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 "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode *inode,
50 return jbd2_journal_begin_ordered_truncate(
51 EXT4_SB(inode->i_sb)->s_journal,
52 &EXT4_I(inode)->jinode,
56 static void ext4_invalidatepage(struct page *page, unsigned long offset);
59 * Test whether an inode is a fast symlink.
61 static int ext4_inode_is_fast_symlink(struct inode *inode)
63 int ea_blocks = EXT4_I(inode)->i_file_acl ?
64 (inode->i_sb->s_blocksize >> 9) : 0;
66 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
70 * The ext4 forget function must perform a revoke if we are freeing data
71 * which has been journaled. Metadata (eg. indirect blocks) must be
72 * revoked in all cases.
74 * "bh" may be NULL: a metadata block may have been freed from memory
75 * but there may still be a record of it in the journal, and that record
76 * still needs to be revoked.
78 * If the handle isn't valid we're not journaling so there's nothing to do.
80 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
81 struct buffer_head *bh, ext4_fsblk_t blocknr)
85 if (!ext4_handle_valid(handle))
90 BUFFER_TRACE(bh, "enter");
92 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 bh, is_metadata, inode->i_mode,
95 test_opt(inode->i_sb, DATA_FLAGS));
97 /* Never use the revoke function if we are doing full data
98 * journaling: there is no need to, and a V1 superblock won't
99 * support it. Otherwise, only skip the revoke on un-journaled
102 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
103 (!is_metadata && !ext4_should_journal_data(inode))) {
105 BUFFER_TRACE(bh, "call jbd2_journal_forget");
106 return ext4_journal_forget(handle, bh);
112 * data!=journal && (is_metadata || should_journal_data(inode))
114 BUFFER_TRACE(bh, "call ext4_journal_revoke");
115 err = ext4_journal_revoke(handle, blocknr, bh);
117 ext4_abort(inode->i_sb, __func__,
118 "error %d when attempting revoke", err);
119 BUFFER_TRACE(bh, "exit");
124 * Work out how many blocks we need to proceed with the next chunk of a
125 * truncate transaction.
127 static unsigned long blocks_for_truncate(struct inode *inode)
131 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
133 /* Give ourselves just enough room to cope with inodes in which
134 * i_blocks is corrupt: we've seen disk corruptions in the past
135 * which resulted in random data in an inode which looked enough
136 * like a regular file for ext4 to try to delete it. Things
137 * will go a bit crazy if that happens, but at least we should
138 * try not to panic the whole kernel. */
142 /* But we need to bound the transaction so we don't overflow the
144 if (needed > EXT4_MAX_TRANS_DATA)
145 needed = EXT4_MAX_TRANS_DATA;
147 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
151 * Truncate transactions can be complex and absolutely huge. So we need to
152 * be able to restart the transaction at a conventient checkpoint to make
153 * sure we don't overflow the journal.
155 * start_transaction gets us a new handle for a truncate transaction,
156 * and extend_transaction tries to extend the existing one a bit. If
157 * extend fails, we need to propagate the failure up and restart the
158 * transaction in the top-level truncate loop. --sct
160 static handle_t *start_transaction(struct inode *inode)
164 result = ext4_journal_start(inode, blocks_for_truncate(inode));
168 ext4_std_error(inode->i_sb, PTR_ERR(result));
173 * Try to extend this transaction for the purposes of truncation.
175 * Returns 0 if we managed to create more room. If we can't create more
176 * room, and the transaction must be restarted we return 1.
178 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
180 if (!ext4_handle_valid(handle))
182 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
184 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
190 * Restart the transaction associated with *handle. This does a commit,
191 * so before we call here everything must be consistently dirtied against
194 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
196 BUG_ON(EXT4_JOURNAL(inode) == NULL);
197 jbd_debug(2, "restarting handle %p\n", handle);
198 return ext4_journal_restart(handle, blocks_for_truncate(inode));
202 * Called at the last iput() if i_nlink is zero.
204 void ext4_delete_inode(struct inode *inode)
209 if (ext4_should_order_data(inode))
210 ext4_begin_ordered_truncate(inode, 0);
211 truncate_inode_pages(&inode->i_data, 0);
213 if (is_bad_inode(inode))
216 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
217 if (IS_ERR(handle)) {
218 ext4_std_error(inode->i_sb, PTR_ERR(handle));
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext4_orphan_del(NULL, inode);
229 ext4_handle_sync(handle);
231 err = ext4_mark_inode_dirty(handle, inode);
233 ext4_warning(inode->i_sb, __func__,
234 "couldn't mark inode dirty (err %d)", err);
238 ext4_truncate(inode);
241 * ext4_ext_truncate() doesn't reserve any slop when it
242 * restarts journal transactions; therefore there may not be
243 * enough credits left in the handle to remove the inode from
244 * the orphan list and set the dtime field.
246 if (!ext4_handle_has_enough_credits(handle, 3)) {
247 err = ext4_journal_extend(handle, 3);
249 err = ext4_journal_restart(handle, 3);
251 ext4_warning(inode->i_sb, __func__,
252 "couldn't extend journal (err %d)", err);
254 ext4_journal_stop(handle);
260 * Kill off the orphan record which ext4_truncate created.
261 * AKPM: I think this can be inside the above `if'.
262 * Note that ext4_orphan_del() has to be able to cope with the
263 * deletion of a non-existent orphan - this is because we don't
264 * know if ext4_truncate() actually created an orphan record.
265 * (Well, we could do this if we need to, but heck - it works)
267 ext4_orphan_del(handle, inode);
268 EXT4_I(inode)->i_dtime = get_seconds();
271 * One subtle ordering requirement: if anything has gone wrong
272 * (transaction abort, IO errors, whatever), then we can still
273 * do these next steps (the fs will already have been marked as
274 * having errors), but we can't free the inode if the mark_dirty
277 if (ext4_mark_inode_dirty(handle, inode))
278 /* If that failed, just do the required in-core inode clear. */
281 ext4_free_inode(handle, inode);
282 ext4_journal_stop(handle);
285 clear_inode(inode); /* We must guarantee clearing of inode... */
291 struct buffer_head *bh;
294 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
296 p->key = *(p->p = v);
301 * ext4_block_to_path - parse the block number into array of offsets
302 * @inode: inode in question (we are only interested in its superblock)
303 * @i_block: block number to be parsed
304 * @offsets: array to store the offsets in
305 * @boundary: set this non-zero if the referred-to block is likely to be
306 * followed (on disk) by an indirect block.
308 * To store the locations of file's data ext4 uses a data structure common
309 * for UNIX filesystems - tree of pointers anchored in the inode, with
310 * data blocks at leaves and indirect blocks in intermediate nodes.
311 * This function translates the block number into path in that tree -
312 * return value is the path length and @offsets[n] is the offset of
313 * pointer to (n+1)th node in the nth one. If @block is out of range
314 * (negative or too large) warning is printed and zero returned.
316 * Note: function doesn't find node addresses, so no IO is needed. All
317 * we need to know is the capacity of indirect blocks (taken from the
322 * Portability note: the last comparison (check that we fit into triple
323 * indirect block) is spelled differently, because otherwise on an
324 * architecture with 32-bit longs and 8Kb pages we might get into trouble
325 * if our filesystem had 8Kb blocks. We might use long long, but that would
326 * kill us on x86. Oh, well, at least the sign propagation does not matter -
327 * i_block would have to be negative in the very beginning, so we would not
331 static int ext4_block_to_path(struct inode *inode,
333 ext4_lblk_t offsets[4], int *boundary)
335 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
336 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
337 const long direct_blocks = EXT4_NDIR_BLOCKS,
338 indirect_blocks = ptrs,
339 double_blocks = (1 << (ptrs_bits * 2));
344 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
345 } else if (i_block < direct_blocks) {
346 offsets[n++] = i_block;
347 final = direct_blocks;
348 } else if ((i_block -= direct_blocks) < indirect_blocks) {
349 offsets[n++] = EXT4_IND_BLOCK;
350 offsets[n++] = i_block;
352 } else if ((i_block -= indirect_blocks) < double_blocks) {
353 offsets[n++] = EXT4_DIND_BLOCK;
354 offsets[n++] = i_block >> ptrs_bits;
355 offsets[n++] = i_block & (ptrs - 1);
357 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
358 offsets[n++] = EXT4_TIND_BLOCK;
359 offsets[n++] = i_block >> (ptrs_bits * 2);
360 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
361 offsets[n++] = i_block & (ptrs - 1);
364 ext4_warning(inode->i_sb, "ext4_block_to_path",
365 "block %lu > max in inode %lu",
366 i_block + direct_blocks +
367 indirect_blocks + double_blocks, inode->i_ino);
370 *boundary = final - 1 - (i_block & (ptrs - 1));
375 * ext4_get_branch - read the chain of indirect blocks leading to data
376 * @inode: inode in question
377 * @depth: depth of the chain (1 - direct pointer, etc.)
378 * @offsets: offsets of pointers in inode/indirect blocks
379 * @chain: place to store the result
380 * @err: here we store the error value
382 * Function fills the array of triples <key, p, bh> and returns %NULL
383 * if everything went OK or the pointer to the last filled triple
384 * (incomplete one) otherwise. Upon the return chain[i].key contains
385 * the number of (i+1)-th block in the chain (as it is stored in memory,
386 * i.e. little-endian 32-bit), chain[i].p contains the address of that
387 * number (it points into struct inode for i==0 and into the bh->b_data
388 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
389 * block for i>0 and NULL for i==0. In other words, it holds the block
390 * numbers of the chain, addresses they were taken from (and where we can
391 * verify that chain did not change) and buffer_heads hosting these
394 * Function stops when it stumbles upon zero pointer (absent block)
395 * (pointer to last triple returned, *@err == 0)
396 * or when it gets an IO error reading an indirect block
397 * (ditto, *@err == -EIO)
398 * or when it reads all @depth-1 indirect blocks successfully and finds
399 * the whole chain, all way to the data (returns %NULL, *err == 0).
401 * Need to be called with
402 * down_read(&EXT4_I(inode)->i_data_sem)
404 static Indirect *ext4_get_branch(struct inode *inode, int depth,
405 ext4_lblk_t *offsets,
406 Indirect chain[4], int *err)
408 struct super_block *sb = inode->i_sb;
410 struct buffer_head *bh;
413 /* i_data is not going away, no lock needed */
414 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
418 bh = sb_bread(sb, le32_to_cpu(p->key));
421 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
435 * ext4_find_near - find a place for allocation with sufficient locality
437 * @ind: descriptor of indirect block.
439 * This function returns the preferred place for block allocation.
440 * It is used when heuristic for sequential allocation fails.
442 * + if there is a block to the left of our position - allocate near it.
443 * + if pointer will live in indirect block - allocate near that block.
444 * + if pointer will live in inode - allocate in the same
447 * In the latter case we colour the starting block by the callers PID to
448 * prevent it from clashing with concurrent allocations for a different inode
449 * in the same block group. The PID is used here so that functionally related
450 * files will be close-by on-disk.
452 * Caller must make sure that @ind is valid and will stay that way.
454 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
456 struct ext4_inode_info *ei = EXT4_I(inode);
457 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
459 ext4_fsblk_t bg_start;
460 ext4_fsblk_t last_block;
461 ext4_grpblk_t colour;
462 ext4_group_t block_group;
463 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
465 /* Try to find previous block */
466 for (p = ind->p - 1; p >= start; p--) {
468 return le32_to_cpu(*p);
471 /* No such thing, so let's try location of indirect block */
473 return ind->bh->b_blocknr;
476 * It is going to be referred to from the inode itself? OK, just put it
477 * into the same cylinder group then.
479 block_group = ei->i_block_group;
480 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
481 block_group &= ~(flex_size-1);
482 if (S_ISREG(inode->i_mode))
485 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
486 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
489 * If we are doing delayed allocation, we don't need take
490 * colour into account.
492 if (test_opt(inode->i_sb, DELALLOC))
495 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
496 colour = (current->pid % 16) *
497 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
499 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
500 return bg_start + colour;
504 * ext4_find_goal - find a preferred place for allocation.
506 * @block: block we want
507 * @partial: pointer to the last triple within a chain
509 * Normally this function find the preferred place for block allocation,
512 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
516 * XXX need to get goal block from mballoc's data structures
519 return ext4_find_near(inode, partial);
523 * ext4_blks_to_allocate: Look up the block map and count the number
524 * of direct blocks need to be allocated for the given branch.
526 * @branch: chain of indirect blocks
527 * @k: number of blocks need for indirect blocks
528 * @blks: number of data blocks to be mapped.
529 * @blocks_to_boundary: the offset in the indirect block
531 * return the total number of blocks to be allocate, including the
532 * direct and indirect blocks.
534 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
535 int blocks_to_boundary)
537 unsigned int count = 0;
540 * Simple case, [t,d]Indirect block(s) has not allocated yet
541 * then it's clear blocks on that path have not allocated
544 /* right now we don't handle cross boundary allocation */
545 if (blks < blocks_to_boundary + 1)
548 count += blocks_to_boundary + 1;
553 while (count < blks && count <= blocks_to_boundary &&
554 le32_to_cpu(*(branch[0].p + count)) == 0) {
561 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
562 * @indirect_blks: the number of blocks need to allocate for indirect
565 * @new_blocks: on return it will store the new block numbers for
566 * the indirect blocks(if needed) and the first direct block,
567 * @blks: on return it will store the total number of allocated
570 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
571 ext4_lblk_t iblock, ext4_fsblk_t goal,
572 int indirect_blks, int blks,
573 ext4_fsblk_t new_blocks[4], int *err)
575 struct ext4_allocation_request ar;
577 unsigned long count = 0, blk_allocated = 0;
579 ext4_fsblk_t current_block = 0;
583 * Here we try to allocate the requested multiple blocks at once,
584 * on a best-effort basis.
585 * To build a branch, we should allocate blocks for
586 * the indirect blocks(if not allocated yet), and at least
587 * the first direct block of this branch. That's the
588 * minimum number of blocks need to allocate(required)
590 /* first we try to allocate the indirect blocks */
591 target = indirect_blks;
594 /* allocating blocks for indirect blocks and direct blocks */
595 current_block = ext4_new_meta_blocks(handle, inode,
601 /* allocate blocks for indirect blocks */
602 while (index < indirect_blks && count) {
603 new_blocks[index++] = current_block++;
608 * save the new block number
609 * for the first direct block
611 new_blocks[index] = current_block;
612 printk(KERN_INFO "%s returned more blocks than "
613 "requested\n", __func__);
619 target = blks - count ;
620 blk_allocated = count;
623 /* Now allocate data blocks */
624 memset(&ar, 0, sizeof(ar));
629 if (S_ISREG(inode->i_mode))
630 /* enable in-core preallocation only for regular files */
631 ar.flags = EXT4_MB_HINT_DATA;
633 current_block = ext4_mb_new_blocks(handle, &ar, err);
635 if (*err && (target == blks)) {
637 * if the allocation failed and we didn't allocate
643 if (target == blks) {
645 * save the new block number
646 * for the first direct block
648 new_blocks[index] = current_block;
650 blk_allocated += ar.len;
653 /* total number of blocks allocated for direct blocks */
658 for (i = 0; i < index; i++)
659 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
664 * ext4_alloc_branch - allocate and set up a chain of blocks.
666 * @indirect_blks: number of allocated indirect blocks
667 * @blks: number of allocated direct blocks
668 * @offsets: offsets (in the blocks) to store the pointers to next.
669 * @branch: place to store the chain in.
671 * This function allocates blocks, zeroes out all but the last one,
672 * links them into chain and (if we are synchronous) writes them to disk.
673 * In other words, it prepares a branch that can be spliced onto the
674 * inode. It stores the information about that chain in the branch[], in
675 * the same format as ext4_get_branch() would do. We are calling it after
676 * we had read the existing part of chain and partial points to the last
677 * triple of that (one with zero ->key). Upon the exit we have the same
678 * picture as after the successful ext4_get_block(), except that in one
679 * place chain is disconnected - *branch->p is still zero (we did not
680 * set the last link), but branch->key contains the number that should
681 * be placed into *branch->p to fill that gap.
683 * If allocation fails we free all blocks we've allocated (and forget
684 * their buffer_heads) and return the error value the from failed
685 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
686 * as described above and return 0.
688 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
689 ext4_lblk_t iblock, int indirect_blks,
690 int *blks, ext4_fsblk_t goal,
691 ext4_lblk_t *offsets, Indirect *branch)
693 int blocksize = inode->i_sb->s_blocksize;
696 struct buffer_head *bh;
698 ext4_fsblk_t new_blocks[4];
699 ext4_fsblk_t current_block;
701 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
702 *blks, new_blocks, &err);
706 branch[0].key = cpu_to_le32(new_blocks[0]);
708 * metadata blocks and data blocks are allocated.
710 for (n = 1; n <= indirect_blks; n++) {
712 * Get buffer_head for parent block, zero it out
713 * and set the pointer to new one, then send
716 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
719 BUFFER_TRACE(bh, "call get_create_access");
720 err = ext4_journal_get_create_access(handle, bh);
727 memset(bh->b_data, 0, blocksize);
728 branch[n].p = (__le32 *) bh->b_data + offsets[n];
729 branch[n].key = cpu_to_le32(new_blocks[n]);
730 *branch[n].p = branch[n].key;
731 if (n == indirect_blks) {
732 current_block = new_blocks[n];
734 * End of chain, update the last new metablock of
735 * the chain to point to the new allocated
736 * data blocks numbers
738 for (i=1; i < num; i++)
739 *(branch[n].p + i) = cpu_to_le32(++current_block);
741 BUFFER_TRACE(bh, "marking uptodate");
742 set_buffer_uptodate(bh);
745 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
746 err = ext4_handle_dirty_metadata(handle, inode, bh);
753 /* Allocation failed, free what we already allocated */
754 for (i = 1; i <= n ; i++) {
755 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
756 ext4_journal_forget(handle, branch[i].bh);
758 for (i = 0; i < indirect_blks; i++)
759 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
761 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
767 * ext4_splice_branch - splice the allocated branch onto inode.
769 * @block: (logical) number of block we are adding
770 * @chain: chain of indirect blocks (with a missing link - see
772 * @where: location of missing link
773 * @num: number of indirect blocks we are adding
774 * @blks: number of direct blocks we are adding
776 * This function fills the missing link and does all housekeeping needed in
777 * inode (->i_blocks, etc.). In case of success we end up with the full
778 * chain to new block and return 0.
780 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
781 ext4_lblk_t block, Indirect *where, int num, int blks)
785 ext4_fsblk_t current_block;
788 * If we're splicing into a [td]indirect block (as opposed to the
789 * inode) then we need to get write access to the [td]indirect block
793 BUFFER_TRACE(where->bh, "get_write_access");
794 err = ext4_journal_get_write_access(handle, where->bh);
800 *where->p = where->key;
803 * Update the host buffer_head or inode to point to more just allocated
804 * direct blocks blocks
806 if (num == 0 && blks > 1) {
807 current_block = le32_to_cpu(where->key) + 1;
808 for (i = 1; i < blks; i++)
809 *(where->p + i) = cpu_to_le32(current_block++);
812 /* We are done with atomic stuff, now do the rest of housekeeping */
814 inode->i_ctime = ext4_current_time(inode);
815 ext4_mark_inode_dirty(handle, inode);
817 /* had we spliced it onto indirect block? */
820 * If we spliced it onto an indirect block, we haven't
821 * altered the inode. Note however that if it is being spliced
822 * onto an indirect block at the very end of the file (the
823 * file is growing) then we *will* alter the inode to reflect
824 * the new i_size. But that is not done here - it is done in
825 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
827 jbd_debug(5, "splicing indirect only\n");
828 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
829 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
834 * OK, we spliced it into the inode itself on a direct block.
835 * Inode was dirtied above.
837 jbd_debug(5, "splicing direct\n");
842 for (i = 1; i <= num; i++) {
843 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
844 ext4_journal_forget(handle, where[i].bh);
845 ext4_free_blocks(handle, inode,
846 le32_to_cpu(where[i-1].key), 1, 0);
848 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
854 * Allocation strategy is simple: if we have to allocate something, we will
855 * have to go the whole way to leaf. So let's do it before attaching anything
856 * to tree, set linkage between the newborn blocks, write them if sync is
857 * required, recheck the path, free and repeat if check fails, otherwise
858 * set the last missing link (that will protect us from any truncate-generated
859 * removals - all blocks on the path are immune now) and possibly force the
860 * write on the parent block.
861 * That has a nice additional property: no special recovery from the failed
862 * allocations is needed - we simply release blocks and do not touch anything
863 * reachable from inode.
865 * `handle' can be NULL if create == 0.
867 * return > 0, # of blocks mapped or allocated.
868 * return = 0, if plain lookup failed.
869 * return < 0, error case.
872 * Need to be called with
873 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
874 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
876 static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
877 ext4_lblk_t iblock, unsigned int maxblocks,
878 struct buffer_head *bh_result,
879 int create, int extend_disksize)
882 ext4_lblk_t offsets[4];
887 int blocks_to_boundary = 0;
889 struct ext4_inode_info *ei = EXT4_I(inode);
891 ext4_fsblk_t first_block = 0;
895 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
896 J_ASSERT(handle != NULL || create == 0);
897 depth = ext4_block_to_path(inode, iblock, offsets,
898 &blocks_to_boundary);
903 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
905 /* Simplest case - block found, no allocation needed */
907 first_block = le32_to_cpu(chain[depth - 1].key);
908 clear_buffer_new(bh_result);
911 while (count < maxblocks && count <= blocks_to_boundary) {
914 blk = le32_to_cpu(*(chain[depth-1].p + count));
916 if (blk == first_block + count)
924 /* Next simple case - plain lookup or failed read of indirect block */
925 if (!create || err == -EIO)
929 * Okay, we need to do block allocation.
931 goal = ext4_find_goal(inode, iblock, partial);
933 /* the number of blocks need to allocate for [d,t]indirect blocks */
934 indirect_blks = (chain + depth) - partial - 1;
937 * Next look up the indirect map to count the totoal number of
938 * direct blocks to allocate for this branch.
940 count = ext4_blks_to_allocate(partial, indirect_blks,
941 maxblocks, blocks_to_boundary);
943 * Block out ext4_truncate while we alter the tree
945 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
947 offsets + (partial - chain), partial);
950 * The ext4_splice_branch call will free and forget any buffers
951 * on the new chain if there is a failure, but that risks using
952 * up transaction credits, especially for bitmaps where the
953 * credits cannot be returned. Can we handle this somehow? We
954 * may need to return -EAGAIN upwards in the worst case. --sct
957 err = ext4_splice_branch(handle, inode, iblock,
958 partial, indirect_blks, count);
960 * i_disksize growing is protected by i_data_sem. Don't forget to
961 * protect it if you're about to implement concurrent
962 * ext4_get_block() -bzzz
964 if (!err && extend_disksize) {
965 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
966 if (disksize > i_size_read(inode))
967 disksize = i_size_read(inode);
968 if (disksize > ei->i_disksize)
969 ei->i_disksize = disksize;
974 set_buffer_new(bh_result);
976 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
977 if (count > blocks_to_boundary)
978 set_buffer_boundary(bh_result);
980 /* Clean up and exit */
981 partial = chain + depth - 1; /* the whole chain */
983 while (partial > chain) {
984 BUFFER_TRACE(partial->bh, "call brelse");
988 BUFFER_TRACE(bh_result, "returned");
993 qsize_t ext4_get_reserved_space(struct inode *inode)
995 unsigned long long total;
997 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
998 total = EXT4_I(inode)->i_reserved_data_blocks +
999 EXT4_I(inode)->i_reserved_meta_blocks;
1000 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1005 * Calculate the number of metadata blocks need to reserve
1006 * to allocate @blocks for non extent file based file
1008 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1010 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1011 int ind_blks, dind_blks, tind_blks;
1013 /* number of new indirect blocks needed */
1014 ind_blks = (blocks + icap - 1) / icap;
1016 dind_blks = (ind_blks + icap - 1) / icap;
1020 return ind_blks + dind_blks + tind_blks;
1024 * Calculate the number of metadata blocks need to reserve
1025 * to allocate given number of blocks
1027 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1032 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1033 return ext4_ext_calc_metadata_amount(inode, blocks);
1035 return ext4_indirect_calc_metadata_amount(inode, blocks);
1038 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1040 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1041 int total, mdb, mdb_free;
1043 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1044 /* recalculate the number of metablocks still need to be reserved */
1045 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1046 mdb = ext4_calc_metadata_amount(inode, total);
1048 /* figure out how many metablocks to release */
1049 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1050 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1053 /* Account for allocated meta_blocks */
1054 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1056 /* update fs dirty blocks counter */
1057 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1058 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1059 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1062 /* update per-inode reservations */
1063 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1064 EXT4_I(inode)->i_reserved_data_blocks -= used;
1065 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1068 * free those over-booking quota for metadata blocks
1072 vfs_dq_release_reservation_block(inode, mdb_free);
1076 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1077 * and returns if the blocks are already mapped.
1079 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1080 * and store the allocated blocks in the result buffer head and mark it
1083 * If file type is extents based, it will call ext4_ext_get_blocks(),
1084 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1087 * On success, it returns the number of blocks being mapped or allocate.
1088 * if create==0 and the blocks are pre-allocated and uninitialized block,
1089 * the result buffer head is unmapped. If the create ==1, it will make sure
1090 * the buffer head is mapped.
1092 * It returns 0 if plain look up failed (blocks have not been allocated), in
1093 * that casem, buffer head is unmapped
1095 * It returns the error in case of allocation failure.
1097 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1098 unsigned int max_blocks, struct buffer_head *bh,
1099 int create, int extend_disksize, int flag)
1103 clear_buffer_mapped(bh);
1106 * Try to see if we can get the block without requesting
1107 * for new file system block.
1109 down_read((&EXT4_I(inode)->i_data_sem));
1110 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1111 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1114 retval = ext4_get_blocks_handle(handle,
1115 inode, block, max_blocks, bh, 0, 0);
1117 up_read((&EXT4_I(inode)->i_data_sem));
1119 /* If it is only a block(s) look up */
1124 * Returns if the blocks have already allocated
1126 * Note that if blocks have been preallocated
1127 * ext4_ext_get_block() returns th create = 0
1128 * with buffer head unmapped.
1130 if (retval > 0 && buffer_mapped(bh))
1134 * New blocks allocate and/or writing to uninitialized extent
1135 * will possibly result in updating i_data, so we take
1136 * the write lock of i_data_sem, and call get_blocks()
1137 * with create == 1 flag.
1139 down_write((&EXT4_I(inode)->i_data_sem));
1142 * if the caller is from delayed allocation writeout path
1143 * we have already reserved fs blocks for allocation
1144 * let the underlying get_block() function know to
1145 * avoid double accounting
1148 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1150 * We need to check for EXT4 here because migrate
1151 * could have changed the inode type in between
1153 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1154 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1155 bh, create, extend_disksize);
1157 retval = ext4_get_blocks_handle(handle, inode, block,
1158 max_blocks, bh, create, extend_disksize);
1160 if (retval > 0 && buffer_new(bh)) {
1162 * We allocated new blocks which will result in
1163 * i_data's format changing. Force the migrate
1164 * to fail by clearing migrate flags
1166 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1172 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1174 * Update reserved blocks/metadata blocks
1175 * after successful block allocation
1176 * which were deferred till now
1178 if ((retval > 0) && buffer_delay(bh))
1179 ext4_da_update_reserve_space(inode, retval);
1182 up_write((&EXT4_I(inode)->i_data_sem));
1186 /* Maximum number of blocks we map for direct IO at once. */
1187 #define DIO_MAX_BLOCKS 4096
1189 int ext4_get_block(struct inode *inode, sector_t iblock,
1190 struct buffer_head *bh_result, int create)
1192 handle_t *handle = ext4_journal_current_handle();
1193 int ret = 0, started = 0;
1194 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1197 if (create && !handle) {
1198 /* Direct IO write... */
1199 if (max_blocks > DIO_MAX_BLOCKS)
1200 max_blocks = DIO_MAX_BLOCKS;
1201 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1202 handle = ext4_journal_start(inode, dio_credits);
1203 if (IS_ERR(handle)) {
1204 ret = PTR_ERR(handle);
1210 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1211 max_blocks, bh_result, create, 0, 0);
1213 bh_result->b_size = (ret << inode->i_blkbits);
1217 ext4_journal_stop(handle);
1223 * `handle' can be NULL if create is zero
1225 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1226 ext4_lblk_t block, int create, int *errp)
1228 struct buffer_head dummy;
1231 J_ASSERT(handle != NULL || create == 0);
1234 dummy.b_blocknr = -1000;
1235 buffer_trace_init(&dummy.b_history);
1236 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1237 &dummy, create, 1, 0);
1239 * ext4_get_blocks_handle() returns number of blocks
1240 * mapped. 0 in case of a HOLE.
1248 if (!err && buffer_mapped(&dummy)) {
1249 struct buffer_head *bh;
1250 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1255 if (buffer_new(&dummy)) {
1256 J_ASSERT(create != 0);
1257 J_ASSERT(handle != NULL);
1260 * Now that we do not always journal data, we should
1261 * keep in mind whether this should always journal the
1262 * new buffer as metadata. For now, regular file
1263 * writes use ext4_get_block instead, so it's not a
1267 BUFFER_TRACE(bh, "call get_create_access");
1268 fatal = ext4_journal_get_create_access(handle, bh);
1269 if (!fatal && !buffer_uptodate(bh)) {
1270 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1271 set_buffer_uptodate(bh);
1274 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1275 err = ext4_handle_dirty_metadata(handle, inode, bh);
1279 BUFFER_TRACE(bh, "not a new buffer");
1292 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1293 ext4_lblk_t block, int create, int *err)
1295 struct buffer_head *bh;
1297 bh = ext4_getblk(handle, inode, block, create, err);
1300 if (buffer_uptodate(bh))
1302 ll_rw_block(READ_META, 1, &bh);
1304 if (buffer_uptodate(bh))
1311 static int walk_page_buffers(handle_t *handle,
1312 struct buffer_head *head,
1316 int (*fn)(handle_t *handle,
1317 struct buffer_head *bh))
1319 struct buffer_head *bh;
1320 unsigned block_start, block_end;
1321 unsigned blocksize = head->b_size;
1323 struct buffer_head *next;
1325 for (bh = head, block_start = 0;
1326 ret == 0 && (bh != head || !block_start);
1327 block_start = block_end, bh = next)
1329 next = bh->b_this_page;
1330 block_end = block_start + blocksize;
1331 if (block_end <= from || block_start >= to) {
1332 if (partial && !buffer_uptodate(bh))
1336 err = (*fn)(handle, bh);
1344 * To preserve ordering, it is essential that the hole instantiation and
1345 * the data write be encapsulated in a single transaction. We cannot
1346 * close off a transaction and start a new one between the ext4_get_block()
1347 * and the commit_write(). So doing the jbd2_journal_start at the start of
1348 * prepare_write() is the right place.
1350 * Also, this function can nest inside ext4_writepage() ->
1351 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1352 * has generated enough buffer credits to do the whole page. So we won't
1353 * block on the journal in that case, which is good, because the caller may
1356 * By accident, ext4 can be reentered when a transaction is open via
1357 * quota file writes. If we were to commit the transaction while thus
1358 * reentered, there can be a deadlock - we would be holding a quota
1359 * lock, and the commit would never complete if another thread had a
1360 * transaction open and was blocking on the quota lock - a ranking
1363 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1364 * will _not_ run commit under these circumstances because handle->h_ref
1365 * is elevated. We'll still have enough credits for the tiny quotafile
1368 static int do_journal_get_write_access(handle_t *handle,
1369 struct buffer_head *bh)
1371 if (!buffer_mapped(bh) || buffer_freed(bh))
1373 return ext4_journal_get_write_access(handle, bh);
1376 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1377 loff_t pos, unsigned len, unsigned flags,
1378 struct page **pagep, void **fsdata)
1380 struct inode *inode = mapping->host;
1381 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1388 trace_mark(ext4_write_begin,
1389 "dev %s ino %lu pos %llu len %u flags %u",
1390 inode->i_sb->s_id, inode->i_ino,
1391 (unsigned long long) pos, len, flags);
1392 index = pos >> PAGE_CACHE_SHIFT;
1393 from = pos & (PAGE_CACHE_SIZE - 1);
1397 handle = ext4_journal_start(inode, needed_blocks);
1398 if (IS_ERR(handle)) {
1399 ret = PTR_ERR(handle);
1403 /* We cannot recurse into the filesystem as the transaction is already
1405 flags |= AOP_FLAG_NOFS;
1407 page = grab_cache_page_write_begin(mapping, index, flags);
1409 ext4_journal_stop(handle);
1415 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1418 if (!ret && ext4_should_journal_data(inode)) {
1419 ret = walk_page_buffers(handle, page_buffers(page),
1420 from, to, NULL, do_journal_get_write_access);
1425 ext4_journal_stop(handle);
1426 page_cache_release(page);
1428 * block_write_begin may have instantiated a few blocks
1429 * outside i_size. Trim these off again. Don't need
1430 * i_size_read because we hold i_mutex.
1432 if (pos + len > inode->i_size)
1433 vmtruncate(inode, inode->i_size);
1436 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1442 /* For write_end() in data=journal mode */
1443 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1445 if (!buffer_mapped(bh) || buffer_freed(bh))
1447 set_buffer_uptodate(bh);
1448 return ext4_handle_dirty_metadata(handle, NULL, bh);
1452 * We need to pick up the new inode size which generic_commit_write gave us
1453 * `file' can be NULL - eg, when called from page_symlink().
1455 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1456 * buffers are managed internally.
1458 static int ext4_ordered_write_end(struct file *file,
1459 struct address_space *mapping,
1460 loff_t pos, unsigned len, unsigned copied,
1461 struct page *page, void *fsdata)
1463 handle_t *handle = ext4_journal_current_handle();
1464 struct inode *inode = mapping->host;
1467 trace_mark(ext4_ordered_write_end,
1468 "dev %s ino %lu pos %llu len %u copied %u",
1469 inode->i_sb->s_id, inode->i_ino,
1470 (unsigned long long) pos, len, copied);
1471 ret = ext4_jbd2_file_inode(handle, inode);
1476 new_i_size = pos + copied;
1477 if (new_i_size > EXT4_I(inode)->i_disksize) {
1478 ext4_update_i_disksize(inode, new_i_size);
1479 /* We need to mark inode dirty even if
1480 * new_i_size is less that inode->i_size
1481 * bu greater than i_disksize.(hint delalloc)
1483 ext4_mark_inode_dirty(handle, inode);
1486 ret2 = generic_write_end(file, mapping, pos, len, copied,
1492 ret2 = ext4_journal_stop(handle);
1496 return ret ? ret : copied;
1499 static int ext4_writeback_write_end(struct file *file,
1500 struct address_space *mapping,
1501 loff_t pos, unsigned len, unsigned copied,
1502 struct page *page, void *fsdata)
1504 handle_t *handle = ext4_journal_current_handle();
1505 struct inode *inode = mapping->host;
1509 trace_mark(ext4_writeback_write_end,
1510 "dev %s ino %lu pos %llu len %u copied %u",
1511 inode->i_sb->s_id, inode->i_ino,
1512 (unsigned long long) pos, len, copied);
1513 new_i_size = pos + copied;
1514 if (new_i_size > EXT4_I(inode)->i_disksize) {
1515 ext4_update_i_disksize(inode, new_i_size);
1516 /* We need to mark inode dirty even if
1517 * new_i_size is less that inode->i_size
1518 * bu greater than i_disksize.(hint delalloc)
1520 ext4_mark_inode_dirty(handle, inode);
1523 ret2 = generic_write_end(file, mapping, pos, len, copied,
1529 ret2 = ext4_journal_stop(handle);
1533 return ret ? ret : copied;
1536 static int ext4_journalled_write_end(struct file *file,
1537 struct address_space *mapping,
1538 loff_t pos, unsigned len, unsigned copied,
1539 struct page *page, void *fsdata)
1541 handle_t *handle = ext4_journal_current_handle();
1542 struct inode *inode = mapping->host;
1548 trace_mark(ext4_journalled_write_end,
1549 "dev %s ino %lu pos %llu len %u copied %u",
1550 inode->i_sb->s_id, inode->i_ino,
1551 (unsigned long long) pos, len, copied);
1552 from = pos & (PAGE_CACHE_SIZE - 1);
1556 if (!PageUptodate(page))
1558 page_zero_new_buffers(page, from+copied, to);
1561 ret = walk_page_buffers(handle, page_buffers(page), from,
1562 to, &partial, write_end_fn);
1564 SetPageUptodate(page);
1565 new_i_size = pos + copied;
1566 if (new_i_size > inode->i_size)
1567 i_size_write(inode, pos+copied);
1568 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1569 if (new_i_size > EXT4_I(inode)->i_disksize) {
1570 ext4_update_i_disksize(inode, new_i_size);
1571 ret2 = ext4_mark_inode_dirty(handle, inode);
1577 ret2 = ext4_journal_stop(handle);
1580 page_cache_release(page);
1582 return ret ? ret : copied;
1585 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1588 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1589 unsigned long md_needed, mdblocks, total = 0;
1592 * recalculate the amount of metadata blocks to reserve
1593 * in order to allocate nrblocks
1594 * worse case is one extent per block
1597 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1598 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1599 mdblocks = ext4_calc_metadata_amount(inode, total);
1600 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1602 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1603 total = md_needed + nrblocks;
1606 * Make quota reservation here to prevent quota overflow
1607 * later. Real quota accounting is done at pages writeout
1610 if (vfs_dq_reserve_block(inode, total)) {
1611 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1615 if (ext4_claim_free_blocks(sbi, total)) {
1616 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1617 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1621 vfs_dq_release_reservation_block(inode, total);
1624 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1625 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1627 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1628 return 0; /* success */
1631 static void ext4_da_release_space(struct inode *inode, int to_free)
1633 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1634 int total, mdb, mdb_free, release;
1637 return; /* Nothing to release, exit */
1639 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1641 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1643 * if there is no reserved blocks, but we try to free some
1644 * then the counter is messed up somewhere.
1645 * but since this function is called from invalidate
1646 * page, it's harmless to return without any action
1648 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1649 "blocks for inode %lu, but there is no reserved "
1650 "data blocks\n", to_free, inode->i_ino);
1651 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1655 /* recalculate the number of metablocks still need to be reserved */
1656 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1657 mdb = ext4_calc_metadata_amount(inode, total);
1659 /* figure out how many metablocks to release */
1660 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1661 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1663 release = to_free + mdb_free;
1665 /* update fs dirty blocks counter for truncate case */
1666 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1668 /* update per-inode reservations */
1669 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1670 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1672 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1673 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1674 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1676 vfs_dq_release_reservation_block(inode, release);
1679 static void ext4_da_page_release_reservation(struct page *page,
1680 unsigned long offset)
1683 struct buffer_head *head, *bh;
1684 unsigned int curr_off = 0;
1686 head = page_buffers(page);
1689 unsigned int next_off = curr_off + bh->b_size;
1691 if ((offset <= curr_off) && (buffer_delay(bh))) {
1693 clear_buffer_delay(bh);
1695 curr_off = next_off;
1696 } while ((bh = bh->b_this_page) != head);
1697 ext4_da_release_space(page->mapping->host, to_release);
1701 * Delayed allocation stuff
1704 struct mpage_da_data {
1705 struct inode *inode;
1706 sector_t b_blocknr; /* start block number of extent */
1707 size_t b_size; /* size of extent */
1708 unsigned long b_state; /* state of the extent */
1709 unsigned long first_page, next_page; /* extent of pages */
1710 struct writeback_control *wbc;
1717 * mpage_da_submit_io - walks through extent of pages and try to write
1718 * them with writepage() call back
1720 * @mpd->inode: inode
1721 * @mpd->first_page: first page of the extent
1722 * @mpd->next_page: page after the last page of the extent
1724 * By the time mpage_da_submit_io() is called we expect all blocks
1725 * to be allocated. this may be wrong if allocation failed.
1727 * As pages are already locked by write_cache_pages(), we can't use it
1729 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1732 struct pagevec pvec;
1733 unsigned long index, end;
1734 int ret = 0, err, nr_pages, i;
1735 struct inode *inode = mpd->inode;
1736 struct address_space *mapping = inode->i_mapping;
1738 BUG_ON(mpd->next_page <= mpd->first_page);
1740 * We need to start from the first_page to the next_page - 1
1741 * to make sure we also write the mapped dirty buffer_heads.
1742 * If we look at mpd->b_blocknr we would only be looking
1743 * at the currently mapped buffer_heads.
1745 index = mpd->first_page;
1746 end = mpd->next_page - 1;
1748 pagevec_init(&pvec, 0);
1749 while (index <= end) {
1750 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1753 for (i = 0; i < nr_pages; i++) {
1754 struct page *page = pvec.pages[i];
1756 index = page->index;
1761 BUG_ON(!PageLocked(page));
1762 BUG_ON(PageWriteback(page));
1764 pages_skipped = mpd->wbc->pages_skipped;
1765 err = mapping->a_ops->writepage(page, mpd->wbc);
1766 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1768 * have successfully written the page
1769 * without skipping the same
1771 mpd->pages_written++;
1773 * In error case, we have to continue because
1774 * remaining pages are still locked
1775 * XXX: unlock and re-dirty them?
1780 pagevec_release(&pvec);
1786 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1788 * @mpd->inode - inode to walk through
1789 * @exbh->b_blocknr - first block on a disk
1790 * @exbh->b_size - amount of space in bytes
1791 * @logical - first logical block to start assignment with
1793 * the function goes through all passed space and put actual disk
1794 * block numbers into buffer heads, dropping BH_Delay
1796 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1797 struct buffer_head *exbh)
1799 struct inode *inode = mpd->inode;
1800 struct address_space *mapping = inode->i_mapping;
1801 int blocks = exbh->b_size >> inode->i_blkbits;
1802 sector_t pblock = exbh->b_blocknr, cur_logical;
1803 struct buffer_head *head, *bh;
1805 struct pagevec pvec;
1808 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1809 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1810 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1812 pagevec_init(&pvec, 0);
1814 while (index <= end) {
1815 /* XXX: optimize tail */
1816 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1819 for (i = 0; i < nr_pages; i++) {
1820 struct page *page = pvec.pages[i];
1822 index = page->index;
1827 BUG_ON(!PageLocked(page));
1828 BUG_ON(PageWriteback(page));
1829 BUG_ON(!page_has_buffers(page));
1831 bh = page_buffers(page);
1834 /* skip blocks out of the range */
1836 if (cur_logical >= logical)
1839 } while ((bh = bh->b_this_page) != head);
1842 if (cur_logical >= logical + blocks)
1844 if (buffer_delay(bh)) {
1845 bh->b_blocknr = pblock;
1846 clear_buffer_delay(bh);
1847 bh->b_bdev = inode->i_sb->s_bdev;
1848 } else if (buffer_unwritten(bh)) {
1849 bh->b_blocknr = pblock;
1850 clear_buffer_unwritten(bh);
1851 set_buffer_mapped(bh);
1853 bh->b_bdev = inode->i_sb->s_bdev;
1854 } else if (buffer_mapped(bh))
1855 BUG_ON(bh->b_blocknr != pblock);
1859 } while ((bh = bh->b_this_page) != head);
1861 pagevec_release(&pvec);
1867 * __unmap_underlying_blocks - just a helper function to unmap
1868 * set of blocks described by @bh
1870 static inline void __unmap_underlying_blocks(struct inode *inode,
1871 struct buffer_head *bh)
1873 struct block_device *bdev = inode->i_sb->s_bdev;
1876 blocks = bh->b_size >> inode->i_blkbits;
1877 for (i = 0; i < blocks; i++)
1878 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1881 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1882 sector_t logical, long blk_cnt)
1886 struct pagevec pvec;
1887 struct inode *inode = mpd->inode;
1888 struct address_space *mapping = inode->i_mapping;
1890 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1891 end = (logical + blk_cnt - 1) >>
1892 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1893 while (index <= end) {
1894 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1897 for (i = 0; i < nr_pages; i++) {
1898 struct page *page = pvec.pages[i];
1899 index = page->index;
1904 BUG_ON(!PageLocked(page));
1905 BUG_ON(PageWriteback(page));
1906 block_invalidatepage(page, 0);
1907 ClearPageUptodate(page);
1914 static void ext4_print_free_blocks(struct inode *inode)
1916 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1917 printk(KERN_EMERG "Total free blocks count %lld\n",
1918 ext4_count_free_blocks(inode->i_sb));
1919 printk(KERN_EMERG "Free/Dirty block details\n");
1920 printk(KERN_EMERG "free_blocks=%lld\n",
1921 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1922 printk(KERN_EMERG "dirty_blocks=%lld\n",
1923 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1924 printk(KERN_EMERG "Block reservation details\n");
1925 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1926 EXT4_I(inode)->i_reserved_data_blocks);
1927 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1928 EXT4_I(inode)->i_reserved_meta_blocks);
1932 #define EXT4_DELALLOC_RSVED 1
1933 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
1934 struct buffer_head *bh_result, int create)
1937 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1938 loff_t disksize = EXT4_I(inode)->i_disksize;
1939 handle_t *handle = NULL;
1941 handle = ext4_journal_current_handle();
1943 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
1944 bh_result, create, 0, EXT4_DELALLOC_RSVED);
1948 bh_result->b_size = (ret << inode->i_blkbits);
1950 if (ext4_should_order_data(inode)) {
1952 retval = ext4_jbd2_file_inode(handle, inode);
1955 * Failed to add inode for ordered mode. Don't
1962 * Update on-disk size along with block allocation we don't
1963 * use 'extend_disksize' as size may change within already
1964 * allocated block -bzzz
1966 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
1967 if (disksize > i_size_read(inode))
1968 disksize = i_size_read(inode);
1969 if (disksize > EXT4_I(inode)->i_disksize) {
1970 ext4_update_i_disksize(inode, disksize);
1971 ret = ext4_mark_inode_dirty(handle, inode);
1978 * mpage_da_map_blocks - go through given space
1980 * @mpd - bh describing space
1982 * The function skips space we know is already mapped to disk blocks.
1985 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1988 struct buffer_head new;
1992 * We consider only non-mapped and non-allocated blocks
1994 if ((mpd->b_state & (1 << BH_Mapped)) &&
1995 !(mpd->b_state & (1 << BH_Delay)))
1997 new.b_state = mpd->b_state;
1999 new.b_size = mpd->b_size;
2000 next = mpd->b_blocknr;
2002 * If we didn't accumulate anything
2003 * to write simply return
2008 err = ext4_da_get_block_write(mpd->inode, next, &new, 1);
2011 * If get block returns with error we simply
2012 * return. Later writepage will redirty the page and
2013 * writepages will find the dirty page again
2018 if (err == -ENOSPC &&
2019 ext4_count_free_blocks(mpd->inode->i_sb)) {
2025 * get block failure will cause us to loop in
2026 * writepages, because a_ops->writepage won't be able
2027 * to make progress. The page will be redirtied by
2028 * writepage and writepages will again try to write
2031 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2032 "at logical offset %llu with max blocks "
2033 "%zd with error %d\n",
2034 __func__, mpd->inode->i_ino,
2035 (unsigned long long)next,
2036 mpd->b_size >> mpd->inode->i_blkbits, err);
2037 printk(KERN_EMERG "This should not happen.!! "
2038 "Data will be lost\n");
2039 if (err == -ENOSPC) {
2040 ext4_print_free_blocks(mpd->inode);
2042 /* invlaidate all the pages */
2043 ext4_da_block_invalidatepages(mpd, next,
2044 mpd->b_size >> mpd->inode->i_blkbits);
2047 BUG_ON(new.b_size == 0);
2049 if (buffer_new(&new))
2050 __unmap_underlying_blocks(mpd->inode, &new);
2053 * If blocks are delayed marked, we need to
2054 * put actual blocknr and drop delayed bit
2056 if ((mpd->b_state & (1 << BH_Delay)) ||
2057 (mpd->b_state & (1 << BH_Unwritten)))
2058 mpage_put_bnr_to_bhs(mpd, next, &new);
2063 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2064 (1 << BH_Delay) | (1 << BH_Unwritten))
2067 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2069 * @mpd->lbh - extent of blocks
2070 * @logical - logical number of the block in the file
2071 * @bh - bh of the block (used to access block's state)
2073 * the function is used to collect contig. blocks in same state
2075 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2076 sector_t logical, size_t b_size,
2077 unsigned long b_state)
2080 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2082 /* check if thereserved journal credits might overflow */
2083 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2084 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2086 * With non-extent format we are limited by the journal
2087 * credit available. Total credit needed to insert
2088 * nrblocks contiguous blocks is dependent on the
2089 * nrblocks. So limit nrblocks.
2092 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2093 EXT4_MAX_TRANS_DATA) {
2095 * Adding the new buffer_head would make it cross the
2096 * allowed limit for which we have journal credit
2097 * reserved. So limit the new bh->b_size
2099 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2100 mpd->inode->i_blkbits;
2101 /* we will do mpage_da_submit_io in the next loop */
2105 * First block in the extent
2107 if (mpd->b_size == 0) {
2108 mpd->b_blocknr = logical;
2109 mpd->b_size = b_size;
2110 mpd->b_state = b_state & BH_FLAGS;
2114 next = mpd->b_blocknr + nrblocks;
2116 * Can we merge the block to our big extent?
2118 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2119 mpd->b_size += b_size;
2125 * We couldn't merge the block to our extent, so we
2126 * need to flush current extent and start new one
2128 if (mpage_da_map_blocks(mpd) == 0)
2129 mpage_da_submit_io(mpd);
2135 * __mpage_da_writepage - finds extent of pages and blocks
2137 * @page: page to consider
2138 * @wbc: not used, we just follow rules
2141 * The function finds extents of pages and scan them for all blocks.
2143 static int __mpage_da_writepage(struct page *page,
2144 struct writeback_control *wbc, void *data)
2146 struct mpage_da_data *mpd = data;
2147 struct inode *inode = mpd->inode;
2148 struct buffer_head *bh, *head;
2153 * Rest of the page in the page_vec
2154 * redirty then and skip then. We will
2155 * try to to write them again after
2156 * starting a new transaction
2158 redirty_page_for_writepage(wbc, page);
2160 return MPAGE_DA_EXTENT_TAIL;
2163 * Can we merge this page to current extent?
2165 if (mpd->next_page != page->index) {
2167 * Nope, we can't. So, we map non-allocated blocks
2168 * and start IO on them using writepage()
2170 if (mpd->next_page != mpd->first_page) {
2171 if (mpage_da_map_blocks(mpd) == 0)
2172 mpage_da_submit_io(mpd);
2174 * skip rest of the page in the page_vec
2177 redirty_page_for_writepage(wbc, page);
2179 return MPAGE_DA_EXTENT_TAIL;
2183 * Start next extent of pages ...
2185 mpd->first_page = page->index;
2195 mpd->next_page = page->index + 1;
2196 logical = (sector_t) page->index <<
2197 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2199 if (!page_has_buffers(page)) {
2200 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2201 (1 << BH_Dirty) | (1 << BH_Uptodate));
2203 return MPAGE_DA_EXTENT_TAIL;
2206 * Page with regular buffer heads, just add all dirty ones
2208 head = page_buffers(page);
2211 BUG_ON(buffer_locked(bh));
2213 * We need to try to allocate
2214 * unmapped blocks in the same page.
2215 * Otherwise we won't make progress
2216 * with the page in ext4_da_writepage
2218 if (buffer_dirty(bh) &&
2219 (!buffer_mapped(bh) || buffer_delay(bh))) {
2220 mpage_add_bh_to_extent(mpd, logical,
2224 return MPAGE_DA_EXTENT_TAIL;
2225 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2227 * mapped dirty buffer. We need to update
2228 * the b_state because we look at
2229 * b_state in mpage_da_map_blocks. We don't
2230 * update b_size because if we find an
2231 * unmapped buffer_head later we need to
2232 * use the b_state flag of that buffer_head.
2234 if (mpd->b_size == 0)
2235 mpd->b_state = bh->b_state & BH_FLAGS;
2238 } while ((bh = bh->b_this_page) != head);
2245 * mpage_da_writepages - walk the list of dirty pages of the given
2246 * address space, allocates non-allocated blocks, maps newly-allocated
2247 * blocks to existing bhs and issue IO them
2249 * @mapping: address space structure to write
2250 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2252 * This is a library function, which implements the writepages()
2253 * address_space_operation.
2255 static int mpage_da_writepages(struct address_space *mapping,
2256 struct writeback_control *wbc,
2257 struct mpage_da_data *mpd)
2264 mpd->first_page = 0;
2267 mpd->pages_written = 0;
2270 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2272 * Handle last extent of pages
2274 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2275 if (mpage_da_map_blocks(mpd) == 0)
2276 mpage_da_submit_io(mpd);
2279 ret = MPAGE_DA_EXTENT_TAIL;
2281 wbc->nr_to_write -= mpd->pages_written;
2286 * this is a special callback for ->write_begin() only
2287 * it's intention is to return mapped block or reserve space
2289 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2290 struct buffer_head *bh_result, int create)
2294 BUG_ON(create == 0);
2295 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2298 * first, we need to know whether the block is allocated already
2299 * preallocated blocks are unmapped but should treated
2300 * the same as allocated blocks.
2302 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2303 if ((ret == 0) && !buffer_delay(bh_result)) {
2304 /* the block isn't (pre)allocated yet, let's reserve space */
2306 * XXX: __block_prepare_write() unmaps passed block,
2309 ret = ext4_da_reserve_space(inode, 1);
2311 /* not enough space to reserve */
2314 map_bh(bh_result, inode->i_sb, 0);
2315 set_buffer_new(bh_result);
2316 set_buffer_delay(bh_result);
2317 } else if (ret > 0) {
2318 bh_result->b_size = (ret << inode->i_blkbits);
2325 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2328 * unmapped buffer is possible for holes.
2329 * delay buffer is possible with delayed allocation
2331 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2334 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2335 struct buffer_head *bh_result, int create)
2338 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2341 * we don't want to do block allocation in writepage
2342 * so call get_block_wrap with create = 0
2344 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2345 bh_result, 0, 0, 0);
2347 bh_result->b_size = (ret << inode->i_blkbits);
2354 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2355 * get called via journal_submit_inode_data_buffers (no journal handle)
2356 * get called via shrink_page_list via pdflush (no journal handle)
2357 * or grab_page_cache when doing write_begin (have journal handle)
2359 static int ext4_da_writepage(struct page *page,
2360 struct writeback_control *wbc)
2365 struct buffer_head *page_bufs;
2366 struct inode *inode = page->mapping->host;
2368 trace_mark(ext4_da_writepage,
2369 "dev %s ino %lu page_index %lu",
2370 inode->i_sb->s_id, inode->i_ino, page->index);
2371 size = i_size_read(inode);
2372 if (page->index == size >> PAGE_CACHE_SHIFT)
2373 len = size & ~PAGE_CACHE_MASK;
2375 len = PAGE_CACHE_SIZE;
2377 if (page_has_buffers(page)) {
2378 page_bufs = page_buffers(page);
2379 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2380 ext4_bh_unmapped_or_delay)) {
2382 * We don't want to do block allocation
2383 * So redirty the page and return
2384 * We may reach here when we do a journal commit
2385 * via journal_submit_inode_data_buffers.
2386 * If we don't have mapping block we just ignore
2387 * them. We can also reach here via shrink_page_list
2389 redirty_page_for_writepage(wbc, page);
2395 * The test for page_has_buffers() is subtle:
2396 * We know the page is dirty but it lost buffers. That means
2397 * that at some moment in time after write_begin()/write_end()
2398 * has been called all buffers have been clean and thus they
2399 * must have been written at least once. So they are all
2400 * mapped and we can happily proceed with mapping them
2401 * and writing the page.
2403 * Try to initialize the buffer_heads and check whether
2404 * all are mapped and non delay. We don't want to
2405 * do block allocation here.
2407 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2408 ext4_normal_get_block_write);
2410 page_bufs = page_buffers(page);
2411 /* check whether all are mapped and non delay */
2412 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2413 ext4_bh_unmapped_or_delay)) {
2414 redirty_page_for_writepage(wbc, page);
2420 * We can't do block allocation here
2421 * so just redity the page and unlock
2424 redirty_page_for_writepage(wbc, page);
2428 /* now mark the buffer_heads as dirty and uptodate */
2429 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2432 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2433 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2435 ret = block_write_full_page(page,
2436 ext4_normal_get_block_write,
2443 * This is called via ext4_da_writepages() to
2444 * calulate the total number of credits to reserve to fit
2445 * a single extent allocation into a single transaction,
2446 * ext4_da_writpeages() will loop calling this before
2447 * the block allocation.
2450 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2452 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2455 * With non-extent format the journal credit needed to
2456 * insert nrblocks contiguous block is dependent on
2457 * number of contiguous block. So we will limit
2458 * number of contiguous block to a sane value
2460 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2461 (max_blocks > EXT4_MAX_TRANS_DATA))
2462 max_blocks = EXT4_MAX_TRANS_DATA;
2464 return ext4_chunk_trans_blocks(inode, max_blocks);
2467 static int ext4_da_writepages(struct address_space *mapping,
2468 struct writeback_control *wbc)
2471 int range_whole = 0;
2472 handle_t *handle = NULL;
2473 struct mpage_da_data mpd;
2474 struct inode *inode = mapping->host;
2475 int no_nrwrite_index_update;
2476 int pages_written = 0;
2478 int range_cyclic, cycled = 1, io_done = 0;
2479 int needed_blocks, ret = 0, nr_to_writebump = 0;
2480 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2482 trace_mark(ext4_da_writepages,
2483 "dev %s ino %lu nr_t_write %ld "
2484 "pages_skipped %ld range_start %llu "
2485 "range_end %llu nonblocking %d "
2486 "for_kupdate %d for_reclaim %d "
2487 "for_writepages %d range_cyclic %d",
2488 inode->i_sb->s_id, inode->i_ino,
2489 wbc->nr_to_write, wbc->pages_skipped,
2490 (unsigned long long) wbc->range_start,
2491 (unsigned long long) wbc->range_end,
2492 wbc->nonblocking, wbc->for_kupdate,
2493 wbc->for_reclaim, wbc->for_writepages,
2497 * No pages to write? This is mainly a kludge to avoid starting
2498 * a transaction for special inodes like journal inode on last iput()
2499 * because that could violate lock ordering on umount
2501 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2505 * If the filesystem has aborted, it is read-only, so return
2506 * right away instead of dumping stack traces later on that
2507 * will obscure the real source of the problem. We test
2508 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2509 * the latter could be true if the filesystem is mounted
2510 * read-only, and in that case, ext4_da_writepages should
2511 * *never* be called, so if that ever happens, we would want
2514 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2518 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2519 * This make sure small files blocks are allocated in
2520 * single attempt. This ensure that small files
2521 * get less fragmented.
2523 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2524 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2525 wbc->nr_to_write = sbi->s_mb_stream_request;
2527 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2530 range_cyclic = wbc->range_cyclic;
2531 if (wbc->range_cyclic) {
2532 index = mapping->writeback_index;
2535 wbc->range_start = index << PAGE_CACHE_SHIFT;
2536 wbc->range_end = LLONG_MAX;
2537 wbc->range_cyclic = 0;
2539 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2542 mpd.inode = mapping->host;
2545 * we don't want write_cache_pages to update
2546 * nr_to_write and writeback_index
2548 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2549 wbc->no_nrwrite_index_update = 1;
2550 pages_skipped = wbc->pages_skipped;
2553 while (!ret && wbc->nr_to_write > 0) {
2556 * we insert one extent at a time. So we need
2557 * credit needed for single extent allocation.
2558 * journalled mode is currently not supported
2561 BUG_ON(ext4_should_journal_data(inode));
2562 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2564 /* start a new transaction*/
2565 handle = ext4_journal_start(inode, needed_blocks);
2566 if (IS_ERR(handle)) {
2567 ret = PTR_ERR(handle);
2568 printk(KERN_CRIT "%s: jbd2_start: "
2569 "%ld pages, ino %lu; err %d\n", __func__,
2570 wbc->nr_to_write, inode->i_ino, ret);
2572 goto out_writepages;
2574 ret = mpage_da_writepages(mapping, wbc, &mpd);
2576 ext4_journal_stop(handle);
2578 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2579 /* commit the transaction which would
2580 * free blocks released in the transaction
2583 jbd2_journal_force_commit_nested(sbi->s_journal);
2584 wbc->pages_skipped = pages_skipped;
2586 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2588 * got one extent now try with
2591 pages_written += mpd.pages_written;
2592 wbc->pages_skipped = pages_skipped;
2595 } else if (wbc->nr_to_write)
2597 * There is no more writeout needed
2598 * or we requested for a noblocking writeout
2599 * and we found the device congested
2603 if (!io_done && !cycled) {
2606 wbc->range_start = index << PAGE_CACHE_SHIFT;
2607 wbc->range_end = mapping->writeback_index - 1;
2610 if (pages_skipped != wbc->pages_skipped)
2611 printk(KERN_EMERG "This should not happen leaving %s "
2612 "with nr_to_write = %ld ret = %d\n",
2613 __func__, wbc->nr_to_write, ret);
2616 index += pages_written;
2617 wbc->range_cyclic = range_cyclic;
2618 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2620 * set the writeback_index so that range_cyclic
2621 * mode will write it back later
2623 mapping->writeback_index = index;
2626 if (!no_nrwrite_index_update)
2627 wbc->no_nrwrite_index_update = 0;
2628 wbc->nr_to_write -= nr_to_writebump;
2629 trace_mark(ext4_da_writepage_result,
2630 "dev %s ino %lu ret %d pages_written %d "
2631 "pages_skipped %ld congestion %d "
2632 "more_io %d no_nrwrite_index_update %d",
2633 inode->i_sb->s_id, inode->i_ino, ret,
2634 pages_written, wbc->pages_skipped,
2635 wbc->encountered_congestion, wbc->more_io,
2636 wbc->no_nrwrite_index_update);
2640 #define FALL_BACK_TO_NONDELALLOC 1
2641 static int ext4_nonda_switch(struct super_block *sb)
2643 s64 free_blocks, dirty_blocks;
2644 struct ext4_sb_info *sbi = EXT4_SB(sb);
2647 * switch to non delalloc mode if we are running low
2648 * on free block. The free block accounting via percpu
2649 * counters can get slightly wrong with percpu_counter_batch getting
2650 * accumulated on each CPU without updating global counters
2651 * Delalloc need an accurate free block accounting. So switch
2652 * to non delalloc when we are near to error range.
2654 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2655 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2656 if (2 * free_blocks < 3 * dirty_blocks ||
2657 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2659 * free block count is less that 150% of dirty blocks
2660 * or free blocks is less that watermark
2667 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2668 loff_t pos, unsigned len, unsigned flags,
2669 struct page **pagep, void **fsdata)
2671 int ret, retries = 0;
2675 struct inode *inode = mapping->host;
2678 index = pos >> PAGE_CACHE_SHIFT;
2679 from = pos & (PAGE_CACHE_SIZE - 1);
2682 if (ext4_nonda_switch(inode->i_sb)) {
2683 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2684 return ext4_write_begin(file, mapping, pos,
2685 len, flags, pagep, fsdata);
2687 *fsdata = (void *)0;
2689 trace_mark(ext4_da_write_begin,
2690 "dev %s ino %lu pos %llu len %u flags %u",
2691 inode->i_sb->s_id, inode->i_ino,
2692 (unsigned long long) pos, len, flags);
2695 * With delayed allocation, we don't log the i_disksize update
2696 * if there is delayed block allocation. But we still need
2697 * to journalling the i_disksize update if writes to the end
2698 * of file which has an already mapped buffer.
2700 handle = ext4_journal_start(inode, 1);
2701 if (IS_ERR(handle)) {
2702 ret = PTR_ERR(handle);
2705 /* We cannot recurse into the filesystem as the transaction is already
2707 flags |= AOP_FLAG_NOFS;
2709 page = grab_cache_page_write_begin(mapping, index, flags);
2711 ext4_journal_stop(handle);
2717 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2718 ext4_da_get_block_prep);
2721 ext4_journal_stop(handle);
2722 page_cache_release(page);
2724 * block_write_begin may have instantiated a few blocks
2725 * outside i_size. Trim these off again. Don't need
2726 * i_size_read because we hold i_mutex.
2728 if (pos + len > inode->i_size)
2729 vmtruncate(inode, inode->i_size);
2732 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2739 * Check if we should update i_disksize
2740 * when write to the end of file but not require block allocation
2742 static int ext4_da_should_update_i_disksize(struct page *page,
2743 unsigned long offset)
2745 struct buffer_head *bh;
2746 struct inode *inode = page->mapping->host;
2750 bh = page_buffers(page);
2751 idx = offset >> inode->i_blkbits;
2753 for (i = 0; i < idx; i++)
2754 bh = bh->b_this_page;
2756 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2761 static int ext4_da_write_end(struct file *file,
2762 struct address_space *mapping,
2763 loff_t pos, unsigned len, unsigned copied,
2764 struct page *page, void *fsdata)
2766 struct inode *inode = mapping->host;
2768 handle_t *handle = ext4_journal_current_handle();
2770 unsigned long start, end;
2771 int write_mode = (int)(unsigned long)fsdata;
2773 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2774 if (ext4_should_order_data(inode)) {
2775 return ext4_ordered_write_end(file, mapping, pos,
2776 len, copied, page, fsdata);
2777 } else if (ext4_should_writeback_data(inode)) {
2778 return ext4_writeback_write_end(file, mapping, pos,
2779 len, copied, page, fsdata);
2785 trace_mark(ext4_da_write_end,
2786 "dev %s ino %lu pos %llu len %u copied %u",
2787 inode->i_sb->s_id, inode->i_ino,
2788 (unsigned long long) pos, len, copied);
2789 start = pos & (PAGE_CACHE_SIZE - 1);
2790 end = start + copied - 1;
2793 * generic_write_end() will run mark_inode_dirty() if i_size
2794 * changes. So let's piggyback the i_disksize mark_inode_dirty
2798 new_i_size = pos + copied;
2799 if (new_i_size > EXT4_I(inode)->i_disksize) {
2800 if (ext4_da_should_update_i_disksize(page, end)) {
2801 down_write(&EXT4_I(inode)->i_data_sem);
2802 if (new_i_size > EXT4_I(inode)->i_disksize) {
2804 * Updating i_disksize when extending file
2805 * without needing block allocation
2807 if (ext4_should_order_data(inode))
2808 ret = ext4_jbd2_file_inode(handle,
2811 EXT4_I(inode)->i_disksize = new_i_size;
2813 up_write(&EXT4_I(inode)->i_data_sem);
2814 /* We need to mark inode dirty even if
2815 * new_i_size is less that inode->i_size
2816 * bu greater than i_disksize.(hint delalloc)
2818 ext4_mark_inode_dirty(handle, inode);
2821 ret2 = generic_write_end(file, mapping, pos, len, copied,
2826 ret2 = ext4_journal_stop(handle);
2830 return ret ? ret : copied;
2833 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2836 * Drop reserved blocks
2838 BUG_ON(!PageLocked(page));
2839 if (!page_has_buffers(page))
2842 ext4_da_page_release_reservation(page, offset);
2845 ext4_invalidatepage(page, offset);
2852 * bmap() is special. It gets used by applications such as lilo and by
2853 * the swapper to find the on-disk block of a specific piece of data.
2855 * Naturally, this is dangerous if the block concerned is still in the
2856 * journal. If somebody makes a swapfile on an ext4 data-journaling
2857 * filesystem and enables swap, then they may get a nasty shock when the
2858 * data getting swapped to that swapfile suddenly gets overwritten by
2859 * the original zero's written out previously to the journal and
2860 * awaiting writeback in the kernel's buffer cache.
2862 * So, if we see any bmap calls here on a modified, data-journaled file,
2863 * take extra steps to flush any blocks which might be in the cache.
2865 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2867 struct inode *inode = mapping->host;
2871 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2872 test_opt(inode->i_sb, DELALLOC)) {
2874 * With delalloc we want to sync the file
2875 * so that we can make sure we allocate
2878 filemap_write_and_wait(mapping);
2881 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2883 * This is a REALLY heavyweight approach, but the use of
2884 * bmap on dirty files is expected to be extremely rare:
2885 * only if we run lilo or swapon on a freshly made file
2886 * do we expect this to happen.
2888 * (bmap requires CAP_SYS_RAWIO so this does not
2889 * represent an unprivileged user DOS attack --- we'd be
2890 * in trouble if mortal users could trigger this path at
2893 * NB. EXT4_STATE_JDATA is not set on files other than
2894 * regular files. If somebody wants to bmap a directory
2895 * or symlink and gets confused because the buffer
2896 * hasn't yet been flushed to disk, they deserve
2897 * everything they get.
2900 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2901 journal = EXT4_JOURNAL(inode);
2902 jbd2_journal_lock_updates(journal);
2903 err = jbd2_journal_flush(journal);
2904 jbd2_journal_unlock_updates(journal);
2910 return generic_block_bmap(mapping, block, ext4_get_block);
2913 static int bget_one(handle_t *handle, struct buffer_head *bh)
2919 static int bput_one(handle_t *handle, struct buffer_head *bh)
2926 * Note that we don't need to start a transaction unless we're journaling data
2927 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2928 * need to file the inode to the transaction's list in ordered mode because if
2929 * we are writing back data added by write(), the inode is already there and if
2930 * we are writing back data modified via mmap(), noone guarantees in which
2931 * transaction the data will hit the disk. In case we are journaling data, we
2932 * cannot start transaction directly because transaction start ranks above page
2933 * lock so we have to do some magic.
2935 * In all journaling modes block_write_full_page() will start the I/O.
2939 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2944 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2946 * Same applies to ext4_get_block(). We will deadlock on various things like
2947 * lock_journal and i_data_sem
2949 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2952 * 16May01: If we're reentered then journal_current_handle() will be
2953 * non-zero. We simply *return*.
2955 * 1 July 2001: @@@ FIXME:
2956 * In journalled data mode, a data buffer may be metadata against the
2957 * current transaction. But the same file is part of a shared mapping
2958 * and someone does a writepage() on it.
2960 * We will move the buffer onto the async_data list, but *after* it has
2961 * been dirtied. So there's a small window where we have dirty data on
2964 * Note that this only applies to the last partial page in the file. The
2965 * bit which block_write_full_page() uses prepare/commit for. (That's
2966 * broken code anyway: it's wrong for msync()).
2968 * It's a rare case: affects the final partial page, for journalled data
2969 * where the file is subject to bith write() and writepage() in the same
2970 * transction. To fix it we'll need a custom block_write_full_page().
2971 * We'll probably need that anyway for journalling writepage() output.
2973 * We don't honour synchronous mounts for writepage(). That would be
2974 * disastrous. Any write() or metadata operation will sync the fs for
2978 static int __ext4_normal_writepage(struct page *page,
2979 struct writeback_control *wbc)
2981 struct inode *inode = page->mapping->host;
2983 if (test_opt(inode->i_sb, NOBH))
2984 return nobh_writepage(page,
2985 ext4_normal_get_block_write, wbc);
2987 return block_write_full_page(page,
2988 ext4_normal_get_block_write,
2992 static int ext4_normal_writepage(struct page *page,
2993 struct writeback_control *wbc)
2995 struct inode *inode = page->mapping->host;
2996 loff_t size = i_size_read(inode);
2999 trace_mark(ext4_normal_writepage,
3000 "dev %s ino %lu page_index %lu",
3001 inode->i_sb->s_id, inode->i_ino, page->index);
3002 J_ASSERT(PageLocked(page));
3003 if (page->index == size >> PAGE_CACHE_SHIFT)
3004 len = size & ~PAGE_CACHE_MASK;
3006 len = PAGE_CACHE_SIZE;
3008 if (page_has_buffers(page)) {
3009 /* if page has buffers it should all be mapped
3010 * and allocated. If there are not buffers attached
3011 * to the page we know the page is dirty but it lost
3012 * buffers. That means that at some moment in time
3013 * after write_begin() / write_end() has been called
3014 * all buffers have been clean and thus they must have been
3015 * written at least once. So they are all mapped and we can
3016 * happily proceed with mapping them and writing the page.
3018 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3019 ext4_bh_unmapped_or_delay));
3022 if (!ext4_journal_current_handle())
3023 return __ext4_normal_writepage(page, wbc);
3025 redirty_page_for_writepage(wbc, page);
3030 static int __ext4_journalled_writepage(struct page *page,
3031 struct writeback_control *wbc)
3033 struct address_space *mapping = page->mapping;
3034 struct inode *inode = mapping->host;
3035 struct buffer_head *page_bufs;
3036 handle_t *handle = NULL;
3040 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3041 ext4_normal_get_block_write);
3045 page_bufs = page_buffers(page);
3046 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3048 /* As soon as we unlock the page, it can go away, but we have
3049 * references to buffers so we are safe */
3052 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3053 if (IS_ERR(handle)) {
3054 ret = PTR_ERR(handle);
3058 ret = walk_page_buffers(handle, page_bufs, 0,
3059 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3061 err = walk_page_buffers(handle, page_bufs, 0,
3062 PAGE_CACHE_SIZE, NULL, write_end_fn);
3065 err = ext4_journal_stop(handle);
3069 walk_page_buffers(handle, page_bufs, 0,
3070 PAGE_CACHE_SIZE, NULL, bput_one);
3071 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3080 static int ext4_journalled_writepage(struct page *page,
3081 struct writeback_control *wbc)
3083 struct inode *inode = page->mapping->host;
3084 loff_t size = i_size_read(inode);
3087 trace_mark(ext4_journalled_writepage,
3088 "dev %s ino %lu page_index %lu",
3089 inode->i_sb->s_id, inode->i_ino, page->index);
3090 J_ASSERT(PageLocked(page));
3091 if (page->index == size >> PAGE_CACHE_SHIFT)
3092 len = size & ~PAGE_CACHE_MASK;
3094 len = PAGE_CACHE_SIZE;
3096 if (page_has_buffers(page)) {
3097 /* if page has buffers it should all be mapped
3098 * and allocated. If there are not buffers attached
3099 * to the page we know the page is dirty but it lost
3100 * buffers. That means that at some moment in time
3101 * after write_begin() / write_end() has been called
3102 * all buffers have been clean and thus they must have been
3103 * written at least once. So they are all mapped and we can
3104 * happily proceed with mapping them and writing the page.
3106 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3107 ext4_bh_unmapped_or_delay));
3110 if (ext4_journal_current_handle())
3113 if (PageChecked(page)) {
3115 * It's mmapped pagecache. Add buffers and journal it. There
3116 * doesn't seem much point in redirtying the page here.
3118 ClearPageChecked(page);
3119 return __ext4_journalled_writepage(page, wbc);
3122 * It may be a page full of checkpoint-mode buffers. We don't
3123 * really know unless we go poke around in the buffer_heads.
3124 * But block_write_full_page will do the right thing.
3126 return block_write_full_page(page,
3127 ext4_normal_get_block_write,
3131 redirty_page_for_writepage(wbc, page);
3136 static int ext4_readpage(struct file *file, struct page *page)
3138 return mpage_readpage(page, ext4_get_block);
3142 ext4_readpages(struct file *file, struct address_space *mapping,
3143 struct list_head *pages, unsigned nr_pages)
3145 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3148 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3150 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3153 * If it's a full truncate we just forget about the pending dirtying
3156 ClearPageChecked(page);
3159 jbd2_journal_invalidatepage(journal, page, offset);
3161 block_invalidatepage(page, offset);
3164 static int ext4_releasepage(struct page *page, gfp_t wait)
3166 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3168 WARN_ON(PageChecked(page));
3169 if (!page_has_buffers(page))
3172 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3174 return try_to_free_buffers(page);
3178 * If the O_DIRECT write will extend the file then add this inode to the
3179 * orphan list. So recovery will truncate it back to the original size
3180 * if the machine crashes during the write.
3182 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3183 * crashes then stale disk data _may_ be exposed inside the file. But current
3184 * VFS code falls back into buffered path in that case so we are safe.
3186 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3187 const struct iovec *iov, loff_t offset,
3188 unsigned long nr_segs)
3190 struct file *file = iocb->ki_filp;
3191 struct inode *inode = file->f_mapping->host;
3192 struct ext4_inode_info *ei = EXT4_I(inode);
3196 size_t count = iov_length(iov, nr_segs);
3199 loff_t final_size = offset + count;
3201 if (final_size > inode->i_size) {
3202 /* Credits for sb + inode write */
3203 handle = ext4_journal_start(inode, 2);
3204 if (IS_ERR(handle)) {
3205 ret = PTR_ERR(handle);
3208 ret = ext4_orphan_add(handle, inode);
3210 ext4_journal_stop(handle);
3214 ei->i_disksize = inode->i_size;
3215 ext4_journal_stop(handle);
3219 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3221 ext4_get_block, NULL);
3226 /* Credits for sb + inode write */
3227 handle = ext4_journal_start(inode, 2);
3228 if (IS_ERR(handle)) {
3229 /* This is really bad luck. We've written the data
3230 * but cannot extend i_size. Bail out and pretend
3231 * the write failed... */
3232 ret = PTR_ERR(handle);
3236 ext4_orphan_del(handle, inode);
3238 loff_t end = offset + ret;
3239 if (end > inode->i_size) {
3240 ei->i_disksize = end;
3241 i_size_write(inode, end);
3243 * We're going to return a positive `ret'
3244 * here due to non-zero-length I/O, so there's
3245 * no way of reporting error returns from
3246 * ext4_mark_inode_dirty() to userspace. So
3249 ext4_mark_inode_dirty(handle, inode);
3252 err = ext4_journal_stop(handle);
3261 * Pages can be marked dirty completely asynchronously from ext4's journalling
3262 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3263 * much here because ->set_page_dirty is called under VFS locks. The page is
3264 * not necessarily locked.
3266 * We cannot just dirty the page and leave attached buffers clean, because the
3267 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3268 * or jbddirty because all the journalling code will explode.
3270 * So what we do is to mark the page "pending dirty" and next time writepage
3271 * is called, propagate that into the buffers appropriately.
3273 static int ext4_journalled_set_page_dirty(struct page *page)
3275 SetPageChecked(page);
3276 return __set_page_dirty_nobuffers(page);
3279 static const struct address_space_operations ext4_ordered_aops = {
3280 .readpage = ext4_readpage,
3281 .readpages = ext4_readpages,
3282 .writepage = ext4_normal_writepage,
3283 .sync_page = block_sync_page,
3284 .write_begin = ext4_write_begin,
3285 .write_end = ext4_ordered_write_end,
3287 .invalidatepage = ext4_invalidatepage,
3288 .releasepage = ext4_releasepage,
3289 .direct_IO = ext4_direct_IO,
3290 .migratepage = buffer_migrate_page,
3291 .is_partially_uptodate = block_is_partially_uptodate,
3294 static const struct address_space_operations ext4_writeback_aops = {
3295 .readpage = ext4_readpage,
3296 .readpages = ext4_readpages,
3297 .writepage = ext4_normal_writepage,
3298 .sync_page = block_sync_page,
3299 .write_begin = ext4_write_begin,
3300 .write_end = ext4_writeback_write_end,
3302 .invalidatepage = ext4_invalidatepage,
3303 .releasepage = ext4_releasepage,
3304 .direct_IO = ext4_direct_IO,
3305 .migratepage = buffer_migrate_page,
3306 .is_partially_uptodate = block_is_partially_uptodate,
3309 static const struct address_space_operations ext4_journalled_aops = {
3310 .readpage = ext4_readpage,
3311 .readpages = ext4_readpages,
3312 .writepage = ext4_journalled_writepage,
3313 .sync_page = block_sync_page,
3314 .write_begin = ext4_write_begin,
3315 .write_end = ext4_journalled_write_end,
3316 .set_page_dirty = ext4_journalled_set_page_dirty,
3318 .invalidatepage = ext4_invalidatepage,
3319 .releasepage = ext4_releasepage,
3320 .is_partially_uptodate = block_is_partially_uptodate,
3323 static const struct address_space_operations ext4_da_aops = {
3324 .readpage = ext4_readpage,
3325 .readpages = ext4_readpages,
3326 .writepage = ext4_da_writepage,
3327 .writepages = ext4_da_writepages,
3328 .sync_page = block_sync_page,
3329 .write_begin = ext4_da_write_begin,
3330 .write_end = ext4_da_write_end,
3332 .invalidatepage = ext4_da_invalidatepage,
3333 .releasepage = ext4_releasepage,
3334 .direct_IO = ext4_direct_IO,
3335 .migratepage = buffer_migrate_page,
3336 .is_partially_uptodate = block_is_partially_uptodate,
3339 void ext4_set_aops(struct inode *inode)
3341 if (ext4_should_order_data(inode) &&
3342 test_opt(inode->i_sb, DELALLOC))
3343 inode->i_mapping->a_ops = &ext4_da_aops;
3344 else if (ext4_should_order_data(inode))
3345 inode->i_mapping->a_ops = &ext4_ordered_aops;
3346 else if (ext4_should_writeback_data(inode) &&
3347 test_opt(inode->i_sb, DELALLOC))
3348 inode->i_mapping->a_ops = &ext4_da_aops;
3349 else if (ext4_should_writeback_data(inode))
3350 inode->i_mapping->a_ops = &ext4_writeback_aops;
3352 inode->i_mapping->a_ops = &ext4_journalled_aops;
3356 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3357 * up to the end of the block which corresponds to `from'.
3358 * This required during truncate. We need to physically zero the tail end
3359 * of that block so it doesn't yield old data if the file is later grown.
3361 int ext4_block_truncate_page(handle_t *handle,
3362 struct address_space *mapping, loff_t from)
3364 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3365 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3366 unsigned blocksize, length, pos;
3368 struct inode *inode = mapping->host;
3369 struct buffer_head *bh;
3373 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3377 blocksize = inode->i_sb->s_blocksize;
3378 length = blocksize - (offset & (blocksize - 1));
3379 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3382 * For "nobh" option, we can only work if we don't need to
3383 * read-in the page - otherwise we create buffers to do the IO.
3385 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3386 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3387 zero_user(page, offset, length);
3388 set_page_dirty(page);
3392 if (!page_has_buffers(page))
3393 create_empty_buffers(page, blocksize, 0);
3395 /* Find the buffer that contains "offset" */
3396 bh = page_buffers(page);
3398 while (offset >= pos) {
3399 bh = bh->b_this_page;
3405 if (buffer_freed(bh)) {
3406 BUFFER_TRACE(bh, "freed: skip");
3410 if (!buffer_mapped(bh)) {
3411 BUFFER_TRACE(bh, "unmapped");
3412 ext4_get_block(inode, iblock, bh, 0);
3413 /* unmapped? It's a hole - nothing to do */
3414 if (!buffer_mapped(bh)) {
3415 BUFFER_TRACE(bh, "still unmapped");
3420 /* Ok, it's mapped. Make sure it's up-to-date */
3421 if (PageUptodate(page))
3422 set_buffer_uptodate(bh);
3424 if (!buffer_uptodate(bh)) {
3426 ll_rw_block(READ, 1, &bh);
3428 /* Uhhuh. Read error. Complain and punt. */
3429 if (!buffer_uptodate(bh))
3433 if (ext4_should_journal_data(inode)) {
3434 BUFFER_TRACE(bh, "get write access");
3435 err = ext4_journal_get_write_access(handle, bh);
3440 zero_user(page, offset, length);
3442 BUFFER_TRACE(bh, "zeroed end of block");
3445 if (ext4_should_journal_data(inode)) {
3446 err = ext4_handle_dirty_metadata(handle, inode, bh);
3448 if (ext4_should_order_data(inode))
3449 err = ext4_jbd2_file_inode(handle, inode);
3450 mark_buffer_dirty(bh);
3455 page_cache_release(page);
3460 * Probably it should be a library function... search for first non-zero word
3461 * or memcmp with zero_page, whatever is better for particular architecture.
3464 static inline int all_zeroes(__le32 *p, __le32 *q)
3473 * ext4_find_shared - find the indirect blocks for partial truncation.
3474 * @inode: inode in question
3475 * @depth: depth of the affected branch
3476 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3477 * @chain: place to store the pointers to partial indirect blocks
3478 * @top: place to the (detached) top of branch
3480 * This is a helper function used by ext4_truncate().
3482 * When we do truncate() we may have to clean the ends of several
3483 * indirect blocks but leave the blocks themselves alive. Block is
3484 * partially truncated if some data below the new i_size is refered
3485 * from it (and it is on the path to the first completely truncated
3486 * data block, indeed). We have to free the top of that path along
3487 * with everything to the right of the path. Since no allocation
3488 * past the truncation point is possible until ext4_truncate()
3489 * finishes, we may safely do the latter, but top of branch may
3490 * require special attention - pageout below the truncation point
3491 * might try to populate it.
3493 * We atomically detach the top of branch from the tree, store the
3494 * block number of its root in *@top, pointers to buffer_heads of
3495 * partially truncated blocks - in @chain[].bh and pointers to
3496 * their last elements that should not be removed - in
3497 * @chain[].p. Return value is the pointer to last filled element
3500 * The work left to caller to do the actual freeing of subtrees:
3501 * a) free the subtree starting from *@top
3502 * b) free the subtrees whose roots are stored in
3503 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3504 * c) free the subtrees growing from the inode past the @chain[0].
3505 * (no partially truncated stuff there). */
3507 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3508 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3510 Indirect *partial, *p;
3514 /* Make k index the deepest non-null offest + 1 */
3515 for (k = depth; k > 1 && !offsets[k-1]; k--)
3517 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3518 /* Writer: pointers */
3520 partial = chain + k-1;
3522 * If the branch acquired continuation since we've looked at it -
3523 * fine, it should all survive and (new) top doesn't belong to us.
3525 if (!partial->key && *partial->p)
3528 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3531 * OK, we've found the last block that must survive. The rest of our
3532 * branch should be detached before unlocking. However, if that rest
3533 * of branch is all ours and does not grow immediately from the inode
3534 * it's easier to cheat and just decrement partial->p.
3536 if (p == chain + k - 1 && p > chain) {
3540 /* Nope, don't do this in ext4. Must leave the tree intact */
3547 while (partial > p) {
3548 brelse(partial->bh);
3556 * Zero a number of block pointers in either an inode or an indirect block.
3557 * If we restart the transaction we must again get write access to the
3558 * indirect block for further modification.
3560 * We release `count' blocks on disk, but (last - first) may be greater
3561 * than `count' because there can be holes in there.
3563 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3564 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3565 unsigned long count, __le32 *first, __le32 *last)
3568 if (try_to_extend_transaction(handle, inode)) {
3570 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3571 ext4_handle_dirty_metadata(handle, inode, bh);
3573 ext4_mark_inode_dirty(handle, inode);
3574 ext4_journal_test_restart(handle, inode);
3576 BUFFER_TRACE(bh, "retaking write access");
3577 ext4_journal_get_write_access(handle, bh);
3582 * Any buffers which are on the journal will be in memory. We find
3583 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3584 * on them. We've already detached each block from the file, so
3585 * bforget() in jbd2_journal_forget() should be safe.
3587 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3589 for (p = first; p < last; p++) {
3590 u32 nr = le32_to_cpu(*p);
3592 struct buffer_head *tbh;
3595 tbh = sb_find_get_block(inode->i_sb, nr);
3596 ext4_forget(handle, 0, inode, tbh, nr);
3600 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3604 * ext4_free_data - free a list of data blocks
3605 * @handle: handle for this transaction
3606 * @inode: inode we are dealing with
3607 * @this_bh: indirect buffer_head which contains *@first and *@last
3608 * @first: array of block numbers
3609 * @last: points immediately past the end of array
3611 * We are freeing all blocks refered from that array (numbers are stored as
3612 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3614 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3615 * blocks are contiguous then releasing them at one time will only affect one
3616 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3617 * actually use a lot of journal space.
3619 * @this_bh will be %NULL if @first and @last point into the inode's direct
3622 static void ext4_free_data(handle_t *handle, struct inode *inode,
3623 struct buffer_head *this_bh,
3624 __le32 *first, __le32 *last)
3626 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3627 unsigned long count = 0; /* Number of blocks in the run */
3628 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3631 ext4_fsblk_t nr; /* Current block # */
3632 __le32 *p; /* Pointer into inode/ind
3633 for current block */
3636 if (this_bh) { /* For indirect block */
3637 BUFFER_TRACE(this_bh, "get_write_access");
3638 err = ext4_journal_get_write_access(handle, this_bh);
3639 /* Important: if we can't update the indirect pointers
3640 * to the blocks, we can't free them. */
3645 for (p = first; p < last; p++) {
3646 nr = le32_to_cpu(*p);
3648 /* accumulate blocks to free if they're contiguous */
3651 block_to_free_p = p;
3653 } else if (nr == block_to_free + count) {
3656 ext4_clear_blocks(handle, inode, this_bh,
3658 count, block_to_free_p, p);
3660 block_to_free_p = p;
3667 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3668 count, block_to_free_p, p);
3671 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3674 * The buffer head should have an attached journal head at this
3675 * point. However, if the data is corrupted and an indirect
3676 * block pointed to itself, it would have been detached when
3677 * the block was cleared. Check for this instead of OOPSing.
3679 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3680 ext4_handle_dirty_metadata(handle, inode, this_bh);
3682 ext4_error(inode->i_sb, __func__,
3683 "circular indirect block detected, "
3684 "inode=%lu, block=%llu",
3686 (unsigned long long) this_bh->b_blocknr);
3691 * ext4_free_branches - free an array of branches
3692 * @handle: JBD handle for this transaction
3693 * @inode: inode we are dealing with
3694 * @parent_bh: the buffer_head which contains *@first and *@last
3695 * @first: array of block numbers
3696 * @last: pointer immediately past the end of array
3697 * @depth: depth of the branches to free
3699 * We are freeing all blocks refered from these branches (numbers are
3700 * stored as little-endian 32-bit) and updating @inode->i_blocks
3703 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3704 struct buffer_head *parent_bh,
3705 __le32 *first, __le32 *last, int depth)
3710 if (ext4_handle_is_aborted(handle))
3714 struct buffer_head *bh;
3715 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3717 while (--p >= first) {
3718 nr = le32_to_cpu(*p);
3720 continue; /* A hole */
3722 /* Go read the buffer for the next level down */
3723 bh = sb_bread(inode->i_sb, nr);
3726 * A read failure? Report error and clear slot
3730 ext4_error(inode->i_sb, "ext4_free_branches",
3731 "Read failure, inode=%lu, block=%llu",
3736 /* This zaps the entire block. Bottom up. */
3737 BUFFER_TRACE(bh, "free child branches");
3738 ext4_free_branches(handle, inode, bh,
3739 (__le32 *) bh->b_data,
3740 (__le32 *) bh->b_data + addr_per_block,
3744 * We've probably journalled the indirect block several
3745 * times during the truncate. But it's no longer
3746 * needed and we now drop it from the transaction via
3747 * jbd2_journal_revoke().
3749 * That's easy if it's exclusively part of this
3750 * transaction. But if it's part of the committing
3751 * transaction then jbd2_journal_forget() will simply
3752 * brelse() it. That means that if the underlying
3753 * block is reallocated in ext4_get_block(),
3754 * unmap_underlying_metadata() will find this block
3755 * and will try to get rid of it. damn, damn.
3757 * If this block has already been committed to the
3758 * journal, a revoke record will be written. And
3759 * revoke records must be emitted *before* clearing
3760 * this block's bit in the bitmaps.
3762 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3765 * Everything below this this pointer has been
3766 * released. Now let this top-of-subtree go.
3768 * We want the freeing of this indirect block to be
3769 * atomic in the journal with the updating of the
3770 * bitmap block which owns it. So make some room in
3773 * We zero the parent pointer *after* freeing its
3774 * pointee in the bitmaps, so if extend_transaction()
3775 * for some reason fails to put the bitmap changes and
3776 * the release into the same transaction, recovery
3777 * will merely complain about releasing a free block,
3778 * rather than leaking blocks.
3780 if (ext4_handle_is_aborted(handle))
3782 if (try_to_extend_transaction(handle, inode)) {
3783 ext4_mark_inode_dirty(handle, inode);
3784 ext4_journal_test_restart(handle, inode);
3787 ext4_free_blocks(handle, inode, nr, 1, 1);
3791 * The block which we have just freed is
3792 * pointed to by an indirect block: journal it
3794 BUFFER_TRACE(parent_bh, "get_write_access");
3795 if (!ext4_journal_get_write_access(handle,
3798 BUFFER_TRACE(parent_bh,
3799 "call ext4_handle_dirty_metadata");
3800 ext4_handle_dirty_metadata(handle,
3807 /* We have reached the bottom of the tree. */
3808 BUFFER_TRACE(parent_bh, "free data blocks");
3809 ext4_free_data(handle, inode, parent_bh, first, last);
3813 int ext4_can_truncate(struct inode *inode)
3815 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3817 if (S_ISREG(inode->i_mode))
3819 if (S_ISDIR(inode->i_mode))
3821 if (S_ISLNK(inode->i_mode))
3822 return !ext4_inode_is_fast_symlink(inode);
3829 * We block out ext4_get_block() block instantiations across the entire
3830 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3831 * simultaneously on behalf of the same inode.
3833 * As we work through the truncate and commmit bits of it to the journal there
3834 * is one core, guiding principle: the file's tree must always be consistent on
3835 * disk. We must be able to restart the truncate after a crash.
3837 * The file's tree may be transiently inconsistent in memory (although it
3838 * probably isn't), but whenever we close off and commit a journal transaction,
3839 * the contents of (the filesystem + the journal) must be consistent and
3840 * restartable. It's pretty simple, really: bottom up, right to left (although
3841 * left-to-right works OK too).
3843 * Note that at recovery time, journal replay occurs *before* the restart of
3844 * truncate against the orphan inode list.
3846 * The committed inode has the new, desired i_size (which is the same as
3847 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3848 * that this inode's truncate did not complete and it will again call
3849 * ext4_truncate() to have another go. So there will be instantiated blocks
3850 * to the right of the truncation point in a crashed ext4 filesystem. But
3851 * that's fine - as long as they are linked from the inode, the post-crash
3852 * ext4_truncate() run will find them and release them.
3854 void ext4_truncate(struct inode *inode)
3857 struct ext4_inode_info *ei = EXT4_I(inode);
3858 __le32 *i_data = ei->i_data;
3859 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3860 struct address_space *mapping = inode->i_mapping;
3861 ext4_lblk_t offsets[4];
3866 ext4_lblk_t last_block;
3867 unsigned blocksize = inode->i_sb->s_blocksize;
3869 if (!ext4_can_truncate(inode))
3872 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3873 ext4_ext_truncate(inode);
3877 handle = start_transaction(inode);
3879 return; /* AKPM: return what? */
3881 last_block = (inode->i_size + blocksize-1)
3882 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3884 if (inode->i_size & (blocksize - 1))
3885 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3888 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3890 goto out_stop; /* error */
3893 * OK. This truncate is going to happen. We add the inode to the
3894 * orphan list, so that if this truncate spans multiple transactions,
3895 * and we crash, we will resume the truncate when the filesystem
3896 * recovers. It also marks the inode dirty, to catch the new size.
3898 * Implication: the file must always be in a sane, consistent
3899 * truncatable state while each transaction commits.
3901 if (ext4_orphan_add(handle, inode))
3905 * From here we block out all ext4_get_block() callers who want to
3906 * modify the block allocation tree.
3908 down_write(&ei->i_data_sem);
3910 ext4_discard_preallocations(inode);
3913 * The orphan list entry will now protect us from any crash which
3914 * occurs before the truncate completes, so it is now safe to propagate
3915 * the new, shorter inode size (held for now in i_size) into the
3916 * on-disk inode. We do this via i_disksize, which is the value which
3917 * ext4 *really* writes onto the disk inode.
3919 ei->i_disksize = inode->i_size;
3921 if (n == 1) { /* direct blocks */
3922 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3923 i_data + EXT4_NDIR_BLOCKS);
3927 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3928 /* Kill the top of shared branch (not detached) */
3930 if (partial == chain) {
3931 /* Shared branch grows from the inode */
3932 ext4_free_branches(handle, inode, NULL,
3933 &nr, &nr+1, (chain+n-1) - partial);
3936 * We mark the inode dirty prior to restart,
3937 * and prior to stop. No need for it here.
3940 /* Shared branch grows from an indirect block */
3941 BUFFER_TRACE(partial->bh, "get_write_access");
3942 ext4_free_branches(handle, inode, partial->bh,
3944 partial->p+1, (chain+n-1) - partial);
3947 /* Clear the ends of indirect blocks on the shared branch */
3948 while (partial > chain) {
3949 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3950 (__le32*)partial->bh->b_data+addr_per_block,
3951 (chain+n-1) - partial);
3952 BUFFER_TRACE(partial->bh, "call brelse");
3953 brelse (partial->bh);
3957 /* Kill the remaining (whole) subtrees */
3958 switch (offsets[0]) {
3960 nr = i_data[EXT4_IND_BLOCK];
3962 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3963 i_data[EXT4_IND_BLOCK] = 0;
3965 case EXT4_IND_BLOCK:
3966 nr = i_data[EXT4_DIND_BLOCK];
3968 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3969 i_data[EXT4_DIND_BLOCK] = 0;
3971 case EXT4_DIND_BLOCK:
3972 nr = i_data[EXT4_TIND_BLOCK];
3974 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3975 i_data[EXT4_TIND_BLOCK] = 0;
3977 case EXT4_TIND_BLOCK:
3981 up_write(&ei->i_data_sem);
3982 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3983 ext4_mark_inode_dirty(handle, inode);
3986 * In a multi-transaction truncate, we only make the final transaction
3990 ext4_handle_sync(handle);
3993 * If this was a simple ftruncate(), and the file will remain alive
3994 * then we need to clear up the orphan record which we created above.
3995 * However, if this was a real unlink then we were called by
3996 * ext4_delete_inode(), and we allow that function to clean up the
3997 * orphan info for us.
4000 ext4_orphan_del(handle, inode);
4002 ext4_journal_stop(handle);
4006 * ext4_get_inode_loc returns with an extra refcount against the inode's
4007 * underlying buffer_head on success. If 'in_mem' is true, we have all
4008 * data in memory that is needed to recreate the on-disk version of this
4011 static int __ext4_get_inode_loc(struct inode *inode,
4012 struct ext4_iloc *iloc, int in_mem)
4014 struct ext4_group_desc *gdp;
4015 struct buffer_head *bh;
4016 struct super_block *sb = inode->i_sb;
4018 int inodes_per_block, inode_offset;
4021 if (!ext4_valid_inum(sb, inode->i_ino))
4024 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4025 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4030 * Figure out the offset within the block group inode table
4032 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4033 inode_offset = ((inode->i_ino - 1) %
4034 EXT4_INODES_PER_GROUP(sb));
4035 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4036 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4038 bh = sb_getblk(sb, block);
4040 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4041 "inode block - inode=%lu, block=%llu",
4042 inode->i_ino, block);
4045 if (!buffer_uptodate(bh)) {
4049 * If the buffer has the write error flag, we have failed
4050 * to write out another inode in the same block. In this
4051 * case, we don't have to read the block because we may
4052 * read the old inode data successfully.
4054 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4055 set_buffer_uptodate(bh);
4057 if (buffer_uptodate(bh)) {
4058 /* someone brought it uptodate while we waited */
4064 * If we have all information of the inode in memory and this
4065 * is the only valid inode in the block, we need not read the
4069 struct buffer_head *bitmap_bh;
4072 start = inode_offset & ~(inodes_per_block - 1);
4074 /* Is the inode bitmap in cache? */
4075 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4080 * If the inode bitmap isn't in cache then the
4081 * optimisation may end up performing two reads instead
4082 * of one, so skip it.
4084 if (!buffer_uptodate(bitmap_bh)) {
4088 for (i = start; i < start + inodes_per_block; i++) {
4089 if (i == inode_offset)
4091 if (ext4_test_bit(i, bitmap_bh->b_data))
4095 if (i == start + inodes_per_block) {
4096 /* all other inodes are free, so skip I/O */
4097 memset(bh->b_data, 0, bh->b_size);
4098 set_buffer_uptodate(bh);
4106 * If we need to do any I/O, try to pre-readahead extra
4107 * blocks from the inode table.
4109 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4110 ext4_fsblk_t b, end, table;
4113 table = ext4_inode_table(sb, gdp);
4114 /* Make sure s_inode_readahead_blks is a power of 2 */
4115 while (EXT4_SB(sb)->s_inode_readahead_blks &
4116 (EXT4_SB(sb)->s_inode_readahead_blks-1))
4117 EXT4_SB(sb)->s_inode_readahead_blks =
4118 (EXT4_SB(sb)->s_inode_readahead_blks &
4119 (EXT4_SB(sb)->s_inode_readahead_blks-1));
4120 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4123 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4124 num = EXT4_INODES_PER_GROUP(sb);
4125 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4126 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4127 num -= ext4_itable_unused_count(sb, gdp);
4128 table += num / inodes_per_block;
4132 sb_breadahead(sb, b++);
4136 * There are other valid inodes in the buffer, this inode
4137 * has in-inode xattrs, or we don't have this inode in memory.
4138 * Read the block from disk.
4141 bh->b_end_io = end_buffer_read_sync;
4142 submit_bh(READ_META, bh);
4144 if (!buffer_uptodate(bh)) {
4145 ext4_error(sb, __func__,
4146 "unable to read inode block - inode=%lu, "
4147 "block=%llu", inode->i_ino, block);
4157 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4159 /* We have all inode data except xattrs in memory here. */
4160 return __ext4_get_inode_loc(inode, iloc,
4161 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4164 void ext4_set_inode_flags(struct inode *inode)
4166 unsigned int flags = EXT4_I(inode)->i_flags;
4168 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4169 if (flags & EXT4_SYNC_FL)
4170 inode->i_flags |= S_SYNC;
4171 if (flags & EXT4_APPEND_FL)
4172 inode->i_flags |= S_APPEND;
4173 if (flags & EXT4_IMMUTABLE_FL)
4174 inode->i_flags |= S_IMMUTABLE;
4175 if (flags & EXT4_NOATIME_FL)
4176 inode->i_flags |= S_NOATIME;
4177 if (flags & EXT4_DIRSYNC_FL)
4178 inode->i_flags |= S_DIRSYNC;
4181 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4182 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4184 unsigned int flags = ei->vfs_inode.i_flags;
4186 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4187 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4189 ei->i_flags |= EXT4_SYNC_FL;
4190 if (flags & S_APPEND)
4191 ei->i_flags |= EXT4_APPEND_FL;
4192 if (flags & S_IMMUTABLE)
4193 ei->i_flags |= EXT4_IMMUTABLE_FL;
4194 if (flags & S_NOATIME)
4195 ei->i_flags |= EXT4_NOATIME_FL;
4196 if (flags & S_DIRSYNC)
4197 ei->i_flags |= EXT4_DIRSYNC_FL;
4199 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4200 struct ext4_inode_info *ei)
4203 struct inode *inode = &(ei->vfs_inode);
4204 struct super_block *sb = inode->i_sb;
4206 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4207 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4208 /* we are using combined 48 bit field */
4209 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4210 le32_to_cpu(raw_inode->i_blocks_lo);
4211 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4212 /* i_blocks represent file system block size */
4213 return i_blocks << (inode->i_blkbits - 9);
4218 return le32_to_cpu(raw_inode->i_blocks_lo);
4222 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4224 struct ext4_iloc iloc;
4225 struct ext4_inode *raw_inode;
4226 struct ext4_inode_info *ei;
4227 struct buffer_head *bh;
4228 struct inode *inode;
4232 inode = iget_locked(sb, ino);
4234 return ERR_PTR(-ENOMEM);
4235 if (!(inode->i_state & I_NEW))
4239 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4240 ei->i_acl = EXT4_ACL_NOT_CACHED;
4241 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4244 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4248 raw_inode = ext4_raw_inode(&iloc);
4249 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4250 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4251 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4252 if (!(test_opt(inode->i_sb, NO_UID32))) {
4253 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4254 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4256 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4259 ei->i_dir_start_lookup = 0;
4260 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4261 /* We now have enough fields to check if the inode was active or not.
4262 * This is needed because nfsd might try to access dead inodes
4263 * the test is that same one that e2fsck uses
4264 * NeilBrown 1999oct15
4266 if (inode->i_nlink == 0) {
4267 if (inode->i_mode == 0 ||
4268 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4269 /* this inode is deleted */
4274 /* The only unlinked inodes we let through here have
4275 * valid i_mode and are being read by the orphan
4276 * recovery code: that's fine, we're about to complete
4277 * the process of deleting those. */
4279 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4280 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4281 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4282 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4283 cpu_to_le32(EXT4_OS_HURD)) {
4285 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4287 inode->i_size = ext4_isize(raw_inode);
4288 ei->i_disksize = inode->i_size;
4289 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4290 ei->i_block_group = iloc.block_group;
4291 ei->i_last_alloc_group = ~0;
4293 * NOTE! The in-memory inode i_data array is in little-endian order
4294 * even on big-endian machines: we do NOT byteswap the block numbers!
4296 for (block = 0; block < EXT4_N_BLOCKS; block++)
4297 ei->i_data[block] = raw_inode->i_block[block];
4298 INIT_LIST_HEAD(&ei->i_orphan);
4300 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4301 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4302 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4303 EXT4_INODE_SIZE(inode->i_sb)) {
4308 if (ei->i_extra_isize == 0) {
4309 /* The extra space is currently unused. Use it. */
4310 ei->i_extra_isize = sizeof(struct ext4_inode) -
4311 EXT4_GOOD_OLD_INODE_SIZE;
4313 __le32 *magic = (void *)raw_inode +
4314 EXT4_GOOD_OLD_INODE_SIZE +
4316 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4317 ei->i_state |= EXT4_STATE_XATTR;
4320 ei->i_extra_isize = 0;
4322 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4323 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4324 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4325 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4327 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4328 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4329 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4331 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4334 if (ei->i_flags & EXT4_EXTENTS_FL) {
4335 /* Validate extent which is part of inode */
4336 ret = ext4_ext_check_inode(inode);
4344 if (S_ISREG(inode->i_mode)) {
4345 inode->i_op = &ext4_file_inode_operations;
4346 inode->i_fop = &ext4_file_operations;
4347 ext4_set_aops(inode);
4348 } else if (S_ISDIR(inode->i_mode)) {
4349 inode->i_op = &ext4_dir_inode_operations;
4350 inode->i_fop = &ext4_dir_operations;
4351 } else if (S_ISLNK(inode->i_mode)) {
4352 if (ext4_inode_is_fast_symlink(inode)) {
4353 inode->i_op = &ext4_fast_symlink_inode_operations;
4354 nd_terminate_link(ei->i_data, inode->i_size,
4355 sizeof(ei->i_data) - 1);
4357 inode->i_op = &ext4_symlink_inode_operations;
4358 ext4_set_aops(inode);
4361 inode->i_op = &ext4_special_inode_operations;
4362 if (raw_inode->i_block[0])
4363 init_special_inode(inode, inode->i_mode,
4364 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4366 init_special_inode(inode, inode->i_mode,
4367 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4370 ext4_set_inode_flags(inode);
4371 unlock_new_inode(inode);
4376 return ERR_PTR(ret);
4379 static int ext4_inode_blocks_set(handle_t *handle,
4380 struct ext4_inode *raw_inode,
4381 struct ext4_inode_info *ei)
4383 struct inode *inode = &(ei->vfs_inode);
4384 u64 i_blocks = inode->i_blocks;
4385 struct super_block *sb = inode->i_sb;
4387 if (i_blocks <= ~0U) {
4389 * i_blocks can be represnted in a 32 bit variable
4390 * as multiple of 512 bytes
4392 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4393 raw_inode->i_blocks_high = 0;
4394 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4397 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4400 if (i_blocks <= 0xffffffffffffULL) {
4402 * i_blocks can be represented in a 48 bit variable
4403 * as multiple of 512 bytes
4405 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4406 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4407 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4409 ei->i_flags |= EXT4_HUGE_FILE_FL;
4410 /* i_block is stored in file system block size */
4411 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4412 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4413 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4419 * Post the struct inode info into an on-disk inode location in the
4420 * buffer-cache. This gobbles the caller's reference to the
4421 * buffer_head in the inode location struct.
4423 * The caller must have write access to iloc->bh.
4425 static int ext4_do_update_inode(handle_t *handle,
4426 struct inode *inode,
4427 struct ext4_iloc *iloc)
4429 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4430 struct ext4_inode_info *ei = EXT4_I(inode);
4431 struct buffer_head *bh = iloc->bh;
4432 int err = 0, rc, block;
4434 /* For fields not not tracking in the in-memory inode,
4435 * initialise them to zero for new inodes. */
4436 if (ei->i_state & EXT4_STATE_NEW)
4437 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4439 ext4_get_inode_flags(ei);
4440 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4441 if (!(test_opt(inode->i_sb, NO_UID32))) {
4442 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4443 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4445 * Fix up interoperability with old kernels. Otherwise, old inodes get
4446 * re-used with the upper 16 bits of the uid/gid intact
4449 raw_inode->i_uid_high =
4450 cpu_to_le16(high_16_bits(inode->i_uid));
4451 raw_inode->i_gid_high =
4452 cpu_to_le16(high_16_bits(inode->i_gid));
4454 raw_inode->i_uid_high = 0;
4455 raw_inode->i_gid_high = 0;
4458 raw_inode->i_uid_low =
4459 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4460 raw_inode->i_gid_low =
4461 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4462 raw_inode->i_uid_high = 0;
4463 raw_inode->i_gid_high = 0;
4465 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4467 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4468 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4469 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4470 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4472 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4474 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4475 /* clear the migrate flag in the raw_inode */
4476 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4477 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4478 cpu_to_le32(EXT4_OS_HURD))
4479 raw_inode->i_file_acl_high =
4480 cpu_to_le16(ei->i_file_acl >> 32);
4481 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4482 ext4_isize_set(raw_inode, ei->i_disksize);
4483 if (ei->i_disksize > 0x7fffffffULL) {
4484 struct super_block *sb = inode->i_sb;
4485 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4486 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4487 EXT4_SB(sb)->s_es->s_rev_level ==
4488 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4489 /* If this is the first large file
4490 * created, add a flag to the superblock.
4492 err = ext4_journal_get_write_access(handle,
4493 EXT4_SB(sb)->s_sbh);
4496 ext4_update_dynamic_rev(sb);
4497 EXT4_SET_RO_COMPAT_FEATURE(sb,
4498 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4500 ext4_handle_sync(handle);
4501 err = ext4_handle_dirty_metadata(handle, inode,
4502 EXT4_SB(sb)->s_sbh);
4505 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4506 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4507 if (old_valid_dev(inode->i_rdev)) {
4508 raw_inode->i_block[0] =
4509 cpu_to_le32(old_encode_dev(inode->i_rdev));
4510 raw_inode->i_block[1] = 0;
4512 raw_inode->i_block[0] = 0;
4513 raw_inode->i_block[1] =
4514 cpu_to_le32(new_encode_dev(inode->i_rdev));
4515 raw_inode->i_block[2] = 0;
4517 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4518 raw_inode->i_block[block] = ei->i_data[block];
4520 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4521 if (ei->i_extra_isize) {
4522 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4523 raw_inode->i_version_hi =
4524 cpu_to_le32(inode->i_version >> 32);
4525 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4528 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4529 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4532 ei->i_state &= ~EXT4_STATE_NEW;
4536 ext4_std_error(inode->i_sb, err);
4541 * ext4_write_inode()
4543 * We are called from a few places:
4545 * - Within generic_file_write() for O_SYNC files.
4546 * Here, there will be no transaction running. We wait for any running
4547 * trasnaction to commit.
4549 * - Within sys_sync(), kupdate and such.
4550 * We wait on commit, if tol to.
4552 * - Within prune_icache() (PF_MEMALLOC == true)
4553 * Here we simply return. We can't afford to block kswapd on the
4556 * In all cases it is actually safe for us to return without doing anything,
4557 * because the inode has been copied into a raw inode buffer in
4558 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4561 * Note that we are absolutely dependent upon all inode dirtiers doing the
4562 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4563 * which we are interested.
4565 * It would be a bug for them to not do this. The code:
4567 * mark_inode_dirty(inode)
4569 * inode->i_size = expr;
4571 * is in error because a kswapd-driven write_inode() could occur while
4572 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4573 * will no longer be on the superblock's dirty inode list.
4575 int ext4_write_inode(struct inode *inode, int wait)
4577 if (current->flags & PF_MEMALLOC)
4580 if (ext4_journal_current_handle()) {
4581 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4589 return ext4_force_commit(inode->i_sb);
4592 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4596 mark_buffer_dirty(bh);
4597 if (inode && inode_needs_sync(inode)) {
4598 sync_dirty_buffer(bh);
4599 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4600 ext4_error(inode->i_sb, __func__,
4601 "IO error syncing inode, "
4602 "inode=%lu, block=%llu",
4604 (unsigned long long)bh->b_blocknr);
4614 * Called from notify_change.
4616 * We want to trap VFS attempts to truncate the file as soon as
4617 * possible. In particular, we want to make sure that when the VFS
4618 * shrinks i_size, we put the inode on the orphan list and modify
4619 * i_disksize immediately, so that during the subsequent flushing of
4620 * dirty pages and freeing of disk blocks, we can guarantee that any
4621 * commit will leave the blocks being flushed in an unused state on
4622 * disk. (On recovery, the inode will get truncated and the blocks will
4623 * be freed, so we have a strong guarantee that no future commit will
4624 * leave these blocks visible to the user.)
4626 * Another thing we have to assure is that if we are in ordered mode
4627 * and inode is still attached to the committing transaction, we must
4628 * we start writeout of all the dirty pages which are being truncated.
4629 * This way we are sure that all the data written in the previous
4630 * transaction are already on disk (truncate waits for pages under
4633 * Called with inode->i_mutex down.
4635 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4637 struct inode *inode = dentry->d_inode;
4639 const unsigned int ia_valid = attr->ia_valid;
4641 error = inode_change_ok(inode, attr);
4645 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4646 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4649 /* (user+group)*(old+new) structure, inode write (sb,
4650 * inode block, ? - but truncate inode update has it) */
4651 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4652 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4653 if (IS_ERR(handle)) {
4654 error = PTR_ERR(handle);
4657 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4659 ext4_journal_stop(handle);
4662 /* Update corresponding info in inode so that everything is in
4663 * one transaction */
4664 if (attr->ia_valid & ATTR_UID)
4665 inode->i_uid = attr->ia_uid;
4666 if (attr->ia_valid & ATTR_GID)
4667 inode->i_gid = attr->ia_gid;
4668 error = ext4_mark_inode_dirty(handle, inode);
4669 ext4_journal_stop(handle);
4672 if (attr->ia_valid & ATTR_SIZE) {
4673 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4674 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4676 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4683 if (S_ISREG(inode->i_mode) &&
4684 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4687 handle = ext4_journal_start(inode, 3);
4688 if (IS_ERR(handle)) {
4689 error = PTR_ERR(handle);
4693 error = ext4_orphan_add(handle, inode);
4694 EXT4_I(inode)->i_disksize = attr->ia_size;
4695 rc = ext4_mark_inode_dirty(handle, inode);
4698 ext4_journal_stop(handle);
4700 if (ext4_should_order_data(inode)) {
4701 error = ext4_begin_ordered_truncate(inode,
4704 /* Do as much error cleanup as possible */
4705 handle = ext4_journal_start(inode, 3);
4706 if (IS_ERR(handle)) {
4707 ext4_orphan_del(NULL, inode);
4710 ext4_orphan_del(handle, inode);
4711 ext4_journal_stop(handle);
4717 rc = inode_setattr(inode, attr);
4719 /* If inode_setattr's call to ext4_truncate failed to get a
4720 * transaction handle at all, we need to clean up the in-core
4721 * orphan list manually. */
4723 ext4_orphan_del(NULL, inode);
4725 if (!rc && (ia_valid & ATTR_MODE))
4726 rc = ext4_acl_chmod(inode);
4729 ext4_std_error(inode->i_sb, error);
4735 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4738 struct inode *inode;
4739 unsigned long delalloc_blocks;
4741 inode = dentry->d_inode;
4742 generic_fillattr(inode, stat);
4745 * We can't update i_blocks if the block allocation is delayed
4746 * otherwise in the case of system crash before the real block
4747 * allocation is done, we will have i_blocks inconsistent with
4748 * on-disk file blocks.
4749 * We always keep i_blocks updated together with real
4750 * allocation. But to not confuse with user, stat
4751 * will return the blocks that include the delayed allocation
4752 * blocks for this file.
4754 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4755 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4756 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4758 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4762 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4767 /* if nrblocks are contiguous */
4770 * With N contiguous data blocks, it need at most
4771 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4772 * 2 dindirect blocks
4775 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4776 return indirects + 3;
4779 * if nrblocks are not contiguous, worse case, each block touch
4780 * a indirect block, and each indirect block touch a double indirect
4781 * block, plus a triple indirect block
4783 indirects = nrblocks * 2 + 1;
4787 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4789 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4790 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4791 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4795 * Account for index blocks, block groups bitmaps and block group
4796 * descriptor blocks if modify datablocks and index blocks
4797 * worse case, the indexs blocks spread over different block groups
4799 * If datablocks are discontiguous, they are possible to spread over
4800 * different block groups too. If they are contiugous, with flexbg,
4801 * they could still across block group boundary.
4803 * Also account for superblock, inode, quota and xattr blocks
4805 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4807 int groups, gdpblocks;
4812 * How many index blocks need to touch to modify nrblocks?
4813 * The "Chunk" flag indicating whether the nrblocks is
4814 * physically contiguous on disk
4816 * For Direct IO and fallocate, they calls get_block to allocate
4817 * one single extent at a time, so they could set the "Chunk" flag
4819 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4824 * Now let's see how many group bitmaps and group descriptors need
4834 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4835 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4836 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4837 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4839 /* bitmaps and block group descriptor blocks */
4840 ret += groups + gdpblocks;
4842 /* Blocks for super block, inode, quota and xattr blocks */
4843 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4849 * Calulate the total number of credits to reserve to fit
4850 * the modification of a single pages into a single transaction,
4851 * which may include multiple chunks of block allocations.
4853 * This could be called via ext4_write_begin()
4855 * We need to consider the worse case, when
4856 * one new block per extent.
4858 int ext4_writepage_trans_blocks(struct inode *inode)
4860 int bpp = ext4_journal_blocks_per_page(inode);
4863 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4865 /* Account for data blocks for journalled mode */
4866 if (ext4_should_journal_data(inode))
4872 * Calculate the journal credits for a chunk of data modification.
4874 * This is called from DIO, fallocate or whoever calling
4875 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4877 * journal buffers for data blocks are not included here, as DIO
4878 * and fallocate do no need to journal data buffers.
4880 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4882 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4886 * The caller must have previously called ext4_reserve_inode_write().
4887 * Give this, we know that the caller already has write access to iloc->bh.
4889 int ext4_mark_iloc_dirty(handle_t *handle,
4890 struct inode *inode, struct ext4_iloc *iloc)
4894 if (test_opt(inode->i_sb, I_VERSION))
4895 inode_inc_iversion(inode);
4897 /* the do_update_inode consumes one bh->b_count */
4900 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4901 err = ext4_do_update_inode(handle, inode, iloc);
4907 * On success, We end up with an outstanding reference count against
4908 * iloc->bh. This _must_ be cleaned up later.
4912 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4913 struct ext4_iloc *iloc)
4917 err = ext4_get_inode_loc(inode, iloc);
4919 BUFFER_TRACE(iloc->bh, "get_write_access");
4920 err = ext4_journal_get_write_access(handle, iloc->bh);
4926 ext4_std_error(inode->i_sb, err);
4931 * Expand an inode by new_extra_isize bytes.
4932 * Returns 0 on success or negative error number on failure.
4934 static int ext4_expand_extra_isize(struct inode *inode,
4935 unsigned int new_extra_isize,
4936 struct ext4_iloc iloc,
4939 struct ext4_inode *raw_inode;
4940 struct ext4_xattr_ibody_header *header;
4941 struct ext4_xattr_entry *entry;
4943 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4946 raw_inode = ext4_raw_inode(&iloc);
4948 header = IHDR(inode, raw_inode);
4949 entry = IFIRST(header);
4951 /* No extended attributes present */
4952 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4953 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4954 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4956 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4960 /* try to expand with EAs present */
4961 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4966 * What we do here is to mark the in-core inode as clean with respect to inode
4967 * dirtiness (it may still be data-dirty).
4968 * This means that the in-core inode may be reaped by prune_icache
4969 * without having to perform any I/O. This is a very good thing,
4970 * because *any* task may call prune_icache - even ones which
4971 * have a transaction open against a different journal.
4973 * Is this cheating? Not really. Sure, we haven't written the
4974 * inode out, but prune_icache isn't a user-visible syncing function.
4975 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4976 * we start and wait on commits.
4978 * Is this efficient/effective? Well, we're being nice to the system
4979 * by cleaning up our inodes proactively so they can be reaped
4980 * without I/O. But we are potentially leaving up to five seconds'
4981 * worth of inodes floating about which prune_icache wants us to
4982 * write out. One way to fix that would be to get prune_icache()
4983 * to do a write_super() to free up some memory. It has the desired
4986 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4988 struct ext4_iloc iloc;
4989 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4990 static unsigned int mnt_count;
4994 err = ext4_reserve_inode_write(handle, inode, &iloc);
4995 if (ext4_handle_valid(handle) &&
4996 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4997 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4999 * We need extra buffer credits since we may write into EA block
5000 * with this same handle. If journal_extend fails, then it will
5001 * only result in a minor loss of functionality for that inode.
5002 * If this is felt to be critical, then e2fsck should be run to
5003 * force a large enough s_min_extra_isize.
5005 if ((jbd2_journal_extend(handle,
5006 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5007 ret = ext4_expand_extra_isize(inode,
5008 sbi->s_want_extra_isize,
5011 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5013 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5014 ext4_warning(inode->i_sb, __func__,
5015 "Unable to expand inode %lu. Delete"
5016 " some EAs or run e2fsck.",
5019 le16_to_cpu(sbi->s_es->s_mnt_count);
5025 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5030 * ext4_dirty_inode() is called from __mark_inode_dirty()
5032 * We're really interested in the case where a file is being extended.
5033 * i_size has been changed by generic_commit_write() and we thus need
5034 * to include the updated inode in the current transaction.
5036 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5037 * are allocated to the file.
5039 * If the inode is marked synchronous, we don't honour that here - doing
5040 * so would cause a commit on atime updates, which we don't bother doing.
5041 * We handle synchronous inodes at the highest possible level.
5043 void ext4_dirty_inode(struct inode *inode)
5045 handle_t *current_handle = ext4_journal_current_handle();
5048 if (!ext4_handle_valid(current_handle)) {
5049 ext4_mark_inode_dirty(current_handle, inode);
5053 handle = ext4_journal_start(inode, 2);
5056 if (current_handle &&
5057 current_handle->h_transaction != handle->h_transaction) {
5058 /* This task has a transaction open against a different fs */
5059 printk(KERN_EMERG "%s: transactions do not match!\n",
5062 jbd_debug(5, "marking dirty. outer handle=%p\n",
5064 ext4_mark_inode_dirty(handle, inode);
5066 ext4_journal_stop(handle);
5073 * Bind an inode's backing buffer_head into this transaction, to prevent
5074 * it from being flushed to disk early. Unlike
5075 * ext4_reserve_inode_write, this leaves behind no bh reference and
5076 * returns no iloc structure, so the caller needs to repeat the iloc
5077 * lookup to mark the inode dirty later.
5079 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5081 struct ext4_iloc iloc;
5085 err = ext4_get_inode_loc(inode, &iloc);
5087 BUFFER_TRACE(iloc.bh, "get_write_access");
5088 err = jbd2_journal_get_write_access(handle, iloc.bh);
5090 err = ext4_handle_dirty_metadata(handle,
5096 ext4_std_error(inode->i_sb, err);
5101 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5108 * We have to be very careful here: changing a data block's
5109 * journaling status dynamically is dangerous. If we write a
5110 * data block to the journal, change the status and then delete
5111 * that block, we risk forgetting to revoke the old log record
5112 * from the journal and so a subsequent replay can corrupt data.
5113 * So, first we make sure that the journal is empty and that
5114 * nobody is changing anything.
5117 journal = EXT4_JOURNAL(inode);
5120 if (is_journal_aborted(journal))
5123 jbd2_journal_lock_updates(journal);
5124 jbd2_journal_flush(journal);
5127 * OK, there are no updates running now, and all cached data is
5128 * synced to disk. We are now in a completely consistent state
5129 * which doesn't have anything in the journal, and we know that
5130 * no filesystem updates are running, so it is safe to modify
5131 * the inode's in-core data-journaling state flag now.
5135 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5137 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5138 ext4_set_aops(inode);
5140 jbd2_journal_unlock_updates(journal);
5142 /* Finally we can mark the inode as dirty. */
5144 handle = ext4_journal_start(inode, 1);
5146 return PTR_ERR(handle);
5148 err = ext4_mark_inode_dirty(handle, inode);
5149 ext4_handle_sync(handle);
5150 ext4_journal_stop(handle);
5151 ext4_std_error(inode->i_sb, err);
5156 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5158 return !buffer_mapped(bh);
5161 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
5167 struct file *file = vma->vm_file;
5168 struct inode *inode = file->f_path.dentry->d_inode;
5169 struct address_space *mapping = inode->i_mapping;
5172 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5173 * get i_mutex because we are already holding mmap_sem.
5175 down_read(&inode->i_alloc_sem);
5176 size = i_size_read(inode);
5177 if (page->mapping != mapping || size <= page_offset(page)
5178 || !PageUptodate(page)) {
5179 /* page got truncated from under us? */
5183 if (PageMappedToDisk(page))
5186 if (page->index == size >> PAGE_CACHE_SHIFT)
5187 len = size & ~PAGE_CACHE_MASK;
5189 len = PAGE_CACHE_SIZE;
5191 if (page_has_buffers(page)) {
5192 /* return if we have all the buffers mapped */
5193 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5198 * OK, we need to fill the hole... Do write_begin write_end
5199 * to do block allocation/reservation.We are not holding
5200 * inode.i__mutex here. That allow * parallel write_begin,
5201 * write_end call. lock_page prevent this from happening
5202 * on the same page though
5204 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5205 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5208 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5209 len, len, page, fsdata);
5214 up_read(&inode->i_alloc_sem);