (page_has_buffers(page)) {
1186 bh = head = page_buffers(page);
1188 if (!buffer_delay(bh) &&
1189 !buffer_unwritten(bh))
1191 bh = bh->b_this_page;
1192 } while (!done && (bh != head));
1199 if (num >= max_pages)
1202 pagevec_release(&pvec);
1208 * The ext4_get_blocks() function tries to look up the requested blocks,
1209 * and returns if the blocks are already mapped.
1211 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1212 * and store the allocated blocks in the result buffer head and mark it
1215 * If file type is extents based, it will call ext4_ext_get_blocks(),
1216 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1219 * On success, it returns the number of blocks being mapped or allocate.
1220 * if create==0 and the blocks are pre-allocated and uninitialized block,
1221 * the result buffer head is unmapped. If the create ==1, it will make sure
1222 * the buffer head is mapped.
1224 * It returns 0 if plain look up failed (blocks have not been allocated), in
1225 * that casem, buffer head is unmapped
1227 * It returns the error in case of allocation failure.
1229 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1230 unsigned int max_blocks, struct buffer_head *bh,
1235 clear_buffer_mapped(bh);
1236 clear_buffer_unwritten(bh);
1238 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1239 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1240 (unsigned long)block);
1242 * Try to see if we can get the block without requesting a new
1243 * file system block.
1245 down_read((&EXT4_I(inode)->i_data_sem));
1246 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1247 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1250 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1253 up_read((&EXT4_I(inode)->i_data_sem));
1255 if (retval > 0 && buffer_mapped(bh)) {
1256 int ret = check_block_validity(inode, "file system corruption",
1257 block, bh->b_blocknr, retval);
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 && buffer_mapped(bh))
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 clear_buffer_unwritten(bh);
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_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1309 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1312 retval = ext4_ind_get_blocks(handle, inode, block,
1313 max_blocks, bh, flags);
1315 if (retval > 0 && buffer_new(bh)) {
1317 * We allocated new blocks which will result in
1318 * i_data's format changing. Force the migrate
1319 * to fail by clearing migrate flags
1321 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
1325 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1326 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1329 * Update reserved blocks/metadata blocks after successful
1330 * block allocation which had been deferred till now.
1332 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1333 ext4_da_update_reserve_space(inode, retval);
1335 up_write((&EXT4_I(inode)->i_data_sem));
1336 if (retval > 0 && buffer_mapped(bh)) {
1337 int ret = check_block_validity(inode, "file system "
1338 "corruption after allocation",
1339 block, bh->b_blocknr, retval);
1346 /* Maximum number of blocks we map for direct IO at once. */
1347 #define DIO_MAX_BLOCKS 4096
1349 int ext4_get_block(struct inode *inode, sector_t iblock,
1350 struct buffer_head *bh_result, int create)
1352 handle_t *handle = ext4_journal_current_handle();
1353 int ret = 0, started = 0;
1354 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1357 if (create && !handle) {
1358 /* Direct IO write... */
1359 if (max_blocks > DIO_MAX_BLOCKS)
1360 max_blocks = DIO_MAX_BLOCKS;
1361 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1362 handle = ext4_journal_start(inode, dio_credits);
1363 if (IS_ERR(handle)) {
1364 ret = PTR_ERR(handle);
1370 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1371 create ? EXT4_GET_BLOCKS_CREATE : 0);
1373 bh_result->b_size = (ret << inode->i_blkbits);
1377 ext4_journal_stop(handle);
1383 * `handle' can be NULL if create is zero
1385 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1386 ext4_lblk_t block, int create, int *errp)
1388 struct buffer_head dummy;
1392 J_ASSERT(handle != NULL || create == 0);
1395 dummy.b_blocknr = -1000;
1396 buffer_trace_init(&dummy.b_history);
1398 flags |= EXT4_GET_BLOCKS_CREATE;
1399 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1401 * ext4_get_blocks() returns number of blocks mapped. 0 in
1410 if (!err && buffer_mapped(&dummy)) {
1411 struct buffer_head *bh;
1412 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1417 if (buffer_new(&dummy)) {
1418 J_ASSERT(create != 0);
1419 J_ASSERT(handle != NULL);
1422 * Now that we do not always journal data, we should
1423 * keep in mind whether this should always journal the
1424 * new buffer as metadata. For now, regular file
1425 * writes use ext4_get_block instead, so it's not a
1429 BUFFER_TRACE(bh, "call get_create_access");
1430 fatal = ext4_journal_get_create_access(handle, bh);
1431 if (!fatal && !buffer_uptodate(bh)) {
1432 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1433 set_buffer_uptodate(bh);
1436 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1437 err = ext4_handle_dirty_metadata(handle, inode, bh);
1441 BUFFER_TRACE(bh, "not a new buffer");
1454 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1455 ext4_lblk_t block, int create, int *err)
1457 struct buffer_head *bh;
1459 bh = ext4_getblk(handle, inode, block, create, err);
1462 if (buffer_uptodate(bh))
1464 ll_rw_block(READ_META, 1, &bh);
1466 if (buffer_uptodate(bh))
1473 static int walk_page_buffers(handle_t *handle,
1474 struct buffer_head *head,
1478 int (*fn)(handle_t *handle,
1479 struct buffer_head *bh))
1481 struct buffer_head *bh;
1482 unsigned block_start, block_end;
1483 unsigned blocksize = head->b_size;
1485 struct buffer_head *next;
1487 for (bh = head, block_start = 0;
1488 ret == 0 && (bh != head || !block_start);
1489 block_start = block_end, bh = next) {
1490 next = bh->b_this_page;
1491 block_end = block_start + blocksize;
1492 if (block_end <= from || block_start >= to) {
1493 if (partial && !buffer_uptodate(bh))
1497 err = (*fn)(handle, bh);
1505 * To preserve ordering, it is essential that the hole instantiation and
1506 * the data write be encapsulated in a single transaction. We cannot
1507 * close off a transaction and start a new one between the ext4_get_block()
1508 * and the commit_write(). So doing the jbd2_journal_start at the start of
1509 * prepare_write() is the right place.
1511 * Also, this function can nest inside ext4_writepage() ->
1512 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1513 * has generated enough buffer credits to do the whole page. So we won't
1514 * block on the journal in that case, which is good, because the caller may
1517 * By accident, ext4 can be reentered when a transaction is open via
1518 * quota file writes. If we were to commit the transaction while thus
1519 * reentered, there can be a deadlock - we would be holding a quota
1520 * lock, and the commit would never complete if another thread had a
1521 * transaction open and was blocking on the quota lock - a ranking
1524 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1525 * will _not_ run commit under these circumstances because handle->h_ref
1526 * is elevated. We'll still have enough credits for the tiny quotafile
1529 static int do_journal_get_write_access(handle_t *handle,
1530 struct buffer_head *bh)
1532 if (!buffer_mapped(bh) || buffer_freed(bh))
1534 return ext4_journal_get_write_access(handle, bh);
1538 * Truncate blocks that were not used by write. We have to truncate the
1539 * pagecache as well so that corresponding buffers get properly unmapped.
1541 static void ext4_truncate_failed_write(struct inode *inode)
1543 truncate_inode_pages(inode->i_mapping, inode->i_size);
1544 ext4_truncate(inode);
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 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1591 if (!ret && ext4_should_journal_data(inode)) {
1592 ret = walk_page_buffers(handle, page_buffers(page),
1593 from, to, NULL, do_journal_get_write_access);
1598 page_cache_release(page);
1600 * block_write_begin may have instantiated a few blocks
1601 * outside i_size. Trim these off again. Don't need
1602 * i_size_read because we hold i_mutex.
1604 * Add inode to orphan list in case we crash before
1607 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1608 ext4_orphan_add(handle, inode);
1610 ext4_journal_stop(handle);
1611 if (pos + len > inode->i_size) {
1612 ext4_truncate_failed_write(inode);
1614 * If truncate failed early the inode might
1615 * still be on the orphan list; we need to
1616 * make sure the inode is removed from the
1617 * orphan list in that case.
1620 ext4_orphan_del(NULL, inode);
1624 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1630 /* For write_end() in data=journal mode */
1631 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1633 if (!buffer_mapped(bh) || buffer_freed(bh))
1635 set_buffer_uptodate(bh);
1636 return ext4_handle_dirty_metadata(handle, NULL, bh);
1639 static int ext4_generic_write_end(struct file *file,
1640 struct address_space *mapping,
1641 loff_t pos, unsigned len, unsigned copied,
1642 struct page *page, void *fsdata)
1644 int i_size_changed = 0;
1645 struct inode *inode = mapping->host;
1646 handle_t *handle = ext4_journal_current_handle();
1648 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1651 * No need to use i_size_read() here, the i_size
1652 * cannot change under us because we hold i_mutex.
1654 * But it's important to update i_size while still holding page lock:
1655 * page writeout could otherwise come in and zero beyond i_size.
1657 if (pos + copied > inode->i_size) {
1658 i_size_write(inode, pos + copied);
1662 if (pos + copied > EXT4_I(inode)->i_disksize) {
1663 /* We need to mark inode dirty even if
1664 * new_i_size is less that inode->i_size
1665 * bu greater than i_disksize.(hint delalloc)
1667 ext4_update_i_disksize(inode, (pos + copied));
1671 page_cache_release(page);
1674 * Don't mark the inode dirty under page lock. First, it unnecessarily
1675 * makes the holding time of page lock longer. Second, it forces lock
1676 * ordering of page lock and transaction start for journaling
1680 ext4_mark_inode_dirty(handle, inode);
1686 * We need to pick up the new inode size which generic_commit_write gave us
1687 * `file' can be NULL - eg, when called from page_symlink().
1689 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1690 * buffers are managed internally.
1692 static int ext4_ordered_write_end(struct file *file,
1693 struct address_space *mapping,
1694 loff_t pos, unsigned len, unsigned copied,
1695 struct page *page, void *fsdata)
1697 handle_t *handle = ext4_journal_current_handle();
1698 struct inode *inode = mapping->host;
1701 trace_ext4_ordered_write_end(inode, pos, len, copied);
1702 ret = ext4_jbd2_file_inode(handle, inode);
1705 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1708 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1709 /* if we have allocated more blocks and copied
1710 * less. We will have blocks allocated outside
1711 * inode->i_size. So truncate them
1713 ext4_orphan_add(handle, inode);
1717 ret2 = ext4_journal_stop(handle);
1721 if (pos + len > inode->i_size) {
1722 ext4_truncate_failed_write(inode);
1724 * If truncate failed early the inode might still be
1725 * on the orphan list; we need to make sure the inode
1726 * is removed from the orphan list in that case.
1729 ext4_orphan_del(NULL, inode);
1733 return ret ? ret : copied;
1736 static int ext4_writeback_write_end(struct file *file,
1737 struct address_space *mapping,
1738 loff_t pos, unsigned len, unsigned copied,
1739 struct page *page, void *fsdata)
1741 handle_t *handle = ext4_journal_current_handle();
1742 struct inode *inode = mapping->host;
1745 trace_ext4_writeback_write_end(inode, pos, len, copied);
1746 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1749 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1750 /* if we have allocated more blocks and copied
1751 * less. We will have blocks allocated outside
1752 * inode->i_size. So truncate them
1754 ext4_orphan_add(handle, inode);
1759 ret2 = ext4_journal_stop(handle);
1763 if (pos + len > inode->i_size) {
1764 ext4_truncate_failed_write(inode);
1766 * If truncate failed early the inode might still be
1767 * on the orphan list; we need to make sure the inode
1768 * is removed from the orphan list in that case.
1771 ext4_orphan_del(NULL, inode);
1774 return ret ? ret : copied;
1777 static int ext4_journalled_write_end(struct file *file,
1778 struct address_space *mapping,
1779 loff_t pos, unsigned len, unsigned copied,
1780 struct page *page, void *fsdata)
1782 handle_t *handle = ext4_journal_current_handle();
1783 struct inode *inode = mapping->host;
1789 trace_ext4_journalled_write_end(inode, pos, len, copied);
1790 from = pos & (PAGE_CACHE_SIZE - 1);
1794 if (!PageUptodate(page))
1796 page_zero_new_buffers(page, from+copied, to);
1799 ret = walk_page_buffers(handle, page_buffers(page), from,
1800 to, &partial, write_end_fn);
1802 SetPageUptodate(page);
1803 new_i_size = pos + copied;
1804 if (new_i_size > inode->i_size)
1805 i_size_write(inode, pos+copied);
1806 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1807 if (new_i_size > EXT4_I(inode)->i_disksize) {
1808 ext4_update_i_disksize(inode, new_i_size);
1809 ret2 = ext4_mark_inode_dirty(handle, inode);
1815 page_cache_release(page);
1816 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1817 /* if we have allocated more blocks and copied
1818 * less. We will have blocks allocated outside
1819 * inode->i_size. So truncate them
1821 ext4_orphan_add(handle, inode);
1823 ret2 = ext4_journal_stop(handle);
1826 if (pos + len > inode->i_size) {
1827 ext4_truncate_failed_write(inode);
1829 * If truncate failed early the inode might still be
1830 * on the orphan list; we need to make sure the inode
1831 * is removed from the orphan list in that case.
1834 ext4_orphan_del(NULL, inode);
1837 return ret ? ret : copied;
1840 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1843 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1844 unsigned long md_needed, mdblocks, total = 0;
1847 * recalculate the amount of metadata blocks to reserve
1848 * in order to allocate nrblocks
1849 * worse case is one extent per block
1852 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1853 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1854 mdblocks = ext4_calc_metadata_amount(inode, total);
1855 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1857 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1858 total = md_needed + nrblocks;
1861 * Make quota reservation here to prevent quota overflow
1862 * later. Real quota accounting is done at pages writeout
1865 if (vfs_dq_reserve_block(inode, total)) {
1866 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1870 if (ext4_claim_free_blocks(sbi, total)) {
1871 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1872 vfs_dq_release_reservation_block(inode, total);
1873 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1879 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1880 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1882 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1883 return 0; /* success */
1886 static void ext4_da_release_space(struct inode *inode, int to_free)
1888 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1889 int total, mdb, mdb_free, release;
1892 return; /* Nothing to release, exit */
1894 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1896 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1898 * if there is no reserved blocks, but we try to free some
1899 * then the counter is messed up somewhere.
1900 * but since this function is called from invalidate
1901 * page, it's harmless to return without any action
1903 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1904 "blocks for inode %lu, but there is no reserved "
1905 "data blocks\n", to_free, inode->i_ino);
1906 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1910 /* recalculate the number of metablocks still need to be reserved */
1911 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1912 mdb = ext4_calc_metadata_amount(inode, total);
1914 /* figure out how many metablocks to release */
1915 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1916 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1918 release = to_free + mdb_free;
1920 /* update fs dirty blocks counter for truncate case */
1921 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1923 /* update per-inode reservations */
1924 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1925 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1927 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1928 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1929 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1931 vfs_dq_release_reservation_block(inode, release);
1934 static void ext4_da_page_release_reservation(struct page *page,
1935 unsigned long offset)
1938 struct buffer_head *head, *bh;
1939 unsigned int curr_off = 0;
1941 head = page_buffers(page);
1944 unsigned int next_off = curr_off + bh->b_size;
1946 if ((offset <= curr_off) && (buffer_delay(bh))) {
1948 clear_buffer_delay(bh);
1950 curr_off = next_off;
1951 } while ((bh = bh->b_this_page) != head);
1952 ext4_da_release_space(page->mapping->host, to_release);
1956 * Delayed allocation stuff
1960 * mpage_da_submit_io - walks through extent of pages and try to write
1961 * them with writepage() call back
1963 * @mpd->inode: inode
1964 * @mpd->first_page: first page of the extent
1965 * @mpd->next_page: page after the last page of the extent
1967 * By the time mpage_da_submit_io() is called we expect all blocks
1968 * to be allocated. this may be wrong if allocation failed.
1970 * As pages are already locked by write_cache_pages(), we can't use it
1972 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1975 struct pagevec pvec;
1976 unsigned long index, end;
1977 int ret = 0, err, nr_pages, i;
1978 struct inode *inode = mpd->inode;
1979 struct address_space *mapping = inode->i_mapping;
1981 BUG_ON(mpd->next_page <= mpd->first_page);
1983 * We need to start from the first_page to the next_page - 1
1984 * to make sure we also write the mapped dirty buffer_heads.
1985 * If we look at mpd->b_blocknr we would only be looking
1986 * at the currently mapped buffer_heads.
1988 index = mpd->first_page;
1989 end = mpd->next_page - 1;
1991 pagevec_init(&pvec, 0);
1992 while (index <= end) {
1993 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1996 for (i = 0; i < nr_pages; i++) {
1997 struct page *page = pvec.pages[i];
1999 index = page->index;
2004 BUG_ON(!PageLocked(page));
2005 BUG_ON(PageWriteback(page));
2007 pages_skipped = mpd->wbc->pages_skipped;
2008 err = mapping->a_ops->writepage(page, mpd->wbc);
2009 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2011 * have successfully written the page
2012 * without skipping the same
2014 mpd->pages_written++;
2016 * In error case, we have to continue because
2017 * remaining pages are still locked
2018 * XXX: unlock and re-dirty them?
2023 pagevec_release(&pvec);
2029 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2031 * @mpd->inode - inode to walk through
2032 * @exbh->b_blocknr - first block on a disk
2033 * @exbh->b_size - amount of space in bytes
2034 * @logical - first logical block to start assignment with
2036 * the function goes through all passed space and put actual disk
2037 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2039 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2040 struct buffer_head *exbh)
2042 struct inode *inode = mpd->inode;
2043 struct address_space *mapping = inode->i_mapping;
2044 int blocks = exbh->b_size >> inode->i_blkbits;
2045 sector_t pblock = exbh->b_blocknr, cur_logical;
2046 struct buffer_head *head, *bh;
2048 struct pagevec pvec;
2051 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2052 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2053 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2055 pagevec_init(&pvec, 0);
2057 while (index <= end) {
2058 /* XXX: optimize tail */
2059 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2062 for (i = 0; i < nr_pages; i++) {
2063 struct page *page = pvec.pages[i];
2065 index = page->index;
2070 BUG_ON(!PageLocked(page));
2071 BUG_ON(PageWriteback(page));
2072 BUG_ON(!page_has_buffers(page));
2074 bh = page_buffers(page);
2077 /* skip blocks out of the range */
2079 if (cur_logical >= logical)
2082 } while ((bh = bh->b_this_page) != head);
2085 if (cur_logical >= logical + blocks)
2088 if (buffer_delay(bh) ||
2089 buffer_unwritten(bh)) {
2091 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2093 if (buffer_delay(bh)) {
2094 clear_buffer_delay(bh);
2095 bh->b_blocknr = pblock;
2098 * unwritten already should have
2099 * blocknr assigned. Verify that
2101 clear_buffer_unwritten(bh);
2102 BUG_ON(bh->b_blocknr != pblock);
2105 } else if (buffer_mapped(bh))
2106 BUG_ON(bh->b_blocknr != pblock);
2110 } while ((bh = bh->b_this_page) != head);
2112 pagevec_release(&pvec);
2118 * __unmap_underlying_blocks - just a helper function to unmap
2119 * set of blocks described by @bh
2121 static inline void __unmap_underlying_blocks(struct inode *inode,
2122 struct buffer_head *bh)
2124 struct block_device *bdev = inode->i_sb->s_bdev;
2127 blocks = bh->b_size >> inode->i_blkbits;
2128 for (i = 0; i < blocks; i++)
2129 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2132 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2133 sector_t logical, long blk_cnt)
2137 struct pagevec pvec;
2138 struct inode *inode = mpd->inode;
2139 struct address_space *mapping = inode->i_mapping;
2141 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2142 end = (logical + blk_cnt - 1) >>
2143 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2144 while (index <= end) {
2145 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2148 for (i = 0; i < nr_pages; i++) {
2149 struct page *page = pvec.pages[i];
2150 index = page->index;
2155 BUG_ON(!PageLocked(page));
2156 BUG_ON(PageWriteback(page));
2157 block_invalidatepage(page, 0);
2158 ClearPageUptodate(page);
2165 static void ext4_print_free_blocks(struct inode *inode)
2167 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2168 printk(KERN_CRIT "Total free blocks count %lld\n",
2169 ext4_count_free_blocks(inode->i_sb));
2170 printk(KERN_CRIT "Free/Dirty block details\n");
2171 printk(KERN_CRIT "free_blocks=%lld\n",
2172 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2173 printk(KERN_CRIT "dirty_blocks=%lld\n",
2174 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2175 printk(KERN_CRIT "Block reservation details\n");
2176 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2177 EXT4_I(inode)->i_reserved_data_blocks);
2178 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2179 EXT4_I(inode)->i_reserved_meta_blocks);
2184 * mpage_da_map_blocks - go through given space
2186 * @mpd - bh describing space
2188 * The function skips space we know is already mapped to disk blocks.
2191 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2193 int err, blks, get_blocks_flags;
2194 struct buffer_head new;
2195 sector_t next = mpd->b_blocknr;
2196 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2197 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2198 handle_t *handle = NULL;
2201 * We consider only non-mapped and non-allocated blocks
2203 if ((mpd->b_state & (1 << BH_Mapped)) &&
2204 !(mpd->b_state & (1 << BH_Delay)) &&
2205 !(mpd->b_state & (1 << BH_Unwritten)))
2209 * If we didn't accumulate anything to write simply return
2214 handle = ext4_journal_current_handle();
2218 * Call ext4_get_blocks() to allocate any delayed allocation
2219 * blocks, or to convert an uninitialized extent to be
2220 * initialized (in the case where we have written into
2221 * one or more preallocated blocks).
2223 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2224 * indicate that we are on the delayed allocation path. This
2225 * affects functions in many different parts of the allocation
2226 * call path. This flag exists primarily because we don't
2227 * want to change *many* call functions, so ext4_get_blocks()
2228 * will set the magic i_delalloc_reserved_flag once the
2229 * inode's allocation semaphore is taken.
2231 * If the blocks in questions were delalloc blocks, set
2232 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2233 * variables are updated after the blocks have been allocated.
2236 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2237 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2238 if (mpd->b_state & (1 << BH_Delay))
2239 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2240 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2241 &new, get_blocks_flags);
2245 * If get block returns with error we simply
2246 * return. Later writepage will redirty the page and
2247 * writepages will find the dirty page again
2252 if (err == -ENOSPC &&
2253 ext4_count_free_blocks(mpd->inode->i_sb)) {
2259 * get block failure will cause us to loop in
2260 * writepages, because a_ops->writepage won't be able
2261 * to make progress. The page will be redirtied by
2262 * writepage and writepages will again try to write
2265 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2266 "delayed block allocation failed for inode %lu at "
2267 "logical offset %llu with max blocks %zd with "
2268 "error %d\n", mpd->inode->i_ino,
2269 (unsigned long long) next,
2270 mpd->b_size >> mpd->inode->i_blkbits, err);
2271 printk(KERN_CRIT "This should not happen!! "
2272 "Data will be lost\n");
2273 if (err == -ENOSPC) {
2274 ext4_print_free_blocks(mpd->inode);
2276 /* invalidate all the pages */
2277 ext4_da_block_invalidatepages(mpd, next,
2278 mpd->b_size >> mpd->inode->i_blkbits);
2283 new.b_size = (blks << mpd->inode->i_blkbits);
2285 if (buffer_new(&new))
2286 __unmap_underlying_blocks(mpd->inode, &new);
2289 * If blocks are delayed marked, we need to
2290 * put actual blocknr and drop delayed bit
2292 if ((mpd->b_state & (1 << BH_Delay)) ||
2293 (mpd->b_state & (1 << BH_Unwritten)))
2294 mpage_put_bnr_to_bhs(mpd, next, &new);
2296 if (ext4_should_order_data(mpd->inode)) {
2297 err = ext4_jbd2_file_inode(handle, mpd->inode);
2303 * Update on-disk size along with block allocation.
2305 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2306 if (disksize > i_size_read(mpd->inode))
2307 disksize = i_size_read(mpd->inode);
2308 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2309 ext4_update_i_disksize(mpd->inode, disksize);
2310 return ext4_mark_inode_dirty(handle, mpd->inode);
2316 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2317 (1 << BH_Delay) | (1 << BH_Unwritten))
2320 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2322 * @mpd->lbh - extent of blocks
2323 * @logical - logical number of the block in the file
2324 * @bh - bh of the block (used to access block's state)
2326 * the function is used to collect contig. blocks in same state
2328 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2329 sector_t logical, size_t b_size,
2330 unsigned long b_state)
2333 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2335 /* check if thereserved journal credits might overflow */
2336 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2337 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2339 * With non-extent format we are limited by the journal
2340 * credit available. Total credit needed to insert
2341 * nrblocks contiguous blocks is dependent on the
2342 * nrblocks. So limit nrblocks.
2345 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2346 EXT4_MAX_TRANS_DATA) {
2348 * Adding the new buffer_head would make it cross the
2349 * allowed limit for which we have journal credit
2350 * reserved. So limit the new bh->b_size
2352 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2353 mpd->inode->i_blkbits;
2354 /* we will do mpage_da_submit_io in the next loop */
2358 * First block in the extent
2360 if (mpd->b_size == 0) {
2361 mpd->b_blocknr = logical;
2362 mpd->b_size = b_size;
2363 mpd->b_state = b_state & BH_FLAGS;
2367 next = mpd->b_blocknr + nrblocks;
2369 * Can we merge the block to our big extent?
2371 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2372 mpd->b_size += b_size;
2378 * We couldn't merge the block to our extent, so we
2379 * need to flush current extent and start new one
2381 if (mpage_da_map_blocks(mpd) == 0)
2382 mpage_da_submit_io(mpd);
2387 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2389 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2393 * __mpage_da_writepage - finds extent of pages and blocks
2395 * @page: page to consider
2396 * @wbc: not used, we just follow rules
2399 * The function finds extents of pages and scan them for all blocks.
2401 static int __mpage_da_writepage(struct page *page,
2402 struct writeback_control *wbc, void *data)
2404 struct mpage_da_data *mpd = data;
2405 struct inode *inode = mpd->inode;
2406 struct buffer_head *bh, *head;
2411 * Rest of the page in the page_vec
2412 * redirty then and skip then. We will
2413 * try to write them again after
2414 * starting a new transaction
2416 redirty_page_for_writepage(wbc, page);
2418 return MPAGE_DA_EXTENT_TAIL;
2421 * Can we merge this page to current extent?
2423 if (mpd->next_page != page->index) {
2425 * Nope, we can't. So, we map non-allocated blocks
2426 * and start IO on them using writepage()
2428 if (mpd->next_page != mpd->first_page) {
2429 if (mpage_da_map_blocks(mpd) == 0)
2430 mpage_da_submit_io(mpd);
2432 * skip rest of the page in the page_vec
2435 redirty_page_for_writepage(wbc, page);
2437 return MPAGE_DA_EXTENT_TAIL;
2441 * Start next extent of pages ...
2443 mpd->first_page = page->index;
2453 mpd->next_page = page->index + 1;
2454 logical = (sector_t) page->index <<
2455 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2457 if (!page_has_buffers(page)) {
2458 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2459 (1 << BH_Dirty) | (1 << BH_Uptodate));
2461 return MPAGE_DA_EXTENT_TAIL;
2464 * Page with regular buffer heads, just add all dirty ones
2466 head = page_buffers(page);
2469 BUG_ON(buffer_locked(bh));
2471 * We need to try to allocate
2472 * unmapped blocks in the same page.
2473 * Otherwise we won't make progress
2474 * with the page in ext4_writepage
2476 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2477 mpage_add_bh_to_extent(mpd, logical,
2481 return MPAGE_DA_EXTENT_TAIL;
2482 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2484 * mapped dirty buffer. We need to update
2485 * the b_state because we look at
2486 * b_state in mpage_da_map_blocks. We don't
2487 * update b_size because if we find an
2488 * unmapped buffer_head later we need to
2489 * use the b_state flag of that buffer_head.
2491 if (mpd->b_size == 0)
2492 mpd->b_state = bh->b_state & BH_FLAGS;
2495 } while ((bh = bh->b_this_page) != head);
2502 * This is a special get_blocks_t callback which is used by
2503 * ext4_da_write_begin(). It will either return mapped block or
2504 * reserve space for a single block.
2506 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2507 * We also have b_blocknr = -1 and b_bdev initialized properly
2509 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2510 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2511 * initialized properly.
2513 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2514 struct buffer_head *bh_result, int create)
2517 sector_t invalid_block = ~((sector_t) 0xffff);
2519 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2522 BUG_ON(create == 0);
2523 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2526 * first, we need to know whether the block is allocated already
2527 * preallocated blocks are unmapped but should treated
2528 * the same as allocated blocks.
2530 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2531 if ((ret == 0) && !buffer_delay(bh_result)) {
2532 /* the block isn't (pre)allocated yet, let's reserve space */
2534 * XXX: __block_prepare_write() unmaps passed block,
2537 ret = ext4_da_reserve_space(inode, 1);
2539 /* not enough space to reserve */
2542 map_bh(bh_result, inode->i_sb, invalid_block);
2543 set_buffer_new(bh_result);
2544 set_buffer_delay(bh_result);
2545 } else if (ret > 0) {
2546 bh_result->b_size = (ret << inode->i_blkbits);
2547 if (buffer_unwritten(bh_result)) {
2548 /* A delayed write to unwritten bh should
2549 * be marked new and mapped. Mapped ensures
2550 * that we don't do get_block multiple times
2551 * when we write to the same offset and new
2552 * ensures that we do proper zero out for
2555 set_buffer_new(bh_result);
2556 set_buffer_mapped(bh_result);
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(), nobh_writepage(), and
2568 * block_write_full_page(). These functions should only try to map a
2569 * single block at a time.
2571 * Since this function doesn't do block allocations even if the caller
2572 * requests it by passing in create=1, it is critically important that
2573 * any caller checks to make sure that any buffer heads are returned
2574 * by this function are either all already mapped or marked for
2575 * delayed allocation before calling nobh_writepage() or
2576 * block_write_full_page(). Otherwise, b_blocknr could be left
2577 * unitialized, and the page write functions will be taken by
2580 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2581 struct buffer_head *bh_result, int create)
2584 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2586 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2589 * we don't want to do block allocation in writepage
2590 * so call get_block_wrap with create = 0
2592 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2594 bh_result->b_size = (ret << inode->i_blkbits);
2600 static int bget_one(handle_t *handle, struct buffer_head *bh)
2606 static int bput_one(handle_t *handle, struct buffer_head *bh)
2612 static int __ext4_journalled_writepage(struct page *page,
2613 struct writeback_control *wbc,
2616 struct address_space *mapping = page->mapping;
2617 struct inode *inode = mapping->host;
2618 struct buffer_head *page_bufs;
2619 handle_t *handle = NULL;
2623 page_bufs = page_buffers(page);
2625 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2626 /* As soon as we unlock the page, it can go away, but we have
2627 * references to buffers so we are safe */
2630 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2631 if (IS_ERR(handle)) {
2632 ret = PTR_ERR(handle);
2636 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2637 do_journal_get_write_access);
2639 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2643 err = ext4_journal_stop(handle);
2647 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2648 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2654 * Note that we don't need to start a transaction unless we're journaling data
2655 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2656 * need to file the inode to the transaction's list in ordered mode because if
2657 * we are writing back data added by write(), the inode is already there and if
2658 * we are writing back data modified via mmap(), noone guarantees in which
2659 * transaction the data will hit the disk. In case we are journaling data, we
2660 * cannot start transaction directly because transaction start ranks above page
2661 * lock so we have to do some magic.
2663 * This function can get called via...
2664 * - ext4_da_writepages after taking page lock (have journal handle)
2665 * - journal_submit_inode_data_buffers (no journal handle)
2666 * - shrink_page_list via pdflush (no journal handle)
2667 * - grab_page_cache when doing write_begin (have journal handle)
2669 * We don't do any block allocation in this function. If we have page with
2670 * multiple blocks we need to write those buffer_heads that are mapped. This
2671 * is important for mmaped based write. So if we do with blocksize 1K
2672 * truncate(f, 1024);
2673 * a = mmap(f, 0, 4096);
2675 * truncate(f, 4096);
2676 * we have in the page first buffer_head mapped via page_mkwrite call back
2677 * but other bufer_heads would be unmapped but dirty(dirty done via the
2678 * do_wp_page). So writepage should write the first block. If we modify
2679 * the mmap area beyond 1024 we will again get a page_fault and the
2680 * page_mkwrite callback will do the block allocation and mark the
2681 * buffer_heads mapped.
2683 * We redirty the page if we have any buffer_heads that is either delay or
2684 * unwritten in the page.
2686 * We can get recursively called as show below.
2688 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2691 * But since we don't do any block allocation we should not deadlock.
2692 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2694 static int ext4_writepage(struct page *page,
2695 struct writeback_control *wbc)
2700 struct buffer_head *page_bufs;
2701 struct inode *inode = page->mapping->host;
2703 trace_ext4_writepage(inode, page);
2704 size = i_size_read(inode);
2705 if (page->index == size >> PAGE_CACHE_SHIFT)
2706 len = size & ~PAGE_CACHE_MASK;
2708 len = PAGE_CACHE_SIZE;
2710 if (page_has_buffers(page)) {
2711 page_bufs = page_buffers(page);
2712 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2713 ext4_bh_delay_or_unwritten)) {
2715 * We don't want to do block allocation
2716 * So redirty the page and return
2717 * We may reach here when we do a journal commit
2718 * via journal_submit_inode_data_buffers.
2719 * If we don't have mapping block we just ignore
2720 * them. We can also reach here via shrink_page_list
2722 redirty_page_for_writepage(wbc, page);
2728 * The test for page_has_buffers() is subtle:
2729 * We know the page is dirty but it lost buffers. That means
2730 * that at some moment in time after write_begin()/write_end()
2731 * has been called all buffers have been clean and thus they
2732 * must have been written at least once. So they are all
2733 * mapped and we can happily proceed with mapping them
2734 * and writing the page.
2736 * Try to initialize the buffer_heads and check whether
2737 * all are mapped and non delay. We don't want to
2738 * do block allocation here.
2740 ret = block_prepare_write(page, 0, len,
2741 noalloc_get_block_write);
2743 page_bufs = page_buffers(page);
2744 /* check whether all are mapped and non delay */
2745 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2746 ext4_bh_delay_or_unwritten)) {
2747 redirty_page_for_writepage(wbc, page);
2753 * We can't do block allocation here
2754 * so just redity the page and unlock
2757 redirty_page_for_writepage(wbc, page);
2761 /* now mark the buffer_heads as dirty and uptodate */
2762 block_commit_write(page, 0, len);
2765 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2767 * It's mmapped pagecache. Add buffers and journal it. There
2768 * doesn't seem much point in redirtying the page here.
2770 ClearPageChecked(page);
2771 return __ext4_journalled_writepage(page, wbc, len);
2774 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2775 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2777 ret = block_write_full_page(page, noalloc_get_block_write,
2784 * This is called via ext4_da_writepages() to
2785 * calulate the total number of credits to reserve to fit
2786 * a single extent allocation into a single transaction,
2787 * ext4_da_writpeages() will loop calling this before
2788 * the block allocation.
2791 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2793 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2796 * With non-extent format the journal credit needed to
2797 * insert nrblocks contiguous block is dependent on
2798 * number of contiguous block. So we will limit
2799 * number of contiguous block to a sane value
2801 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) &&
2802 (max_blocks > EXT4_MAX_TRANS_DATA))
2803 max_blocks = EXT4_MAX_TRANS_DATA;
2805 return ext4_chunk_trans_blocks(inode, max_blocks);
2808 static int ext4_da_writepages(struct address_space *mapping,
2809 struct writeback_control *wbc)
2812 int range_whole = 0;
2813 handle_t *handle = NULL;
2814 struct mpage_da_data mpd;
2815 struct inode *inode = mapping->host;
2816 int no_nrwrite_index_update;
2817 int pages_written = 0;
2819 unsigned int max_pages;
2820 int range_cyclic, cycled = 1, io_done = 0;
2821 int needed_blocks, ret = 0;
2822 long desired_nr_to_write, nr_to_writebump = 0;
2823 loff_t range_start = wbc->range_start;
2824 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2826 trace_ext4_da_writepages(inode, wbc);
2829 * No pages to write? This is mainly a kludge to avoid starting
2830 * a transaction for special inodes like journal inode on last iput()
2831 * because that could violate lock ordering on umount
2833 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2837 * If the filesystem has aborted, it is read-only, so return
2838 * right away instead of dumping stack traces later on that
2839 * will obscure the real source of the problem. We test
2840 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2841 * the latter could be true if the filesystem is mounted
2842 * read-only, and in that case, ext4_da_writepages should
2843 * *never* be called, so if that ever happens, we would want
2846 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2849 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2852 range_cyclic = wbc->range_cyclic;
2853 if (wbc->range_cyclic) {
2854 index = mapping->writeback_index;
2857 wbc->range_start = index << PAGE_CACHE_SHIFT;
2858 wbc->range_end = LLONG_MAX;
2859 wbc->range_cyclic = 0;
2861 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2864 * This works around two forms of stupidity. The first is in
2865 * the writeback code, which caps the maximum number of pages
2866 * written to be 1024 pages. This is wrong on multiple
2867 * levels; different architectues have a different page size,
2868 * which changes the maximum amount of data which gets
2869 * written. Secondly, 4 megabytes is way too small. XFS
2870 * forces this value to be 16 megabytes by multiplying
2871 * nr_to_write parameter by four, and then relies on its
2872 * allocator to allocate larger extents to make them
2873 * contiguous. Unfortunately this brings us to the second
2874 * stupidity, which is that ext4's mballoc code only allocates
2875 * at most 2048 blocks. So we force contiguous writes up to
2876 * the number of dirty blocks in the inode, or
2877 * sbi->max_writeback_mb_bump whichever is smaller.
2879 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2880 if (!range_cyclic && range_whole)
2881 desired_nr_to_write = wbc->nr_to_write * 8;
2883 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2885 if (desired_nr_to_write > max_pages)
2886 desired_nr_to_write = max_pages;
2888 if (wbc->nr_to_write < desired_nr_to_write) {
2889 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2890 wbc->nr_to_write = desired_nr_to_write;
2894 mpd.inode = mapping->host;
2897 * we don't want write_cache_pages to update
2898 * nr_to_write and writeback_index
2900 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2901 wbc->no_nrwrite_index_update = 1;
2902 pages_skipped = wbc->pages_skipped;
2905 while (!ret && wbc->nr_to_write > 0) {
2908 * we insert one extent at a time. So we need
2909 * credit needed for single extent allocation.
2910 * journalled mode is currently not supported
2913 BUG_ON(ext4_should_journal_data(inode));
2914 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2916 /* start a new transaction*/
2917 handle = ext4_journal_start(inode, needed_blocks);
2918 if (IS_ERR(handle)) {
2919 ret = PTR_ERR(handle);
2920 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2921 "%ld pages, ino %lu; err %d\n", __func__,
2922 wbc->nr_to_write, inode->i_ino, ret);
2923 goto out_writepages;
2927 * Now call __mpage_da_writepage to find the next
2928 * contiguous region of logical blocks that need
2929 * blocks to be allocated by ext4. We don't actually
2930 * submit the blocks for I/O here, even though
2931 * write_cache_pages thinks it will, and will set the
2932 * pages as clean for write before calling
2933 * __mpage_da_writepage().
2941 mpd.pages_written = 0;
2943 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2946 * If we have a contigous extent of pages and we
2947 * haven't done the I/O yet, map the blocks and submit
2950 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2951 if (mpage_da_map_blocks(&mpd) == 0)
2952 mpage_da_submit_io(&mpd);
2954 ret = MPAGE_DA_EXTENT_TAIL;
2956 trace_ext4_da_write_pages(inode, &mpd);
2957 wbc->nr_to_write -= mpd.pages_written;
2959 ext4_journal_stop(handle);
2961 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2962 /* commit the transaction which would
2963 * free blocks released in the transaction
2966 jbd2_journal_force_commit_nested(sbi->s_journal);
2967 wbc->pages_skipped = pages_skipped;
2969 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2971 * got one extent now try with
2974 pages_written += mpd.pages_written;
2975 wbc->pages_skipped = pages_skipped;
2978 } else if (wbc->nr_to_write)
2980 * There is no more writeout needed
2981 * or we requested for a noblocking writeout
2982 * and we found the device congested
2986 if (!io_done && !cycled) {
2989 wbc->range_start = index << PAGE_CACHE_SHIFT;
2990 wbc->range_end = mapping->writeback_index - 1;
2993 if (pages_skipped != wbc->pages_skipped)
2994 ext4_msg(inode->i_sb, KERN_CRIT,
2995 "This should not happen leaving %s "
2996 "with nr_to_write = %ld ret = %d\n",
2997 __func__, wbc->nr_to_write, ret);
3000 index += pages_written;
3001 wbc->range_cyclic = range_cyclic;
3002 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3004 * set the writeback_index so that range_cyclic
3005 * mode will write it back later
3007 mapping->writeback_index = index;
3010 if (!no_nrwrite_index_update)
3011 wbc->no_nrwrite_index_update = 0;
3012 if (wbc->nr_to_write > nr_to_writebump)
3013 wbc->nr_to_write -= nr_to_writebump;
3014 wbc->range_start = range_start;
3015 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3019 #define FALL_BACK_TO_NONDELALLOC 1
3020 static int ext4_nonda_switch(struct super_block *sb)
3022 s64 free_blocks, dirty_blocks;
3023 struct ext4_sb_info *sbi = EXT4_SB(sb);
3026 * switch to non delalloc mode if we are running low
3027 * on free block. The free block accounting via percpu
3028 * counters can get slightly wrong with percpu_counter_batch getting
3029 * accumulated on each CPU without updating global counters
3030 * Delalloc need an accurate free block accounting. So switch
3031 * to non delalloc when we are near to error range.
3033 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3034 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3035 if (2 * free_blocks < 3 * dirty_blocks ||
3036 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3038 * free block count is less that 150% of dirty blocks
3039 * or free blocks is less that watermark
3046 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3047 loff_t pos, unsigned len, unsigned flags,
3048 struct page **pagep, void **fsdata)
3050 int ret, retries = 0;
3054 struct inode *inode = mapping->host;
3057 index = pos >> PAGE_CACHE_SHIFT;
3058 from = pos & (PAGE_CACHE_SIZE - 1);
3061 if (ext4_nonda_switch(inode->i_sb)) {
3062 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3063 return ext4_write_begin(file, mapping, pos,
3064 len, flags, pagep, fsdata);
3066 *fsdata = (void *)0;
3067 trace_ext4_da_write_begin(inode, pos, len, flags);
3070 * With delayed allocation, we don't log the i_disksize update
3071 * if there is delayed block allocation. But we still need
3072 * to journalling the i_disksize update if writes to the end
3073 * of file which has an already mapped buffer.
3075 handle = ext4_journal_start(inode, 1);
3076 if (IS_ERR(handle)) {
3077 ret = PTR_ERR(handle);
3080 /* We cannot recurse into the filesystem as the transaction is already
3082 flags |= AOP_FLAG_NOFS;
3084 page = grab_cache_page_write_begin(mapping, index, flags);
3086 ext4_journal_stop(handle);
3092 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3093 ext4_da_get_block_prep);
3096 ext4_journal_stop(handle);
3097 page_cache_release(page);
3099 * block_write_begin may have instantiated a few blocks
3100 * outside i_size. Trim these off again. Don't need
3101 * i_size_read because we hold i_mutex.
3103 if (pos + len > inode->i_size)
3104 ext4_truncate_failed_write(inode);
3107 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3114 * Check if we should update i_disksize
3115 * when write to the end of file but not require block allocation
3117 static int ext4_da_should_update_i_disksize(struct page *page,
3118 unsigned long offset)
3120 struct buffer_head *bh;
3121 struct inode *inode = page->mapping->host;
3125 bh = page_buffers(page);
3126 idx = offset >> inode->i_blkbits;
3128 for (i = 0; i < idx; i++)
3129 bh = bh->b_this_page;
3131 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3136 static int ext4_da_write_end(struct file *file,
3137 struct address_space *mapping,
3138 loff_t pos, unsigned len, unsigned copied,
3139 struct page *page, void *fsdata)
3141 struct inode *inode = mapping->host;
3143 handle_t *handle = ext4_journal_current_handle();
3145 unsigned long start, end;
3146 int write_mode = (int)(unsigned long)fsdata;
3148 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3149 if (ext4_should_order_data(inode)) {
3150 return ext4_ordered_write_end(file, mapping, pos,
3151 len, copied, page, fsdata);
3152 } else if (ext4_should_writeback_data(inode)) {
3153 return ext4_writeback_write_end(file, mapping, pos,
3154 len, copied, page, fsdata);
3160 trace_ext4_da_write_end(inode, pos, len, copied);
3161 start = pos & (PAGE_CACHE_SIZE - 1);
3162 end = start + copied - 1;
3165 * generic_write_end() will run mark_inode_dirty() if i_size
3166 * changes. So let's piggyback the i_disksize mark_inode_dirty
3170 new_i_size = pos + copied;
3171 if (new_i_size > EXT4_I(inode)->i_disksize) {
3172 if (ext4_da_should_update_i_disksize(page, end)) {
3173 down_write(&EXT4_I(inode)->i_data_sem);
3174 if (new_i_size > EXT4_I(inode)->i_disksize) {
3176 * Updating i_disksize when extending file
3177 * without needing block allocation
3179 if (ext4_should_order_data(inode))
3180 ret = ext4_jbd2_file_inode(handle,
3183 EXT4_I(inode)->i_disksize = new_i_size;
3185 up_write(&EXT4_I(inode)->i_data_sem);
3186 /* We need to mark inode dirty even if
3187 * new_i_size is less that inode->i_size
3188 * bu greater than i_disksize.(hint delalloc)
3190 ext4_mark_inode_dirty(handle, inode);
3193 ret2 = generic_write_end(file, mapping, pos, len, copied,
3198 ret2 = ext4_journal_stop(handle);
3202 return ret ? ret : copied;
3205 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3208 * Drop reserved blocks
3210 BUG_ON(!PageLocked(page));
3211 if (!page_has_buffers(page))
3214 ext4_da_page_release_reservation(page, offset);
3217 ext4_invalidatepage(page, offset);
3223 * Force all delayed allocation blocks to be allocated for a given inode.
3225 int ext4_alloc_da_blocks(struct inode *inode)
3227 trace_ext4_alloc_da_blocks(inode);
3229 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3230 !EXT4_I(inode)->i_reserved_meta_blocks)
3234 * We do something simple for now. The filemap_flush() will
3235 * also start triggering a write of the data blocks, which is
3236 * not strictly speaking necessary (and for users of
3237 * laptop_mode, not even desirable). However, to do otherwise
3238 * would require replicating code paths in:
3240 * ext4_da_writepages() ->
3241 * write_cache_pages() ---> (via passed in callback function)
3242 * __mpage_da_writepage() -->
3243 * mpage_add_bh_to_extent()
3244 * mpage_da_map_blocks()
3246 * The problem is that write_cache_pages(), located in
3247 * mm/page-writeback.c, marks pages clean in preparation for
3248 * doing I/O, which is not desirable if we're not planning on
3251 * We could call write_cache_pages(), and then redirty all of
3252 * the pages by calling redirty_page_for_writeback() but that
3253 * would be ugly in the extreme. So instead we would need to
3254 * replicate parts of the code in the above functions,
3255 * simplifying them becuase we wouldn't actually intend to
3256 * write out the pages, but rather only collect contiguous
3257 * logical block extents, call the multi-block allocator, and
3258 * then update the buffer heads with the block allocations.
3260 * For now, though, we'll cheat by calling filemap_flush(),
3261 * which will map the blocks, and start the I/O, but not
3262 * actually wait for the I/O to complete.
3264 return filemap_flush(inode->i_mapping);
3268 * bmap() is special. It gets used by applications such as lilo and by
3269 * the swapper to find the on-disk block of a specific piece of data.
3271 * Naturally, this is dangerous if the block concerned is still in the
3272 * journal. If somebody makes a swapfile on an ext4 data-journaling
3273 * filesystem and enables swap, then they may get a nasty shock when the
3274 * data getting swapped to that swapfile suddenly gets overwritten by
3275 * the original zero's written out previously to the journal and
3276 * awaiting writeback in the kernel's buffer cache.
3278 * So, if we see any bmap calls here on a modified, data-journaled file,
3279 * take extra steps to flush any blocks which might be in the cache.
3281 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3283 struct inode *inode = mapping->host;
3287 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3288 test_opt(inode->i_sb, DELALLOC)) {
3290 * With delalloc we want to sync the file
3291 * so that we can make sure we allocate
3294 filemap_write_and_wait(mapping);
3297 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3299 * This is a REALLY heavyweight approach, but the use of
3300 * bmap on dirty files is expected to be extremely rare:
3301 * only if we run lilo or swapon on a freshly made file
3302 * do we expect this to happen.
3304 * (bmap requires CAP_SYS_RAWIO so this does not
3305 * represent an unprivileged user DOS attack --- we'd be
3306 * in trouble if mortal users could trigger this path at
3309 * NB. EXT4_STATE_JDATA is not set on files other than
3310 * regular files. If somebody wants to bmap a directory
3311 * or symlink and gets confused because the buffer
3312 * hasn't yet been flushed to disk, they deserve
3313 * everything they get.
3316 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3317 journal = EXT4_JOURNAL(inode);
3318 jbd2_journal_lock_updates(journal);
3319 err = jbd2_journal_flush(journal);
3320 jbd2_journal_unlock_updates(journal);
3326 return generic_block_bmap(mapping, block, ext4_get_block);
3329 static int ext4_readpage(struct file *file, struct page *page)
3331 return mpage_readpage(page, ext4_get_block);
3335 ext4_readpages(struct file *file, struct address_space *mapping,
3336 struct list_head *pages, unsigned nr_pages)
3338 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3341 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3343 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3346 * If it's a full truncate we just forget about the pending dirtying
3349 ClearPageChecked(page);
3352 jbd2_journal_invalidatepage(journal, page, offset);
3354 block_invalidatepage(page, offset);
3357 static int ext4_releasepage(struct page *page, gfp_t wait)
3359 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3361 WARN_ON(PageChecked(page));
3362 if (!page_has_buffers(page))
3365 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3367 return try_to_free_buffers(page);
3371 * O_DIRECT for ext3 (or indirect map) based files
3373 * If the O_DIRECT write will extend the file then add this inode to the
3374 * orphan list. So recovery will truncate it back to the original size
3375 * if the machine crashes during the write.
3377 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3378 * crashes then stale disk data _may_ be exposed inside the file. But current
3379 * VFS code falls back into buffered path in that case so we are safe.
3381 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3382 const struct iovec *iov, loff_t offset,
3383 unsigned long nr_segs)
3385 struct file *file = iocb->ki_filp;
3386 struct inode *inode = file->f_mapping->host;
3387 struct ext4_inode_info *ei = EXT4_I(inode);
3391 size_t count = iov_length(iov, nr_segs);
3395 loff_t final_size = offset + count;
3397 if (final_size > inode->i_size) {
3398 /* Credits for sb + inode write */
3399 handle = ext4_journal_start(inode, 2);
3400 if (IS_ERR(handle)) {
3401 ret = PTR_ERR(handle);
3404 ret = ext4_orphan_add(handle, inode);
3406 ext4_journal_stop(handle);
3410 ei->i_disksize = inode->i_size;
3411 ext4_journal_stop(handle);
3416 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3418 ext4_get_block, NULL);
3419 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3425 /* Credits for sb + inode write */
3426 handle = ext4_journal_start(inode, 2);
3427 if (IS_ERR(handle)) {
3428 /* This is really bad luck. We've written the data
3429 * but cannot extend i_size. Bail out and pretend
3430 * the write failed... */
3431 ret = PTR_ERR(handle);
3435 ext4_orphan_del(handle, inode);
3437 loff_t end = offset + ret;
3438 if (end > inode->i_size) {
3439 ei->i_disksize = end;
3440 i_size_write(inode, end);
3442 * We're going to return a positive `ret'
3443 * here due to non-zero-length I/O, so there's
3444 * no way of reporting error returns from
3445 * ext4_mark_inode_dirty() to userspace. So
3448 ext4_mark_inode_dirty(handle, inode);
3451 err = ext4_journal_stop(handle);
3459 static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
3460 struct buffer_head *bh_result, int create)
3462 handle_t *handle = NULL;
3464 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3467 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3468 inode->i_ino, create);
3470 * DIO VFS code passes create = 0 flag for write to
3471 * the middle of file. It does this to avoid block
3472 * allocation for holes, to prevent expose stale data
3473 * out when there is parallel buffered read (which does
3474 * not hold the i_mutex lock) while direct IO write has
3475 * not completed. DIO request on holes finally falls back
3476 * to buffered IO for this reason.
3478 * For ext4 extent based file, since we support fallocate,
3479 * new allocated extent as uninitialized, for holes, we
3480 * could fallocate blocks for holes, thus parallel
3481 * buffered IO read will zero out the page when read on
3482 * a hole while parallel DIO write to the hole has not completed.
3484 * when we come here, we know it's a direct IO write to
3485 * to the middle of file (<i_size)
3486 * so it's safe to override the create flag from VFS.
3488 create = EXT4_GET_BLOCKS_DIO_CREATE_EXT;
3490 if (max_blocks > DIO_MAX_BLOCKS)
3491 max_blocks = DIO_MAX_BLOCKS;
3492 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3493 handle = ext4_journal_start(inode, dio_credits);
3494 if (IS_ERR(handle)) {
3495 ret = PTR_ERR(handle);
3498 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3501 bh_result->b_size = (ret << inode->i_blkbits);
3504 ext4_journal_stop(handle);
3509 static void ext4_free_io_end(ext4_io_end_t *io)
3515 static void dump_aio_dio_list(struct inode * inode)
3518 struct list_head *cur, *before, *after;
3519 ext4_io_end_t *io, *io0, *io1;
3521 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3522 ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
3526 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
3527 list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
3530 io0 = container_of(before, ext4_io_end_t, list);
3532 io1 = container_of(after, ext4_io_end_t, list);
3534 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3535 io, inode->i_ino, io0, io1);
3541 * check a range of space and convert unwritten extents to written.
3543 static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
3545 struct inode *inode = io->inode;
3546 loff_t offset = io->offset;
3547 size_t size = io->size;
3550 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3551 "list->prev 0x%p\n",
3552 io, inode->i_ino, io->list.next, io->list.prev);
3554 if (list_empty(&io->list))
3557 if (io->flag != DIO_AIO_UNWRITTEN)
3560 if (offset + size <= i_size_read(inode))
3561 ret = ext4_convert_unwritten_extents(inode, offset, size);
3564 printk(KERN_EMERG "%s: failed to convert unwritten"
3565 "extents to written extents, error is %d"
3566 " io is still on inode %lu aio dio list\n",
3567 __func__, ret, inode->i_ino);
3571 /* clear the DIO AIO unwritten flag */
3576 * work on completed aio dio IO, to convert unwritten extents to extents
3578 static void ext4_end_aio_dio_work(struct work_struct *work)
3580 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3581 struct inode *inode = io->inode;
3584 mutex_lock(&inode->i_mutex);
3585 ret = ext4_end_aio_dio_nolock(io);
3587 if (!list_empty(&io->list))
3588 list_del_init(&io->list);
3589 ext4_free_io_end(io);
3591 mutex_unlock(&inode->i_mutex);
3594 * This function is called from ext4_sync_file().
3596 * When AIO DIO IO is completed, the work to convert unwritten
3597 * extents to written is queued on workqueue but may not get immediately
3598 * scheduled. When fsync is called, we need to ensure the
3599 * conversion is complete before fsync returns.
3600 * The inode keeps track of a list of completed AIO from DIO path
3601 * that might needs to do the conversion. This function walks through
3602 * the list and convert the related unwritten extents to written.
3604 int flush_aio_dio_completed_IO(struct inode *inode)
3610 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
3613 dump_aio_dio_list(inode);
3614 while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3615 io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
3616 ext4_io_end_t, list);
3618 * Calling ext4_end_aio_dio_nolock() to convert completed
3621 * When ext4_sync_file() is called, run_queue() may already
3622 * about to flush the work corresponding to this io structure.
3623 * It will be upset if it founds the io structure related
3624 * to the work-to-be schedule is freed.
3626 * Thus we need to keep the io structure still valid here after
3627 * convertion finished. The io structure has a flag to
3628 * avoid double converting from both fsync and background work
3631 ret = ext4_end_aio_dio_nolock(io);
3635 list_del_init(&io->list);
3637 return (ret2 < 0) ? ret2 : 0;
3640 static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
3642 ext4_io_end_t *io = NULL;
3644 io = kmalloc(sizeof(*io), GFP_NOFS);
3653 INIT_WORK(&io->work, ext4_end_aio_dio_work);
3654 INIT_LIST_HEAD(&io->list);
3660 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3661 ssize_t size, void *private)
3663 ext4_io_end_t *io_end = iocb->private;
3664 struct workqueue_struct *wq;
3666 /* if not async direct IO or dio with 0 bytes write, just return */
3667 if (!io_end || !size)
3670 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3671 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3672 iocb->private, io_end->inode->i_ino, iocb, offset,
3675 /* if not aio dio with unwritten extents, just free io and return */
3676 if (io_end->flag != DIO_AIO_UNWRITTEN){
3677 ext4_free_io_end(io_end);
3678 iocb->private = NULL;
3682 io_end->offset = offset;
3683 io_end->size = size;
3684 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3686 /* queue the work to convert unwritten extents to written */
3687 queue_work(wq, &io_end->work);
3689 /* Add the io_end to per-inode completed aio dio list*/
3690 list_add_tail(&io_end->list,
3691 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
3692 iocb->private = NULL;
3695 * For ext4 extent files, ext4 will do direct-io write to holes,
3696 * preallocated extents, and those write extend the file, no need to
3697 * fall back to buffered IO.
3699 * For holes, we fallocate those blocks, mark them as unintialized
3700 * If those blocks were preallocated, we mark sure they are splited, but
3701 * still keep the range to write as unintialized.
3703 * The unwrritten extents will be converted to written when DIO is completed.
3704 * For async direct IO, since the IO may still pending when return, we
3705 * set up an end_io call back function, which will do the convertion
3706 * when async direct IO completed.
3708 * If the O_DIRECT write will extend the file then add this inode to the
3709 * orphan list. So recovery will truncate it back to the original size
3710 * if the machine crashes during the write.
3713 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3714 const struct iovec *iov, loff_t offset,
3715 unsigned long nr_segs)
3717 struct file *file = iocb->ki_filp;
3718 struct inode *inode = file->f_mapping->host;
3720 size_t count = iov_length(iov, nr_segs);
3722 loff_t final_size = offset + count;
3723 if (rw == WRITE && final_size <= inode->i_size) {
3725 * We could direct write to holes and fallocate.
3727 * Allocated blocks to fill the hole are marked as uninitialized
3728 * to prevent paralel buffered read to expose the stale data
3729 * before DIO complete the data IO.
3731 * As to previously fallocated extents, ext4 get_block
3732 * will just simply mark the buffer mapped but still
3733 * keep the extents uninitialized.
3735 * for non AIO case, we will convert those unwritten extents
3736 * to written after return back from blockdev_direct_IO.
3738 * for async DIO, the conversion needs to be defered when
3739 * the IO is completed. The ext4 end_io callback function
3740 * will be called to take care of the conversion work.
3741 * Here for async case, we allocate an io_end structure to
3744 iocb->private = NULL;
3745 EXT4_I(inode)->cur_aio_dio = NULL;
3746 if (!is_sync_kiocb(iocb)) {
3747 iocb->private = ext4_init_io_end(inode);
3751 * we save the io structure for current async
3752 * direct IO, so that later ext4_get_blocks()
3753 * could flag the io structure whether there
3754 * is a unwritten extents needs to be converted
3755 * when IO is completed.
3757 EXT4_I(inode)->cur_aio_dio = iocb->private;
3760 ret = blockdev_direct_IO(rw, iocb, inode,
3761 inode->i_sb->s_bdev, iov,
3763 ext4_get_block_dio_write,
3766 EXT4_I(inode)->cur_aio_dio = NULL;
3768 * The io_end structure takes a reference to the inode,
3769 * that structure needs to be destroyed and the
3770 * reference to the inode need to be dropped, when IO is
3771 * complete, even with 0 byte write, or failed.
3773 * In the successful AIO DIO case, the io_end structure will be
3774 * desctroyed and the reference to the inode will be dropped
3775 * after the end_io call back function is called.
3777 * In the case there is 0 byte write, or error case, since
3778 * VFS direct IO won't invoke the end_io call back function,
3779 * we need to free the end_io structure here.
3781 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3782 ext4_free_io_end(iocb->private);
3783 iocb->private = NULL;
3784 } else if (ret > 0 && (EXT4_I(inode)->i_state &
3785 EXT4_STATE_DIO_UNWRITTEN)) {
3788 * for non AIO case, since the IO is already
3789 * completed, we could do the convertion right here
3791 err = ext4_convert_unwritten_extents(inode,
3795 EXT4_I(inode)->i_state &= ~EXT4_STATE_DIO_UNWRITTEN;
3800 /* for write the the end of file case, we fall back to old way */
3801 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3804 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3805 const struct iovec *iov, loff_t offset,
3806 unsigned long nr_segs)
3808 struct file *file = iocb->ki_filp;
3809 struct inode *inode = file->f_mapping->host;
3811 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3812 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3814 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3818 * Pages can be marked dirty completely asynchronously from ext4's journalling
3819 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3820 * much here because ->set_page_dirty is called under VFS locks. The page is
3821 * not necessarily locked.
3823 * We cannot just dirty the page and leave attached buffers clean, because the
3824 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3825 * or jbddirty because all the journalling code will explode.
3827 * So what we do is to mark the page "pending dirty" and next time writepage
3828 * is called, propagate that into the buffers appropriately.
3830 static int ext4_journalled_set_page_dirty(struct page *page)
3832 SetPageChecked(page);
3833 return __set_page_dirty_nobuffers(page);
3836 static const struct address_space_operations ext4_ordered_aops = {
3837 .readpage = ext4_readpage,
3838 .readpages = ext4_readpages,
3839 .writepage = ext4_writepage,
3840 .sync_page = block_sync_page,
3841 .write_begin = ext4_write_begin,
3842 .write_end = ext4_ordered_write_end,
3844 .invalidatepage = ext4_invalidatepage,
3845 .releasepage = ext4_releasepage,
3846 .direct_IO = ext4_direct_IO,
3847 .migratepage = buffer_migrate_page,
3848 .is_partially_uptodate = block_is_partially_uptodate,
3849 .error_remove_page = generic_error_remove_page,
3852 static const struct address_space_operations ext4_writeback_aops = {
3853 .readpage = ext4_readpage,
3854 .readpages = ext4_readpages,
3855 .writepage = ext4_writepage,
3856 .sync_page = block_sync_page,
3857 .write_begin = ext4_write_begin,
3858 .write_end = ext4_writeback_write_end,
3860 .invalidatepage = ext4_invalidatepage,
3861 .releasepage = ext4_releasepage,
3862 .direct_IO = ext4_direct_IO,
3863 .migratepage = buffer_migrate_page,
3864 .is_partially_uptodate = block_is_partially_uptodate,
3865 .error_remove_page = generic_error_remove_page,
3868 static const struct address_space_operations ext4_journalled_aops = {
3869 .readpage = ext4_readpage,
3870 .readpages = ext4_readpages,
3871 .writepage = ext4_writepage,
3872 .sync_page = block_sync_page,
3873 .write_begin = ext4_write_begin,
3874 .write_end = ext4_journalled_write_end,
3875 .set_page_dirty = ext4_journalled_set_page_dirty,
3877 .invalidatepage = ext4_invalidatepage,
3878 .releasepage = ext4_releasepage,
3879 .is_partially_uptodate = block_is_partially_uptodate,
3880 .error_remove_page = generic_error_remove_page,
3883 static const struct address_space_operations ext4_da_aops = {
3884 .readpage = ext4_readpage,
3885 .readpages = ext4_readpages,
3886 .writepage = ext4_writepage,
3887 .writepages = ext4_da_writepages,
3888 .sync_page = block_sync_page,
3889 .write_begin = ext4_da_write_begin,
3890 .write_end = ext4_da_write_end,
3892 .invalidatepage = ext4_da_invalidatepage,
3893 .releasepage = ext4_releasepage,
3894 .direct_IO = ext4_direct_IO,
3895 .migratepage = buffer_migrate_page,
3896 .is_partially_uptodate = block_is_partially_uptodate,
3897 .error_remove_page = generic_error_remove_page,
3900 void ext4_set_aops(struct inode *inode)
3902 if (ext4_should_order_data(inode) &&
3903 test_opt(inode->i_sb, DELALLOC))
3904 inode->i_mapping->a_ops = &ext4_da_aops;
3905 else if (ext4_should_order_data(inode))
3906 inode->i_mapping->a_ops = &ext4_ordered_aops;
3907 else if (ext4_should_writeback_data(inode) &&
3908 test_opt(inode->i_sb, DELALLOC))
3909 inode->i_mapping->a_ops = &ext4_da_aops;
3910 else if (ext4_should_writeback_data(inode))
3911 inode->i_mapping->a_ops = &ext4_writeback_aops;
3913 inode->i_mapping->a_ops = &ext4_journalled_aops;
3917 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3918 * up to the end of the block which corresponds to `from'.
3919 * This required during truncate. We need to physically zero the tail end
3920 * of that block so it doesn't yield old data if the file is later grown.
3922 int ext4_block_truncate_page(handle_t *handle,
3923 struct address_space *mapping, loff_t from)
3925 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3926 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3927 unsigned blocksize, length, pos;
3929 struct inode *inode = mapping->host;
3930 struct buffer_head *bh;
3934 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3935 mapping_gfp_mask(mapping) & ~__GFP_FS);
3939 blocksize = inode->i_sb->s_blocksize;
3940 length = blocksize - (offset & (blocksize - 1));
3941 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3944 * For "nobh" option, we can only work if we don't need to
3945 * read-in the page - otherwise we create buffers to do the IO.
3947 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3948 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3949 zero_user(page, offset, length);
3950 set_page_dirty(page);
3954 if (!page_has_buffers(page))
3955 create_empty_buffers(page, blocksize, 0);
3957 /* Find the buffer that contains "offset" */
3958 bh = page_buffers(page);
3960 while (offset >= pos) {
3961 bh = bh->b_this_page;
3967 if (buffer_freed(bh)) {
3968 BUFFER_TRACE(bh, "freed: skip");
3972 if (!buffer_mapped(bh)) {
3973 BUFFER_TRACE(bh, "unmapped");
3974 ext4_get_block(inode, iblock, bh, 0);
3975 /* unmapped? It's a hole - nothing to do */
3976 if (!buffer_mapped(bh)) {
3977 BUFFER_TRACE(bh, "still unmapped");
3982 /* Ok, it's mapped. Make sure it's up-to-date */
3983 if (PageUptodate(page))
3984 set_buffer_uptodate(bh);
3986 if (!buffer_uptodate(bh)) {
3988 ll_rw_block(READ, 1, &bh);
3990 /* Uhhuh. Read error. Complain and punt. */
3991 if (!buffer_uptodate(bh))
3995 if (ext4_should_journal_data(inode)) {
3996 BUFFER_TRACE(bh, "get write access");
3997 err = ext4_journal_get_write_access(handle, bh);
4002 zero_user(page, offset, length);
4004 BUFFER_TRACE(bh, "zeroed end of block");
4007 if (ext4_should_journal_data(inode)) {
4008 err = ext4_handle_dirty_metadata(handle, inode, bh);
4010 if (ext4_should_order_data(inode))
4011 err = ext4_jbd2_file_inode(handle, inode);
4012 mark_buffer_dirty(bh);
4017 page_cache_release(page);
4022 * Probably it should be a library function... search for first non-zero word
4023 * or memcmp with zero_page, whatever is better for particular architecture.
4026 static inline int all_zeroes(__le32 *p, __le32 *q)
4035 * ext4_find_shared - find the indirect blocks for partial truncation.
4036 * @inode: inode in question
4037 * @depth: depth of the affected branch
4038 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4039 * @chain: place to store the pointers to partial indirect blocks
4040 * @top: place to the (detached) top of branch
4042 * This is a helper function used by ext4_truncate().
4044 * When we do truncate() we may have to clean the ends of several
4045 * indirect blocks but leave the blocks themselves alive. Block is
4046 * partially truncated if some data below the new i_size is refered
4047 * from it (and it is on the path to the first completely truncated
4048 * data block, indeed). We have to free the top of that path along
4049 * with everything to the right of the path. Since no allocation
4050 * past the truncation point is possible until ext4_truncate()
4051 * finishes, we may safely do the latter, but top of branch may
4052 * require special attention - pageout below the truncation point
4053 * might try to populate it.
4055 * We atomically detach the top of branch from the tree, store the
4056 * block number of its root in *@top, pointers to buffer_heads of
4057 * partially truncated blocks - in @chain[].bh and pointers to
4058 * their last elements that should not be removed - in
4059 * @chain[].p. Return value is the pointer to last filled element
4062 * The work left to caller to do the actual freeing of subtrees:
4063 * a) free the subtree starting from *@top
4064 * b) free the subtrees whose roots are stored in
4065 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4066 * c) free the subtrees growing from the inode past the @chain[0].
4067 * (no partially truncated stuff there). */
4069 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4070 ext4_lblk_t offsets[4], Indirect chain[4],
4073 Indirect *partial, *p;
4077 /* Make k index the deepest non-null offest + 1 */
4078 for (k = depth; k > 1 && !offsets[k-1]; k--)
4080 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4081 /* Writer: pointers */
4083 partial = chain + k-1;
4085 * If the branch acquired continuation since we've looked at it -
4086 * fine, it should all survive and (new) top doesn't belong to us.
4088 if (!partial->key && *partial->p)
4091 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4094 * OK, we've found the last block that must survive. The rest of our
4095 * branch should be detached before unlocking. However, if that rest
4096 * of branch is all ours and does not grow immediately from the inode
4097 * it's easier to cheat and just decrement partial->p.
4099 if (p == chain + k - 1 && p > chain) {
4103 /* Nope, don't do this in ext4. Must leave the tree intact */
4110 while (partial > p) {
4111 brelse(partial->bh);
4119 * Zero a number of block pointers in either an inode or an indirect block.
4120 * If we restart the transaction we must again get write access to the
4121 * indirect block for further modification.
4123 * We release `count' blocks on disk, but (last - first) may be greater
4124 * than `count' because there can be holes in there.
4126 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
4127 struct buffer_head *bh,
4128 ext4_fsblk_t block_to_free,
4129 unsigned long count, __le32 *first,
4133 int is_metadata = S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode);
4135 if (try_to_extend_transaction(handle, inode)) {
4137 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4138 ext4_handle_dirty_metadata(handle, inode, bh);
4140 ext4_mark_inode_dirty(handle, inode);
4141 ext4_truncate_restart_trans(handle, inode,
4142 blocks_for_truncate(inode));
4144 BUFFER_TRACE(bh, "retaking write access");
4145 ext4_journal_get_write_access(handle, bh);
4150 * Any buffers which are on the journal will be in memory. We
4151 * find them on the hash table so jbd2_journal_revoke() will
4152 * run jbd2_journal_forget() on them. We've already detached
4153 * each block from the file, so bforget() in
4154 * jbd2_journal_forget() should be safe.
4156 * AKPM: turn on bforget in jbd2_journal_forget()!!!
4158 for (p = first; p < last; p++) {
4159 u32 nr = le32_to_cpu(*p);
4161 struct buffer_head *tbh;
4164 tbh = sb_find_get_block(inode->i_sb, nr);
4165 ext4_forget(handle, is_metadata, inode, tbh, nr);
4169 ext4_free_blocks(handle, inode, block_to_free, count, is_metadata);
4173 * ext4_free_data - free a list of data blocks
4174 * @handle: handle for this transaction
4175 * @inode: inode we are dealing with
4176 * @this_bh: indirect buffer_head which contains *@first and *@last
4177 * @first: array of block numbers
4178 * @last: points immediately past the end of array
4180 * We are freeing all blocks refered from that array (numbers are stored as
4181 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4183 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4184 * blocks are contiguous then releasing them at one time will only affect one
4185 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4186 * actually use a lot of journal space.
4188 * @this_bh will be %NULL if @first and @last point into the inode's direct
4191 static void ext4_free_data(handle_t *handle, struct inode *inode,
4192 struct buffer_head *this_bh,
4193 __le32 *first, __le32 *last)
4195 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4196 unsigned long count = 0; /* Number of blocks in the run */
4197 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4200 ext4_fsblk_t nr; /* Current block # */
4201 __le32 *p; /* Pointer into inode/ind
4202 for current block */
4205 if (this_bh) { /* For indirect block */
4206 BUFFER_TRACE(this_bh, "get_write_access");
4207 err = ext4_journal_get_write_access(handle, this_bh);
4208 /* Important: if we can't update the indirect pointers
4209 * to the blocks, we can't free them. */
4214 for (p = first; p < last; p++) {
4215 nr = le32_to_cpu(*p);
4217 /* accumulate blocks to free if they're contiguous */
4220 block_to_free_p = p;
4222 } else if (nr == block_to_free + count) {
4225 ext4_clear_blocks(handle, inode, this_bh,
4227 count, block_to_free_p, p);
4229 block_to_free_p = p;
4236 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4237 count, block_to_free_p, p);
4240 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4243 * The buffer head should have an attached journal head at this
4244 * point. However, if the data is corrupted and an indirect
4245 * block pointed to itself, it would have been detached when
4246 * the block was cleared. Check for this instead of OOPSing.
4248 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4249 ext4_handle_dirty_metadata(handle, inode, this_bh);
4251 ext4_error(inode->i_sb, __func__,
4252 "circular indirect block detected, "
4253 "inode=%lu, block=%llu",
4255 (unsigned long long) this_bh->b_blocknr);
4260 * ext4_free_branches - free an array of branches
4261 * @handle: JBD handle for this transaction
4262 * @inode: inode we are dealing with
4263 * @parent_bh: the buffer_head which contains *@first and *@last
4264 * @first: array of block numbers
4265 * @last: pointer immediately past the end of array
4266 * @depth: depth of the branches to free
4268 * We are freeing all blocks refered from these branches (numbers are
4269 * stored as little-endian 32-bit) and updating @inode->i_blocks
4272 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4273 struct buffer_head *parent_bh,
4274 __le32 *first, __le32 *last, int depth)
4279 if (ext4_handle_is_aborted(handle))
4283 struct buffer_head *bh;
4284 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4286 while (--p >= first) {
4287 nr = le32_to_cpu(*p);
4289 continue; /* A hole */
4291 /* Go read the buffer for the next level down */
4292 bh = sb_bread(inode->i_sb, nr);
4295 * A read failure? Report error and clear slot
4299 ext4_error(inode->i_sb, "ext4_free_branches",
4300 "Read failure, inode=%lu, block=%llu",
4305 /* This zaps the entire block. Bottom up. */
4306 BUFFER_TRACE(bh, "free child branches");
4307 ext4_free_branches(handle, inode, bh,
4308 (__le32 *) bh->b_data,
4309 (__le32 *) bh->b_data + addr_per_block,
4313 * We've probably journalled the indirect block several
4314 * times during the truncate. But it's no longer
4315 * needed and we now drop it from the transaction via
4316 * jbd2_journal_revoke().
4318 * That's easy if it's exclusively part of this
4319 * transaction. But if it's part of the committing
4320 * transaction then jbd2_journal_forget() will simply
4321 * brelse() it. That means that if the underlying
4322 * block is reallocated in ext4_get_block(),
4323 * unmap_underlying_metadata() will find this block
4324 * and will try to get rid of it. damn, damn.
4326 * If this block has already been committed to the
4327 * journal, a revoke record will be written. And
4328 * revoke records must be emitted *before* clearing
4329 * this block's bit in the bitmaps.
4331 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4334 * Everything below this this pointer has been
4335 * released. Now let this top-of-subtree go.
4337 * We want the freeing of this indirect block to be
4338 * atomic in the journal with the updating of the
4339 * bitmap block which owns it. So make some room in
4342 * We zero the parent pointer *after* freeing its
4343 * pointee in the bitmaps, so if extend_transaction()
4344 * for some reason fails to put the bitmap changes and
4345 * the release into the same transaction, recovery
4346 * will merely complain about releasing a free block,
4347 * rather than leaking blocks.
4349 if (ext4_handle_is_aborted(handle))
4351 if (try_to_extend_transaction(handle, inode)) {
4352 ext4_mark_inode_dirty(handle, inode);
4353 ext4_truncate_restart_trans(handle, inode,
4354 blocks_for_truncate(inode));
4357 ext4_free_blocks(handle, inode, nr, 1, 1);
4361 * The block which we have just freed is
4362 * pointed to by an indirect block: journal it
4364 BUFFER_TRACE(parent_bh, "get_write_access");
4365 if (!ext4_journal_get_write_access(handle,
4368 BUFFER_TRACE(parent_bh,
4369 "call ext4_handle_dirty_metadata");
4370 ext4_handle_dirty_metadata(handle,
4377 /* We have reached the bottom of the tree. */
4378 BUFFER_TRACE(parent_bh, "free data blocks");
4379 ext4_free_data(handle, inode, parent_bh, first, last);
4383 int ext4_can_truncate(struct inode *inode)
4385 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4387 if (S_ISREG(inode->i_mode))
4389 if (S_ISDIR(inode->i_mode))
4391 if (S_ISLNK(inode->i_mode))
4392 return !ext4_inode_is_fast_symlink(inode);
4399 * We block out ext4_get_block() block instantiations across the entire
4400 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4401 * simultaneously on behalf of the same inode.
4403 * As we work through the truncate and commmit bits of it to the journal there
4404 * is one core, guiding principle: the file's tree must always be consistent on
4405 * disk. We must be able to restart the truncate after a crash.
4407 * The file's tree may be transiently inconsistent in memory (although it
4408 * probably isn't), but whenever we close off and commit a journal transaction,
4409 * the contents of (the filesystem + the journal) must be consistent and
4410 * restartable. It's pretty simple, really: bottom up, right to left (although
4411 * left-to-right works OK too).
4413 * Note that at recovery time, journal replay occurs *before* the restart of
4414 * truncate against the orphan inode list.
4416 * The committed inode has the new, desired i_size (which is the same as
4417 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4418 * that this inode's truncate did not complete and it will again call
4419 * ext4_truncate() to have another go. So there will be instantiated blocks
4420 * to the right of the truncation point in a crashed ext4 filesystem. But
4421 * that's fine - as long as they are linked from the inode, the post-crash
4422 * ext4_truncate() run will find them and release them.
4424 void ext4_truncate(struct inode *inode)
4427 struct ext4_inode_info *ei = EXT4_I(inode);
4428 __le32 *i_data = ei->i_data;
4429 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4430 struct address_space *mapping = inode->i_mapping;
4431 ext4_lblk_t offsets[4];
4436 ext4_lblk_t last_block;
4437 unsigned blocksize = inode->i_sb->s_blocksize;
4439 if (!ext4_can_truncate(inode))
4442 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4443 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4445 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4446 ext4_ext_truncate(inode);
4450 handle = start_transaction(inode);
4452 return; /* AKPM: return what? */
4454 last_block = (inode->i_size + blocksize-1)
4455 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4457 if (inode->i_size & (blocksize - 1))
4458 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4461 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4463 goto out_stop; /* error */
4466 * OK. This truncate is going to happen. We add the inode to the
4467 * orphan list, so that if this truncate spans multiple transactions,
4468 * and we crash, we will resume the truncate when the filesystem
4469 * recovers. It also marks the inode dirty, to catch the new size.
4471 * Implication: the file must always be in a sane, consistent
4472 * truncatable state while each transaction commits.
4474 if (ext4_orphan_add(handle, inode))
4478 * From here we block out all ext4_get_block() callers who want to
4479 * modify the block allocation tree.
4481 down_write(&ei->i_data_sem);
4483 ext4_discard_preallocations(inode);
4486 * The orphan list entry will now protect us from any crash which
4487 * occurs before the truncate completes, so it is now safe to propagate
4488 * the new, shorter inode size (held for now in i_size) into the
4489 * on-disk inode. We do this via i_disksize, which is the value which
4490 * ext4 *really* writes onto the disk inode.
4492 ei->i_disksize = inode->i_size;
4494 if (n == 1) { /* direct blocks */
4495 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4496 i_data + EXT4_NDIR_BLOCKS);
4500 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4501 /* Kill the top of shared branch (not detached) */
4503 if (partial == chain) {
4504 /* Shared branch grows from the inode */
4505 ext4_free_branches(handle, inode, NULL,
4506 &nr, &nr+1, (chain+n-1) - partial);
4509 * We mark the inode dirty prior to restart,
4510 * and prior to stop. No need for it here.
4513 /* Shared branch grows from an indirect block */
4514 BUFFER_TRACE(partial->bh, "get_write_access");
4515 ext4_free_branches(handle, inode, partial->bh,
4517 partial->p+1, (chain+n-1) - partial);
4520 /* Clear the ends of indirect blocks on the shared branch */
4521 while (partial > chain) {
4522 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4523 (__le32*)partial->bh->b_data+addr_per_block,
4524 (chain+n-1) - partial);
4525 BUFFER_TRACE(partial->bh, "call brelse");
4526 brelse(partial->bh);
4530 /* Kill the remaining (whole) subtrees */
4531 switch (offsets[0]) {
4533 nr = i_data[EXT4_IND_BLOCK];
4535 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4536 i_data[EXT4_IND_BLOCK] = 0;
4538 case EXT4_IND_BLOCK:
4539 nr = i_data[EXT4_DIND_BLOCK];
4541 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4542 i_data[EXT4_DIND_BLOCK] = 0;
4544 case EXT4_DIND_BLOCK:
4545 nr = i_data[EXT4_TIND_BLOCK];
4547 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4548 i_data[EXT4_TIND_BLOCK] = 0;
4550 case EXT4_TIND_BLOCK:
4554 up_write(&ei->i_data_sem);
4555 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4556 ext4_mark_inode_dirty(handle, inode);
4559 * In a multi-transaction truncate, we only make the final transaction
4563 ext4_handle_sync(handle);
4566 * If this was a simple ftruncate(), and the file will remain alive
4567 * then we need to clear up the orphan record which we created above.
4568 * However, if this was a real unlink then we were called by
4569 * ext4_delete_inode(), and we allow that function to clean up the
4570 * orphan info for us.
4573 ext4_orphan_del(handle, inode);
4575 ext4_journal_stop(handle);
4579 * ext4_get_inode_loc returns with an extra refcount against the inode's
4580 * underlying buffer_head on success. If 'in_mem' is true, we have all
4581 * data in memory that is needed to recreate the on-disk version of this
4584 static int __ext4_get_inode_loc(struct inode *inode,
4585 struct ext4_iloc *iloc, int in_mem)
4587 struct ext4_group_desc *gdp;
4588 struct buffer_head *bh;
4589 struct super_block *sb = inode->i_sb;
4591 int inodes_per_block, inode_offset;
4594 if (!ext4_valid_inum(sb, inode->i_ino))
4597 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4598 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4603 * Figure out the offset within the block group inode table
4605 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4606 inode_offset = ((inode->i_ino - 1) %
4607 EXT4_INODES_PER_GROUP(sb));
4608 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4609 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4611 bh = sb_getblk(sb, block);
4613 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4614 "inode block - inode=%lu, block=%llu",
4615 inode->i_ino, block);
4618 if (!buffer_uptodate(bh)) {
4622 * If the buffer has the write error flag, we have failed
4623 * to write out another inode in the same block. In this
4624 * case, we don't have to read the block because we may
4625 * read the old inode data successfully.
4627 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4628 set_buffer_uptodate(bh);
4630 if (buffer_uptodate(bh)) {
4631 /* someone brought it uptodate while we waited */
4637 * If we have all information of the inode in memory and this
4638 * is the only valid inode in the block, we need not read the
4642 struct buffer_head *bitmap_bh;
4645 start = inode_offset & ~(inodes_per_block - 1);
4647 /* Is the inode bitmap in cache? */
4648 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4653 * If the inode bitmap isn't in cache then the
4654 * optimisation may end up performing two reads instead
4655 * of one, so skip it.
4657 if (!buffer_uptodate(bitmap_bh)) {
4661 for (i = start; i < start + inodes_per_block; i++) {
4662 if (i == inode_offset)
4664 if (ext4_test_bit(i, bitmap_bh->b_data))
4668 if (i == start + inodes_per_block) {
4669 /* all other inodes are free, so skip I/O */
4670 memset(bh->b_data, 0, bh->b_size);
4671 set_buffer_uptodate(bh);
4679 * If we need to do any I/O, try to pre-readahead extra
4680 * blocks from the inode table.
4682 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4683 ext4_fsblk_t b, end, table;
4686 table = ext4_inode_table(sb, gdp);
4687 /* s_inode_readahead_blks is always a power of 2 */
4688 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4691 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4692 num = EXT4_INODES_PER_GROUP(sb);
4693 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4694 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4695 num -= ext4_itable_unused_count(sb, gdp);
4696 table += num / inodes_per_block;
4700 sb_breadahead(sb, b++);
4704 * There are other valid inodes in the buffer, this inode
4705 * has in-inode xattrs, or we don't have this inode in memory.
4706 * Read the block from disk.
4709 bh->b_end_io = end_buffer_read_sync;
4710 submit_bh(READ_META, bh);
4712 if (!buffer_uptodate(bh)) {
4713 ext4_error(sb, __func__,
4714 "unable to read inode block - inode=%lu, "
4715 "block=%llu", inode->i_ino, block);
4725 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4727 /* We have all inode data except xattrs in memory here. */
4728 return __ext4_get_inode_loc(inode, iloc,
4729 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4732 void ext4_set_inode_flags(struct inode *inode)
4734 unsigned int flags = EXT4_I(inode)->i_flags;
4736 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4737 if (flags & EXT4_SYNC_FL)
4738 inode->i_flags |= S_SYNC;
4739 if (flags & EXT4_APPEND_FL)
4740 inode->i_flags |= S_APPEND;
4741 if (flags & EXT4_IMMUTABLE_FL)
4742 inode->i_flags |= S_IMMUTABLE;
4743 if (flags & EXT4_NOATIME_FL)
4744 inode->i_flags |= S_NOATIME;
4745 if (flags & EXT4_DIRSYNC_FL)
4746 inode->i_flags |= S_DIRSYNC;
4749 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4750 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4752 unsigned int flags = ei->vfs_inode.i_flags;
4754 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4755 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4757 ei->i_flags |= EXT4_SYNC_FL;
4758 if (flags & S_APPEND)
4759 ei->i_flags |= EXT4_APPEND_FL;
4760 if (flags & S_IMMUTABLE)
4761 ei->i_flags |= EXT4_IMMUTABLE_FL;
4762 if (flags & S_NOATIME)
4763 ei->i_flags |= EXT4_NOATIME_FL;
4764 if (flags & S_DIRSYNC)
4765 ei->i_flags |= EXT4_DIRSYNC_FL;
4768 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4769 struct ext4_inode_info *ei)
4772 struct inode *inode = &(ei->vfs_inode);
4773 struct super_block *sb = inode->i_sb;
4775 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4776 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4777 /* we are using combined 48 bit field */
4778 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4779 le32_to_cpu(raw_inode->i_blocks_lo);
4780 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4781 /* i_blocks represent file system block size */
4782 return i_blocks << (inode->i_blkbits - 9);
4787 return le32_to_cpu(raw_inode->i_blocks_lo);
4791 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4793 struct ext4_iloc iloc;
4794 struct ext4_inode *raw_inode;
4795 struct ext4_inode_info *ei;
4796 struct inode *inode;
4800 inode = iget_locked(sb, ino);
4802 return ERR_PTR(-ENOMEM);
4803 if (!(inode->i_state & I_NEW))
4809 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4812 raw_inode = ext4_raw_inode(&iloc);
4813 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4814 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4815 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4816 if (!(test_opt(inode->i_sb, NO_UID32))) {
4817 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4818 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4820 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4823 ei->i_dir_start_lookup = 0;
4824 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4825 /* We now have enough fields to check if the inode was active or not.
4826 * This is needed because nfsd might try to access dead inodes
4827 * the test is that same one that e2fsck uses
4828 * NeilBrown 1999oct15
4830 if (inode->i_nlink == 0) {
4831 if (inode->i_mode == 0 ||
4832 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4833 /* this inode is deleted */
4837 /* The only unlinked inodes we let through here have
4838 * valid i_mode and are being read by the orphan
4839 * recovery code: that's fine, we're about to complete
4840 * the process of deleting those. */
4842 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4843 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4844 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4845 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4847 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4848 inode->i_size = ext4_isize(raw_inode);
4849 ei->i_disksize = inode->i_size;
4850 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4851 ei->i_block_group = iloc.block_group;
4852 ei->i_last_alloc_group = ~0;
4854 * NOTE! The in-memory inode i_data array is in little-endian order
4855 * even on big-endian machines: we do NOT byteswap the block numbers!
4857 for (block = 0; block < EXT4_N_BLOCKS; block++)
4858 ei->i_data[block] = raw_inode->i_block[block];
4859 INIT_LIST_HEAD(&ei->i_orphan);
4861 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4862 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4863 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4864 EXT4_INODE_SIZE(inode->i_sb)) {
4868 if (ei->i_extra_isize == 0) {
4869 /* The extra space is currently unused. Use it. */
4870 ei->i_extra_isize = sizeof(struct ext4_inode) -
4871 EXT4_GOOD_OLD_INODE_SIZE;
4873 __le32 *magic = (void *)raw_inode +
4874 EXT4_GOOD_OLD_INODE_SIZE +
4876 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4877 ei->i_state |= EXT4_STATE_XATTR;
4880 ei->i_extra_isize = 0;
4882 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4883 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4884 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4885 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4887 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4888 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4889 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4891 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4895 if (ei->i_file_acl &&
4896 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4897 ext4_error(sb, __func__,
4898 "bad extended attribute block %llu in inode #%lu",
4899 ei->i_file_acl, inode->i_ino);
4902 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4903 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4904 (S_ISLNK(inode->i_mode) &&
4905 !ext4_inode_is_fast_symlink(inode)))
4906 /* Validate extent which is part of inode */
4907 ret = ext4_ext_check_inode(inode);
4908 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4909 (S_ISLNK(inode->i_mode) &&
4910 !ext4_inode_is_fast_symlink(inode))) {
4911 /* Validate block references which are part of inode */
4912 ret = ext4_check_inode_blockref(inode);
4917 if (S_ISREG(inode->i_mode)) {
4918 inode->i_op = &ext4_file_inode_operations;
4919 inode->i_fop = &ext4_file_operations;
4920 ext4_set_aops(inode);
4921 } else if (S_ISDIR(inode->i_mode)) {
4922 inode->i_op = &ext4_dir_inode_operations;
4923 inode->i_fop = &ext4_dir_operations;
4924 } else if (S_ISLNK(inode->i_mode)) {
4925 if (ext4_inode_is_fast_symlink(inode)) {
4926 inode->i_op = &ext4_fast_symlink_inode_operations;
4927 nd_terminate_link(ei->i_data, inode->i_size,
4928 sizeof(ei->i_data) - 1);
4930 inode->i_op = &ext4_symlink_inode_operations;
4931 ext4_set_aops(inode);
4933 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4934 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4935 inode->i_op = &ext4_special_inode_operations;
4936 if (raw_inode->i_block[0])
4937 init_special_inode(inode, inode->i_mode,
4938 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4940 init_special_inode(inode, inode->i_mode,
4941 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4944 ext4_error(inode->i_sb, __func__,
4945 "bogus i_mode (%o) for inode=%lu",
4946 inode->i_mode, inode->i_ino);
4950 ext4_set_inode_flags(inode);
4951 unlock_new_inode(inode);
4957 return ERR_PTR(ret);
4960 static int ext4_inode_blocks_set(handle_t *handle,
4961 struct ext4_inode *raw_inode,
4962 struct ext4_inode_info *ei)
4964 struct inode *inode = &(ei->vfs_inode);
4965 u64 i_blocks = inode->i_blocks;
4966 struct super_block *sb = inode->i_sb;
4968 if (i_blocks <= ~0U) {
4970 * i_blocks can be represnted in a 32 bit variable
4971 * as multiple of 512 bytes
4973 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4974 raw_inode->i_blocks_high = 0;
4975 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4978 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4981 if (i_blocks <= 0xffffffffffffULL) {
4983 * i_blocks can be represented in a 48 bit variable
4984 * as multiple of 512 bytes
4986 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4987 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4988 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4990 ei->i_flags |= EXT4_HUGE_FILE_FL;
4991 /* i_block is stored in file system block size */
4992 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4993 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4994 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5000 * Post the struct inode info into an on-disk inode location in the
5001 * buffer-cache. This gobbles the caller's reference to the
5002 * buffer_head in the inode location struct.
5004 * The caller must have write access to iloc->bh.
5006 static int ext4_do_update_inode(handle_t *handle,
5007 struct inode *inode,
5008 struct ext4_iloc *iloc)
5010 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5011 struct ext4_inode_info *ei = EXT4_I(inode);
5012 struct buffer_head *bh = iloc->bh;
5013 int err = 0, rc, block;
5015 /* For fields not not tracking in the in-memory inode,
5016 * initialise them to zero for new inodes. */
5017 if (ei->i_state & EXT4_STATE_NEW)
5018 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5020 ext4_get_inode_flags(ei);
5021 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5022 if (!(test_opt(inode->i_sb, NO_UID32))) {
5023 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5024 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5026 * Fix up interoperability with old kernels. Otherwise, old inodes get
5027 * re-used with the upper 16 bits of the uid/gid intact
5030 raw_inode->i_uid_high =
5031 cpu_to_le16(high_16_bits(inode->i_uid));
5032 raw_inode->i_gid_high =
5033 cpu_to_le16(high_16_bits(inode->i_gid));
5035 raw_inode->i_uid_high = 0;
5036 raw_inode->i_gid_high = 0;
5039 raw_inode->i_uid_low =
5040 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5041 raw_inode->i_gid_low =
5042 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5043 raw_inode->i_uid_high = 0;
5044 raw_inode->i_gid_high = 0;
5046 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5048 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5049 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5050 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5051 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5053 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5055 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5056 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5057 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5058 cpu_to_le32(EXT4_OS_HURD))
5059 raw_inode->i_file_acl_high =
5060 cpu_to_le16(ei->i_file_acl >> 32);
5061 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5062 ext4_isize_set(raw_inode, ei->i_disksize);
5063 if (ei->i_disksize > 0x7fffffffULL) {
5064 struct super_block *sb = inode->i_sb;
5065 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5066 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5067 EXT4_SB(sb)->s_es->s_rev_level ==
5068 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5069 /* If this is the first large file
5070 * created, add a flag to the superblock.
5072 err = ext4_journal_get_write_access(handle,
5073 EXT4_SB(sb)->s_sbh);
5076 ext4_update_dynamic_rev(sb);
5077 EXT4_SET_RO_COMPAT_FEATURE(sb,
5078 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5080 ext4_handle_sync(handle);
5081 err = ext4_handle_dirty_metadata(handle, inode,
5082 EXT4_SB(sb)->s_sbh);
5085 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5086 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5087 if (old_valid_dev(inode->i_rdev)) {
5088 raw_inode->i_block[0] =
5089 cpu_to_le32(old_encode_dev(inode->i_rdev));
5090 raw_inode->i_block[1] = 0;
5092 raw_inode->i_block[0] = 0;
5093 raw_inode->i_block[1] =
5094 cpu_to_le32(new_encode_dev(inode->i_rdev));
5095 raw_inode->i_block[2] = 0;
5098 for (block = 0; block < EXT4_N_BLOCKS; block++)
5099 raw_inode->i_block[block] = ei->i_data[block];
5101 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5102 if (ei->i_extra_isize) {
5103 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5104 raw_inode->i_version_hi =
5105 cpu_to_le32(inode->i_version >> 32);
5106 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5109 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5110 rc = ext4_handle_dirty_metadata(handle, inode, bh);
5113 ei->i_state &= ~EXT4_STATE_NEW;
5117 ext4_std_error(inode->i_sb, err);
5122 * ext4_write_inode()
5124 * We are called from a few places:
5126 * - Within generic_file_write() for O_SYNC files.
5127 * Here, there will be no transaction running. We wait for any running
5128 * trasnaction to commit.
5130 * - Within sys_sync(), kupdate and such.
5131 * We wait on commit, if tol to.
5133 * - Within prune_icache() (PF_MEMALLOC == true)
5134 * Here we simply return. We can't afford to block kswapd on the
5137 * In all cases it is actually safe for us to return without doing anything,
5138 * because the inode has been copied into a raw inode buffer in
5139 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5142 * Note that we are absolutely dependent upon all inode dirtiers doing the
5143 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5144 * which we are interested.
5146 * It would be a bug for them to not do this. The code:
5148 * mark_inode_dirty(inode)
5150 * inode->i_size = expr;
5152 * is in error because a kswapd-driven write_inode() could occur while
5153 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5154 * will no longer be on the superblock's dirty inode list.
5156 int ext4_write_inode(struct inode *inode, int wait)
5160 if (current->flags & PF_MEMALLOC)
5163 if (EXT4_SB(inode->i_sb)->s_journal) {
5164 if (ext4_journal_current_handle()) {
5165 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5173 err = ext4_force_commit(inode->i_sb);
5175 struct ext4_iloc iloc;
5177 err = ext4_get_inode_loc(inode, &iloc);
5181 sync_dirty_buffer(iloc.bh);
5182 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5183 ext4_error(inode->i_sb, __func__,
5184 "IO error syncing inode, "
5185 "inode=%lu, block=%llu",
5187 (unsigned long long)iloc.bh->b_blocknr);
5197 * Called from notify_change.
5199 * We want to trap VFS attempts to truncate the file as soon as
5200 * possible. In particular, we want to make sure that when the VFS
5201 * shrinks i_size, we put the inode on the orphan list and modify
5202 * i_disksize immediately, so that during the subsequent flushing of
5203 * dirty pages and freeing of disk blocks, we can guarantee that any
5204 * commit will leave the blocks being flushed in an unused state on
5205 * disk. (On recovery, the inode will get truncated and the blocks will
5206 * be freed, so we have a strong guarantee that no future commit will
5207 * leave these blocks visible to the user.)
5209 * Another thing we have to assure is that if we are in ordered mode
5210 * and inode is still attached to the committing transaction, we must
5211 * we start writeout of all the dirty pages which are being truncated.
5212 * This way we are sure that all the data written in the previous
5213 * transaction are already on disk (truncate waits for pages under
5216 * Called with inode->i_mutex down.
5218 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5220 struct inode *inode = dentry->d_inode;
5222 const unsigned int ia_valid = attr->ia_valid;
5224 error = inode_change_ok(inode, attr);
5228 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5229 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5232 /* (user+group)*(old+new) structure, inode write (sb,
5233 * inode block, ? - but truncate inode update has it) */
5234 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5235 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
5236 if (IS_ERR(handle)) {
5237 error = PTR_ERR(handle);
5240 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
5242 ext4_journal_stop(handle);
5245 /* Update corresponding info in inode so that everything is in
5246 * one transaction */
5247 if (attr->ia_valid & ATTR_UID)
5248 inode->i_uid = attr->ia_uid;
5249 if (attr->ia_valid & ATTR_GID)
5250 inode->i_gid = attr->ia_gid;
5251 error = ext4_mark_inode_dirty(handle, inode);
5252 ext4_journal_stop(handle);
5255 if (attr->ia_valid & ATTR_SIZE) {
5256 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5257 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5259 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5266 if (S_ISREG(inode->i_mode) &&
5267 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
5270 handle = ext4_journal_start(inode, 3);
5271 if (IS_ERR(handle)) {
5272 error = PTR_ERR(handle);
5276 error = ext4_orphan_add(handle, inode);
5277 EXT4_I(inode)->i_disksize = attr->ia_size;
5278 rc = ext4_mark_inode_dirty(handle, inode);
5281 ext4_journal_stop(handle);
5283 if (ext4_should_order_data(inode)) {
5284 error = ext4_begin_ordered_truncate(inode,
5287 /* Do as much error cleanup as possible */
5288 handle = ext4_journal_start(inode, 3);
5289 if (IS_ERR(handle)) {
5290 ext4_orphan_del(NULL, inode);
5293 ext4_orphan_del(handle, inode);
5294 ext4_journal_stop(handle);
5300 rc = inode_setattr(inode, attr);
5302 /* If inode_setattr's call to ext4_truncate failed to get a
5303 * transaction handle at all, we need to clean up the in-core
5304 * orphan list manually. */
5306 ext4_orphan_del(NULL, inode);
5308 if (!rc && (ia_valid & ATTR_MODE))
5309 rc = ext4_acl_chmod(inode);
5312 ext4_std_error(inode->i_sb, error);
5318 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5321 struct inode *inode;
5322 unsigned long delalloc_blocks;
5324 inode = dentry->d_inode;
5325 generic_fillattr(inode, stat);
5328 * We can't update i_blocks if the block allocation is delayed
5329 * otherwise in the case of system crash before the real block
5330 * allocation is done, we will have i_blocks inconsistent with
5331 * on-disk file blocks.
5332 * We always keep i_blocks updated together with real
5333 * allocation. But to not confuse with user, stat
5334 * will return the blocks that include the delayed allocation
5335 * blocks for this file.
5337 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5338 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5339 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5341 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5345 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5350 /* if nrblocks are contiguous */
5353 * With N contiguous data blocks, it need at most
5354 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5355 * 2 dindirect blocks
5358 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5359 return indirects + 3;
5362 * if nrblocks are not contiguous, worse case, each block touch
5363 * a indirect block, and each indirect block touch a double indirect
5364 * block, plus a triple indirect block
5366 indirects = nrblocks * 2 + 1;
5370 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5372 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5373 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5374 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5378 * Account for index blocks, block groups bitmaps and block group
5379 * descriptor blocks if modify datablocks and index blocks
5380 * worse case, the indexs blocks spread over different block groups
5382 * If datablocks are discontiguous, they are possible to spread over
5383 * different block groups too. If they are contiugous, with flexbg,
5384 * they could still across block group boundary.
5386 * Also account for superblock, inode, quota and xattr blocks
5388 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5390 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5396 * How many index blocks need to touch to modify nrblocks?
5397 * The "Chunk" flag indicating whether the nrblocks is
5398 * physically contiguous on disk
5400 * For Direct IO and fallocate, they calls get_block to allocate
5401 * one single extent at a time, so they could set the "Chunk" flag
5403 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5408 * Now let's see how many group bitmaps and group descriptors need
5418 if (groups > ngroups)
5420 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5421 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5423 /* bitmaps and block group descriptor blocks */
5424 ret += groups + gdpblocks;
5426 /* Blocks for super block, inode, quota and xattr blocks */
5427 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5433 * Calulate the total number of credits to reserve to fit
5434 * the modification of a single pages into a single transaction,
5435 * which may include multiple chunks of block allocations.
5437 * This could be called via ext4_write_begin()
5439 * We need to consider the worse case, when
5440 * one new block per extent.
5442 int ext4_writepage_trans_blocks(struct inode *inode)
5444 int bpp = ext4_journal_blocks_per_page(inode);
5447 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5449 /* Account for data blocks for journalled mode */
5450 if (ext4_should_journal_data(inode))
5456 * Calculate the journal credits for a chunk of data modification.
5458 * This is called from DIO, fallocate or whoever calling
5459 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5461 * journal buffers for data blocks are not included here, as DIO
5462 * and fallocate do no need to journal data buffers.
5464 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5466 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5470 * The caller must have previously called ext4_reserve_inode_write().
5471 * Give this, we know that the caller already has write access to iloc->bh.
5473 int ext4_mark_iloc_dirty(handle_t *handle,
5474 struct inode *inode, struct ext4_iloc *iloc)
5478 if (test_opt(inode->i_sb, I_VERSION))
5479 inode_inc_iversion(inode);
5481 /* the do_update_inode consumes one bh->b_count */
5484 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5485 err = ext4_do_update_inode(handle, inode, iloc);
5491 * On success, We end up with an outstanding reference count against
5492 * iloc->bh. This _must_ be cleaned up later.
5496 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5497 struct ext4_iloc *iloc)
5501 err = ext4_get_inode_loc(inode, iloc);
5503 BUFFER_TRACE(iloc->bh, "get_write_access");
5504 err = ext4_journal_get_write_access(handle, iloc->bh);
5510 ext4_std_error(inode->i_sb, err);
5515 * Expand an inode by new_extra_isize bytes.
5516 * Returns 0 on success or negative error number on failure.
5518 static int ext4_expand_extra_isize(struct inode *inode,
5519 unsigned int new_extra_isize,
5520 struct ext4_iloc iloc,
5523 struct ext4_inode *raw_inode;
5524 struct ext4_xattr_ibody_header *header;
5525 struct ext4_xattr_entry *entry;
5527 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5530 raw_inode = ext4_raw_inode(&iloc);
5532 header = IHDR(inode, raw_inode);
5533 entry = IFIRST(header);
5535 /* No extended attributes present */
5536 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5537 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5538 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5540 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5544 /* try to expand with EAs present */
5545 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5550 * What we do here is to mark the in-core inode as clean with respect to inode
5551 * dirtiness (it may still be data-dirty).
5552 * This means that the in-core inode may be reaped by prune_icache
5553 * without having to perform any I/O. This is a very good thing,
5554 * because *any* task may call prune_icache - even ones which
5555 * have a transaction open against a different journal.
5557 * Is this cheating? Not really. Sure, we haven't written the
5558 * inode out, but prune_icache isn't a user-visible syncing function.
5559 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5560 * we start and wait on commits.
5562 * Is this efficient/effective? Well, we're being nice to the system
5563 * by cleaning up our inodes proactively so they can be reaped
5564 * without I/O. But we are potentially leaving up to five seconds'
5565 * worth of inodes floating about which prune_icache wants us to
5566 * write out. One way to fix that would be to get prune_icache()
5567 * to do a write_super() to free up some memory. It has the desired
5570 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5572 struct ext4_iloc iloc;
5573 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5574 static unsigned int mnt_count;
5578 err = ext4_reserve_inode_write(handle, inode, &iloc);
5579 if (ext4_handle_valid(handle) &&
5580 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5581 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5583 * We need extra buffer credits since we may write into EA block
5584 * with this same handle. If journal_extend fails, then it will
5585 * only result in a minor loss of functionality for that inode.
5586 * If this is felt to be critical, then e2fsck should be run to
5587 * force a large enough s_min_extra_isize.
5589 if ((jbd2_journal_extend(handle,
5590 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5591 ret = ext4_expand_extra_isize(inode,
5592 sbi->s_want_extra_isize,
5595 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5597 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5598 ext4_warning(inode->i_sb, __func__,
5599 "Unable to expand inode %lu. Delete"
5600 " some EAs or run e2fsck.",
5603 le16_to_cpu(sbi->s_es->s_mnt_count);
5609 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5614 * ext4_dirty_inode() is called from __mark_inode_dirty()
5616 * We're really interested in the case where a file is being extended.
5617 * i_size has been changed by generic_commit_write() and we thus need
5618 * to include the updated inode in the current transaction.
5620 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5621 * are allocated to the file.
5623 * If the inode is marked synchronous, we don't honour that here - doing
5624 * so would cause a commit on atime updates, which we don't bother doing.
5625 * We handle synchronous inodes at the highest possible level.
5627 void ext4_dirty_inode(struct inode *inode)
5631 handle = ext4_journal_start(inode, 2);
5635 ext4_mark_inode_dirty(handle, inode);
5637 ext4_journal_stop(handle);
5644 * Bind an inode's backing buffer_head into this transaction, to prevent
5645 * it from being flushed to disk early. Unlike
5646 * ext4_reserve_inode_write, this leaves behind no bh reference and
5647 * returns no iloc structure, so the caller needs to repeat the iloc
5648 * lookup to mark the inode dirty later.
5650 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5652 struct ext4_iloc iloc;
5656 err = ext4_get_inode_loc(inode, &iloc);
5658 BUFFER_TRACE(iloc.bh, "get_write_access");
5659 err = jbd2_journal_get_write_access(handle, iloc.bh);
5661 err = ext4_handle_dirty_metadata(handle,
5667 ext4_std_error(inode->i_sb, err);
5672 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5679 * We have to be very careful here: changing a data block's
5680 * journaling status dynamically is dangerous. If we write a
5681 * data block to the journal, change the status and then delete
5682 * that block, we risk forgetting to revoke the old log record
5683 * from the journal and so a subsequent replay can corrupt data.
5684 * So, first we make sure that the journal is empty and that
5685 * nobody is changing anything.
5688 journal = EXT4_JOURNAL(inode);
5691 if (is_journal_aborted(journal))
5694 jbd2_journal_lock_updates(journal);
5695 jbd2_journal_flush(journal);
5698 * OK, there are no updates running now, and all cached data is
5699 * synced to disk. We are now in a completely consistent state
5700 * which doesn't have anything in the journal, and we know that
5701 * no filesystem updates are running, so it is safe to modify
5702 * the inode's in-core data-journaling state flag now.
5706 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5708 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5709 ext4_set_aops(inode);
5711 jbd2_journal_unlock_updates(journal);
5713 /* Finally we can mark the inode as dirty. */
5715 handle = ext4_journal_start(inode, 1);
5717 return PTR_ERR(handle);
5719 err = ext4_mark_inode_dirty(handle, inode);
5720 ext4_handle_sync(handle);
5721 ext4_journal_stop(handle);
5722 ext4_std_error(inode->i_sb, err);
5727 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5729 return !buffer_mapped(bh);
5732 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5734 struct page *page = vmf->page;
5739 struct file *file = vma->vm_file;
5740 struct inode *inode = file->f_path.dentry->d_inode;
5741 struct address_space *mapping = inode->i_mapping;
5744 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5745 * get i_mutex because we are already holding mmap_sem.
5747 down_read(&inode->i_alloc_sem);
5748 size = i_size_read(inode);
5749 if (page->mapping != mapping || size <= page_offset(page)
5750 || !PageUptodate(page)) {
5751 /* page got truncated from under us? */
5755 if (PageMappedToDisk(page))
5758 if (page->index == size >> PAGE_CACHE_SHIFT)
5759 len = size & ~PAGE_CACHE_MASK;
5761 len = PAGE_CACHE_SIZE;
5765 * return if we have all the buffers mapped. This avoid
5766 * the need to call write_begin/write_end which does a
5767 * journal_start/journal_stop which can block and take
5770 if (page_has_buffers(page)) {
5771 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5772 ext4_bh_unmapped)) {
5779 * OK, we need to fill the hole... Do write_begin write_end
5780 * to do block allocation/reservation.We are not holding
5781 * inode.i__mutex here. That allow * parallel write_begin,
5782 * write_end call. lock_page prevent this from happening
5783 * on the same page though
5785 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5786 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5789 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5790 len, len, page, fsdata);
5796 ret = VM_FAULT_SIGBUS;
5797 up_read(&inode->i_alloc_sem);