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);
217 * Called at the last iput() if i_nlink is zero.
219 void ext4_delete_inode(struct inode *inode)
224 if (ext4_should_order_data(inode))
225 ext4_begin_ordered_truncate(inode, 0);
226 truncate_inode_pages(&inode->i_data, 0);
228 if (is_bad_inode(inode))
231 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
232 if (IS_ERR(handle)) {
233 ext4_std_error(inode->i_sb, PTR_ERR(handle));
235 * If we're going to skip the normal cleanup, we still need to
236 * make sure that the in-core orphan linked list is properly
239 ext4_orphan_del(NULL, inode);
244 ext4_handle_sync(handle);
246 err = ext4_mark_inode_dirty(handle, inode);
248 ext4_warning(inode->i_sb, __func__,
249 "couldn't mark inode dirty (err %d)", err);
253 ext4_truncate(inode);
256 * ext4_ext_truncate() doesn't reserve any slop when it
257 * restarts journal transactions; therefore there may not be
258 * enough credits left in the handle to remove the inode from
259 * the orphan list and set the dtime field.
261 if (!ext4_handle_has_enough_credits(handle, 3)) {
262 err = ext4_journal_extend(handle, 3);
264 err = ext4_journal_restart(handle, 3);
266 ext4_warning(inode->i_sb, __func__,
267 "couldn't extend journal (err %d)", err);
269 ext4_journal_stop(handle);
275 * Kill off the orphan record which ext4_truncate created.
276 * AKPM: I think this can be inside the above `if'.
277 * Note that ext4_orphan_del() has to be able to cope with the
278 * deletion of a non-existent orphan - this is because we don't
279 * know if ext4_truncate() actually created an orphan record.
280 * (Well, we could do this if we need to, but heck - it works)
282 ext4_orphan_del(handle, inode);
283 EXT4_I(inode)->i_dtime = get_seconds();
286 * One subtle ordering requirement: if anything has gone wrong
287 * (transaction abort, IO errors, whatever), then we can still
288 * do these next steps (the fs will already have been marked as
289 * having errors), but we can't free the inode if the mark_dirty
292 if (ext4_mark_inode_dirty(handle, inode))
293 /* If that failed, just do the required in-core inode clear. */
296 ext4_free_inode(handle, inode);
297 ext4_journal_stop(handle);
300 clear_inode(inode); /* We must guarantee clearing of inode... */
306 struct buffer_head *bh;
309 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
311 p->key = *(p->p = v);
316 * ext4_block_to_path - parse the block number into array of offsets
317 * @inode: inode in question (we are only interested in its superblock)
318 * @i_block: block number to be parsed
319 * @offsets: array to store the offsets in
320 * @boundary: set this non-zero if the referred-to block is likely to be
321 * followed (on disk) by an indirect block.
323 * To store the locations of file's data ext4 uses a data structure common
324 * for UNIX filesystems - tree of pointers anchored in the inode, with
325 * data blocks at leaves and indirect blocks in intermediate nodes.
326 * This function translates the block number into path in that tree -
327 * return value is the path length and @offsets[n] is the offset of
328 * pointer to (n+1)th node in the nth one. If @block is out of range
329 * (negative or too large) warning is printed and zero returned.
331 * Note: function doesn't find node addresses, so no IO is needed. All
332 * we need to know is the capacity of indirect blocks (taken from the
337 * Portability note: the last comparison (check that we fit into triple
338 * indirect block) is spelled differently, because otherwise on an
339 * architecture with 32-bit longs and 8Kb pages we might get into trouble
340 * if our filesystem had 8Kb blocks. We might use long long, but that would
341 * kill us on x86. Oh, well, at least the sign propagation does not matter -
342 * i_block would have to be negative in the very beginning, so we would not
346 static int ext4_block_to_path(struct inode *inode,
348 ext4_lblk_t offsets[4], int *boundary)
350 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
351 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
352 const long direct_blocks = EXT4_NDIR_BLOCKS,
353 indirect_blocks = ptrs,
354 double_blocks = (1 << (ptrs_bits * 2));
358 if (i_block < direct_blocks) {
359 offsets[n++] = i_block;
360 final = direct_blocks;
361 } else if ((i_block -= direct_blocks) < indirect_blocks) {
362 offsets[n++] = EXT4_IND_BLOCK;
363 offsets[n++] = i_block;
365 } else if ((i_block -= indirect_blocks) < double_blocks) {
366 offsets[n++] = EXT4_DIND_BLOCK;
367 offsets[n++] = i_block >> ptrs_bits;
368 offsets[n++] = i_block & (ptrs - 1);
370 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
371 offsets[n++] = EXT4_TIND_BLOCK;
372 offsets[n++] = i_block >> (ptrs_bits * 2);
373 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
374 offsets[n++] = i_block & (ptrs - 1);
377 ext4_warning(inode->i_sb, "ext4_block_to_path",
378 "block %lu > max in inode %lu",
379 i_block + direct_blocks +
380 indirect_blocks + double_blocks, inode->i_ino);
383 *boundary = final - 1 - (i_block & (ptrs - 1));
387 static int __ext4_check_blockref(const char *function, struct inode *inode,
388 __le32 *p, unsigned int max)
393 while (bref < p+max) {
394 blk = le32_to_cpu(*bref++);
396 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
398 ext4_error(inode->i_sb, function,
399 "invalid block reference %u "
400 "in inode #%lu", blk, inode->i_ino);
408 #define ext4_check_indirect_blockref(inode, bh) \
409 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
410 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
412 #define ext4_check_inode_blockref(inode) \
413 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
417 * ext4_get_branch - read the chain of indirect blocks leading to data
418 * @inode: inode in question
419 * @depth: depth of the chain (1 - direct pointer, etc.)
420 * @offsets: offsets of pointers in inode/indirect blocks
421 * @chain: place to store the result
422 * @err: here we store the error value
424 * Function fills the array of triples <key, p, bh> and returns %NULL
425 * if everything went OK or the pointer to the last filled triple
426 * (incomplete one) otherwise. Upon the return chain[i].key contains
427 * the number of (i+1)-th block in the chain (as it is stored in memory,
428 * i.e. little-endian 32-bit), chain[i].p contains the address of that
429 * number (it points into struct inode for i==0 and into the bh->b_data
430 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
431 * block for i>0 and NULL for i==0. In other words, it holds the block
432 * numbers of the chain, addresses they were taken from (and where we can
433 * verify that chain did not change) and buffer_heads hosting these
436 * Function stops when it stumbles upon zero pointer (absent block)
437 * (pointer to last triple returned, *@err == 0)
438 * or when it gets an IO error reading an indirect block
439 * (ditto, *@err == -EIO)
440 * or when it reads all @depth-1 indirect blocks successfully and finds
441 * the whole chain, all way to the data (returns %NULL, *err == 0).
443 * Need to be called with
444 * down_read(&EXT4_I(inode)->i_data_sem)
446 static Indirect *ext4_get_branch(struct inode *inode, int depth,
447 ext4_lblk_t *offsets,
448 Indirect chain[4], int *err)
450 struct super_block *sb = inode->i_sb;
452 struct buffer_head *bh;
455 /* i_data is not going away, no lock needed */
456 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
460 bh = sb_getblk(sb, le32_to_cpu(p->key));
464 if (!bh_uptodate_or_lock(bh)) {
465 if (bh_submit_read(bh) < 0) {
469 /* validate block references */
470 if (ext4_check_indirect_blockref(inode, bh)) {
476 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
490 * ext4_find_near - find a place for allocation with sufficient locality
492 * @ind: descriptor of indirect block.
494 * This function returns the preferred place for block allocation.
495 * It is used when heuristic for sequential allocation fails.
497 * + if there is a block to the left of our position - allocate near it.
498 * + if pointer will live in indirect block - allocate near that block.
499 * + if pointer will live in inode - allocate in the same
502 * In the latter case we colour the starting block by the callers PID to
503 * prevent it from clashing with concurrent allocations for a different inode
504 * in the same block group. The PID is used here so that functionally related
505 * files will be close-by on-disk.
507 * Caller must make sure that @ind is valid and will stay that way.
509 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
511 struct ext4_inode_info *ei = EXT4_I(inode);
512 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
514 ext4_fsblk_t bg_start;
515 ext4_fsblk_t last_block;
516 ext4_grpblk_t colour;
517 ext4_group_t block_group;
518 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
520 /* Try to find previous block */
521 for (p = ind->p - 1; p >= start; p--) {
523 return le32_to_cpu(*p);
526 /* No such thing, so let's try location of indirect block */
528 return ind->bh->b_blocknr;
531 * It is going to be referred to from the inode itself? OK, just put it
532 * into the same cylinder group then.
534 block_group = ei->i_block_group;
535 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
536 block_group &= ~(flex_size-1);
537 if (S_ISREG(inode->i_mode))
540 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
541 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
544 * If we are doing delayed allocation, we don't need take
545 * colour into account.
547 if (test_opt(inode->i_sb, DELALLOC))
550 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
551 colour = (current->pid % 16) *
552 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
554 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
555 return bg_start + colour;
559 * ext4_find_goal - find a preferred place for allocation.
561 * @block: block we want
562 * @partial: pointer to the last triple within a chain
564 * Normally this function find the preferred place for block allocation,
566 * Because this is only used for non-extent files, we limit the block nr
569 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
575 * XXX need to get goal block from mballoc's data structures
578 goal = ext4_find_near(inode, partial);
579 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
584 * ext4_blks_to_allocate: Look up the block map and count the number
585 * of direct blocks need to be allocated for the given branch.
587 * @branch: chain of indirect blocks
588 * @k: number of blocks need for indirect blocks
589 * @blks: number of data blocks to be mapped.
590 * @blocks_to_boundary: the offset in the indirect block
592 * return the total number of blocks to be allocate, including the
593 * direct and indirect blocks.
595 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
596 int blocks_to_boundary)
598 unsigned int count = 0;
601 * Simple case, [t,d]Indirect block(s) has not allocated yet
602 * then it's clear blocks on that path have not allocated
605 /* right now we don't handle cross boundary allocation */
606 if (blks < blocks_to_boundary + 1)
609 count += blocks_to_boundary + 1;
614 while (count < blks && count <= blocks_to_boundary &&
615 le32_to_cpu(*(branch[0].p + count)) == 0) {
622 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
623 * @indirect_blks: the number of blocks need to allocate for indirect
626 * @new_blocks: on return it will store the new block numbers for
627 * the indirect blocks(if needed) and the first direct block,
628 * @blks: on return it will store the total number of allocated
631 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
632 ext4_lblk_t iblock, ext4_fsblk_t goal,
633 int indirect_blks, int blks,
634 ext4_fsblk_t new_blocks[4], int *err)
636 struct ext4_allocation_request ar;
638 unsigned long count = 0, blk_allocated = 0;
640 ext4_fsblk_t current_block = 0;
644 * Here we try to allocate the requested multiple blocks at once,
645 * on a best-effort basis.
646 * To build a branch, we should allocate blocks for
647 * the indirect blocks(if not allocated yet), and at least
648 * the first direct block of this branch. That's the
649 * minimum number of blocks need to allocate(required)
651 /* first we try to allocate the indirect blocks */
652 target = indirect_blks;
655 /* allocating blocks for indirect blocks and direct blocks */
656 current_block = ext4_new_meta_blocks(handle, inode,
661 BUG_ON(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS);
664 /* allocate blocks for indirect blocks */
665 while (index < indirect_blks && count) {
666 new_blocks[index++] = current_block++;
671 * save the new block number
672 * for the first direct block
674 new_blocks[index] = current_block;
675 printk(KERN_INFO "%s returned more blocks than "
676 "requested\n", __func__);
682 target = blks - count ;
683 blk_allocated = count;
686 /* Now allocate data blocks */
687 memset(&ar, 0, sizeof(ar));
692 if (S_ISREG(inode->i_mode))
693 /* enable in-core preallocation only for regular files */
694 ar.flags = EXT4_MB_HINT_DATA;
696 current_block = ext4_mb_new_blocks(handle, &ar, err);
697 BUG_ON(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS);
699 if (*err && (target == blks)) {
701 * if the allocation failed and we didn't allocate
707 if (target == blks) {
709 * save the new block number
710 * for the first direct block
712 new_blocks[index] = current_block;
714 blk_allocated += ar.len;
717 /* total number of blocks allocated for direct blocks */
722 for (i = 0; i < index; i++)
723 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
730 * @indirect_blks: number of allocated indirect blocks
731 * @blks: number of allocated direct blocks
732 * @offsets: offsets (in the blocks) to store the pointers to next.
733 * @branch: place to store the chain in.
735 * This function allocates blocks, zeroes out all but the last one,
736 * links them into chain and (if we are synchronous) writes them to disk.
737 * In other words, it prepares a branch that can be spliced onto the
738 * inode. It stores the information about that chain in the branch[], in
739 * the same format as ext4_get_branch() would do. We are calling it after
740 * we had read the existing part of chain and partial points to the last
741 * triple of that (one with zero ->key). Upon the exit we have the same
742 * picture as after the successful ext4_get_block(), except that in one
743 * place chain is disconnected - *branch->p is still zero (we did not
744 * set the last link), but branch->key contains the number that should
745 * be placed into *branch->p to fill that gap.
747 * If allocation fails we free all blocks we've allocated (and forget
748 * their buffer_heads) and return the error value the from failed
749 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
750 * as described above and return 0.
752 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
753 ext4_lblk_t iblock, int indirect_blks,
754 int *blks, ext4_fsblk_t goal,
755 ext4_lblk_t *offsets, Indirect *branch)
757 int blocksize = inode->i_sb->s_blocksize;
760 struct buffer_head *bh;
762 ext4_fsblk_t new_blocks[4];
763 ext4_fsblk_t current_block;
765 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
766 *blks, new_blocks, &err);
770 branch[0].key = cpu_to_le32(new_blocks[0]);
772 * metadata blocks and data blocks are allocated.
774 for (n = 1; n <= indirect_blks; n++) {
776 * Get buffer_head for parent block, zero it out
777 * and set the pointer to new one, then send
780 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
783 BUFFER_TRACE(bh, "call get_create_access");
784 err = ext4_journal_get_create_access(handle, bh);
786 /* Don't brelse(bh) here; it's done in
787 * ext4_journal_forget() below */
792 memset(bh->b_data, 0, blocksize);
793 branch[n].p = (__le32 *) bh->b_data + offsets[n];
794 branch[n].key = cpu_to_le32(new_blocks[n]);
795 *branch[n].p = branch[n].key;
796 if (n == indirect_blks) {
797 current_block = new_blocks[n];
799 * End of chain, update the last new metablock of
800 * the chain to point to the new allocated
801 * data blocks numbers
803 for (i = 1; i < num; i++)
804 *(branch[n].p + i) = cpu_to_le32(++current_block);
806 BUFFER_TRACE(bh, "marking uptodate");
807 set_buffer_uptodate(bh);
810 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
811 err = ext4_handle_dirty_metadata(handle, inode, bh);
818 /* Allocation failed, free what we already allocated */
819 for (i = 1; i <= n ; i++) {
820 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
821 ext4_journal_forget(handle, branch[i].bh);
823 for (i = 0; i < indirect_blks; i++)
824 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
826 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
832 * ext4_splice_branch - splice the allocated branch onto inode.
834 * @block: (logical) number of block we are adding
835 * @chain: chain of indirect blocks (with a missing link - see
837 * @where: location of missing link
838 * @num: number of indirect blocks we are adding
839 * @blks: number of direct blocks we are adding
841 * This function fills the missing link and does all housekeeping needed in
842 * inode (->i_blocks, etc.). In case of success we end up with the full
843 * chain to new block and return 0.
845 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
846 ext4_lblk_t block, Indirect *where, int num,
851 ext4_fsblk_t current_block;
854 * If we're splicing into a [td]indirect block (as opposed to the
855 * inode) then we need to get write access to the [td]indirect block
859 BUFFER_TRACE(where->bh, "get_write_access");
860 err = ext4_journal_get_write_access(handle, where->bh);
866 *where->p = where->key;
869 * Update the host buffer_head or inode to point to more just allocated
870 * direct blocks blocks
872 if (num == 0 && blks > 1) {
873 current_block = le32_to_cpu(where->key) + 1;
874 for (i = 1; i < blks; i++)
875 *(where->p + i) = cpu_to_le32(current_block++);
878 /* We are done with atomic stuff, now do the rest of housekeeping */
879 /* had we spliced it onto indirect block? */
882 * If we spliced it onto an indirect block, we haven't
883 * altered the inode. Note however that if it is being spliced
884 * onto an indirect block at the very end of the file (the
885 * file is growing) then we *will* alter the inode to reflect
886 * the new i_size. But that is not done here - it is done in
887 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
889 jbd_debug(5, "splicing indirect only\n");
890 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
891 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
896 * OK, we spliced it into the inode itself on a direct block.
898 ext4_mark_inode_dirty(handle, inode);
899 jbd_debug(5, "splicing direct\n");
904 for (i = 1; i <= num; i++) {
905 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
906 ext4_journal_forget(handle, where[i].bh);
907 ext4_free_blocks(handle, inode,
908 le32_to_cpu(where[i-1].key), 1, 0);
910 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
916 * The ext4_ind_get_blocks() function handles non-extents inodes
917 * (i.e., using the traditional indirect/double-indirect i_blocks
918 * scheme) for ext4_get_blocks().
920 * Allocation strategy is simple: if we have to allocate something, we will
921 * have to go the whole way to leaf. So let's do it before attaching anything
922 * to tree, set linkage between the newborn blocks, write them if sync is
923 * required, recheck the path, free and repeat if check fails, otherwise
924 * set the last missing link (that will protect us from any truncate-generated
925 * removals - all blocks on the path are immune now) and possibly force the
926 * write on the parent block.
927 * That has a nice additional property: no special recovery from the failed
928 * allocations is needed - we simply release blocks and do not touch anything
929 * reachable from inode.
931 * `handle' can be NULL if create == 0.
933 * return > 0, # of blocks mapped or allocated.
934 * return = 0, if plain lookup failed.
935 * return < 0, error case.
937 * The ext4_ind_get_blocks() function should be called with
938 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
939 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
940 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
943 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
944 ext4_lblk_t iblock, unsigned int maxblocks,
945 struct buffer_head *bh_result,
949 ext4_lblk_t offsets[4];
954 int blocks_to_boundary = 0;
957 ext4_fsblk_t first_block = 0;
959 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
960 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
961 depth = ext4_block_to_path(inode, iblock, offsets,
962 &blocks_to_boundary);
967 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
969 /* Simplest case - block found, no allocation needed */
971 first_block = le32_to_cpu(chain[depth - 1].key);
972 clear_buffer_new(bh_result);
975 while (count < maxblocks && count <= blocks_to_boundary) {
978 blk = le32_to_cpu(*(chain[depth-1].p + count));
980 if (blk == first_block + count)
988 /* Next simple case - plain lookup or failed read of indirect block */
989 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
993 * Okay, we need to do block allocation.
995 goal = ext4_find_goal(inode, iblock, partial);
997 /* the number of blocks need to allocate for [d,t]indirect blocks */
998 indirect_blks = (chain + depth) - partial - 1;
1001 * Next look up the indirect map to count the totoal number of
1002 * direct blocks to allocate for this branch.
1004 count = ext4_blks_to_allocate(partial, indirect_blks,
1005 maxblocks, blocks_to_boundary);
1007 * Block out ext4_truncate while we alter the tree
1009 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
1011 offsets + (partial - chain), partial);
1014 * The ext4_splice_branch call will free and forget any buffers
1015 * on the new chain if there is a failure, but that risks using
1016 * up transaction credits, especially for bitmaps where the
1017 * credits cannot be returned. Can we handle this somehow? We
1018 * may need to return -EAGAIN upwards in the worst case. --sct
1021 err = ext4_splice_branch(handle, inode, iblock,
1022 partial, indirect_blks, count);
1026 set_buffer_new(bh_result);
1028 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1029 if (count > blocks_to_boundary)
1030 set_buffer_boundary(bh_result);
1032 /* Clean up and exit */
1033 partial = chain + depth - 1; /* the whole chain */
1035 while (partial > chain) {
1036 BUFFER_TRACE(partial->bh, "call brelse");
1037 brelse(partial->bh);
1040 BUFFER_TRACE(bh_result, "returned");
1045 qsize_t ext4_get_reserved_space(struct inode *inode)
1047 unsigned long long total;
1049 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1050 total = EXT4_I(inode)->i_reserved_data_blocks +
1051 EXT4_I(inode)->i_reserved_meta_blocks;
1052 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1057 * Calculate the number of metadata blocks need to reserve
1058 * to allocate @blocks for non extent file based file
1060 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1062 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1063 int ind_blks, dind_blks, tind_blks;
1065 /* number of new indirect blocks needed */
1066 ind_blks = (blocks + icap - 1) / icap;
1068 dind_blks = (ind_blks + icap - 1) / icap;
1072 return ind_blks + dind_blks + tind_blks;
1076 * Calculate the number of metadata blocks need to reserve
1077 * to allocate given number of blocks
1079 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1084 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1085 return ext4_ext_calc_metadata_amount(inode, blocks);
1087 return ext4_indirect_calc_metadata_amount(inode, blocks);
1090 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1092 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1093 int total, mdb, mdb_free;
1095 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1096 /* recalculate the number of metablocks still need to be reserved */
1097 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1098 mdb = ext4_calc_metadata_amount(inode, total);
1100 /* figure out how many metablocks to release */
1101 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1102 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1105 /* Account for allocated meta_blocks */
1106 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1108 /* update fs dirty blocks counter */
1109 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1110 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1111 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1114 /* update per-inode reservations */
1115 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1116 EXT4_I(inode)->i_reserved_data_blocks -= used;
1117 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1120 * free those over-booking quota for metadata blocks
1123 vfs_dq_release_reservation_block(inode, mdb_free);
1126 * If we have done all the pending block allocations and if
1127 * there aren't any writers on the inode, we can discard the
1128 * inode's preallocations.
1130 if (!total && (atomic_read(&inode->i_writecount) == 0))
1131 ext4_discard_preallocations(inode);
1134 static int check_block_validity(struct inode *inode, const char *msg,
1135 sector_t logical, sector_t phys, int len)
1137 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), phys, len)) {
1138 ext4_error(inode->i_sb, msg,
1139 "inode #%lu logical block %llu mapped to %llu "
1140 "(size %d)", inode->i_ino,
1141 (unsigned long long) logical,
1142 (unsigned long long) phys, len);
1149 * Return the number of dirty pages in the given inode starting at
1152 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1153 unsigned int max_pages)
1155 struct address_space *mapping = inode->i_mapping;
1157 struct pagevec pvec;
1159 int i, nr_pages, done = 0;
1163 pagevec_init(&pvec, 0);
1166 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1167 PAGECACHE_TAG_DIRTY,
1168 (pgoff_t)PAGEVEC_SIZE);
1171 for (i = 0; i < nr_pages; i++) {
1172 struct page *page = pvec.pages[i];
1173 struct buffer_head *bh, *head;
1176 if (unlikely(page->mapping != mapping) ||
1178 PageWriteback(page) ||
1179 page->index != idx) {
1184 head = page_buffers(page);
1187 if (!buffer_delay(bh) &&
1188 !buffer_unwritten(bh)) {
1192 } while ((bh = bh->b_this_page) != head);
1198 if (num >= max_pages)
1201 pagevec_release(&pvec);
1207 * The ext4_get_blocks() function tries to look up the requested blocks,
1208 * and returns if the blocks are already mapped.
1210 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1211 * and store the allocated blocks in the result buffer head and mark it
1214 * If file type is extents based, it will call ext4_ext_get_blocks(),
1215 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1218 * On success, it returns the number of blocks being mapped or allocate.
1219 * if create==0 and the blocks are pre-allocated and uninitialized block,
1220 * the result buffer head is unmapped. If the create ==1, it will make sure
1221 * the buffer head is mapped.
1223 * It returns 0 if plain look up failed (blocks have not been allocated), in
1224 * that casem, buffer head is unmapped
1226 * It returns the error in case of allocation failure.
1228 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1229 unsigned int max_blocks, struct buffer_head *bh,
1234 clear_buffer_mapped(bh);
1235 clear_buffer_unwritten(bh);
1237 ext_debug("ext4_get_blocks(): inode %lu, flag %d, max_blocks %u,"
1238 "logical block %lu\n", inode->i_ino, flags, max_blocks,
1239 (unsigned long)block);
1241 * Try to see if we can get the block without requesting a new
1242 * file system block.
1244 down_read((&EXT4_I(inode)->i_data_sem));
1245 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1246 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1249 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1252 up_read((&EXT4_I(inode)->i_data_sem));
1254 if (retval > 0 && buffer_mapped(bh)) {
1255 int ret = check_block_validity(inode, "file system corruption",
1256 block, bh->b_blocknr, retval);
1261 /* If it is only a block(s) look up */
1262 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1266 * Returns if the blocks have already allocated
1268 * Note that if blocks have been preallocated
1269 * ext4_ext_get_block() returns th create = 0
1270 * with buffer head unmapped.
1272 if (retval > 0 && buffer_mapped(bh))
1276 * When we call get_blocks without the create flag, the
1277 * BH_Unwritten flag could have gotten set if the blocks
1278 * requested were part of a uninitialized extent. We need to
1279 * clear this flag now that we are committed to convert all or
1280 * part of the uninitialized extent to be an initialized
1281 * extent. This is because we need to avoid the combination
1282 * of BH_Unwritten and BH_Mapped flags being simultaneously
1283 * set on the buffer_head.
1285 clear_buffer_unwritten(bh);
1288 * New blocks allocate and/or writing to uninitialized extent
1289 * will possibly result in updating i_data, so we take
1290 * the write lock of i_data_sem, and call get_blocks()
1291 * with create == 1 flag.
1293 down_write((&EXT4_I(inode)->i_data_sem));
1296 * if the caller is from delayed allocation writeout path
1297 * we have already reserved fs blocks for allocation
1298 * let the underlying get_block() function know to
1299 * avoid double accounting
1301 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1302 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1304 * We need to check for EXT4 here because migrate
1305 * could have changed the inode type in between
1307 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1308 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1311 retval = ext4_ind_get_blocks(handle, inode, block,
1312 max_blocks, bh, flags);
1314 if (retval > 0 && buffer_new(bh)) {
1316 * We allocated new blocks which will result in
1317 * i_data's format changing. Force the migrate
1318 * to fail by clearing migrate flags
1320 EXT4_I(inode)->i_state &= ~EXT4_STATE_EXT_MIGRATE;
1324 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1325 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1328 * Update reserved blocks/metadata blocks after successful
1329 * block allocation which had been deferred till now.
1331 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1332 ext4_da_update_reserve_space(inode, retval);
1334 up_write((&EXT4_I(inode)->i_data_sem));
1335 if (retval > 0 && buffer_mapped(bh)) {
1336 int ret = check_block_validity(inode, "file system "
1337 "corruption after allocation",
1338 block, bh->b_blocknr, retval);
1345 /* Maximum number of blocks we map for direct IO at once. */
1346 #define DIO_MAX_BLOCKS 4096
1348 int ext4_get_block(struct inode *inode, sector_t iblock,
1349 struct buffer_head *bh_result, int create)
1351 handle_t *handle = ext4_journal_current_handle();
1352 int ret = 0, started = 0;
1353 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1356 if (create && !handle) {
1357 /* Direct IO write... */
1358 if (max_blocks > DIO_MAX_BLOCKS)
1359 max_blocks = DIO_MAX_BLOCKS;
1360 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1361 handle = ext4_journal_start(inode, dio_credits);
1362 if (IS_ERR(handle)) {
1363 ret = PTR_ERR(handle);
1369 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1370 create ? EXT4_GET_BLOCKS_CREATE : 0);
1372 bh_result->b_size = (ret << inode->i_blkbits);
1376 ext4_journal_stop(handle);
1382 * `handle' can be NULL if create is zero
1384 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1385 ext4_lblk_t block, int create, int *errp)
1387 struct buffer_head dummy;
1391 J_ASSERT(handle != NULL || create == 0);
1394 dummy.b_blocknr = -1000;
1395 buffer_trace_init(&dummy.b_history);
1397 flags |= EXT4_GET_BLOCKS_CREATE;
1398 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1400 * ext4_get_blocks() returns number of blocks mapped. 0 in
1409 if (!err && buffer_mapped(&dummy)) {
1410 struct buffer_head *bh;
1411 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1416 if (buffer_new(&dummy)) {
1417 J_ASSERT(create != 0);
1418 J_ASSERT(handle != NULL);
1421 * Now that we do not always journal data, we should
1422 * keep in mind whether this should always journal the
1423 * new buffer as metadata. For now, regular file
1424 * writes use ext4_get_block instead, so it's not a
1428 BUFFER_TRACE(bh, "call get_create_access");
1429 fatal = ext4_journal_get_create_access(handle, bh);
1430 if (!fatal && !buffer_uptodate(bh)) {
1431 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1432 set_buffer_uptodate(bh);
1435 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1436 err = ext4_handle_dirty_metadata(handle, inode, bh);
1440 BUFFER_TRACE(bh, "not a new buffer");
1453 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1454 ext4_lblk_t block, int create, int *err)
1456 struct buffer_head *bh;
1458 bh = ext4_getblk(handle, inode, block, create, err);
1461 if (buffer_uptodate(bh))
1463 ll_rw_block(READ_META, 1, &bh);
1465 if (buffer_uptodate(bh))
1472 static int walk_page_buffers(handle_t *handle,
1473 struct buffer_head *head,
1477 int (*fn)(handle_t *handle,
1478 struct buffer_head *bh))
1480 struct buffer_head *bh;
1481 unsigned block_start, block_end;
1482 unsigned blocksize = head->b_size;
1484 struct buffer_head *next;
1486 for (bh = head, block_start = 0;
1487 ret == 0 && (bh != head || !block_start);
1488 block_start = block_end, bh = next) {
1489 next = bh->b_this_page;
1490 block_end = block_start + blocksize;
1491 if (block_end <= from || block_start >= to) {
1492 if (partial && !buffer_uptodate(bh))
1496 err = (*fn)(handle, bh);
1504 * To preserve ordering, it is essential that the hole instantiation and
1505 * the data write be encapsulated in a single transaction. We cannot
1506 * close off a transaction and start a new one between the ext4_get_block()
1507 * and the commit_write(). So doing the jbd2_journal_start at the start of
1508 * prepare_write() is the right place.
1510 * Also, this function can nest inside ext4_writepage() ->
1511 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1512 * has generated enough buffer credits to do the whole page. So we won't
1513 * block on the journal in that case, which is good, because the caller may
1516 * By accident, ext4 can be reentered when a transaction is open via
1517 * quota file writes. If we were to commit the transaction while thus
1518 * reentered, there can be a deadlock - we would be holding a quota
1519 * lock, and the commit would never complete if another thread had a
1520 * transaction open and was blocking on the quota lock - a ranking
1523 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1524 * will _not_ run commit under these circumstances because handle->h_ref
1525 * is elevated. We'll still have enough credits for the tiny quotafile
1528 static int do_journal_get_write_access(handle_t *handle,
1529 struct buffer_head *bh)
1531 if (!buffer_mapped(bh) || buffer_freed(bh))
1533 return ext4_journal_get_write_access(handle, bh);
1536 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1537 loff_t pos, unsigned len, unsigned flags,
1538 struct page **pagep, void **fsdata)
1540 struct inode *inode = mapping->host;
1541 int ret, needed_blocks;
1548 trace_ext4_write_begin(inode, pos, len, flags);
1550 * Reserve one block more for addition to orphan list in case
1551 * we allocate blocks but write fails for some reason
1553 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1554 index = pos >> PAGE_CACHE_SHIFT;
1555 from = pos & (PAGE_CACHE_SIZE - 1);
1559 handle = ext4_journal_start(inode, needed_blocks);
1560 if (IS_ERR(handle)) {
1561 ret = PTR_ERR(handle);
1565 /* We cannot recurse into the filesystem as the transaction is already
1567 flags |= AOP_FLAG_NOFS;
1569 page = grab_cache_page_write_begin(mapping, index, flags);
1571 ext4_journal_stop(handle);
1577 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1580 if (!ret && ext4_should_journal_data(inode)) {
1581 ret = walk_page_buffers(handle, page_buffers(page),
1582 from, to, NULL, do_journal_get_write_access);
1587 page_cache_release(page);
1589 * block_write_begin may have instantiated a few blocks
1590 * outside i_size. Trim these off again. Don't need
1591 * i_size_read because we hold i_mutex.
1593 * Add inode to orphan list in case we crash before
1596 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1597 ext4_orphan_add(handle, inode);
1599 ext4_journal_stop(handle);
1600 if (pos + len > inode->i_size) {
1601 ext4_truncate(inode);
1603 * If truncate failed early the inode might
1604 * still be on the orphan list; we need to
1605 * make sure the inode is removed from the
1606 * orphan list in that case.
1609 ext4_orphan_del(NULL, inode);
1613 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1619 /* For write_end() in data=journal mode */
1620 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1622 if (!buffer_mapped(bh) || buffer_freed(bh))
1624 set_buffer_uptodate(bh);
1625 return ext4_handle_dirty_metadata(handle, NULL, bh);
1628 static int ext4_generic_write_end(struct file *file,
1629 struct address_space *mapping,
1630 loff_t pos, unsigned len, unsigned copied,
1631 struct page *page, void *fsdata)
1633 int i_size_changed = 0;
1634 struct inode *inode = mapping->host;
1635 handle_t *handle = ext4_journal_current_handle();
1637 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1640 * No need to use i_size_read() here, the i_size
1641 * cannot change under us because we hold i_mutex.
1643 * But it's important to update i_size while still holding page lock:
1644 * page writeout could otherwise come in and zero beyond i_size.
1646 if (pos + copied > inode->i_size) {
1647 i_size_write(inode, pos + copied);
1651 if (pos + copied > EXT4_I(inode)->i_disksize) {
1652 /* We need to mark inode dirty even if
1653 * new_i_size is less that inode->i_size
1654 * bu greater than i_disksize.(hint delalloc)
1656 ext4_update_i_disksize(inode, (pos + copied));
1660 page_cache_release(page);
1663 * Don't mark the inode dirty under page lock. First, it unnecessarily
1664 * makes the holding time of page lock longer. Second, it forces lock
1665 * ordering of page lock and transaction start for journaling
1669 ext4_mark_inode_dirty(handle, inode);
1675 * We need to pick up the new inode size which generic_commit_write gave us
1676 * `file' can be NULL - eg, when called from page_symlink().
1678 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1679 * buffers are managed internally.
1681 static int ext4_ordered_write_end(struct file *file,
1682 struct address_space *mapping,
1683 loff_t pos, unsigned len, unsigned copied,
1684 struct page *page, void *fsdata)
1686 handle_t *handle = ext4_journal_current_handle();
1687 struct inode *inode = mapping->host;
1690 trace_ext4_ordered_write_end(inode, pos, len, copied);
1691 ret = ext4_jbd2_file_inode(handle, inode);
1694 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1697 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1698 /* if we have allocated more blocks and copied
1699 * less. We will have blocks allocated outside
1700 * inode->i_size. So truncate them
1702 ext4_orphan_add(handle, inode);
1706 ret2 = ext4_journal_stop(handle);
1710 if (pos + len > inode->i_size) {
1711 ext4_truncate(inode);
1713 * If truncate failed early the inode might still be
1714 * on the orphan list; we need to make sure the inode
1715 * is removed from the orphan list in that case.
1718 ext4_orphan_del(NULL, inode);
1722 return ret ? ret : copied;
1725 static int ext4_writeback_write_end(struct file *file,
1726 struct address_space *mapping,
1727 loff_t pos, unsigned len, unsigned copied,
1728 struct page *page, void *fsdata)
1730 handle_t *handle = ext4_journal_current_handle();
1731 struct inode *inode = mapping->host;
1734 trace_ext4_writeback_write_end(inode, pos, len, copied);
1735 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1738 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1739 /* if we have allocated more blocks and copied
1740 * less. We will have blocks allocated outside
1741 * inode->i_size. So truncate them
1743 ext4_orphan_add(handle, inode);
1748 ret2 = ext4_journal_stop(handle);
1752 if (pos + len > inode->i_size) {
1753 ext4_truncate(inode);
1755 * If truncate failed early the inode might still be
1756 * on the orphan list; we need to make sure the inode
1757 * is removed from the orphan list in that case.
1760 ext4_orphan_del(NULL, inode);
1763 return ret ? ret : copied;
1766 static int ext4_journalled_write_end(struct file *file,
1767 struct address_space *mapping,
1768 loff_t pos, unsigned len, unsigned copied,
1769 struct page *page, void *fsdata)
1771 handle_t *handle = ext4_journal_current_handle();
1772 struct inode *inode = mapping->host;
1778 trace_ext4_journalled_write_end(inode, pos, len, copied);
1779 from = pos & (PAGE_CACHE_SIZE - 1);
1783 if (!PageUptodate(page))
1785 page_zero_new_buffers(page, from+copied, to);
1788 ret = walk_page_buffers(handle, page_buffers(page), from,
1789 to, &partial, write_end_fn);
1791 SetPageUptodate(page);
1792 new_i_size = pos + copied;
1793 if (new_i_size > inode->i_size)
1794 i_size_write(inode, pos+copied);
1795 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1796 if (new_i_size > EXT4_I(inode)->i_disksize) {
1797 ext4_update_i_disksize(inode, new_i_size);
1798 ret2 = ext4_mark_inode_dirty(handle, inode);
1804 page_cache_release(page);
1805 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1806 /* if we have allocated more blocks and copied
1807 * less. We will have blocks allocated outside
1808 * inode->i_size. So truncate them
1810 ext4_orphan_add(handle, inode);
1812 ret2 = ext4_journal_stop(handle);
1815 if (pos + len > inode->i_size) {
1816 ext4_truncate(inode);
1818 * If truncate failed early the inode might still be
1819 * on the orphan list; we need to make sure the inode
1820 * is removed from the orphan list in that case.
1823 ext4_orphan_del(NULL, inode);
1826 return ret ? ret : copied;
1829 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1833 unsigned long md_needed, mdblocks, total = 0;
1836 * recalculate the amount of metadata blocks to reserve
1837 * in order to allocate nrblocks
1838 * worse case is one extent per block
1841 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1842 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1843 mdblocks = ext4_calc_metadata_amount(inode, total);
1844 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1846 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1847 total = md_needed + nrblocks;
1850 * Make quota reservation here to prevent quota overflow
1851 * later. Real quota accounting is done at pages writeout
1854 if (vfs_dq_reserve_block(inode, total)) {
1855 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1859 if (ext4_claim_free_blocks(sbi, total)) {
1860 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1861 vfs_dq_release_reservation_block(inode, total);
1862 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1868 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1869 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1871 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1872 return 0; /* success */
1875 static void ext4_da_release_space(struct inode *inode, int to_free)
1877 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1878 int total, mdb, mdb_free, release;
1881 return; /* Nothing to release, exit */
1883 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1885 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1887 * if there is no reserved blocks, but we try to free some
1888 * then the counter is messed up somewhere.
1889 * but since this function is called from invalidate
1890 * page, it's harmless to return without any action
1892 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1893 "blocks for inode %lu, but there is no reserved "
1894 "data blocks\n", to_free, inode->i_ino);
1895 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1899 /* recalculate the number of metablocks still need to be reserved */
1900 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1901 mdb = ext4_calc_metadata_amount(inode, total);
1903 /* figure out how many metablocks to release */
1904 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1905 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1907 release = to_free + mdb_free;
1909 /* update fs dirty blocks counter for truncate case */
1910 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1912 /* update per-inode reservations */
1913 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1914 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1916 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1917 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1918 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1920 vfs_dq_release_reservation_block(inode, release);
1923 static void ext4_da_page_release_reservation(struct page *page,
1924 unsigned long offset)
1927 struct buffer_head *head, *bh;
1928 unsigned int curr_off = 0;
1930 head = page_buffers(page);
1933 unsigned int next_off = curr_off + bh->b_size;
1935 if ((offset <= curr_off) && (buffer_delay(bh))) {
1937 clear_buffer_delay(bh);
1939 curr_off = next_off;
1940 } while ((bh = bh->b_this_page) != head);
1941 ext4_da_release_space(page->mapping->host, to_release);
1945 * Delayed allocation stuff
1949 * mpage_da_submit_io - walks through extent of pages and try to write
1950 * them with writepage() call back
1952 * @mpd->inode: inode
1953 * @mpd->first_page: first page of the extent
1954 * @mpd->next_page: page after the last page of the extent
1956 * By the time mpage_da_submit_io() is called we expect all blocks
1957 * to be allocated. this may be wrong if allocation failed.
1959 * As pages are already locked by write_cache_pages(), we can't use it
1961 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1964 struct pagevec pvec;
1965 unsigned long index, end;
1966 int ret = 0, err, nr_pages, i;
1967 struct inode *inode = mpd->inode;
1968 struct address_space *mapping = inode->i_mapping;
1970 BUG_ON(mpd->next_page <= mpd->first_page);
1972 * We need to start from the first_page to the next_page - 1
1973 * to make sure we also write the mapped dirty buffer_heads.
1974 * If we look at mpd->b_blocknr we would only be looking
1975 * at the currently mapped buffer_heads.
1977 index = mpd->first_page;
1978 end = mpd->next_page - 1;
1980 pagevec_init(&pvec, 0);
1981 while (index <= end) {
1982 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1985 for (i = 0; i < nr_pages; i++) {
1986 struct page *page = pvec.pages[i];
1988 index = page->index;
1993 BUG_ON(!PageLocked(page));
1994 BUG_ON(PageWriteback(page));
1996 pages_skipped = mpd->wbc->pages_skipped;
1997 err = mapping->a_ops->writepage(page, mpd->wbc);
1998 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2000 * have successfully written the page
2001 * without skipping the same
2003 mpd->pages_written++;
2005 * In error case, we have to continue because
2006 * remaining pages are still locked
2007 * XXX: unlock and re-dirty them?
2012 pagevec_release(&pvec);
2018 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2020 * @mpd->inode - inode to walk through
2021 * @exbh->b_blocknr - first block on a disk
2022 * @exbh->b_size - amount of space in bytes
2023 * @logical - first logical block to start assignment with
2025 * the function goes through all passed space and put actual disk
2026 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2028 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
2029 struct buffer_head *exbh)
2031 struct inode *inode = mpd->inode;
2032 struct address_space *mapping = inode->i_mapping;
2033 int blocks = exbh->b_size >> inode->i_blkbits;
2034 sector_t pblock = exbh->b_blocknr, cur_logical;
2035 struct buffer_head *head, *bh;
2037 struct pagevec pvec;
2040 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2041 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2042 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2044 pagevec_init(&pvec, 0);
2046 while (index <= end) {
2047 /* XXX: optimize tail */
2048 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2051 for (i = 0; i < nr_pages; i++) {
2052 struct page *page = pvec.pages[i];
2054 index = page->index;
2059 BUG_ON(!PageLocked(page));
2060 BUG_ON(PageWriteback(page));
2061 BUG_ON(!page_has_buffers(page));
2063 bh = page_buffers(page);
2066 /* skip blocks out of the range */
2068 if (cur_logical >= logical)
2071 } while ((bh = bh->b_this_page) != head);
2074 if (cur_logical >= logical + blocks)
2077 if (buffer_delay(bh) ||
2078 buffer_unwritten(bh)) {
2080 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2082 if (buffer_delay(bh)) {
2083 clear_buffer_delay(bh);
2084 bh->b_blocknr = pblock;
2087 * unwritten already should have
2088 * blocknr assigned. Verify that
2090 clear_buffer_unwritten(bh);
2091 BUG_ON(bh->b_blocknr != pblock);
2094 } else if (buffer_mapped(bh))
2095 BUG_ON(bh->b_blocknr != pblock);
2099 } while ((bh = bh->b_this_page) != head);
2101 pagevec_release(&pvec);
2107 * __unmap_underlying_blocks - just a helper function to unmap
2108 * set of blocks described by @bh
2110 static inline void __unmap_underlying_blocks(struct inode *inode,
2111 struct buffer_head *bh)
2113 struct block_device *bdev = inode->i_sb->s_bdev;
2116 blocks = bh->b_size >> inode->i_blkbits;
2117 for (i = 0; i < blocks; i++)
2118 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
2121 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2122 sector_t logical, long blk_cnt)
2126 struct pagevec pvec;
2127 struct inode *inode = mpd->inode;
2128 struct address_space *mapping = inode->i_mapping;
2130 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2131 end = (logical + blk_cnt - 1) >>
2132 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2133 while (index <= end) {
2134 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2137 for (i = 0; i < nr_pages; i++) {
2138 struct page *page = pvec.pages[i];
2139 index = page->index;
2144 BUG_ON(!PageLocked(page));
2145 BUG_ON(PageWriteback(page));
2146 block_invalidatepage(page, 0);
2147 ClearPageUptodate(page);
2154 static void ext4_print_free_blocks(struct inode *inode)
2156 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2157 printk(KERN_CRIT "Total free blocks count %lld\n",
2158 ext4_count_free_blocks(inode->i_sb));
2159 printk(KERN_CRIT "Free/Dirty block details\n");
2160 printk(KERN_CRIT "free_blocks=%lld\n",
2161 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2162 printk(KERN_CRIT "dirty_blocks=%lld\n",
2163 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2164 printk(KERN_CRIT "Block reservation details\n");
2165 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2166 EXT4_I(inode)->i_reserved_data_blocks);
2167 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2168 EXT4_I(inode)->i_reserved_meta_blocks);
2173 * mpage_da_map_blocks - go through given space
2175 * @mpd - bh describing space
2177 * The function skips space we know is already mapped to disk blocks.
2180 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2182 int err, blks, get_blocks_flags;
2183 struct buffer_head new;
2184 sector_t next = mpd->b_blocknr;
2185 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2186 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2187 handle_t *handle = NULL;
2190 * We consider only non-mapped and non-allocated blocks
2192 if ((mpd->b_state & (1 << BH_Mapped)) &&
2193 !(mpd->b_state & (1 << BH_Delay)) &&
2194 !(mpd->b_state & (1 << BH_Unwritten)))
2198 * If we didn't accumulate anything to write simply return
2203 handle = ext4_journal_current_handle();
2207 * Call ext4_get_blocks() to allocate any delayed allocation
2208 * blocks, or to convert an uninitialized extent to be
2209 * initialized (in the case where we have written into
2210 * one or more preallocated blocks).
2212 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2213 * indicate that we are on the delayed allocation path. This
2214 * affects functions in many different parts of the allocation
2215 * call path. This flag exists primarily because we don't
2216 * want to change *many* call functions, so ext4_get_blocks()
2217 * will set the magic i_delalloc_reserved_flag once the
2218 * inode's allocation semaphore is taken.
2220 * If the blocks in questions were delalloc blocks, set
2221 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2222 * variables are updated after the blocks have been allocated.
2225 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2226 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2227 if (mpd->b_state & (1 << BH_Delay))
2228 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2229 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2230 &new, get_blocks_flags);
2234 * If get block returns with error we simply
2235 * return. Later writepage will redirty the page and
2236 * writepages will find the dirty page again
2241 if (err == -ENOSPC &&
2242 ext4_count_free_blocks(mpd->inode->i_sb)) {
2248 * get block failure will cause us to loop in
2249 * writepages, because a_ops->writepage won't be able
2250 * to make progress. The page will be redirtied by
2251 * writepage and writepages will again try to write
2254 ext4_msg(mpd->inode->i_sb, KERN_CRIT,
2255 "delayed block allocation failed for inode %lu at "
2256 "logical offset %llu with max blocks %zd with "
2257 "error %d\n", mpd->inode->i_ino,
2258 (unsigned long long) next,
2259 mpd->b_size >> mpd->inode->i_blkbits, err);
2260 printk(KERN_CRIT "This should not happen!! "
2261 "Data will be lost\n");
2262 if (err == -ENOSPC) {
2263 ext4_print_free_blocks(mpd->inode);
2265 /* invalidate all the pages */
2266 ext4_da_block_invalidatepages(mpd, next,
2267 mpd->b_size >> mpd->inode->i_blkbits);
2272 new.b_size = (blks << mpd->inode->i_blkbits);
2274 if (buffer_new(&new))
2275 __unmap_underlying_blocks(mpd->inode, &new);
2278 * If blocks are delayed marked, we need to
2279 * put actual blocknr and drop delayed bit
2281 if ((mpd->b_state & (1 << BH_Delay)) ||
2282 (mpd->b_state & (1 << BH_Unwritten)))
2283 mpage_put_bnr_to_bhs(mpd, next, &new);
2285 if (ext4_should_order_data(mpd->inode)) {
2286 err = ext4_jbd2_file_inode(handle, mpd->inode);
2292 * Update on-disk size along with block allocation.
2294 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2295 if (disksize > i_size_read(mpd->inode))
2296 disksize = i_size_read(mpd->inode);
2297 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2298 ext4_update_i_disksize(mpd->inode, disksize);
2299 return ext4_mark_inode_dirty(handle, mpd->inode);
2305 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2306 (1 << BH_Delay) | (1 << BH_Unwritten))
2309 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2311 * @mpd->lbh - extent of blocks
2312 * @logical - logical number of the block in the file
2313 * @bh - bh of the block (used to access block's state)
2315 * the function is used to collect contig. blocks in same state
2317 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2318 sector_t logical, size_t b_size,
2319 unsigned long b_state)
2322 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2324 /* check if thereserved journal credits might overflow */
2325 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2326 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2328 * With non-extent format we are limited by the journal
2329 * credit available. Total credit needed to insert
2330 * nrblocks contiguous blocks is dependent on the
2331 * nrblocks. So limit nrblocks.
2334 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2335 EXT4_MAX_TRANS_DATA) {
2337 * Adding the new buffer_head would make it cross the
2338 * allowed limit for which we have journal credit
2339 * reserved. So limit the new bh->b_size
2341 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2342 mpd->inode->i_blkbits;
2343 /* we will do mpage_da_submit_io in the next loop */
2347 * First block in the extent
2349 if (mpd->b_size == 0) {
2350 mpd->b_blocknr = logical;
2351 mpd->b_size = b_size;
2352 mpd->b_state = b_state & BH_FLAGS;
2356 next = mpd->b_blocknr + nrblocks;
2358 * Can we merge the block to our big extent?
2360 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2361 mpd->b_size += b_size;
2367 * We couldn't merge the block to our extent, so we
2368 * need to flush current extent and start new one
2370 if (mpage_da_map_blocks(mpd) == 0)
2371 mpage_da_submit_io(mpd);
2376 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2378 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2382 * __mpage_da_writepage - finds extent of pages and blocks
2384 * @page: page to consider
2385 * @wbc: not used, we just follow rules
2388 * The function finds extents of pages and scan them for all blocks.
2390 static int __mpage_da_writepage(struct page *page,
2391 struct writeback_control *wbc, void *data)
2393 struct mpage_da_data *mpd = data;
2394 struct inode *inode = mpd->inode;
2395 struct buffer_head *bh, *head;
2400 * Rest of the page in the page_vec
2401 * redirty then and skip then. We will
2402 * try to write them again after
2403 * starting a new transaction
2405 redirty_page_for_writepage(wbc, page);
2407 return MPAGE_DA_EXTENT_TAIL;
2410 * Can we merge this page to current extent?
2412 if (mpd->next_page != page->index) {
2414 * Nope, we can't. So, we map non-allocated blocks
2415 * and start IO on them using writepage()
2417 if (mpd->next_page != mpd->first_page) {
2418 if (mpage_da_map_blocks(mpd) == 0)
2419 mpage_da_submit_io(mpd);
2421 * skip rest of the page in the page_vec
2424 redirty_page_for_writepage(wbc, page);
2426 return MPAGE_DA_EXTENT_TAIL;
2430 * Start next extent of pages ...
2432 mpd->first_page = page->index;
2442 mpd->next_page = page->index + 1;
2443 logical = (sector_t) page->index <<
2444 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2446 if (!page_has_buffers(page)) {
2447 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2448 (1 << BH_Dirty) | (1 << BH_Uptodate));
2450 return MPAGE_DA_EXTENT_TAIL;
2453 * Page with regular buffer heads, just add all dirty ones
2455 head = page_buffers(page);
2458 BUG_ON(buffer_locked(bh));
2460 * We need to try to allocate
2461 * unmapped blocks in the same page.
2462 * Otherwise we won't make progress
2463 * with the page in ext4_writepage
2465 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2466 mpage_add_bh_to_extent(mpd, logical,
2470 return MPAGE_DA_EXTENT_TAIL;
2471 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2473 * mapped dirty buffer. We need to update
2474 * the b_state because we look at
2475 * b_state in mpage_da_map_blocks. We don't
2476 * update b_size because if we find an
2477 * unmapped buffer_head later we need to
2478 * use the b_state flag of that buffer_head.
2480 if (mpd->b_size == 0)
2481 mpd->b_state = bh->b_state & BH_FLAGS;
2484 } while ((bh = bh->b_this_page) != head);
2491 * This is a special get_blocks_t callback which is used by
2492 * ext4_da_write_begin(). It will either return mapped block or
2493 * reserve space for a single block.
2495 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2496 * We also have b_blocknr = -1 and b_bdev initialized properly
2498 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2499 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2500 * initialized properly.
2502 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2503 struct buffer_head *bh_result, int create)
2506 sector_t invalid_block = ~((sector_t) 0xffff);
2508 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2511 BUG_ON(create == 0);
2512 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2515 * first, we need to know whether the block is allocated already
2516 * preallocated blocks are unmapped but should treated
2517 * the same as allocated blocks.
2519 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2520 if ((ret == 0) && !buffer_delay(bh_result)) {
2521 /* the block isn't (pre)allocated yet, let's reserve space */
2523 * XXX: __block_prepare_write() unmaps passed block,
2526 ret = ext4_da_reserve_space(inode, 1);
2528 /* not enough space to reserve */
2531 map_bh(bh_result, inode->i_sb, invalid_block);
2532 set_buffer_new(bh_result);
2533 set_buffer_delay(bh_result);
2534 } else if (ret > 0) {
2535 bh_result->b_size = (ret << inode->i_blkbits);
2536 if (buffer_unwritten(bh_result)) {
2537 /* A delayed write to unwritten bh should
2538 * be marked new and mapped. Mapped ensures
2539 * that we don't do get_block multiple times
2540 * when we write to the same offset and new
2541 * ensures that we do proper zero out for
2544 set_buffer_new(bh_result);
2545 set_buffer_mapped(bh_result);
2554 * This function is used as a standard get_block_t calback function
2555 * when there is no desire to allocate any blocks. It is used as a
2556 * callback function for block_prepare_write(), nobh_writepage(), and
2557 * block_write_full_page(). These functions should only try to map a
2558 * single block at a time.
2560 * Since this function doesn't do block allocations even if the caller
2561 * requests it by passing in create=1, it is critically important that
2562 * any caller checks to make sure that any buffer heads are returned
2563 * by this function are either all already mapped or marked for
2564 * delayed allocation before calling nobh_writepage() or
2565 * block_write_full_page(). Otherwise, b_blocknr could be left
2566 * unitialized, and the page write functions will be taken by
2569 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2570 struct buffer_head *bh_result, int create)
2573 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2575 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2578 * we don't want to do block allocation in writepage
2579 * so call get_block_wrap with create = 0
2581 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2583 bh_result->b_size = (ret << inode->i_blkbits);
2589 static int bget_one(handle_t *handle, struct buffer_head *bh)
2595 static int bput_one(handle_t *handle, struct buffer_head *bh)
2601 static int __ext4_journalled_writepage(struct page *page,
2602 struct writeback_control *wbc,
2605 struct address_space *mapping = page->mapping;
2606 struct inode *inode = mapping->host;
2607 struct buffer_head *page_bufs;
2608 handle_t *handle = NULL;
2612 page_bufs = page_buffers(page);
2614 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2615 /* As soon as we unlock the page, it can go away, but we have
2616 * references to buffers so we are safe */
2619 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2620 if (IS_ERR(handle)) {
2621 ret = PTR_ERR(handle);
2625 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2626 do_journal_get_write_access);
2628 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2632 err = ext4_journal_stop(handle);
2636 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2637 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2643 * Note that we don't need to start a transaction unless we're journaling data
2644 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2645 * need to file the inode to the transaction's list in ordered mode because if
2646 * we are writing back data added by write(), the inode is already there and if
2647 * we are writing back data modified via mmap(), noone guarantees in which
2648 * transaction the data will hit the disk. In case we are journaling data, we
2649 * cannot start transaction directly because transaction start ranks above page
2650 * lock so we have to do some magic.
2652 * This function can get called via...
2653 * - ext4_da_writepages after taking page lock (have journal handle)
2654 * - journal_submit_inode_data_buffers (no journal handle)
2655 * - shrink_page_list via pdflush (no journal handle)
2656 * - grab_page_cache when doing write_begin (have journal handle)
2658 * We don't do any block allocation in this function. If we have page with
2659 * multiple blocks we need to write those buffer_heads that are mapped. This
2660 * is important for mmaped based write. So if we do with blocksize 1K
2661 * truncate(f, 1024);
2662 * a = mmap(f, 0, 4096);
2664 * truncate(f, 4096);
2665 * we have in the page first buffer_head mapped via page_mkwrite call back
2666 * but other bufer_heads would be unmapped but dirty(dirty done via the
2667 * do_wp_page). So writepage should write the first block. If we modify
2668 * the mmap area beyond 1024 we will again get a page_fault and the
2669 * page_mkwrite callback will do the block allocation and mark the
2670 * buffer_heads mapped.
2672 * We redirty the page if we have any buffer_heads that is either delay or
2673 * unwritten in the page.
2675 * We can get recursively called as show below.
2677 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2680 * But since we don't do any block allocation we should not deadlock.
2681 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2683 static int ext4_writepage(struct page *page,
2684 struct writeback_control *wbc)
2689 struct buffer_head *page_bufs;
2690 struct inode *inode = page->mapping->host;
2692 trace_ext4_writepage(inode, page);
2693 size = i_size_read(inode);
2694 if (page->index == size >> PAGE_CACHE_SHIFT)
2695 len = size & ~PAGE_CACHE_MASK;
2697 len = PAGE_CACHE_SIZE;
2699 if (page_has_buffers(page)) {
2700 page_bufs = page_buffers(page);
2701 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2702 ext4_bh_delay_or_unwritten)) {
2704 * We don't want to do block allocation
2705 * So redirty the page and return
2706 * We may reach here when we do a journal commit
2707 * via journal_submit_inode_data_buffers.
2708 * If we don't have mapping block we just ignore
2709 * them. We can also reach here via shrink_page_list
2711 redirty_page_for_writepage(wbc, page);
2717 * The test for page_has_buffers() is subtle:
2718 * We know the page is dirty but it lost buffers. That means
2719 * that at some moment in time after write_begin()/write_end()
2720 * has been called all buffers have been clean and thus they
2721 * must have been written at least once. So they are all
2722 * mapped and we can happily proceed with mapping them
2723 * and writing the page.
2725 * Try to initialize the buffer_heads and check whether
2726 * all are mapped and non delay. We don't want to
2727 * do block allocation here.
2729 ret = block_prepare_write(page, 0, len,
2730 noalloc_get_block_write);
2732 page_bufs = page_buffers(page);
2733 /* check whether all are mapped and non delay */
2734 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2735 ext4_bh_delay_or_unwritten)) {
2736 redirty_page_for_writepage(wbc, page);
2742 * We can't do block allocation here
2743 * so just redity the page and unlock
2746 redirty_page_for_writepage(wbc, page);
2750 /* now mark the buffer_heads as dirty and uptodate */
2751 block_commit_write(page, 0, len);
2754 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2756 * It's mmapped pagecache. Add buffers and journal it. There
2757 * doesn't seem much point in redirtying the page here.
2759 ClearPageChecked(page);
2760 return __ext4_journalled_writepage(page, wbc, len);
2763 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2764 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2766 ret = block_write_full_page(page, noalloc_get_block_write,
2773 * This is called via ext4_da_writepages() to
2774 * calulate the total number of credits to reserve to fit
2775 * a single extent allocation into a single transaction,
2776 * ext4_da_writpeages() will loop calling this before
2777 * the block allocation.
2780 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2782 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2785 * With non-extent format the journal credit needed to
2786 * insert nrblocks contiguous block is dependent on
2787 * number of contiguous block. So we will limit
2788 * number of contiguous block to a sane value
2790 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2791 (max_blocks > EXT4_MAX_TRANS_DATA))
2792 max_blocks = EXT4_MAX_TRANS_DATA;
2794 return ext4_chunk_trans_blocks(inode, max_blocks);
2797 static int ext4_da_writepages(struct address_space *mapping,
2798 struct writeback_control *wbc)
2801 int range_whole = 0;
2802 handle_t *handle = NULL;
2803 struct mpage_da_data mpd;
2804 struct inode *inode = mapping->host;
2805 int no_nrwrite_index_update;
2806 int pages_written = 0;
2808 unsigned int max_pages;
2809 int range_cyclic, cycled = 1, io_done = 0;
2810 int needed_blocks, ret = 0;
2811 long desired_nr_to_write, nr_to_writebump = 0;
2812 loff_t range_start = wbc->range_start;
2813 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2815 trace_ext4_da_writepages(inode, wbc);
2818 * No pages to write? This is mainly a kludge to avoid starting
2819 * a transaction for special inodes like journal inode on last iput()
2820 * because that could violate lock ordering on umount
2822 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2826 * If the filesystem has aborted, it is read-only, so return
2827 * right away instead of dumping stack traces later on that
2828 * will obscure the real source of the problem. We test
2829 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2830 * the latter could be true if the filesystem is mounted
2831 * read-only, and in that case, ext4_da_writepages should
2832 * *never* be called, so if that ever happens, we would want
2835 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2838 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2841 range_cyclic = wbc->range_cyclic;
2842 if (wbc->range_cyclic) {
2843 index = mapping->writeback_index;
2846 wbc->range_start = index << PAGE_CACHE_SHIFT;
2847 wbc->range_end = LLONG_MAX;
2848 wbc->range_cyclic = 0;
2850 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2853 * This works around two forms of stupidity. The first is in
2854 * the writeback code, which caps the maximum number of pages
2855 * written to be 1024 pages. This is wrong on multiple
2856 * levels; different architectues have a different page size,
2857 * which changes the maximum amount of data which gets
2858 * written. Secondly, 4 megabytes is way too small. XFS
2859 * forces this value to be 16 megabytes by multiplying
2860 * nr_to_write parameter by four, and then relies on its
2861 * allocator to allocate larger extents to make them
2862 * contiguous. Unfortunately this brings us to the second
2863 * stupidity, which is that ext4's mballoc code only allocates
2864 * at most 2048 blocks. So we force contiguous writes up to
2865 * the number of dirty blocks in the inode, or
2866 * sbi->max_writeback_mb_bump whichever is smaller.
2868 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2869 if (!range_cyclic && range_whole)
2870 desired_nr_to_write = wbc->nr_to_write * 8;
2872 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2874 if (desired_nr_to_write > max_pages)
2875 desired_nr_to_write = max_pages;
2877 if (wbc->nr_to_write < desired_nr_to_write) {
2878 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2879 wbc->nr_to_write = desired_nr_to_write;
2883 mpd.inode = mapping->host;
2886 * we don't want write_cache_pages to update
2887 * nr_to_write and writeback_index
2889 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2890 wbc->no_nrwrite_index_update = 1;
2891 pages_skipped = wbc->pages_skipped;
2894 while (!ret && wbc->nr_to_write > 0) {
2897 * we insert one extent at a time. So we need
2898 * credit needed for single extent allocation.
2899 * journalled mode is currently not supported
2902 BUG_ON(ext4_should_journal_data(inode));
2903 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2905 /* start a new transaction*/
2906 handle = ext4_journal_start(inode, needed_blocks);
2907 if (IS_ERR(handle)) {
2908 ret = PTR_ERR(handle);
2909 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2910 "%ld pages, ino %lu; err %d\n", __func__,
2911 wbc->nr_to_write, inode->i_ino, ret);
2912 goto out_writepages;
2916 * Now call __mpage_da_writepage to find the next
2917 * contiguous region of logical blocks that need
2918 * blocks to be allocated by ext4. We don't actually
2919 * submit the blocks for I/O here, even though
2920 * write_cache_pages thinks it will, and will set the
2921 * pages as clean for write before calling
2922 * __mpage_da_writepage().
2930 mpd.pages_written = 0;
2932 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2935 * If we have a contigous extent of pages and we
2936 * haven't done the I/O yet, map the blocks and submit
2939 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2940 if (mpage_da_map_blocks(&mpd) == 0)
2941 mpage_da_submit_io(&mpd);
2943 ret = MPAGE_DA_EXTENT_TAIL;
2945 trace_ext4_da_write_pages(inode, &mpd);
2946 wbc->nr_to_write -= mpd.pages_written;
2948 ext4_journal_stop(handle);
2950 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2951 /* commit the transaction which would
2952 * free blocks released in the transaction
2955 jbd2_journal_force_commit_nested(sbi->s_journal);
2956 wbc->pages_skipped = pages_skipped;
2958 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2960 * got one extent now try with
2963 pages_written += mpd.pages_written;
2964 wbc->pages_skipped = pages_skipped;
2967 } else if (wbc->nr_to_write)
2969 * There is no more writeout needed
2970 * or we requested for a noblocking writeout
2971 * and we found the device congested
2975 if (!io_done && !cycled) {
2978 wbc->range_start = index << PAGE_CACHE_SHIFT;
2979 wbc->range_end = mapping->writeback_index - 1;
2982 if (pages_skipped != wbc->pages_skipped)
2983 ext4_msg(inode->i_sb, KERN_CRIT,
2984 "This should not happen leaving %s "
2985 "with nr_to_write = %ld ret = %d\n",
2986 __func__, wbc->nr_to_write, ret);
2989 index += pages_written;
2990 wbc->range_cyclic = range_cyclic;
2991 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2993 * set the writeback_index so that range_cyclic
2994 * mode will write it back later
2996 mapping->writeback_index = index;
2999 if (!no_nrwrite_index_update)
3000 wbc->no_nrwrite_index_update = 0;
3001 if (wbc->nr_to_write > nr_to_writebump)
3002 wbc->nr_to_write -= nr_to_writebump;
3003 wbc->range_start = range_start;
3004 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3008 #define FALL_BACK_TO_NONDELALLOC 1
3009 static int ext4_nonda_switch(struct super_block *sb)
3011 s64 free_blocks, dirty_blocks;
3012 struct ext4_sb_info *sbi = EXT4_SB(sb);
3015 * switch to non delalloc mode if we are running low
3016 * on free block. The free block accounting via percpu
3017 * counters can get slightly wrong with percpu_counter_batch getting
3018 * accumulated on each CPU without updating global counters
3019 * Delalloc need an accurate free block accounting. So switch
3020 * to non delalloc when we are near to error range.
3022 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3023 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3024 if (2 * free_blocks < 3 * dirty_blocks ||
3025 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3027 * free block count is less that 150% of dirty blocks
3028 * or free blocks is less that watermark
3035 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3036 loff_t pos, unsigned len, unsigned flags,
3037 struct page **pagep, void **fsdata)
3039 int ret, retries = 0;
3043 struct inode *inode = mapping->host;
3046 index = pos >> PAGE_CACHE_SHIFT;
3047 from = pos & (PAGE_CACHE_SIZE - 1);
3050 if (ext4_nonda_switch(inode->i_sb)) {
3051 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3052 return ext4_write_begin(file, mapping, pos,
3053 len, flags, pagep, fsdata);
3055 *fsdata = (void *)0;
3056 trace_ext4_da_write_begin(inode, pos, len, flags);
3059 * With delayed allocation, we don't log the i_disksize update
3060 * if there is delayed block allocation. But we still need
3061 * to journalling the i_disksize update if writes to the end
3062 * of file which has an already mapped buffer.
3064 handle = ext4_journal_start(inode, 1);
3065 if (IS_ERR(handle)) {
3066 ret = PTR_ERR(handle);
3069 /* We cannot recurse into the filesystem as the transaction is already
3071 flags |= AOP_FLAG_NOFS;
3073 page = grab_cache_page_write_begin(mapping, index, flags);
3075 ext4_journal_stop(handle);
3081 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
3082 ext4_da_get_block_prep);
3085 ext4_journal_stop(handle);
3086 page_cache_release(page);
3088 * block_write_begin may have instantiated a few blocks
3089 * outside i_size. Trim these off again. Don't need
3090 * i_size_read because we hold i_mutex.
3092 if (pos + len > inode->i_size)
3093 ext4_truncate(inode);
3096 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3103 * Check if we should update i_disksize
3104 * when write to the end of file but not require block allocation
3106 static int ext4_da_should_update_i_disksize(struct page *page,
3107 unsigned long offset)
3109 struct buffer_head *bh;
3110 struct inode *inode = page->mapping->host;
3114 bh = page_buffers(page);
3115 idx = offset >> inode->i_blkbits;
3117 for (i = 0; i < idx; i++)
3118 bh = bh->b_this_page;
3120 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3125 static int ext4_da_write_end(struct file *file,
3126 struct address_space *mapping,
3127 loff_t pos, unsigned len, unsigned copied,
3128 struct page *page, void *fsdata)
3130 struct inode *inode = mapping->host;
3132 handle_t *handle = ext4_journal_current_handle();
3134 unsigned long start, end;
3135 int write_mode = (int)(unsigned long)fsdata;
3137 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3138 if (ext4_should_order_data(inode)) {
3139 return ext4_ordered_write_end(file, mapping, pos,
3140 len, copied, page, fsdata);
3141 } else if (ext4_should_writeback_data(inode)) {
3142 return ext4_writeback_write_end(file, mapping, pos,
3143 len, copied, page, fsdata);
3149 trace_ext4_da_write_end(inode, pos, len, copied);
3150 start = pos & (PAGE_CACHE_SIZE - 1);
3151 end = start + copied - 1;
3154 * generic_write_end() will run mark_inode_dirty() if i_size
3155 * changes. So let's piggyback the i_disksize mark_inode_dirty
3159 new_i_size = pos + copied;
3160 if (new_i_size > EXT4_I(inode)->i_disksize) {
3161 if (ext4_da_should_update_i_disksize(page, end)) {
3162 down_write(&EXT4_I(inode)->i_data_sem);
3163 if (new_i_size > EXT4_I(inode)->i_disksize) {
3165 * Updating i_disksize when extending file
3166 * without needing block allocation
3168 if (ext4_should_order_data(inode))
3169 ret = ext4_jbd2_file_inode(handle,
3172 EXT4_I(inode)->i_disksize = new_i_size;
3174 up_write(&EXT4_I(inode)->i_data_sem);
3175 /* We need to mark inode dirty even if
3176 * new_i_size is less that inode->i_size
3177 * bu greater than i_disksize.(hint delalloc)
3179 ext4_mark_inode_dirty(handle, inode);
3182 ret2 = generic_write_end(file, mapping, pos, len, copied,
3187 ret2 = ext4_journal_stop(handle);
3191 return ret ? ret : copied;
3194 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3197 * Drop reserved blocks
3199 BUG_ON(!PageLocked(page));
3200 if (!page_has_buffers(page))
3203 ext4_da_page_release_reservation(page, offset);
3206 ext4_invalidatepage(page, offset);
3212 * Force all delayed allocation blocks to be allocated for a given inode.
3214 int ext4_alloc_da_blocks(struct inode *inode)
3216 trace_ext4_alloc_da_blocks(inode);
3218 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3219 !EXT4_I(inode)->i_reserved_meta_blocks)
3223 * We do something simple for now. The filemap_flush() will
3224 * also start triggering a write of the data blocks, which is
3225 * not strictly speaking necessary (and for users of
3226 * laptop_mode, not even desirable). However, to do otherwise
3227 * would require replicating code paths in:
3229 * ext4_da_writepages() ->
3230 * write_cache_pages() ---> (via passed in callback function)
3231 * __mpage_da_writepage() -->
3232 * mpage_add_bh_to_extent()
3233 * mpage_da_map_blocks()
3235 * The problem is that write_cache_pages(), located in
3236 * mm/page-writeback.c, marks pages clean in preparation for
3237 * doing I/O, which is not desirable if we're not planning on
3240 * We could call write_cache_pages(), and then redirty all of
3241 * the pages by calling redirty_page_for_writeback() but that
3242 * would be ugly in the extreme. So instead we would need to
3243 * replicate parts of the code in the above functions,
3244 * simplifying them becuase we wouldn't actually intend to
3245 * write out the pages, but rather only collect contiguous
3246 * logical block extents, call the multi-block allocator, and
3247 * then update the buffer heads with the block allocations.
3249 * For now, though, we'll cheat by calling filemap_flush(),
3250 * which will map the blocks, and start the I/O, but not
3251 * actually wait for the I/O to complete.
3253 return filemap_flush(inode->i_mapping);
3257 * bmap() is special. It gets used by applications such as lilo and by
3258 * the swapper to find the on-disk block of a specific piece of data.
3260 * Naturally, this is dangerous if the block concerned is still in the
3261 * journal. If somebody makes a swapfile on an ext4 data-journaling
3262 * filesystem and enables swap, then they may get a nasty shock when the
3263 * data getting swapped to that swapfile suddenly gets overwritten by
3264 * the original zero's written out previously to the journal and
3265 * awaiting writeback in the kernel's buffer cache.
3267 * So, if we see any bmap calls here on a modified, data-journaled file,
3268 * take extra steps to flush any blocks which might be in the cache.
3270 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3272 struct inode *inode = mapping->host;
3276 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3277 test_opt(inode->i_sb, DELALLOC)) {
3279 * With delalloc we want to sync the file
3280 * so that we can make sure we allocate
3283 filemap_write_and_wait(mapping);
3286 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3288 * This is a REALLY heavyweight approach, but the use of
3289 * bmap on dirty files is expected to be extremely rare:
3290 * only if we run lilo or swapon on a freshly made file
3291 * do we expect this to happen.
3293 * (bmap requires CAP_SYS_RAWIO so this does not
3294 * represent an unprivileged user DOS attack --- we'd be
3295 * in trouble if mortal users could trigger this path at
3298 * NB. EXT4_STATE_JDATA is not set on files other than
3299 * regular files. If somebody wants to bmap a directory
3300 * or symlink and gets confused because the buffer
3301 * hasn't yet been flushed to disk, they deserve
3302 * everything they get.
3305 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3306 journal = EXT4_JOURNAL(inode);
3307 jbd2_journal_lock_updates(journal);
3308 err = jbd2_journal_flush(journal);
3309 jbd2_journal_unlock_updates(journal);
3315 return generic_block_bmap(mapping, block, ext4_get_block);
3318 static int ext4_readpage(struct file *file, struct page *page)
3320 return mpage_readpage(page, ext4_get_block);
3324 ext4_readpages(struct file *file, struct address_space *mapping,
3325 struct list_head *pages, unsigned nr_pages)
3327 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3330 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3332 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3335 * If it's a full truncate we just forget about the pending dirtying
3338 ClearPageChecked(page);
3341 jbd2_journal_invalidatepage(journal, page, offset);
3343 block_invalidatepage(page, offset);
3346 static int ext4_releasepage(struct page *page, gfp_t wait)
3348 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3350 WARN_ON(PageChecked(page));
3351 if (!page_has_buffers(page))
3354 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3356 return try_to_free_buffers(page);
3360 * O_DIRECT for ext3 (or indirect map) based files
3362 * If the O_DIRECT write will extend the file then add this inode to the
3363 * orphan list. So recovery will truncate it back to the original size
3364 * if the machine crashes during the write.
3366 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3367 * crashes then stale disk data _may_ be exposed inside the file. But current
3368 * VFS code falls back into buffered path in that case so we are safe.
3370 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3371 const struct iovec *iov, loff_t offset,
3372 unsigned long nr_segs)
3374 struct file *file = iocb->ki_filp;
3375 struct inode *inode = file->f_mapping->host;
3376 struct ext4_inode_info *ei = EXT4_I(inode);
3380 size_t count = iov_length(iov, nr_segs);
3383 loff_t final_size = offset + count;
3385 if (final_size > inode->i_size) {
3386 /* Credits for sb + inode write */
3387 handle = ext4_journal_start(inode, 2);
3388 if (IS_ERR(handle)) {
3389 ret = PTR_ERR(handle);
3392 ret = ext4_orphan_add(handle, inode);
3394 ext4_journal_stop(handle);
3398 ei->i_disksize = inode->i_size;
3399 ext4_journal_stop(handle);
3403 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3405 ext4_get_block, NULL);
3410 /* Credits for sb + inode write */
3411 handle = ext4_journal_start(inode, 2);
3412 if (IS_ERR(handle)) {
3413 /* This is really bad luck. We've written the data
3414 * but cannot extend i_size. Bail out and pretend
3415 * the write failed... */
3416 ret = PTR_ERR(handle);
3420 ext4_orphan_del(handle, inode);
3422 loff_t end = offset + ret;
3423 if (end > inode->i_size) {
3424 ei->i_disksize = end;
3425 i_size_write(inode, end);
3427 * We're going to return a positive `ret'
3428 * here due to non-zero-length I/O, so there's
3429 * no way of reporting error returns from
3430 * ext4_mark_inode_dirty() to userspace. So
3433 ext4_mark_inode_dirty(handle, inode);
3436 err = ext4_journal_stop(handle);
3444 /* Maximum number of blocks we map for direct IO at once. */
3446 static int ext4_get_block_dio_write(struct inode *inode, sector_t iblock,
3447 struct buffer_head *bh_result, int create)
3449 handle_t *handle = NULL;
3451 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
3454 ext4_debug("ext4_get_block_dio_write: inode %lu, create flag %d\n",
3455 inode->i_ino, create);
3457 * DIO VFS code passes create = 0 flag for write to
3458 * the middle of file. It does this to avoid block
3459 * allocation for holes, to prevent expose stale data
3460 * out when there is parallel buffered read (which does
3461 * not hold the i_mutex lock) while direct IO write has
3462 * not completed. DIO request on holes finally falls back
3463 * to buffered IO for this reason.
3465 * For ext4 extent based file, since we support fallocate,
3466 * new allocated extent as uninitialized, for holes, we
3467 * could fallocate blocks for holes, thus parallel
3468 * buffered IO read will zero out the page when read on
3469 * a hole while parallel DIO write to the hole has not completed.
3471 * when we come here, we know it's a direct IO write to
3472 * to the middle of file (<i_size)
3473 * so it's safe to override the create flag from VFS.
3475 create = EXT4_GET_BLOCKS_DIO_CREATE_EXT;
3477 if (max_blocks > DIO_MAX_BLOCKS)
3478 max_blocks = DIO_MAX_BLOCKS;
3479 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
3480 handle = ext4_journal_start(inode, dio_credits);
3481 if (IS_ERR(handle)) {
3482 ret = PTR_ERR(handle);
3485 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
3488 bh_result->b_size = (ret << inode->i_blkbits);
3491 ext4_journal_stop(handle);
3496 static void ext4_free_io_end(ext4_io_end_t *io)
3502 static void dump_aio_dio_list(struct inode * inode)
3505 struct list_head *cur, *before, *after;
3506 ext4_io_end_t *io, *io0, *io1;
3508 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3509 ext4_debug("inode %lu aio dio list is empty\n", inode->i_ino);
3513 ext4_debug("Dump inode %lu aio_dio_completed_IO list \n", inode->i_ino);
3514 list_for_each_entry(io, &EXT4_I(inode)->i_aio_dio_complete_list, list){
3517 io0 = container_of(before, ext4_io_end_t, list);
3519 io1 = container_of(after, ext4_io_end_t, list);
3521 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3522 io, inode->i_ino, io0, io1);
3528 * check a range of space and convert unwritten extents to written.
3530 static int ext4_end_aio_dio_nolock(ext4_io_end_t *io)
3532 struct inode *inode = io->inode;
3533 loff_t offset = io->offset;
3534 size_t size = io->size;
3537 ext4_debug("end_aio_dio_onlock: io 0x%p from inode %lu,list->next 0x%p,"
3538 "list->prev 0x%p\n",
3539 io, inode->i_ino, io->list.next, io->list.prev);
3541 if (list_empty(&io->list))
3544 if (io->flag != DIO_AIO_UNWRITTEN)
3547 if (offset + size <= i_size_read(inode))
3548 ret = ext4_convert_unwritten_extents(inode, offset, size);
3551 printk(KERN_EMERG "%s: failed to convert unwritten"
3552 "extents to written extents, error is %d"
3553 " io is still on inode %lu aio dio list\n",
3554 __func__, ret, inode->i_ino);
3558 /* clear the DIO AIO unwritten flag */
3563 * work on completed aio dio IO, to convert unwritten extents to extents
3565 static void ext4_end_aio_dio_work(struct work_struct *work)
3567 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3568 struct inode *inode = io->inode;
3571 mutex_lock(&inode->i_mutex);
3572 ret = ext4_end_aio_dio_nolock(io);
3574 if (!list_empty(&io->list))
3575 list_del_init(&io->list);
3576 ext4_free_io_end(io);
3578 mutex_unlock(&inode->i_mutex);
3581 * This function is called from ext4_sync_file().
3583 * When AIO DIO IO is completed, the work to convert unwritten
3584 * extents to written is queued on workqueue but may not get immediately
3585 * scheduled. When fsync is called, we need to ensure the
3586 * conversion is complete before fsync returns.
3587 * The inode keeps track of a list of completed AIO from DIO path
3588 * that might needs to do the conversion. This function walks through
3589 * the list and convert the related unwritten extents to written.
3591 int flush_aio_dio_completed_IO(struct inode *inode)
3597 if (list_empty(&EXT4_I(inode)->i_aio_dio_complete_list))
3600 dump_aio_dio_list(inode);
3601 while (!list_empty(&EXT4_I(inode)->i_aio_dio_complete_list)){
3602 io = list_entry(EXT4_I(inode)->i_aio_dio_complete_list.next,
3603 ext4_io_end_t, list);
3605 * Calling ext4_end_aio_dio_nolock() to convert completed
3608 * When ext4_sync_file() is called, run_queue() may already
3609 * about to flush the work corresponding to this io structure.
3610 * It will be upset if it founds the io structure related
3611 * to the work-to-be schedule is freed.
3613 * Thus we need to keep the io structure still valid here after
3614 * convertion finished. The io structure has a flag to
3615 * avoid double converting from both fsync and background work
3618 ret = ext4_end_aio_dio_nolock(io);
3622 list_del_init(&io->list);
3624 return (ret2 < 0) ? ret2 : 0;
3627 static ext4_io_end_t *ext4_init_io_end (struct inode *inode)
3629 ext4_io_end_t *io = NULL;
3631 io = kmalloc(sizeof(*io), GFP_NOFS);
3640 INIT_WORK(&io->work, ext4_end_aio_dio_work);
3641 INIT_LIST_HEAD(&io->list);
3647 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3648 ssize_t size, void *private)
3650 ext4_io_end_t *io_end = iocb->private;
3651 struct workqueue_struct *wq;
3653 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3654 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3655 iocb->private, io_end->inode->i_ino, iocb, offset,
3657 /* if not async direct IO or dio with 0 bytes write, just return */
3658 if (!io_end || !size)
3661 /* if not aio dio with unwritten extents, just free io and return */
3662 if (io_end->flag != DIO_AIO_UNWRITTEN){
3663 ext4_free_io_end(io_end);
3664 iocb->private = NULL;
3668 io_end->offset = offset;
3669 io_end->size = size;
3670 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3672 /* queue the work to convert unwritten extents to written */
3673 queue_work(wq, &io_end->work);
3675 /* Add the io_end to per-inode completed aio dio list*/
3676 list_add_tail(&io_end->list,
3677 &EXT4_I(io_end->inode)->i_aio_dio_complete_list);
3678 iocb->private = NULL;
3681 * For ext4 extent files, ext4 will do direct-io write to holes,
3682 * preallocated extents, and those write extend the file, no need to
3683 * fall back to buffered IO.
3685 * For holes, we fallocate those blocks, mark them as unintialized
3686 * If those blocks were preallocated, we mark sure they are splited, but
3687 * still keep the range to write as unintialized.
3689 * The unwrritten extents will be converted to written when DIO is completed.
3690 * For async direct IO, since the IO may still pending when return, we
3691 * set up an end_io call back function, which will do the convertion
3692 * when async direct IO completed.
3694 * If the O_DIRECT write will extend the file then add this inode to the
3695 * orphan list. So recovery will truncate it back to the original size
3696 * if the machine crashes during the write.
3699 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3700 const struct iovec *iov, loff_t offset,
3701 unsigned long nr_segs)
3703 struct file *file = iocb->ki_filp;
3704 struct inode *inode = file->f_mapping->host;
3706 size_t count = iov_length(iov, nr_segs);
3708 loff_t final_size = offset + count;
3709 if (rw == WRITE && final_size <= inode->i_size) {
3711 * We could direct write to holes and fallocate.
3713 * Allocated blocks to fill the hole are marked as uninitialized
3714 * to prevent paralel buffered read to expose the stale data
3715 * before DIO complete the data IO.
3717 * As to previously fallocated extents, ext4 get_block
3718 * will just simply mark the buffer mapped but still
3719 * keep the extents uninitialized.
3721 * for non AIO case, we will convert those unwritten extents
3722 * to written after return back from blockdev_direct_IO.
3724 * for async DIO, the conversion needs to be defered when
3725 * the IO is completed. The ext4 end_io callback function
3726 * will be called to take care of the conversion work.
3727 * Here for async case, we allocate an io_end structure to
3730 iocb->private = NULL;
3731 EXT4_I(inode)->cur_aio_dio = NULL;
3732 if (!is_sync_kiocb(iocb)) {
3733 iocb->private = ext4_init_io_end(inode);
3737 * we save the io structure for current async
3738 * direct IO, so that later ext4_get_blocks()
3739 * could flag the io structure whether there
3740 * is a unwritten extents needs to be converted
3741 * when IO is completed.
3743 EXT4_I(inode)->cur_aio_dio = iocb->private;
3746 ret = blockdev_direct_IO(rw, iocb, inode,
3747 inode->i_sb->s_bdev, iov,
3749 ext4_get_block_dio_write,
3752 EXT4_I(inode)->cur_aio_dio = NULL;
3754 * The io_end structure takes a reference to the inode,
3755 * that structure needs to be destroyed and the
3756 * reference to the inode need to be dropped, when IO is
3757 * complete, even with 0 byte write, or failed.
3759 * In the successful AIO DIO case, the io_end structure will be
3760 * desctroyed and the reference to the inode will be dropped
3761 * after the end_io call back function is called.
3763 * In the case there is 0 byte write, or error case, since
3764 * VFS direct IO won't invoke the end_io call back function,
3765 * we need to free the end_io structure here.
3767 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3768 ext4_free_io_end(iocb->private);
3769 iocb->private = NULL;
3772 * for non AIO case, since the IO is already
3773 * completed, we could do the convertion right here
3775 ret = ext4_convert_unwritten_extents(inode,
3780 /* for write the the end of file case, we fall back to old way */
3781 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3784 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3785 const struct iovec *iov, loff_t offset,
3786 unsigned long nr_segs)
3788 struct file *file = iocb->ki_filp;
3789 struct inode *inode = file->f_mapping->host;
3791 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3792 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3794 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3798 * Pages can be marked dirty completely asynchronously from ext4's journalling
3799 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3800 * much here because ->set_page_dirty is called under VFS locks. The page is
3801 * not necessarily locked.
3803 * We cannot just dirty the page and leave attached buffers clean, because the
3804 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3805 * or jbddirty because all the journalling code will explode.
3807 * So what we do is to mark the page "pending dirty" and next time writepage
3808 * is called, propagate that into the buffers appropriately.
3810 static int ext4_journalled_set_page_dirty(struct page *page)
3812 SetPageChecked(page);
3813 return __set_page_dirty_nobuffers(page);
3816 static const struct address_space_operations ext4_ordered_aops = {
3817 .readpage = ext4_readpage,
3818 .readpages = ext4_readpages,
3819 .writepage = ext4_writepage,
3820 .sync_page = block_sync_page,
3821 .write_begin = ext4_write_begin,
3822 .write_end = ext4_ordered_write_end,
3824 .invalidatepage = ext4_invalidatepage,
3825 .releasepage = ext4_releasepage,
3826 .direct_IO = ext4_direct_IO,
3827 .migratepage = buffer_migrate_page,
3828 .is_partially_uptodate = block_is_partially_uptodate,
3829 .error_remove_page = generic_error_remove_page,
3832 static const struct address_space_operations ext4_writeback_aops = {
3833 .readpage = ext4_readpage,
3834 .readpages = ext4_readpages,
3835 .writepage = ext4_writepage,
3836 .sync_page = block_sync_page,
3837 .write_begin = ext4_write_begin,
3838 .write_end = ext4_writeback_write_end,
3840 .invalidatepage = ext4_invalidatepage,
3841 .releasepage = ext4_releasepage,
3842 .direct_IO = ext4_direct_IO,
3843 .migratepage = buffer_migrate_page,
3844 .is_partially_uptodate = block_is_partially_uptodate,
3845 .error_remove_page = generic_error_remove_page,
3848 static const struct address_space_operations ext4_journalled_aops = {
3849 .readpage = ext4_readpage,
3850 .readpages = ext4_readpages,
3851 .writepage = ext4_writepage,
3852 .sync_page = block_sync_page,
3853 .write_begin = ext4_write_begin,
3854 .write_end = ext4_journalled_write_end,
3855 .set_page_dirty = ext4_journalled_set_page_dirty,
3857 .invalidatepage = ext4_invalidatepage,
3858 .releasepage = ext4_releasepage,
3859 .is_partially_uptodate = block_is_partially_uptodate,
3860 .error_remove_page = generic_error_remove_page,
3863 static const struct address_space_operations ext4_da_aops = {
3864 .readpage = ext4_readpage,
3865 .readpages = ext4_readpages,
3866 .writepage = ext4_writepage,
3867 .writepages = ext4_da_writepages,
3868 .sync_page = block_sync_page,
3869 .write_begin = ext4_da_write_begin,
3870 .write_end = ext4_da_write_end,
3872 .invalidatepage = ext4_da_invalidatepage,
3873 .releasepage = ext4_releasepage,
3874 .direct_IO = ext4_direct_IO,
3875 .migratepage = buffer_migrate_page,
3876 .is_partially_uptodate = block_is_partially_uptodate,
3877 .error_remove_page = generic_error_remove_page,
3880 void ext4_set_aops(struct inode *inode)
3882 if (ext4_should_order_data(inode) &&
3883 test_opt(inode->i_sb, DELALLOC))
3884 inode->i_mapping->a_ops = &ext4_da_aops;
3885 else if (ext4_should_order_data(inode))
3886 inode->i_mapping->a_ops = &ext4_ordered_aops;
3887 else if (ext4_should_writeback_data(inode) &&
3888 test_opt(inode->i_sb, DELALLOC))
3889 inode->i_mapping->a_ops = &ext4_da_aops;
3890 else if (ext4_should_writeback_data(inode))
3891 inode->i_mapping->a_ops = &ext4_writeback_aops;
3893 inode->i_mapping->a_ops = &ext4_journalled_aops;
3897 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3898 * up to the end of the block which corresponds to `from'.
3899 * This required during truncate. We need to physically zero the tail end
3900 * of that block so it doesn't yield old data if the file is later grown.
3902 int ext4_block_truncate_page(handle_t *handle,
3903 struct address_space *mapping, loff_t from)
3905 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3906 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3907 unsigned blocksize, length, pos;
3909 struct inode *inode = mapping->host;
3910 struct buffer_head *bh;
3914 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3915 mapping_gfp_mask(mapping) & ~__GFP_FS);
3919 blocksize = inode->i_sb->s_blocksize;
3920 length = blocksize - (offset & (blocksize - 1));
3921 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3924 * For "nobh" option, we can only work if we don't need to
3925 * read-in the page - otherwise we create buffers to do the IO.
3927 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3928 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3929 zero_user(page, offset, length);
3930 set_page_dirty(page);
3934 if (!page_has_buffers(page))
3935 create_empty_buffers(page, blocksize, 0);
3937 /* Find the buffer that contains "offset" */
3938 bh = page_buffers(page);
3940 while (offset >= pos) {
3941 bh = bh->b_this_page;
3947 if (buffer_freed(bh)) {
3948 BUFFER_TRACE(bh, "freed: skip");
3952 if (!buffer_mapped(bh)) {
3953 BUFFER_TRACE(bh, "unmapped");
3954 ext4_get_block(inode, iblock, bh, 0);
3955 /* unmapped? It's a hole - nothing to do */
3956 if (!buffer_mapped(bh)) {
3957 BUFFER_TRACE(bh, "still unmapped");
3962 /* Ok, it's mapped. Make sure it's up-to-date */
3963 if (PageUptodate(page))
3964 set_buffer_uptodate(bh);
3966 if (!buffer_uptodate(bh)) {
3968 ll_rw_block(READ, 1, &bh);
3970 /* Uhhuh. Read error. Complain and punt. */
3971 if (!buffer_uptodate(bh))
3975 if (ext4_should_journal_data(inode)) {
3976 BUFFER_TRACE(bh, "get write access");
3977 err = ext4_journal_get_write_access(handle, bh);
3982 zero_user(page, offset, length);
3984 BUFFER_TRACE(bh, "zeroed end of block");
3987 if (ext4_should_journal_data(inode)) {
3988 err = ext4_handle_dirty_metadata(handle, inode, bh);
3990 if (ext4_should_order_data(inode))
3991 err = ext4_jbd2_file_inode(handle, inode);
3992 mark_buffer_dirty(bh);
3997 page_cache_release(page);
4002 * Probably it should be a library function... search for first non-zero word
4003 * or memcmp with zero_page, whatever is better for particular architecture.
4006 static inline int all_zeroes(__le32 *p, __le32 *q)
4015 * ext4_find_shared - find the indirect blocks for partial truncation.
4016 * @inode: inode in question
4017 * @depth: depth of the affected branch
4018 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4019 * @chain: place to store the pointers to partial indirect blocks
4020 * @top: place to the (detached) top of branch
4022 * This is a helper function used by ext4_truncate().
4024 * When we do truncate() we may have to clean the ends of several
4025 * indirect blocks but leave the blocks themselves alive. Block is
4026 * partially truncated if some data below the new i_size is refered
4027 * from it (and it is on the path to the first completely truncated
4028 * data block, indeed). We have to free the top of that path along
4029 * with everything to the right of the path. Since no allocation
4030 * past the truncation point is possible until ext4_truncate()
4031 * finishes, we may safely do the latter, but top of branch may
4032 * require special attention - pageout below the truncation point
4033 * might try to populate it.
4035 * We atomically detach the top of branch from the tree, store the
4036 * block number of its root in *@top, pointers to buffer_heads of
4037 * partially truncated blocks - in @chain[].bh and pointers to
4038 * their last elements that should not be removed - in
4039 * @chain[].p. Return value is the pointer to last filled element
4042 * The work left to caller to do the actual freeing of subtrees:
4043 * a) free the subtree starting from *@top
4044 * b) free the subtrees whose roots are stored in
4045 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4046 * c) free the subtrees growing from the inode past the @chain[0].
4047 * (no partially truncated stuff there). */
4049 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4050 ext4_lblk_t offsets[4], Indirect chain[4],
4053 Indirect *partial, *p;
4057 /* Make k index the deepest non-null offest + 1 */
4058 for (k = depth; k > 1 && !offsets[k-1]; k--)
4060 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4061 /* Writer: pointers */
4063 partial = chain + k-1;
4065 * If the branch acquired continuation since we've looked at it -
4066 * fine, it should all survive and (new) top doesn't belong to us.
4068 if (!partial->key && *partial->p)
4071 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4074 * OK, we've found the last block that must survive. The rest of our
4075 * branch should be detached before unlocking. However, if that rest
4076 * of branch is all ours and does not grow immediately from the inode
4077 * it's easier to cheat and just decrement partial->p.
4079 if (p == chain + k - 1 && p > chain) {
4083 /* Nope, don't do this in ext4. Must leave the tree intact */
4090 while (partial > p) {
4091 brelse(partial->bh);
4099 * Zero a number of block pointers in either an inode or an indirect block.
4100 * If we restart the transaction we must again get write access to the
4101 * indirect block for further modification.
4103 * We release `count' blocks on disk, but (last - first) may be greater
4104 * than `count' because there can be holes in there.
4106 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
4107 struct buffer_head *bh,
4108 ext4_fsblk_t block_to_free,
4109 unsigned long count, __le32 *first,
4113 if (try_to_extend_transaction(handle, inode)) {
4115 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4116 ext4_handle_dirty_metadata(handle, inode, bh);
4118 ext4_mark_inode_dirty(handle, inode);
4119 ext4_truncate_restart_trans(handle, inode,
4120 blocks_for_truncate(inode));
4122 BUFFER_TRACE(bh, "retaking write access");
4123 ext4_journal_get_write_access(handle, bh);
4128 * Any buffers which are on the journal will be in memory. We
4129 * find them on the hash table so jbd2_journal_revoke() will
4130 * run jbd2_journal_forget() on them. We've already detached
4131 * each block from the file, so bforget() in
4132 * jbd2_journal_forget() should be safe.
4134 * AKPM: turn on bforget in jbd2_journal_forget()!!!
4136 for (p = first; p < last; p++) {
4137 u32 nr = le32_to_cpu(*p);
4139 struct buffer_head *tbh;
4142 tbh = sb_find_get_block(inode->i_sb, nr);
4143 ext4_forget(handle, 0, inode, tbh, nr);
4147 ext4_free_blocks(handle, inode, block_to_free, count, 0);
4151 * ext4_free_data - free a list of data blocks
4152 * @handle: handle for this transaction
4153 * @inode: inode we are dealing with
4154 * @this_bh: indirect buffer_head which contains *@first and *@last
4155 * @first: array of block numbers
4156 * @last: points immediately past the end of array
4158 * We are freeing all blocks refered from that array (numbers are stored as
4159 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4161 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4162 * blocks are contiguous then releasing them at one time will only affect one
4163 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4164 * actually use a lot of journal space.
4166 * @this_bh will be %NULL if @first and @last point into the inode's direct
4169 static void ext4_free_data(handle_t *handle, struct inode *inode,
4170 struct buffer_head *this_bh,
4171 __le32 *first, __le32 *last)
4173 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4174 unsigned long count = 0; /* Number of blocks in the run */
4175 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4178 ext4_fsblk_t nr; /* Current block # */
4179 __le32 *p; /* Pointer into inode/ind
4180 for current block */
4183 if (this_bh) { /* For indirect block */
4184 BUFFER_TRACE(this_bh, "get_write_access");
4185 err = ext4_journal_get_write_access(handle, this_bh);
4186 /* Important: if we can't update the indirect pointers
4187 * to the blocks, we can't free them. */
4192 for (p = first; p < last; p++) {
4193 nr = le32_to_cpu(*p);
4195 /* accumulate blocks to free if they're contiguous */
4198 block_to_free_p = p;
4200 } else if (nr == block_to_free + count) {
4203 ext4_clear_blocks(handle, inode, this_bh,
4205 count, block_to_free_p, p);
4207 block_to_free_p = p;
4214 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4215 count, block_to_free_p, p);
4218 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4221 * The buffer head should have an attached journal head at this
4222 * point. However, if the data is corrupted and an indirect
4223 * block pointed to itself, it would have been detached when
4224 * the block was cleared. Check for this instead of OOPSing.
4226 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4227 ext4_handle_dirty_metadata(handle, inode, this_bh);
4229 ext4_error(inode->i_sb, __func__,
4230 "circular indirect block detected, "
4231 "inode=%lu, block=%llu",
4233 (unsigned long long) this_bh->b_blocknr);
4238 * ext4_free_branches - free an array of branches
4239 * @handle: JBD handle for this transaction
4240 * @inode: inode we are dealing with
4241 * @parent_bh: the buffer_head which contains *@first and *@last
4242 * @first: array of block numbers
4243 * @last: pointer immediately past the end of array
4244 * @depth: depth of the branches to free
4246 * We are freeing all blocks refered from these branches (numbers are
4247 * stored as little-endian 32-bit) and updating @inode->i_blocks
4250 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4251 struct buffer_head *parent_bh,
4252 __le32 *first, __le32 *last, int depth)
4257 if (ext4_handle_is_aborted(handle))
4261 struct buffer_head *bh;
4262 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4264 while (--p >= first) {
4265 nr = le32_to_cpu(*p);
4267 continue; /* A hole */
4269 /* Go read the buffer for the next level down */
4270 bh = sb_bread(inode->i_sb, nr);
4273 * A read failure? Report error and clear slot
4277 ext4_error(inode->i_sb, "ext4_free_branches",
4278 "Read failure, inode=%lu, block=%llu",
4283 /* This zaps the entire block. Bottom up. */
4284 BUFFER_TRACE(bh, "free child branches");
4285 ext4_free_branches(handle, inode, bh,
4286 (__le32 *) bh->b_data,
4287 (__le32 *) bh->b_data + addr_per_block,
4291 * We've probably journalled the indirect block several
4292 * times during the truncate. But it's no longer
4293 * needed and we now drop it from the transaction via
4294 * jbd2_journal_revoke().
4296 * That's easy if it's exclusively part of this
4297 * transaction. But if it's part of the committing
4298 * transaction then jbd2_journal_forget() will simply
4299 * brelse() it. That means that if the underlying
4300 * block is reallocated in ext4_get_block(),
4301 * unmap_underlying_metadata() will find this block
4302 * and will try to get rid of it. damn, damn.
4304 * If this block has already been committed to the
4305 * journal, a revoke record will be written. And
4306 * revoke records must be emitted *before* clearing
4307 * this block's bit in the bitmaps.
4309 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
4312 * Everything below this this pointer has been
4313 * released. Now let this top-of-subtree go.
4315 * We want the freeing of this indirect block to be
4316 * atomic in the journal with the updating of the
4317 * bitmap block which owns it. So make some room in
4320 * We zero the parent pointer *after* freeing its
4321 * pointee in the bitmaps, so if extend_transaction()
4322 * for some reason fails to put the bitmap changes and
4323 * the release into the same transaction, recovery
4324 * will merely complain about releasing a free block,
4325 * rather than leaking blocks.
4327 if (ext4_handle_is_aborted(handle))
4329 if (try_to_extend_transaction(handle, inode)) {
4330 ext4_mark_inode_dirty(handle, inode);
4331 ext4_truncate_restart_trans(handle, inode,
4332 blocks_for_truncate(inode));
4335 ext4_free_blocks(handle, inode, nr, 1, 1);
4339 * The block which we have just freed is
4340 * pointed to by an indirect block: journal it
4342 BUFFER_TRACE(parent_bh, "get_write_access");
4343 if (!ext4_journal_get_write_access(handle,
4346 BUFFER_TRACE(parent_bh,
4347 "call ext4_handle_dirty_metadata");
4348 ext4_handle_dirty_metadata(handle,
4355 /* We have reached the bottom of the tree. */
4356 BUFFER_TRACE(parent_bh, "free data blocks");
4357 ext4_free_data(handle, inode, parent_bh, first, last);
4361 int ext4_can_truncate(struct inode *inode)
4363 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4365 if (S_ISREG(inode->i_mode))
4367 if (S_ISDIR(inode->i_mode))
4369 if (S_ISLNK(inode->i_mode))
4370 return !ext4_inode_is_fast_symlink(inode);
4377 * We block out ext4_get_block() block instantiations across the entire
4378 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4379 * simultaneously on behalf of the same inode.
4381 * As we work through the truncate and commmit bits of it to the journal there
4382 * is one core, guiding principle: the file's tree must always be consistent on
4383 * disk. We must be able to restart the truncate after a crash.
4385 * The file's tree may be transiently inconsistent in memory (although it
4386 * probably isn't), but whenever we close off and commit a journal transaction,
4387 * the contents of (the filesystem + the journal) must be consistent and
4388 * restartable. It's pretty simple, really: bottom up, right to left (although
4389 * left-to-right works OK too).
4391 * Note that at recovery time, journal replay occurs *before* the restart of
4392 * truncate against the orphan inode list.
4394 * The committed inode has the new, desired i_size (which is the same as
4395 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4396 * that this inode's truncate did not complete and it will again call
4397 * ext4_truncate() to have another go. So there will be instantiated blocks
4398 * to the right of the truncation point in a crashed ext4 filesystem. But
4399 * that's fine - as long as they are linked from the inode, the post-crash
4400 * ext4_truncate() run will find them and release them.
4402 void ext4_truncate(struct inode *inode)
4405 struct ext4_inode_info *ei = EXT4_I(inode);
4406 __le32 *i_data = ei->i_data;
4407 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4408 struct address_space *mapping = inode->i_mapping;
4409 ext4_lblk_t offsets[4];
4414 ext4_lblk_t last_block;
4415 unsigned blocksize = inode->i_sb->s_blocksize;
4417 if (!ext4_can_truncate(inode))
4420 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4421 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4423 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4424 ext4_ext_truncate(inode);
4428 handle = start_transaction(inode);
4430 return; /* AKPM: return what? */
4432 last_block = (inode->i_size + blocksize-1)
4433 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4435 if (inode->i_size & (blocksize - 1))
4436 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4439 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4441 goto out_stop; /* error */
4444 * OK. This truncate is going to happen. We add the inode to the
4445 * orphan list, so that if this truncate spans multiple transactions,
4446 * and we crash, we will resume the truncate when the filesystem
4447 * recovers. It also marks the inode dirty, to catch the new size.
4449 * Implication: the file must always be in a sane, consistent
4450 * truncatable state while each transaction commits.
4452 if (ext4_orphan_add(handle, inode))
4456 * From here we block out all ext4_get_block() callers who want to
4457 * modify the block allocation tree.
4459 down_write(&ei->i_data_sem);
4461 ext4_discard_preallocations(inode);
4464 * The orphan list entry will now protect us from any crash which
4465 * occurs before the truncate completes, so it is now safe to propagate
4466 * the new, shorter inode size (held for now in i_size) into the
4467 * on-disk inode. We do this via i_disksize, which is the value which
4468 * ext4 *really* writes onto the disk inode.
4470 ei->i_disksize = inode->i_size;
4472 if (n == 1) { /* direct blocks */
4473 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4474 i_data + EXT4_NDIR_BLOCKS);
4478 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4479 /* Kill the top of shared branch (not detached) */
4481 if (partial == chain) {
4482 /* Shared branch grows from the inode */
4483 ext4_free_branches(handle, inode, NULL,
4484 &nr, &nr+1, (chain+n-1) - partial);
4487 * We mark the inode dirty prior to restart,
4488 * and prior to stop. No need for it here.
4491 /* Shared branch grows from an indirect block */
4492 BUFFER_TRACE(partial->bh, "get_write_access");
4493 ext4_free_branches(handle, inode, partial->bh,
4495 partial->p+1, (chain+n-1) - partial);
4498 /* Clear the ends of indirect blocks on the shared branch */
4499 while (partial > chain) {
4500 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4501 (__le32*)partial->bh->b_data+addr_per_block,
4502 (chain+n-1) - partial);
4503 BUFFER_TRACE(partial->bh, "call brelse");
4504 brelse(partial->bh);
4508 /* Kill the remaining (whole) subtrees */
4509 switch (offsets[0]) {
4511 nr = i_data[EXT4_IND_BLOCK];
4513 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4514 i_data[EXT4_IND_BLOCK] = 0;
4516 case EXT4_IND_BLOCK:
4517 nr = i_data[EXT4_DIND_BLOCK];
4519 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4520 i_data[EXT4_DIND_BLOCK] = 0;
4522 case EXT4_DIND_BLOCK:
4523 nr = i_data[EXT4_TIND_BLOCK];
4525 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4526 i_data[EXT4_TIND_BLOCK] = 0;
4528 case EXT4_TIND_BLOCK:
4532 up_write(&ei->i_data_sem);
4533 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4534 ext4_mark_inode_dirty(handle, inode);
4537 * In a multi-transaction truncate, we only make the final transaction
4541 ext4_handle_sync(handle);
4544 * If this was a simple ftruncate(), and the file will remain alive
4545 * then we need to clear up the orphan record which we created above.
4546 * However, if this was a real unlink then we were called by
4547 * ext4_delete_inode(), and we allow that function to clean up the
4548 * orphan info for us.
4551 ext4_orphan_del(handle, inode);
4553 ext4_journal_stop(handle);
4557 * ext4_get_inode_loc returns with an extra refcount against the inode's
4558 * underlying buffer_head on success. If 'in_mem' is true, we have all
4559 * data in memory that is needed to recreate the on-disk version of this
4562 static int __ext4_get_inode_loc(struct inode *inode,
4563 struct ext4_iloc *iloc, int in_mem)
4565 struct ext4_group_desc *gdp;
4566 struct buffer_head *bh;
4567 struct super_block *sb = inode->i_sb;
4569 int inodes_per_block, inode_offset;
4572 if (!ext4_valid_inum(sb, inode->i_ino))
4575 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4576 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4581 * Figure out the offset within the block group inode table
4583 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4584 inode_offset = ((inode->i_ino - 1) %
4585 EXT4_INODES_PER_GROUP(sb));
4586 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4587 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4589 bh = sb_getblk(sb, block);
4591 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4592 "inode block - inode=%lu, block=%llu",
4593 inode->i_ino, block);
4596 if (!buffer_uptodate(bh)) {
4600 * If the buffer has the write error flag, we have failed
4601 * to write out another inode in the same block. In this
4602 * case, we don't have to read the block because we may
4603 * read the old inode data successfully.
4605 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4606 set_buffer_uptodate(bh);
4608 if (buffer_uptodate(bh)) {
4609 /* someone brought it uptodate while we waited */
4615 * If we have all information of the inode in memory and this
4616 * is the only valid inode in the block, we need not read the
4620 struct buffer_head *bitmap_bh;
4623 start = inode_offset & ~(inodes_per_block - 1);
4625 /* Is the inode bitmap in cache? */
4626 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4631 * If the inode bitmap isn't in cache then the
4632 * optimisation may end up performing two reads instead
4633 * of one, so skip it.
4635 if (!buffer_uptodate(bitmap_bh)) {
4639 for (i = start; i < start + inodes_per_block; i++) {
4640 if (i == inode_offset)
4642 if (ext4_test_bit(i, bitmap_bh->b_data))
4646 if (i == start + inodes_per_block) {
4647 /* all other inodes are free, so skip I/O */
4648 memset(bh->b_data, 0, bh->b_size);
4649 set_buffer_uptodate(bh);
4657 * If we need to do any I/O, try to pre-readahead extra
4658 * blocks from the inode table.
4660 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4661 ext4_fsblk_t b, end, table;
4664 table = ext4_inode_table(sb, gdp);
4665 /* s_inode_readahead_blks is always a power of 2 */
4666 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4669 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4670 num = EXT4_INODES_PER_GROUP(sb);
4671 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4672 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4673 num -= ext4_itable_unused_count(sb, gdp);
4674 table += num / inodes_per_block;
4678 sb_breadahead(sb, b++);
4682 * There are other valid inodes in the buffer, this inode
4683 * has in-inode xattrs, or we don't have this inode in memory.
4684 * Read the block from disk.
4687 bh->b_end_io = end_buffer_read_sync;
4688 submit_bh(READ_META, bh);
4690 if (!buffer_uptodate(bh)) {
4691 ext4_error(sb, __func__,
4692 "unable to read inode block - inode=%lu, "
4693 "block=%llu", inode->i_ino, block);
4703 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4705 /* We have all inode data except xattrs in memory here. */
4706 return __ext4_get_inode_loc(inode, iloc,
4707 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4710 void ext4_set_inode_flags(struct inode *inode)
4712 unsigned int flags = EXT4_I(inode)->i_flags;
4714 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4715 if (flags & EXT4_SYNC_FL)
4716 inode->i_flags |= S_SYNC;
4717 if (flags & EXT4_APPEND_FL)
4718 inode->i_flags |= S_APPEND;
4719 if (flags & EXT4_IMMUTABLE_FL)
4720 inode->i_flags |= S_IMMUTABLE;
4721 if (flags & EXT4_NOATIME_FL)
4722 inode->i_flags |= S_NOATIME;
4723 if (flags & EXT4_DIRSYNC_FL)
4724 inode->i_flags |= S_DIRSYNC;
4727 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4728 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4730 unsigned int flags = ei->vfs_inode.i_flags;
4732 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4733 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4735 ei->i_flags |= EXT4_SYNC_FL;
4736 if (flags & S_APPEND)
4737 ei->i_flags |= EXT4_APPEND_FL;
4738 if (flags & S_IMMUTABLE)
4739 ei->i_flags |= EXT4_IMMUTABLE_FL;
4740 if (flags & S_NOATIME)
4741 ei->i_flags |= EXT4_NOATIME_FL;
4742 if (flags & S_DIRSYNC)
4743 ei->i_flags |= EXT4_DIRSYNC_FL;
4746 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4747 struct ext4_inode_info *ei)
4750 struct inode *inode = &(ei->vfs_inode);
4751 struct super_block *sb = inode->i_sb;
4753 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4754 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4755 /* we are using combined 48 bit field */
4756 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4757 le32_to_cpu(raw_inode->i_blocks_lo);
4758 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4759 /* i_blocks represent file system block size */
4760 return i_blocks << (inode->i_blkbits - 9);
4765 return le32_to_cpu(raw_inode->i_blocks_lo);
4769 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4771 struct ext4_iloc iloc;
4772 struct ext4_inode *raw_inode;
4773 struct ext4_inode_info *ei;
4774 struct buffer_head *bh;
4775 struct inode *inode;
4779 inode = iget_locked(sb, ino);
4781 return ERR_PTR(-ENOMEM);
4782 if (!(inode->i_state & I_NEW))
4787 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4791 raw_inode = ext4_raw_inode(&iloc);
4792 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4793 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4794 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4795 if (!(test_opt(inode->i_sb, NO_UID32))) {
4796 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4797 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4799 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4802 ei->i_dir_start_lookup = 0;
4803 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4804 /* We now have enough fields to check if the inode was active or not.
4805 * This is needed because nfsd might try to access dead inodes
4806 * the test is that same one that e2fsck uses
4807 * NeilBrown 1999oct15
4809 if (inode->i_nlink == 0) {
4810 if (inode->i_mode == 0 ||
4811 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4812 /* this inode is deleted */
4817 /* The only unlinked inodes we let through here have
4818 * valid i_mode and are being read by the orphan
4819 * recovery code: that's fine, we're about to complete
4820 * the process of deleting those. */
4822 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4823 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4824 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4825 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4827 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4828 inode->i_size = ext4_isize(raw_inode);
4829 ei->i_disksize = inode->i_size;
4830 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4831 ei->i_block_group = iloc.block_group;
4832 ei->i_last_alloc_group = ~0;
4834 * NOTE! The in-memory inode i_data array is in little-endian order
4835 * even on big-endian machines: we do NOT byteswap the block numbers!
4837 for (block = 0; block < EXT4_N_BLOCKS; block++)
4838 ei->i_data[block] = raw_inode->i_block[block];
4839 INIT_LIST_HEAD(&ei->i_orphan);
4841 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4842 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4843 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4844 EXT4_INODE_SIZE(inode->i_sb)) {
4849 if (ei->i_extra_isize == 0) {
4850 /* The extra space is currently unused. Use it. */
4851 ei->i_extra_isize = sizeof(struct ext4_inode) -
4852 EXT4_GOOD_OLD_INODE_SIZE;
4854 __le32 *magic = (void *)raw_inode +
4855 EXT4_GOOD_OLD_INODE_SIZE +
4857 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4858 ei->i_state |= EXT4_STATE_XATTR;
4861 ei->i_extra_isize = 0;
4863 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4864 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4865 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4866 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4868 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4869 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4870 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4872 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4876 if (ei->i_file_acl &&
4878 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4879 EXT4_SB(sb)->s_gdb_count)) ||
4880 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4881 ext4_error(sb, __func__,
4882 "bad extended attribute block %llu in inode #%lu",
4883 ei->i_file_acl, inode->i_ino);
4886 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4887 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4888 (S_ISLNK(inode->i_mode) &&
4889 !ext4_inode_is_fast_symlink(inode)))
4890 /* Validate extent which is part of inode */
4891 ret = ext4_ext_check_inode(inode);
4892 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4893 (S_ISLNK(inode->i_mode) &&
4894 !ext4_inode_is_fast_symlink(inode))) {
4895 /* Validate block references which are part of inode */
4896 ret = ext4_check_inode_blockref(inode);
4903 if (S_ISREG(inode->i_mode)) {
4904 inode->i_op = &ext4_file_inode_operations;
4905 inode->i_fop = &ext4_file_operations;
4906 ext4_set_aops(inode);
4907 } else if (S_ISDIR(inode->i_mode)) {
4908 inode->i_op = &ext4_dir_inode_operations;
4909 inode->i_fop = &ext4_dir_operations;
4910 } else if (S_ISLNK(inode->i_mode)) {
4911 if (ext4_inode_is_fast_symlink(inode)) {
4912 inode->i_op = &ext4_fast_symlink_inode_operations;
4913 nd_terminate_link(ei->i_data, inode->i_size,
4914 sizeof(ei->i_data) - 1);
4916 inode->i_op = &ext4_symlink_inode_operations;
4917 ext4_set_aops(inode);
4919 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4920 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4921 inode->i_op = &ext4_special_inode_operations;
4922 if (raw_inode->i_block[0])
4923 init_special_inode(inode, inode->i_mode,
4924 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4926 init_special_inode(inode, inode->i_mode,
4927 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4931 ext4_error(inode->i_sb, __func__,
4932 "bogus i_mode (%o) for inode=%lu",
4933 inode->i_mode, inode->i_ino);
4937 ext4_set_inode_flags(inode);
4938 unlock_new_inode(inode);
4943 return ERR_PTR(ret);
4946 static int ext4_inode_blocks_set(handle_t *handle,
4947 struct ext4_inode *raw_inode,
4948 struct ext4_inode_info *ei)
4950 struct inode *inode = &(ei->vfs_inode);
4951 u64 i_blocks = inode->i_blocks;
4952 struct super_block *sb = inode->i_sb;
4954 if (i_blocks <= ~0U) {
4956 * i_blocks can be represnted in a 32 bit variable
4957 * as multiple of 512 bytes
4959 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4960 raw_inode->i_blocks_high = 0;
4961 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4964 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4967 if (i_blocks <= 0xffffffffffffULL) {
4969 * i_blocks can be represented in a 48 bit variable
4970 * as multiple of 512 bytes
4972 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4973 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4974 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4976 ei->i_flags |= EXT4_HUGE_FILE_FL;
4977 /* i_block is stored in file system block size */
4978 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4979 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4980 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4986 * Post the struct inode info into an on-disk inode location in the
4987 * buffer-cache. This gobbles the caller's reference to the
4988 * buffer_head in the inode location struct.
4990 * The caller must have write access to iloc->bh.
4992 static int ext4_do_update_inode(handle_t *handle,
4993 struct inode *inode,
4994 struct ext4_iloc *iloc)
4996 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4997 struct ext4_inode_info *ei = EXT4_I(inode);
4998 struct buffer_head *bh = iloc->bh;
4999 int err = 0, rc, block;
5001 /* For fields not not tracking in the in-memory inode,
5002 * initialise them to zero for new inodes. */
5003 if (ei->i_state & EXT4_STATE_NEW)
5004 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5006 ext4_get_inode_flags(ei);
5007 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5008 if (!(test_opt(inode->i_sb, NO_UID32))) {
5009 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5010 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5012 * Fix up interoperability with old kernels. Otherwise, old inodes get
5013 * re-used with the upper 16 bits of the uid/gid intact
5016 raw_inode->i_uid_high =
5017 cpu_to_le16(high_16_bits(inode->i_uid));
5018 raw_inode->i_gid_high =
5019 cpu_to_le16(high_16_bits(inode->i_gid));
5021 raw_inode->i_uid_high = 0;
5022 raw_inode->i_gid_high = 0;
5025 raw_inode->i_uid_low =
5026 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5027 raw_inode->i_gid_low =
5028 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5029 raw_inode->i_uid_high = 0;
5030 raw_inode->i_gid_high = 0;
5032 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5034 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5035 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5036 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5037 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5039 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5041 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5042 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5043 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5044 cpu_to_le32(EXT4_OS_HURD))
5045 raw_inode->i_file_acl_high =
5046 cpu_to_le16(ei->i_file_acl >> 32);
5047 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5048 ext4_isize_set(raw_inode, ei->i_disksize);
5049 if (ei->i_disksize > 0x7fffffffULL) {
5050 struct super_block *sb = inode->i_sb;
5051 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5052 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5053 EXT4_SB(sb)->s_es->s_rev_level ==
5054 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5055 /* If this is the first large file
5056 * created, add a flag to the superblock.
5058 err = ext4_journal_get_write_access(handle,
5059 EXT4_SB(sb)->s_sbh);
5062 ext4_update_dynamic_rev(sb);
5063 EXT4_SET_RO_COMPAT_FEATURE(sb,
5064 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5066 ext4_handle_sync(handle);
5067 err = ext4_handle_dirty_metadata(handle, inode,
5068 EXT4_SB(sb)->s_sbh);
5071 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5072 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5073 if (old_valid_dev(inode->i_rdev)) {
5074 raw_inode->i_block[0] =
5075 cpu_to_le32(old_encode_dev(inode->i_rdev));
5076 raw_inode->i_block[1] = 0;
5078 raw_inode->i_block[0] = 0;
5079 raw_inode->i_block[1] =
5080 cpu_to_le32(new_encode_dev(inode->i_rdev));
5081 raw_inode->i_block[2] = 0;
5084 for (block = 0; block < EXT4_N_BLOCKS; block++)
5085 raw_inode->i_block[block] = ei->i_data[block];
5087 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5088 if (ei->i_extra_isize) {
5089 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5090 raw_inode->i_version_hi =
5091 cpu_to_le32(inode->i_version >> 32);
5092 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5095 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5096 rc = ext4_handle_dirty_metadata(handle, inode, bh);
5099 ei->i_state &= ~EXT4_STATE_NEW;
5103 ext4_std_error(inode->i_sb, err);
5108 * ext4_write_inode()
5110 * We are called from a few places:
5112 * - Within generic_file_write() for O_SYNC files.
5113 * Here, there will be no transaction running. We wait for any running
5114 * trasnaction to commit.
5116 * - Within sys_sync(), kupdate and such.
5117 * We wait on commit, if tol to.
5119 * - Within prune_icache() (PF_MEMALLOC == true)
5120 * Here we simply return. We can't afford to block kswapd on the
5123 * In all cases it is actually safe for us to return without doing anything,
5124 * because the inode has been copied into a raw inode buffer in
5125 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5128 * Note that we are absolutely dependent upon all inode dirtiers doing the
5129 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5130 * which we are interested.
5132 * It would be a bug for them to not do this. The code:
5134 * mark_inode_dirty(inode)
5136 * inode->i_size = expr;
5138 * is in error because a kswapd-driven write_inode() could occur while
5139 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5140 * will no longer be on the superblock's dirty inode list.
5142 int ext4_write_inode(struct inode *inode, int wait)
5146 if (current->flags & PF_MEMALLOC)
5149 if (EXT4_SB(inode->i_sb)->s_journal) {
5150 if (ext4_journal_current_handle()) {
5151 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5159 err = ext4_force_commit(inode->i_sb);
5161 struct ext4_iloc iloc;
5163 err = ext4_get_inode_loc(inode, &iloc);
5167 sync_dirty_buffer(iloc.bh);
5168 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5169 ext4_error(inode->i_sb, __func__,
5170 "IO error syncing inode, "
5171 "inode=%lu, block=%llu",
5173 (unsigned long long)iloc.bh->b_blocknr);
5183 * Called from notify_change.
5185 * We want to trap VFS attempts to truncate the file as soon as
5186 * possible. In particular, we want to make sure that when the VFS
5187 * shrinks i_size, we put the inode on the orphan list and modify
5188 * i_disksize immediately, so that during the subsequent flushing of
5189 * dirty pages and freeing of disk blocks, we can guarantee that any
5190 * commit will leave the blocks being flushed in an unused state on
5191 * disk. (On recovery, the inode will get truncated and the blocks will
5192 * be freed, so we have a strong guarantee that no future commit will
5193 * leave these blocks visible to the user.)
5195 * Another thing we have to assure is that if we are in ordered mode
5196 * and inode is still attached to the committing transaction, we must
5197 * we start writeout of all the dirty pages which are being truncated.
5198 * This way we are sure that all the data written in the previous
5199 * transaction are already on disk (truncate waits for pages under
5202 * Called with inode->i_mutex down.
5204 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5206 struct inode *inode = dentry->d_inode;
5208 const unsigned int ia_valid = attr->ia_valid;
5210 error = inode_change_ok(inode, attr);
5214 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5215 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5218 /* (user+group)*(old+new) structure, inode write (sb,
5219 * inode block, ? - but truncate inode update has it) */
5220 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
5221 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
5222 if (IS_ERR(handle)) {
5223 error = PTR_ERR(handle);
5226 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
5228 ext4_journal_stop(handle);
5231 /* Update corresponding info in inode so that everything is in
5232 * one transaction */
5233 if (attr->ia_valid & ATTR_UID)
5234 inode->i_uid = attr->ia_uid;
5235 if (attr->ia_valid & ATTR_GID)
5236 inode->i_gid = attr->ia_gid;
5237 error = ext4_mark_inode_dirty(handle, inode);
5238 ext4_journal_stop(handle);
5241 if (attr->ia_valid & ATTR_SIZE) {
5242 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
5243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5245 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5252 if (S_ISREG(inode->i_mode) &&
5253 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
5256 handle = ext4_journal_start(inode, 3);
5257 if (IS_ERR(handle)) {
5258 error = PTR_ERR(handle);
5262 error = ext4_orphan_add(handle, inode);
5263 EXT4_I(inode)->i_disksize = attr->ia_size;
5264 rc = ext4_mark_inode_dirty(handle, inode);
5267 ext4_journal_stop(handle);
5269 if (ext4_should_order_data(inode)) {
5270 error = ext4_begin_ordered_truncate(inode,
5273 /* Do as much error cleanup as possible */
5274 handle = ext4_journal_start(inode, 3);
5275 if (IS_ERR(handle)) {
5276 ext4_orphan_del(NULL, inode);
5279 ext4_orphan_del(handle, inode);
5280 ext4_journal_stop(handle);
5286 rc = inode_setattr(inode, attr);
5288 /* If inode_setattr's call to ext4_truncate failed to get a
5289 * transaction handle at all, we need to clean up the in-core
5290 * orphan list manually. */
5292 ext4_orphan_del(NULL, inode);
5294 if (!rc && (ia_valid & ATTR_MODE))
5295 rc = ext4_acl_chmod(inode);
5298 ext4_std_error(inode->i_sb, error);
5304 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5307 struct inode *inode;
5308 unsigned long delalloc_blocks;
5310 inode = dentry->d_inode;
5311 generic_fillattr(inode, stat);
5314 * We can't update i_blocks if the block allocation is delayed
5315 * otherwise in the case of system crash before the real block
5316 * allocation is done, we will have i_blocks inconsistent with
5317 * on-disk file blocks.
5318 * We always keep i_blocks updated together with real
5319 * allocation. But to not confuse with user, stat
5320 * will return the blocks that include the delayed allocation
5321 * blocks for this file.
5323 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5324 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5325 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5327 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5331 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5336 /* if nrblocks are contiguous */
5339 * With N contiguous data blocks, it need at most
5340 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5341 * 2 dindirect blocks
5344 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5345 return indirects + 3;
5348 * if nrblocks are not contiguous, worse case, each block touch
5349 * a indirect block, and each indirect block touch a double indirect
5350 * block, plus a triple indirect block
5352 indirects = nrblocks * 2 + 1;
5356 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5358 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
5359 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5360 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5364 * Account for index blocks, block groups bitmaps and block group
5365 * descriptor blocks if modify datablocks and index blocks
5366 * worse case, the indexs blocks spread over different block groups
5368 * If datablocks are discontiguous, they are possible to spread over
5369 * different block groups too. If they are contiugous, with flexbg,
5370 * they could still across block group boundary.
5372 * Also account for superblock, inode, quota and xattr blocks
5374 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5376 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5382 * How many index blocks need to touch to modify nrblocks?
5383 * The "Chunk" flag indicating whether the nrblocks is
5384 * physically contiguous on disk
5386 * For Direct IO and fallocate, they calls get_block to allocate
5387 * one single extent at a time, so they could set the "Chunk" flag
5389 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5394 * Now let's see how many group bitmaps and group descriptors need
5404 if (groups > ngroups)
5406 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5407 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5409 /* bitmaps and block group descriptor blocks */
5410 ret += groups + gdpblocks;
5412 /* Blocks for super block, inode, quota and xattr blocks */
5413 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5419 * Calulate the total number of credits to reserve to fit
5420 * the modification of a single pages into a single transaction,
5421 * which may include multiple chunks of block allocations.
5423 * This could be called via ext4_write_begin()
5425 * We need to consider the worse case, when
5426 * one new block per extent.
5428 int ext4_writepage_trans_blocks(struct inode *inode)
5430 int bpp = ext4_journal_blocks_per_page(inode);
5433 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5435 /* Account for data blocks for journalled mode */
5436 if (ext4_should_journal_data(inode))
5442 * Calculate the journal credits for a chunk of data modification.
5444 * This is called from DIO, fallocate or whoever calling
5445 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5447 * journal buffers for data blocks are not included here, as DIO
5448 * and fallocate do no need to journal data buffers.
5450 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5452 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5456 * The caller must have previously called ext4_reserve_inode_write().
5457 * Give this, we know that the caller already has write access to iloc->bh.
5459 int ext4_mark_iloc_dirty(handle_t *handle,
5460 struct inode *inode, struct ext4_iloc *iloc)
5464 if (test_opt(inode->i_sb, I_VERSION))
5465 inode_inc_iversion(inode);
5467 /* the do_update_inode consumes one bh->b_count */
5470 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5471 err = ext4_do_update_inode(handle, inode, iloc);
5477 * On success, We end up with an outstanding reference count against
5478 * iloc->bh. This _must_ be cleaned up later.
5482 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5483 struct ext4_iloc *iloc)
5487 err = ext4_get_inode_loc(inode, iloc);
5489 BUFFER_TRACE(iloc->bh, "get_write_access");
5490 err = ext4_journal_get_write_access(handle, iloc->bh);
5496 ext4_std_error(inode->i_sb, err);
5501 * Expand an inode by new_extra_isize bytes.
5502 * Returns 0 on success or negative error number on failure.
5504 static int ext4_expand_extra_isize(struct inode *inode,
5505 unsigned int new_extra_isize,
5506 struct ext4_iloc iloc,
5509 struct ext4_inode *raw_inode;
5510 struct ext4_xattr_ibody_header *header;
5511 struct ext4_xattr_entry *entry;
5513 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5516 raw_inode = ext4_raw_inode(&iloc);
5518 header = IHDR(inode, raw_inode);
5519 entry = IFIRST(header);
5521 /* No extended attributes present */
5522 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5523 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5524 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5526 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5530 /* try to expand with EAs present */
5531 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5536 * What we do here is to mark the in-core inode as clean with respect to inode
5537 * dirtiness (it may still be data-dirty).
5538 * This means that the in-core inode may be reaped by prune_icache
5539 * without having to perform any I/O. This is a very good thing,
5540 * because *any* task may call prune_icache - even ones which
5541 * have a transaction open against a different journal.
5543 * Is this cheating? Not really. Sure, we haven't written the
5544 * inode out, but prune_icache isn't a user-visible syncing function.
5545 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5546 * we start and wait on commits.
5548 * Is this efficient/effective? Well, we're being nice to the system
5549 * by cleaning up our inodes proactively so they can be reaped
5550 * without I/O. But we are potentially leaving up to five seconds'
5551 * worth of inodes floating about which prune_icache wants us to
5552 * write out. One way to fix that would be to get prune_icache()
5553 * to do a write_super() to free up some memory. It has the desired
5556 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5558 struct ext4_iloc iloc;
5559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5560 static unsigned int mnt_count;
5564 err = ext4_reserve_inode_write(handle, inode, &iloc);
5565 if (ext4_handle_valid(handle) &&
5566 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5567 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5569 * We need extra buffer credits since we may write into EA block
5570 * with this same handle. If journal_extend fails, then it will
5571 * only result in a minor loss of functionality for that inode.
5572 * If this is felt to be critical, then e2fsck should be run to
5573 * force a large enough s_min_extra_isize.
5575 if ((jbd2_journal_extend(handle,
5576 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5577 ret = ext4_expand_extra_isize(inode,
5578 sbi->s_want_extra_isize,
5581 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5583 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5584 ext4_warning(inode->i_sb, __func__,
5585 "Unable to expand inode %lu. Delete"
5586 " some EAs or run e2fsck.",
5589 le16_to_cpu(sbi->s_es->s_mnt_count);
5595 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5600 * ext4_dirty_inode() is called from __mark_inode_dirty()
5602 * We're really interested in the case where a file is being extended.
5603 * i_size has been changed by generic_commit_write() and we thus need
5604 * to include the updated inode in the current transaction.
5606 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5607 * are allocated to the file.
5609 * If the inode is marked synchronous, we don't honour that here - doing
5610 * so would cause a commit on atime updates, which we don't bother doing.
5611 * We handle synchronous inodes at the highest possible level.
5613 void ext4_dirty_inode(struct inode *inode)
5615 handle_t *current_handle = ext4_journal_current_handle();
5618 handle = ext4_journal_start(inode, 2);
5622 jbd_debug(5, "marking dirty. outer handle=%p\n", current_handle);
5623 ext4_mark_inode_dirty(handle, inode);
5625 ext4_journal_stop(handle);
5632 * Bind an inode's backing buffer_head into this transaction, to prevent
5633 * it from being flushed to disk early. Unlike
5634 * ext4_reserve_inode_write, this leaves behind no bh reference and
5635 * returns no iloc structure, so the caller needs to repeat the iloc
5636 * lookup to mark the inode dirty later.
5638 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5640 struct ext4_iloc iloc;
5644 err = ext4_get_inode_loc(inode, &iloc);
5646 BUFFER_TRACE(iloc.bh, "get_write_access");
5647 err = jbd2_journal_get_write_access(handle, iloc.bh);
5649 err = ext4_handle_dirty_metadata(handle,
5655 ext4_std_error(inode->i_sb, err);
5660 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5667 * We have to be very careful here: changing a data block's
5668 * journaling status dynamically is dangerous. If we write a
5669 * data block to the journal, change the status and then delete
5670 * that block, we risk forgetting to revoke the old log record
5671 * from the journal and so a subsequent replay can corrupt data.
5672 * So, first we make sure that the journal is empty and that
5673 * nobody is changing anything.
5676 journal = EXT4_JOURNAL(inode);
5679 if (is_journal_aborted(journal))
5682 jbd2_journal_lock_updates(journal);
5683 jbd2_journal_flush(journal);
5686 * OK, there are no updates running now, and all cached data is
5687 * synced to disk. We are now in a completely consistent state
5688 * which doesn't have anything in the journal, and we know that
5689 * no filesystem updates are running, so it is safe to modify
5690 * the inode's in-core data-journaling state flag now.
5694 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5696 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5697 ext4_set_aops(inode);
5699 jbd2_journal_unlock_updates(journal);
5701 /* Finally we can mark the inode as dirty. */
5703 handle = ext4_journal_start(inode, 1);
5705 return PTR_ERR(handle);
5707 err = ext4_mark_inode_dirty(handle, inode);
5708 ext4_handle_sync(handle);
5709 ext4_journal_stop(handle);
5710 ext4_std_error(inode->i_sb, err);
5715 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5717 return !buffer_mapped(bh);
5720 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5722 struct page *page = vmf->page;
5727 struct file *file = vma->vm_file;
5728 struct inode *inode = file->f_path.dentry->d_inode;
5729 struct address_space *mapping = inode->i_mapping;
5732 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5733 * get i_mutex because we are already holding mmap_sem.
5735 down_read(&inode->i_alloc_sem);
5736 size = i_size_read(inode);
5737 if (page->mapping != mapping || size <= page_offset(page)
5738 || !PageUptodate(page)) {
5739 /* page got truncated from under us? */
5743 if (PageMappedToDisk(page))
5746 if (page->index == size >> PAGE_CACHE_SHIFT)
5747 len = size & ~PAGE_CACHE_MASK;
5749 len = PAGE_CACHE_SIZE;
5753 * return if we have all the buffers mapped. This avoid
5754 * the need to call write_begin/write_end which does a
5755 * journal_start/journal_stop which can block and take
5758 if (page_has_buffers(page)) {
5759 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5760 ext4_bh_unmapped)) {
5767 * OK, we need to fill the hole... Do write_begin write_end
5768 * to do block allocation/reservation.We are not holding
5769 * inode.i__mutex here. That allow * parallel write_begin,
5770 * write_end call. lock_page prevent this from happening
5771 * on the same page though
5773 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5774 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5777 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5778 len, len, page, fsdata);
5784 ret = VM_FAULT_SIGBUS;
5785 up_read(&inode->i_alloc_sem);