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
1071 vfs_dq_release_reservation_block(inode, mdb_free);
1074 * If we have done all the pending block allocations and if
1075 * there aren't any writers on the inode, we can discard the
1076 * inode's preallocations.
1078 if (!total && (atomic_read(&inode->i_writecount) == 0))
1079 ext4_discard_preallocations(inode);
1083 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1084 * and returns if the blocks are already mapped.
1086 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1087 * and store the allocated blocks in the result buffer head and mark it
1090 * If file type is extents based, it will call ext4_ext_get_blocks(),
1091 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1094 * On success, it returns the number of blocks being mapped or allocate.
1095 * if create==0 and the blocks are pre-allocated and uninitialized block,
1096 * the result buffer head is unmapped. If the create ==1, it will make sure
1097 * the buffer head is mapped.
1099 * It returns 0 if plain look up failed (blocks have not been allocated), in
1100 * that casem, buffer head is unmapped
1102 * It returns the error in case of allocation failure.
1104 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1105 unsigned int max_blocks, struct buffer_head *bh,
1106 int create, int extend_disksize, int flag)
1110 clear_buffer_mapped(bh);
1113 * Try to see if we can get the block without requesting
1114 * for new file system block.
1116 down_read((&EXT4_I(inode)->i_data_sem));
1117 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1118 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1121 retval = ext4_get_blocks_handle(handle,
1122 inode, block, max_blocks, bh, 0, 0);
1124 up_read((&EXT4_I(inode)->i_data_sem));
1126 /* If it is only a block(s) look up */
1131 * Returns if the blocks have already allocated
1133 * Note that if blocks have been preallocated
1134 * ext4_ext_get_block() returns th create = 0
1135 * with buffer head unmapped.
1137 if (retval > 0 && buffer_mapped(bh))
1141 * New blocks allocate and/or writing to uninitialized extent
1142 * will possibly result in updating i_data, so we take
1143 * the write lock of i_data_sem, and call get_blocks()
1144 * with create == 1 flag.
1146 down_write((&EXT4_I(inode)->i_data_sem));
1149 * if the caller is from delayed allocation writeout path
1150 * we have already reserved fs blocks for allocation
1151 * let the underlying get_block() function know to
1152 * avoid double accounting
1155 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1157 * We need to check for EXT4 here because migrate
1158 * could have changed the inode type in between
1160 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1161 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1162 bh, create, extend_disksize);
1164 retval = ext4_get_blocks_handle(handle, inode, block,
1165 max_blocks, bh, create, extend_disksize);
1167 if (retval > 0 && buffer_new(bh)) {
1169 * We allocated new blocks which will result in
1170 * i_data's format changing. Force the migrate
1171 * to fail by clearing migrate flags
1173 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1179 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1181 * Update reserved blocks/metadata blocks
1182 * after successful block allocation
1183 * which were deferred till now
1185 if ((retval > 0) && buffer_delay(bh))
1186 ext4_da_update_reserve_space(inode, retval);
1189 up_write((&EXT4_I(inode)->i_data_sem));
1193 /* Maximum number of blocks we map for direct IO at once. */
1194 #define DIO_MAX_BLOCKS 4096
1196 int ext4_get_block(struct inode *inode, sector_t iblock,
1197 struct buffer_head *bh_result, int create)
1199 handle_t *handle = ext4_journal_current_handle();
1200 int ret = 0, started = 0;
1201 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1204 if (create && !handle) {
1205 /* Direct IO write... */
1206 if (max_blocks > DIO_MAX_BLOCKS)
1207 max_blocks = DIO_MAX_BLOCKS;
1208 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1209 handle = ext4_journal_start(inode, dio_credits);
1210 if (IS_ERR(handle)) {
1211 ret = PTR_ERR(handle);
1217 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1218 max_blocks, bh_result, create, 0, 0);
1220 bh_result->b_size = (ret << inode->i_blkbits);
1224 ext4_journal_stop(handle);
1230 * `handle' can be NULL if create is zero
1232 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1233 ext4_lblk_t block, int create, int *errp)
1235 struct buffer_head dummy;
1238 J_ASSERT(handle != NULL || create == 0);
1241 dummy.b_blocknr = -1000;
1242 buffer_trace_init(&dummy.b_history);
1243 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1244 &dummy, create, 1, 0);
1246 * ext4_get_blocks_handle() returns number of blocks
1247 * mapped. 0 in case of a HOLE.
1255 if (!err && buffer_mapped(&dummy)) {
1256 struct buffer_head *bh;
1257 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1262 if (buffer_new(&dummy)) {
1263 J_ASSERT(create != 0);
1264 J_ASSERT(handle != NULL);
1267 * Now that we do not always journal data, we should
1268 * keep in mind whether this should always journal the
1269 * new buffer as metadata. For now, regular file
1270 * writes use ext4_get_block instead, so it's not a
1274 BUFFER_TRACE(bh, "call get_create_access");
1275 fatal = ext4_journal_get_create_access(handle, bh);
1276 if (!fatal && !buffer_uptodate(bh)) {
1277 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1278 set_buffer_uptodate(bh);
1281 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1282 err = ext4_handle_dirty_metadata(handle, inode, bh);
1286 BUFFER_TRACE(bh, "not a new buffer");
1299 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1300 ext4_lblk_t block, int create, int *err)
1302 struct buffer_head *bh;
1304 bh = ext4_getblk(handle, inode, block, create, err);
1307 if (buffer_uptodate(bh))
1309 ll_rw_block(READ_META, 1, &bh);
1311 if (buffer_uptodate(bh))
1318 static int walk_page_buffers(handle_t *handle,
1319 struct buffer_head *head,
1323 int (*fn)(handle_t *handle,
1324 struct buffer_head *bh))
1326 struct buffer_head *bh;
1327 unsigned block_start, block_end;
1328 unsigned blocksize = head->b_size;
1330 struct buffer_head *next;
1332 for (bh = head, block_start = 0;
1333 ret == 0 && (bh != head || !block_start);
1334 block_start = block_end, bh = next)
1336 next = bh->b_this_page;
1337 block_end = block_start + blocksize;
1338 if (block_end <= from || block_start >= to) {
1339 if (partial && !buffer_uptodate(bh))
1343 err = (*fn)(handle, bh);
1351 * To preserve ordering, it is essential that the hole instantiation and
1352 * the data write be encapsulated in a single transaction. We cannot
1353 * close off a transaction and start a new one between the ext4_get_block()
1354 * and the commit_write(). So doing the jbd2_journal_start at the start of
1355 * prepare_write() is the right place.
1357 * Also, this function can nest inside ext4_writepage() ->
1358 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1359 * has generated enough buffer credits to do the whole page. So we won't
1360 * block on the journal in that case, which is good, because the caller may
1363 * By accident, ext4 can be reentered when a transaction is open via
1364 * quota file writes. If we were to commit the transaction while thus
1365 * reentered, there can be a deadlock - we would be holding a quota
1366 * lock, and the commit would never complete if another thread had a
1367 * transaction open and was blocking on the quota lock - a ranking
1370 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1371 * will _not_ run commit under these circumstances because handle->h_ref
1372 * is elevated. We'll still have enough credits for the tiny quotafile
1375 static int do_journal_get_write_access(handle_t *handle,
1376 struct buffer_head *bh)
1378 if (!buffer_mapped(bh) || buffer_freed(bh))
1380 return ext4_journal_get_write_access(handle, bh);
1383 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1384 loff_t pos, unsigned len, unsigned flags,
1385 struct page **pagep, void **fsdata)
1387 struct inode *inode = mapping->host;
1388 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1395 trace_mark(ext4_write_begin,
1396 "dev %s ino %lu pos %llu len %u flags %u",
1397 inode->i_sb->s_id, inode->i_ino,
1398 (unsigned long long) pos, len, flags);
1399 index = pos >> PAGE_CACHE_SHIFT;
1400 from = pos & (PAGE_CACHE_SIZE - 1);
1404 handle = ext4_journal_start(inode, needed_blocks);
1405 if (IS_ERR(handle)) {
1406 ret = PTR_ERR(handle);
1410 /* We cannot recurse into the filesystem as the transaction is already
1412 flags |= AOP_FLAG_NOFS;
1414 page = grab_cache_page_write_begin(mapping, index, flags);
1416 ext4_journal_stop(handle);
1422 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1425 if (!ret && ext4_should_journal_data(inode)) {
1426 ret = walk_page_buffers(handle, page_buffers(page),
1427 from, to, NULL, do_journal_get_write_access);
1432 ext4_journal_stop(handle);
1433 page_cache_release(page);
1435 * block_write_begin may have instantiated a few blocks
1436 * outside i_size. Trim these off again. Don't need
1437 * i_size_read because we hold i_mutex.
1439 if (pos + len > inode->i_size)
1440 vmtruncate(inode, inode->i_size);
1443 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1449 /* For write_end() in data=journal mode */
1450 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1452 if (!buffer_mapped(bh) || buffer_freed(bh))
1454 set_buffer_uptodate(bh);
1455 return ext4_handle_dirty_metadata(handle, NULL, bh);
1459 * We need to pick up the new inode size which generic_commit_write gave us
1460 * `file' can be NULL - eg, when called from page_symlink().
1462 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1463 * buffers are managed internally.
1465 static int ext4_ordered_write_end(struct file *file,
1466 struct address_space *mapping,
1467 loff_t pos, unsigned len, unsigned copied,
1468 struct page *page, void *fsdata)
1470 handle_t *handle = ext4_journal_current_handle();
1471 struct inode *inode = mapping->host;
1474 trace_mark(ext4_ordered_write_end,
1475 "dev %s ino %lu pos %llu len %u copied %u",
1476 inode->i_sb->s_id, inode->i_ino,
1477 (unsigned long long) pos, len, copied);
1478 ret = ext4_jbd2_file_inode(handle, inode);
1483 new_i_size = pos + copied;
1484 if (new_i_size > EXT4_I(inode)->i_disksize) {
1485 ext4_update_i_disksize(inode, new_i_size);
1486 /* We need to mark inode dirty even if
1487 * new_i_size is less that inode->i_size
1488 * bu greater than i_disksize.(hint delalloc)
1490 ext4_mark_inode_dirty(handle, inode);
1493 ret2 = generic_write_end(file, mapping, pos, len, copied,
1499 ret2 = ext4_journal_stop(handle);
1503 return ret ? ret : copied;
1506 static int ext4_writeback_write_end(struct file *file,
1507 struct address_space *mapping,
1508 loff_t pos, unsigned len, unsigned copied,
1509 struct page *page, void *fsdata)
1511 handle_t *handle = ext4_journal_current_handle();
1512 struct inode *inode = mapping->host;
1516 trace_mark(ext4_writeback_write_end,
1517 "dev %s ino %lu pos %llu len %u copied %u",
1518 inode->i_sb->s_id, inode->i_ino,
1519 (unsigned long long) pos, len, copied);
1520 new_i_size = pos + copied;
1521 if (new_i_size > EXT4_I(inode)->i_disksize) {
1522 ext4_update_i_disksize(inode, new_i_size);
1523 /* We need to mark inode dirty even if
1524 * new_i_size is less that inode->i_size
1525 * bu greater than i_disksize.(hint delalloc)
1527 ext4_mark_inode_dirty(handle, inode);
1530 ret2 = generic_write_end(file, mapping, pos, len, copied,
1536 ret2 = ext4_journal_stop(handle);
1540 return ret ? ret : copied;
1543 static int ext4_journalled_write_end(struct file *file,
1544 struct address_space *mapping,
1545 loff_t pos, unsigned len, unsigned copied,
1546 struct page *page, void *fsdata)
1548 handle_t *handle = ext4_journal_current_handle();
1549 struct inode *inode = mapping->host;
1555 trace_mark(ext4_journalled_write_end,
1556 "dev %s ino %lu pos %llu len %u copied %u",
1557 inode->i_sb->s_id, inode->i_ino,
1558 (unsigned long long) pos, len, copied);
1559 from = pos & (PAGE_CACHE_SIZE - 1);
1563 if (!PageUptodate(page))
1565 page_zero_new_buffers(page, from+copied, to);
1568 ret = walk_page_buffers(handle, page_buffers(page), from,
1569 to, &partial, write_end_fn);
1571 SetPageUptodate(page);
1572 new_i_size = pos + copied;
1573 if (new_i_size > inode->i_size)
1574 i_size_write(inode, pos+copied);
1575 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1576 if (new_i_size > EXT4_I(inode)->i_disksize) {
1577 ext4_update_i_disksize(inode, new_i_size);
1578 ret2 = ext4_mark_inode_dirty(handle, inode);
1584 ret2 = ext4_journal_stop(handle);
1587 page_cache_release(page);
1589 return ret ? ret : copied;
1592 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1595 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1596 unsigned long md_needed, mdblocks, total = 0;
1599 * recalculate the amount of metadata blocks to reserve
1600 * in order to allocate nrblocks
1601 * worse case is one extent per block
1604 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1605 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1606 mdblocks = ext4_calc_metadata_amount(inode, total);
1607 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1609 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1610 total = md_needed + nrblocks;
1613 * Make quota reservation here to prevent quota overflow
1614 * later. Real quota accounting is done at pages writeout
1617 if (vfs_dq_reserve_block(inode, total)) {
1618 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1622 if (ext4_claim_free_blocks(sbi, total)) {
1623 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1624 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1628 vfs_dq_release_reservation_block(inode, total);
1631 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1632 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1634 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1635 return 0; /* success */
1638 static void ext4_da_release_space(struct inode *inode, int to_free)
1640 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1641 int total, mdb, mdb_free, release;
1644 return; /* Nothing to release, exit */
1646 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1648 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1650 * if there is no reserved blocks, but we try to free some
1651 * then the counter is messed up somewhere.
1652 * but since this function is called from invalidate
1653 * page, it's harmless to return without any action
1655 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1656 "blocks for inode %lu, but there is no reserved "
1657 "data blocks\n", to_free, inode->i_ino);
1658 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1662 /* recalculate the number of metablocks still need to be reserved */
1663 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1664 mdb = ext4_calc_metadata_amount(inode, total);
1666 /* figure out how many metablocks to release */
1667 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1668 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1670 release = to_free + mdb_free;
1672 /* update fs dirty blocks counter for truncate case */
1673 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1675 /* update per-inode reservations */
1676 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1677 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1679 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1680 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1681 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1683 vfs_dq_release_reservation_block(inode, release);
1686 static void ext4_da_page_release_reservation(struct page *page,
1687 unsigned long offset)
1690 struct buffer_head *head, *bh;
1691 unsigned int curr_off = 0;
1693 head = page_buffers(page);
1696 unsigned int next_off = curr_off + bh->b_size;
1698 if ((offset <= curr_off) && (buffer_delay(bh))) {
1700 clear_buffer_delay(bh);
1702 curr_off = next_off;
1703 } while ((bh = bh->b_this_page) != head);
1704 ext4_da_release_space(page->mapping->host, to_release);
1708 * Delayed allocation stuff
1711 struct mpage_da_data {
1712 struct inode *inode;
1713 sector_t b_blocknr; /* start block number of extent */
1714 size_t b_size; /* size of extent */
1715 unsigned long b_state; /* state of the extent */
1716 unsigned long first_page, next_page; /* extent of pages */
1717 struct writeback_control *wbc;
1724 * mpage_da_submit_io - walks through extent of pages and try to write
1725 * them with writepage() call back
1727 * @mpd->inode: inode
1728 * @mpd->first_page: first page of the extent
1729 * @mpd->next_page: page after the last page of the extent
1731 * By the time mpage_da_submit_io() is called we expect all blocks
1732 * to be allocated. this may be wrong if allocation failed.
1734 * As pages are already locked by write_cache_pages(), we can't use it
1736 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1739 struct pagevec pvec;
1740 unsigned long index, end;
1741 int ret = 0, err, nr_pages, i;
1742 struct inode *inode = mpd->inode;
1743 struct address_space *mapping = inode->i_mapping;
1745 BUG_ON(mpd->next_page <= mpd->first_page);
1747 * We need to start from the first_page to the next_page - 1
1748 * to make sure we also write the mapped dirty buffer_heads.
1749 * If we look at mpd->b_blocknr we would only be looking
1750 * at the currently mapped buffer_heads.
1752 index = mpd->first_page;
1753 end = mpd->next_page - 1;
1755 pagevec_init(&pvec, 0);
1756 while (index <= end) {
1757 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1760 for (i = 0; i < nr_pages; i++) {
1761 struct page *page = pvec.pages[i];
1763 index = page->index;
1768 BUG_ON(!PageLocked(page));
1769 BUG_ON(PageWriteback(page));
1771 pages_skipped = mpd->wbc->pages_skipped;
1772 err = mapping->a_ops->writepage(page, mpd->wbc);
1773 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1775 * have successfully written the page
1776 * without skipping the same
1778 mpd->pages_written++;
1780 * In error case, we have to continue because
1781 * remaining pages are still locked
1782 * XXX: unlock and re-dirty them?
1787 pagevec_release(&pvec);
1793 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1795 * @mpd->inode - inode to walk through
1796 * @exbh->b_blocknr - first block on a disk
1797 * @exbh->b_size - amount of space in bytes
1798 * @logical - first logical block to start assignment with
1800 * the function goes through all passed space and put actual disk
1801 * block numbers into buffer heads, dropping BH_Delay
1803 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1804 struct buffer_head *exbh)
1806 struct inode *inode = mpd->inode;
1807 struct address_space *mapping = inode->i_mapping;
1808 int blocks = exbh->b_size >> inode->i_blkbits;
1809 sector_t pblock = exbh->b_blocknr, cur_logical;
1810 struct buffer_head *head, *bh;
1812 struct pagevec pvec;
1815 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1816 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1817 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1819 pagevec_init(&pvec, 0);
1821 while (index <= end) {
1822 /* XXX: optimize tail */
1823 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1826 for (i = 0; i < nr_pages; i++) {
1827 struct page *page = pvec.pages[i];
1829 index = page->index;
1834 BUG_ON(!PageLocked(page));
1835 BUG_ON(PageWriteback(page));
1836 BUG_ON(!page_has_buffers(page));
1838 bh = page_buffers(page);
1841 /* skip blocks out of the range */
1843 if (cur_logical >= logical)
1846 } while ((bh = bh->b_this_page) != head);
1849 if (cur_logical >= logical + blocks)
1851 if (buffer_delay(bh)) {
1852 bh->b_blocknr = pblock;
1853 clear_buffer_delay(bh);
1854 bh->b_bdev = inode->i_sb->s_bdev;
1855 } else if (buffer_unwritten(bh)) {
1856 bh->b_blocknr = pblock;
1857 clear_buffer_unwritten(bh);
1858 set_buffer_mapped(bh);
1860 bh->b_bdev = inode->i_sb->s_bdev;
1861 } else if (buffer_mapped(bh))
1862 BUG_ON(bh->b_blocknr != pblock);
1866 } while ((bh = bh->b_this_page) != head);
1868 pagevec_release(&pvec);
1874 * __unmap_underlying_blocks - just a helper function to unmap
1875 * set of blocks described by @bh
1877 static inline void __unmap_underlying_blocks(struct inode *inode,
1878 struct buffer_head *bh)
1880 struct block_device *bdev = inode->i_sb->s_bdev;
1883 blocks = bh->b_size >> inode->i_blkbits;
1884 for (i = 0; i < blocks; i++)
1885 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1888 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1889 sector_t logical, long blk_cnt)
1893 struct pagevec pvec;
1894 struct inode *inode = mpd->inode;
1895 struct address_space *mapping = inode->i_mapping;
1897 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1898 end = (logical + blk_cnt - 1) >>
1899 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1900 while (index <= end) {
1901 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1904 for (i = 0; i < nr_pages; i++) {
1905 struct page *page = pvec.pages[i];
1906 index = page->index;
1911 BUG_ON(!PageLocked(page));
1912 BUG_ON(PageWriteback(page));
1913 block_invalidatepage(page, 0);
1914 ClearPageUptodate(page);
1921 static void ext4_print_free_blocks(struct inode *inode)
1923 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1924 printk(KERN_EMERG "Total free blocks count %lld\n",
1925 ext4_count_free_blocks(inode->i_sb));
1926 printk(KERN_EMERG "Free/Dirty block details\n");
1927 printk(KERN_EMERG "free_blocks=%lld\n",
1928 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1929 printk(KERN_EMERG "dirty_blocks=%lld\n",
1930 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1931 printk(KERN_EMERG "Block reservation details\n");
1932 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1933 EXT4_I(inode)->i_reserved_data_blocks);
1934 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1935 EXT4_I(inode)->i_reserved_meta_blocks);
1939 #define EXT4_DELALLOC_RSVED 1
1940 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
1941 struct buffer_head *bh_result, int create)
1944 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1945 loff_t disksize = EXT4_I(inode)->i_disksize;
1946 handle_t *handle = NULL;
1948 handle = ext4_journal_current_handle();
1950 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
1951 bh_result, create, 0, EXT4_DELALLOC_RSVED);
1955 bh_result->b_size = (ret << inode->i_blkbits);
1957 if (ext4_should_order_data(inode)) {
1959 retval = ext4_jbd2_file_inode(handle, inode);
1962 * Failed to add inode for ordered mode. Don't
1969 * Update on-disk size along with block allocation we don't
1970 * use 'extend_disksize' as size may change within already
1971 * allocated block -bzzz
1973 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
1974 if (disksize > i_size_read(inode))
1975 disksize = i_size_read(inode);
1976 if (disksize > EXT4_I(inode)->i_disksize) {
1977 ext4_update_i_disksize(inode, disksize);
1978 ret = ext4_mark_inode_dirty(handle, inode);
1985 * mpage_da_map_blocks - go through given space
1987 * @mpd - bh describing space
1989 * The function skips space we know is already mapped to disk blocks.
1992 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1995 struct buffer_head new;
1999 * We consider only non-mapped and non-allocated blocks
2001 if ((mpd->b_state & (1 << BH_Mapped)) &&
2002 !(mpd->b_state & (1 << BH_Delay)))
2004 new.b_state = mpd->b_state;
2006 new.b_size = mpd->b_size;
2007 next = mpd->b_blocknr;
2009 * If we didn't accumulate anything
2010 * to write simply return
2015 err = ext4_da_get_block_write(mpd->inode, next, &new, 1);
2018 * If get block returns with error we simply
2019 * return. Later writepage will redirty the page and
2020 * writepages will find the dirty page again
2025 if (err == -ENOSPC &&
2026 ext4_count_free_blocks(mpd->inode->i_sb)) {
2032 * get block failure will cause us to loop in
2033 * writepages, because a_ops->writepage won't be able
2034 * to make progress. The page will be redirtied by
2035 * writepage and writepages will again try to write
2038 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2039 "at logical offset %llu with max blocks "
2040 "%zd with error %d\n",
2041 __func__, mpd->inode->i_ino,
2042 (unsigned long long)next,
2043 mpd->b_size >> mpd->inode->i_blkbits, err);
2044 printk(KERN_EMERG "This should not happen.!! "
2045 "Data will be lost\n");
2046 if (err == -ENOSPC) {
2047 ext4_print_free_blocks(mpd->inode);
2049 /* invlaidate all the pages */
2050 ext4_da_block_invalidatepages(mpd, next,
2051 mpd->b_size >> mpd->inode->i_blkbits);
2054 BUG_ON(new.b_size == 0);
2056 if (buffer_new(&new))
2057 __unmap_underlying_blocks(mpd->inode, &new);
2060 * If blocks are delayed marked, we need to
2061 * put actual blocknr and drop delayed bit
2063 if ((mpd->b_state & (1 << BH_Delay)) ||
2064 (mpd->b_state & (1 << BH_Unwritten)))
2065 mpage_put_bnr_to_bhs(mpd, next, &new);
2070 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2071 (1 << BH_Delay) | (1 << BH_Unwritten))
2074 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2076 * @mpd->lbh - extent of blocks
2077 * @logical - logical number of the block in the file
2078 * @bh - bh of the block (used to access block's state)
2080 * the function is used to collect contig. blocks in same state
2082 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2083 sector_t logical, size_t b_size,
2084 unsigned long b_state)
2087 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2089 /* check if thereserved journal credits might overflow */
2090 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2091 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2093 * With non-extent format we are limited by the journal
2094 * credit available. Total credit needed to insert
2095 * nrblocks contiguous blocks is dependent on the
2096 * nrblocks. So limit nrblocks.
2099 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2100 EXT4_MAX_TRANS_DATA) {
2102 * Adding the new buffer_head would make it cross the
2103 * allowed limit for which we have journal credit
2104 * reserved. So limit the new bh->b_size
2106 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2107 mpd->inode->i_blkbits;
2108 /* we will do mpage_da_submit_io in the next loop */
2112 * First block in the extent
2114 if (mpd->b_size == 0) {
2115 mpd->b_blocknr = logical;
2116 mpd->b_size = b_size;
2117 mpd->b_state = b_state & BH_FLAGS;
2121 next = mpd->b_blocknr + nrblocks;
2123 * Can we merge the block to our big extent?
2125 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2126 mpd->b_size += b_size;
2132 * We couldn't merge the block to our extent, so we
2133 * need to flush current extent and start new one
2135 if (mpage_da_map_blocks(mpd) == 0)
2136 mpage_da_submit_io(mpd);
2142 * __mpage_da_writepage - finds extent of pages and blocks
2144 * @page: page to consider
2145 * @wbc: not used, we just follow rules
2148 * The function finds extents of pages and scan them for all blocks.
2150 static int __mpage_da_writepage(struct page *page,
2151 struct writeback_control *wbc, void *data)
2153 struct mpage_da_data *mpd = data;
2154 struct inode *inode = mpd->inode;
2155 struct buffer_head *bh, *head;
2160 * Rest of the page in the page_vec
2161 * redirty then and skip then. We will
2162 * try to to write them again after
2163 * starting a new transaction
2165 redirty_page_for_writepage(wbc, page);
2167 return MPAGE_DA_EXTENT_TAIL;
2170 * Can we merge this page to current extent?
2172 if (mpd->next_page != page->index) {
2174 * Nope, we can't. So, we map non-allocated blocks
2175 * and start IO on them using writepage()
2177 if (mpd->next_page != mpd->first_page) {
2178 if (mpage_da_map_blocks(mpd) == 0)
2179 mpage_da_submit_io(mpd);
2181 * skip rest of the page in the page_vec
2184 redirty_page_for_writepage(wbc, page);
2186 return MPAGE_DA_EXTENT_TAIL;
2190 * Start next extent of pages ...
2192 mpd->first_page = page->index;
2202 mpd->next_page = page->index + 1;
2203 logical = (sector_t) page->index <<
2204 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2206 if (!page_has_buffers(page)) {
2207 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2208 (1 << BH_Dirty) | (1 << BH_Uptodate));
2210 return MPAGE_DA_EXTENT_TAIL;
2213 * Page with regular buffer heads, just add all dirty ones
2215 head = page_buffers(page);
2218 BUG_ON(buffer_locked(bh));
2220 * We need to try to allocate
2221 * unmapped blocks in the same page.
2222 * Otherwise we won't make progress
2223 * with the page in ext4_da_writepage
2225 if (buffer_dirty(bh) &&
2226 (!buffer_mapped(bh) || buffer_delay(bh))) {
2227 mpage_add_bh_to_extent(mpd, logical,
2231 return MPAGE_DA_EXTENT_TAIL;
2232 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2234 * mapped dirty buffer. We need to update
2235 * the b_state because we look at
2236 * b_state in mpage_da_map_blocks. We don't
2237 * update b_size because if we find an
2238 * unmapped buffer_head later we need to
2239 * use the b_state flag of that buffer_head.
2241 if (mpd->b_size == 0)
2242 mpd->b_state = bh->b_state & BH_FLAGS;
2245 } while ((bh = bh->b_this_page) != head);
2252 * this is a special callback for ->write_begin() only
2253 * it's intention is to return mapped block or reserve space
2255 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2256 struct buffer_head *bh_result, int create)
2260 BUG_ON(create == 0);
2261 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2264 * first, we need to know whether the block is allocated already
2265 * preallocated blocks are unmapped but should treated
2266 * the same as allocated blocks.
2268 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2269 if ((ret == 0) && !buffer_delay(bh_result)) {
2270 /* the block isn't (pre)allocated yet, let's reserve space */
2272 * XXX: __block_prepare_write() unmaps passed block,
2275 ret = ext4_da_reserve_space(inode, 1);
2277 /* not enough space to reserve */
2280 map_bh(bh_result, inode->i_sb, 0);
2281 set_buffer_new(bh_result);
2282 set_buffer_delay(bh_result);
2283 } else if (ret > 0) {
2284 bh_result->b_size = (ret << inode->i_blkbits);
2291 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2294 * unmapped buffer is possible for holes.
2295 * delay buffer is possible with delayed allocation
2297 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2300 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2301 struct buffer_head *bh_result, int create)
2304 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2307 * we don't want to do block allocation in writepage
2308 * so call get_block_wrap with create = 0
2310 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2311 bh_result, 0, 0, 0);
2313 bh_result->b_size = (ret << inode->i_blkbits);
2320 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2321 * get called via journal_submit_inode_data_buffers (no journal handle)
2322 * get called via shrink_page_list via pdflush (no journal handle)
2323 * or grab_page_cache when doing write_begin (have journal handle)
2325 static int ext4_da_writepage(struct page *page,
2326 struct writeback_control *wbc)
2331 struct buffer_head *page_bufs;
2332 struct inode *inode = page->mapping->host;
2334 trace_mark(ext4_da_writepage,
2335 "dev %s ino %lu page_index %lu",
2336 inode->i_sb->s_id, inode->i_ino, page->index);
2337 size = i_size_read(inode);
2338 if (page->index == size >> PAGE_CACHE_SHIFT)
2339 len = size & ~PAGE_CACHE_MASK;
2341 len = PAGE_CACHE_SIZE;
2343 if (page_has_buffers(page)) {
2344 page_bufs = page_buffers(page);
2345 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2346 ext4_bh_unmapped_or_delay)) {
2348 * We don't want to do block allocation
2349 * So redirty the page and return
2350 * We may reach here when we do a journal commit
2351 * via journal_submit_inode_data_buffers.
2352 * If we don't have mapping block we just ignore
2353 * them. We can also reach here via shrink_page_list
2355 redirty_page_for_writepage(wbc, page);
2361 * The test for page_has_buffers() is subtle:
2362 * We know the page is dirty but it lost buffers. That means
2363 * that at some moment in time after write_begin()/write_end()
2364 * has been called all buffers have been clean and thus they
2365 * must have been written at least once. So they are all
2366 * mapped and we can happily proceed with mapping them
2367 * and writing the page.
2369 * Try to initialize the buffer_heads and check whether
2370 * all are mapped and non delay. We don't want to
2371 * do block allocation here.
2373 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2374 ext4_normal_get_block_write);
2376 page_bufs = page_buffers(page);
2377 /* check whether all are mapped and non delay */
2378 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2379 ext4_bh_unmapped_or_delay)) {
2380 redirty_page_for_writepage(wbc, page);
2386 * We can't do block allocation here
2387 * so just redity the page and unlock
2390 redirty_page_for_writepage(wbc, page);
2394 /* now mark the buffer_heads as dirty and uptodate */
2395 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2398 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2399 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2401 ret = block_write_full_page(page,
2402 ext4_normal_get_block_write,
2409 * This is called via ext4_da_writepages() to
2410 * calulate the total number of credits to reserve to fit
2411 * a single extent allocation into a single transaction,
2412 * ext4_da_writpeages() will loop calling this before
2413 * the block allocation.
2416 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2418 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2421 * With non-extent format the journal credit needed to
2422 * insert nrblocks contiguous block is dependent on
2423 * number of contiguous block. So we will limit
2424 * number of contiguous block to a sane value
2426 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2427 (max_blocks > EXT4_MAX_TRANS_DATA))
2428 max_blocks = EXT4_MAX_TRANS_DATA;
2430 return ext4_chunk_trans_blocks(inode, max_blocks);
2433 static int ext4_da_writepages(struct address_space *mapping,
2434 struct writeback_control *wbc)
2437 int range_whole = 0;
2438 handle_t *handle = NULL;
2439 struct mpage_da_data mpd;
2440 struct inode *inode = mapping->host;
2441 int no_nrwrite_index_update;
2442 int pages_written = 0;
2444 int range_cyclic, cycled = 1, io_done = 0;
2445 int needed_blocks, ret = 0, nr_to_writebump = 0;
2446 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2448 trace_mark(ext4_da_writepages,
2449 "dev %s ino %lu nr_t_write %ld "
2450 "pages_skipped %ld range_start %llu "
2451 "range_end %llu nonblocking %d "
2452 "for_kupdate %d for_reclaim %d "
2453 "for_writepages %d range_cyclic %d",
2454 inode->i_sb->s_id, inode->i_ino,
2455 wbc->nr_to_write, wbc->pages_skipped,
2456 (unsigned long long) wbc->range_start,
2457 (unsigned long long) wbc->range_end,
2458 wbc->nonblocking, wbc->for_kupdate,
2459 wbc->for_reclaim, wbc->for_writepages,
2463 * No pages to write? This is mainly a kludge to avoid starting
2464 * a transaction for special inodes like journal inode on last iput()
2465 * because that could violate lock ordering on umount
2467 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2471 * If the filesystem has aborted, it is read-only, so return
2472 * right away instead of dumping stack traces later on that
2473 * will obscure the real source of the problem. We test
2474 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2475 * the latter could be true if the filesystem is mounted
2476 * read-only, and in that case, ext4_da_writepages should
2477 * *never* be called, so if that ever happens, we would want
2480 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2484 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2485 * This make sure small files blocks are allocated in
2486 * single attempt. This ensure that small files
2487 * get less fragmented.
2489 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2490 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2491 wbc->nr_to_write = sbi->s_mb_stream_request;
2493 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2496 range_cyclic = wbc->range_cyclic;
2497 if (wbc->range_cyclic) {
2498 index = mapping->writeback_index;
2501 wbc->range_start = index << PAGE_CACHE_SHIFT;
2502 wbc->range_end = LLONG_MAX;
2503 wbc->range_cyclic = 0;
2505 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2508 mpd.inode = mapping->host;
2511 * we don't want write_cache_pages to update
2512 * nr_to_write and writeback_index
2514 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2515 wbc->no_nrwrite_index_update = 1;
2516 pages_skipped = wbc->pages_skipped;
2519 while (!ret && wbc->nr_to_write > 0) {
2522 * we insert one extent at a time. So we need
2523 * credit needed for single extent allocation.
2524 * journalled mode is currently not supported
2527 BUG_ON(ext4_should_journal_data(inode));
2528 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2530 /* start a new transaction*/
2531 handle = ext4_journal_start(inode, needed_blocks);
2532 if (IS_ERR(handle)) {
2533 ret = PTR_ERR(handle);
2534 printk(KERN_CRIT "%s: jbd2_start: "
2535 "%ld pages, ino %lu; err %d\n", __func__,
2536 wbc->nr_to_write, inode->i_ino, ret);
2538 goto out_writepages;
2542 * Now call __mpage_da_writepage to find the next
2543 * contiguous region of logical blocks that need
2544 * blocks to be allocated by ext4. We don't actually
2545 * submit the blocks for I/O here, even though
2546 * write_cache_pages thinks it will, and will set the
2547 * pages as clean for write before calling
2548 * __mpage_da_writepage().
2556 mpd.pages_written = 0;
2558 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2561 * If we have a contigous extent of pages and we
2562 * haven't done the I/O yet, map the blocks and submit
2565 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2566 if (mpage_da_map_blocks(&mpd) == 0)
2567 mpage_da_submit_io(&mpd);
2569 ret = MPAGE_DA_EXTENT_TAIL;
2571 wbc->nr_to_write -= mpd.pages_written;
2573 ext4_journal_stop(handle);
2575 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2576 /* commit the transaction which would
2577 * free blocks released in the transaction
2580 jbd2_journal_force_commit_nested(sbi->s_journal);
2581 wbc->pages_skipped = pages_skipped;
2583 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2585 * got one extent now try with
2588 pages_written += mpd.pages_written;
2589 wbc->pages_skipped = pages_skipped;
2592 } else if (wbc->nr_to_write)
2594 * There is no more writeout needed
2595 * or we requested for a noblocking writeout
2596 * and we found the device congested
2600 if (!io_done && !cycled) {
2603 wbc->range_start = index << PAGE_CACHE_SHIFT;
2604 wbc->range_end = mapping->writeback_index - 1;
2607 if (pages_skipped != wbc->pages_skipped)
2608 printk(KERN_EMERG "This should not happen leaving %s "
2609 "with nr_to_write = %ld ret = %d\n",
2610 __func__, wbc->nr_to_write, ret);
2613 index += pages_written;
2614 wbc->range_cyclic = range_cyclic;
2615 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2617 * set the writeback_index so that range_cyclic
2618 * mode will write it back later
2620 mapping->writeback_index = index;
2623 if (!no_nrwrite_index_update)
2624 wbc->no_nrwrite_index_update = 0;
2625 wbc->nr_to_write -= nr_to_writebump;
2626 trace_mark(ext4_da_writepage_result,
2627 "dev %s ino %lu ret %d pages_written %d "
2628 "pages_skipped %ld congestion %d "
2629 "more_io %d no_nrwrite_index_update %d",
2630 inode->i_sb->s_id, inode->i_ino, ret,
2631 pages_written, wbc->pages_skipped,
2632 wbc->encountered_congestion, wbc->more_io,
2633 wbc->no_nrwrite_index_update);
2637 #define FALL_BACK_TO_NONDELALLOC 1
2638 static int ext4_nonda_switch(struct super_block *sb)
2640 s64 free_blocks, dirty_blocks;
2641 struct ext4_sb_info *sbi = EXT4_SB(sb);
2644 * switch to non delalloc mode if we are running low
2645 * on free block. The free block accounting via percpu
2646 * counters can get slightly wrong with percpu_counter_batch getting
2647 * accumulated on each CPU without updating global counters
2648 * Delalloc need an accurate free block accounting. So switch
2649 * to non delalloc when we are near to error range.
2651 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2652 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2653 if (2 * free_blocks < 3 * dirty_blocks ||
2654 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2656 * free block count is less that 150% of dirty blocks
2657 * or free blocks is less that watermark
2664 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2665 loff_t pos, unsigned len, unsigned flags,
2666 struct page **pagep, void **fsdata)
2668 int ret, retries = 0;
2672 struct inode *inode = mapping->host;
2675 index = pos >> PAGE_CACHE_SHIFT;
2676 from = pos & (PAGE_CACHE_SIZE - 1);
2679 if (ext4_nonda_switch(inode->i_sb)) {
2680 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2681 return ext4_write_begin(file, mapping, pos,
2682 len, flags, pagep, fsdata);
2684 *fsdata = (void *)0;
2686 trace_mark(ext4_da_write_begin,
2687 "dev %s ino %lu pos %llu len %u flags %u",
2688 inode->i_sb->s_id, inode->i_ino,
2689 (unsigned long long) pos, len, flags);
2692 * With delayed allocation, we don't log the i_disksize update
2693 * if there is delayed block allocation. But we still need
2694 * to journalling the i_disksize update if writes to the end
2695 * of file which has an already mapped buffer.
2697 handle = ext4_journal_start(inode, 1);
2698 if (IS_ERR(handle)) {
2699 ret = PTR_ERR(handle);
2702 /* We cannot recurse into the filesystem as the transaction is already
2704 flags |= AOP_FLAG_NOFS;
2706 page = grab_cache_page_write_begin(mapping, index, flags);
2708 ext4_journal_stop(handle);
2714 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2715 ext4_da_get_block_prep);
2718 ext4_journal_stop(handle);
2719 page_cache_release(page);
2721 * block_write_begin may have instantiated a few blocks
2722 * outside i_size. Trim these off again. Don't need
2723 * i_size_read because we hold i_mutex.
2725 if (pos + len > inode->i_size)
2726 vmtruncate(inode, inode->i_size);
2729 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2736 * Check if we should update i_disksize
2737 * when write to the end of file but not require block allocation
2739 static int ext4_da_should_update_i_disksize(struct page *page,
2740 unsigned long offset)
2742 struct buffer_head *bh;
2743 struct inode *inode = page->mapping->host;
2747 bh = page_buffers(page);
2748 idx = offset >> inode->i_blkbits;
2750 for (i = 0; i < idx; i++)
2751 bh = bh->b_this_page;
2753 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2758 static int ext4_da_write_end(struct file *file,
2759 struct address_space *mapping,
2760 loff_t pos, unsigned len, unsigned copied,
2761 struct page *page, void *fsdata)
2763 struct inode *inode = mapping->host;
2765 handle_t *handle = ext4_journal_current_handle();
2767 unsigned long start, end;
2768 int write_mode = (int)(unsigned long)fsdata;
2770 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2771 if (ext4_should_order_data(inode)) {
2772 return ext4_ordered_write_end(file, mapping, pos,
2773 len, copied, page, fsdata);
2774 } else if (ext4_should_writeback_data(inode)) {
2775 return ext4_writeback_write_end(file, mapping, pos,
2776 len, copied, page, fsdata);
2782 trace_mark(ext4_da_write_end,
2783 "dev %s ino %lu pos %llu len %u copied %u",
2784 inode->i_sb->s_id, inode->i_ino,
2785 (unsigned long long) pos, len, copied);
2786 start = pos & (PAGE_CACHE_SIZE - 1);
2787 end = start + copied - 1;
2790 * generic_write_end() will run mark_inode_dirty() if i_size
2791 * changes. So let's piggyback the i_disksize mark_inode_dirty
2795 new_i_size = pos + copied;
2796 if (new_i_size > EXT4_I(inode)->i_disksize) {
2797 if (ext4_da_should_update_i_disksize(page, end)) {
2798 down_write(&EXT4_I(inode)->i_data_sem);
2799 if (new_i_size > EXT4_I(inode)->i_disksize) {
2801 * Updating i_disksize when extending file
2802 * without needing block allocation
2804 if (ext4_should_order_data(inode))
2805 ret = ext4_jbd2_file_inode(handle,
2808 EXT4_I(inode)->i_disksize = new_i_size;
2810 up_write(&EXT4_I(inode)->i_data_sem);
2811 /* We need to mark inode dirty even if
2812 * new_i_size is less that inode->i_size
2813 * bu greater than i_disksize.(hint delalloc)
2815 ext4_mark_inode_dirty(handle, inode);
2818 ret2 = generic_write_end(file, mapping, pos, len, copied,
2823 ret2 = ext4_journal_stop(handle);
2827 return ret ? ret : copied;
2830 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2833 * Drop reserved blocks
2835 BUG_ON(!PageLocked(page));
2836 if (!page_has_buffers(page))
2839 ext4_da_page_release_reservation(page, offset);
2842 ext4_invalidatepage(page, offset);
2848 * Force all delayed allocation blocks to be allocated for a given inode.
2850 int ext4_alloc_da_blocks(struct inode *inode)
2852 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2853 !EXT4_I(inode)->i_reserved_meta_blocks)
2857 * We do something simple for now. The filemap_flush() will
2858 * also start triggering a write of the data blocks, which is
2859 * not strictly speaking necessary (and for users of
2860 * laptop_mode, not even desirable). However, to do otherwise
2861 * would require replicating code paths in:
2863 * ext4_da_writepages() ->
2864 * write_cache_pages() ---> (via passed in callback function)
2865 * __mpage_da_writepage() -->
2866 * mpage_add_bh_to_extent()
2867 * mpage_da_map_blocks()
2869 * The problem is that write_cache_pages(), located in
2870 * mm/page-writeback.c, marks pages clean in preparation for
2871 * doing I/O, which is not desirable if we're not planning on
2874 * We could call write_cache_pages(), and then redirty all of
2875 * the pages by calling redirty_page_for_writeback() but that
2876 * would be ugly in the extreme. So instead we would need to
2877 * replicate parts of the code in the above functions,
2878 * simplifying them becuase we wouldn't actually intend to
2879 * write out the pages, but rather only collect contiguous
2880 * logical block extents, call the multi-block allocator, and
2881 * then update the buffer heads with the block allocations.
2883 * For now, though, we'll cheat by calling filemap_flush(),
2884 * which will map the blocks, and start the I/O, but not
2885 * actually wait for the I/O to complete.
2887 return filemap_flush(inode->i_mapping);
2891 * bmap() is special. It gets used by applications such as lilo and by
2892 * the swapper to find the on-disk block of a specific piece of data.
2894 * Naturally, this is dangerous if the block concerned is still in the
2895 * journal. If somebody makes a swapfile on an ext4 data-journaling
2896 * filesystem and enables swap, then they may get a nasty shock when the
2897 * data getting swapped to that swapfile suddenly gets overwritten by
2898 * the original zero's written out previously to the journal and
2899 * awaiting writeback in the kernel's buffer cache.
2901 * So, if we see any bmap calls here on a modified, data-journaled file,
2902 * take extra steps to flush any blocks which might be in the cache.
2904 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2906 struct inode *inode = mapping->host;
2910 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2911 test_opt(inode->i_sb, DELALLOC)) {
2913 * With delalloc we want to sync the file
2914 * so that we can make sure we allocate
2917 filemap_write_and_wait(mapping);
2920 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2922 * This is a REALLY heavyweight approach, but the use of
2923 * bmap on dirty files is expected to be extremely rare:
2924 * only if we run lilo or swapon on a freshly made file
2925 * do we expect this to happen.
2927 * (bmap requires CAP_SYS_RAWIO so this does not
2928 * represent an unprivileged user DOS attack --- we'd be
2929 * in trouble if mortal users could trigger this path at
2932 * NB. EXT4_STATE_JDATA is not set on files other than
2933 * regular files. If somebody wants to bmap a directory
2934 * or symlink and gets confused because the buffer
2935 * hasn't yet been flushed to disk, they deserve
2936 * everything they get.
2939 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2940 journal = EXT4_JOURNAL(inode);
2941 jbd2_journal_lock_updates(journal);
2942 err = jbd2_journal_flush(journal);
2943 jbd2_journal_unlock_updates(journal);
2949 return generic_block_bmap(mapping, block, ext4_get_block);
2952 static int bget_one(handle_t *handle, struct buffer_head *bh)
2958 static int bput_one(handle_t *handle, struct buffer_head *bh)
2965 * Note that we don't need to start a transaction unless we're journaling data
2966 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2967 * need to file the inode to the transaction's list in ordered mode because if
2968 * we are writing back data added by write(), the inode is already there and if
2969 * we are writing back data modified via mmap(), noone guarantees in which
2970 * transaction the data will hit the disk. In case we are journaling data, we
2971 * cannot start transaction directly because transaction start ranks above page
2972 * lock so we have to do some magic.
2974 * In all journaling modes block_write_full_page() will start the I/O.
2978 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2983 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2985 * Same applies to ext4_get_block(). We will deadlock on various things like
2986 * lock_journal and i_data_sem
2988 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2991 * 16May01: If we're reentered then journal_current_handle() will be
2992 * non-zero. We simply *return*.
2994 * 1 July 2001: @@@ FIXME:
2995 * In journalled data mode, a data buffer may be metadata against the
2996 * current transaction. But the same file is part of a shared mapping
2997 * and someone does a writepage() on it.
2999 * We will move the buffer onto the async_data list, but *after* it has
3000 * been dirtied. So there's a small window where we have dirty data on
3003 * Note that this only applies to the last partial page in the file. The
3004 * bit which block_write_full_page() uses prepare/commit for. (That's
3005 * broken code anyway: it's wrong for msync()).
3007 * It's a rare case: affects the final partial page, for journalled data
3008 * where the file is subject to bith write() and writepage() in the same
3009 * transction. To fix it we'll need a custom block_write_full_page().
3010 * We'll probably need that anyway for journalling writepage() output.
3012 * We don't honour synchronous mounts for writepage(). That would be
3013 * disastrous. Any write() or metadata operation will sync the fs for
3017 static int __ext4_normal_writepage(struct page *page,
3018 struct writeback_control *wbc)
3020 struct inode *inode = page->mapping->host;
3022 if (test_opt(inode->i_sb, NOBH))
3023 return nobh_writepage(page,
3024 ext4_normal_get_block_write, wbc);
3026 return block_write_full_page(page,
3027 ext4_normal_get_block_write,
3031 static int ext4_normal_writepage(struct page *page,
3032 struct writeback_control *wbc)
3034 struct inode *inode = page->mapping->host;
3035 loff_t size = i_size_read(inode);
3038 trace_mark(ext4_normal_writepage,
3039 "dev %s ino %lu page_index %lu",
3040 inode->i_sb->s_id, inode->i_ino, page->index);
3041 J_ASSERT(PageLocked(page));
3042 if (page->index == size >> PAGE_CACHE_SHIFT)
3043 len = size & ~PAGE_CACHE_MASK;
3045 len = PAGE_CACHE_SIZE;
3047 if (page_has_buffers(page)) {
3048 /* if page has buffers it should all be mapped
3049 * and allocated. If there are not buffers attached
3050 * to the page we know the page is dirty but it lost
3051 * buffers. That means that at some moment in time
3052 * after write_begin() / write_end() has been called
3053 * all buffers have been clean and thus they must have been
3054 * written at least once. So they are all mapped and we can
3055 * happily proceed with mapping them and writing the page.
3057 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3058 ext4_bh_unmapped_or_delay));
3061 if (!ext4_journal_current_handle())
3062 return __ext4_normal_writepage(page, wbc);
3064 redirty_page_for_writepage(wbc, page);
3069 static int __ext4_journalled_writepage(struct page *page,
3070 struct writeback_control *wbc)
3072 struct address_space *mapping = page->mapping;
3073 struct inode *inode = mapping->host;
3074 struct buffer_head *page_bufs;
3075 handle_t *handle = NULL;
3079 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3080 ext4_normal_get_block_write);
3084 page_bufs = page_buffers(page);
3085 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3087 /* As soon as we unlock the page, it can go away, but we have
3088 * references to buffers so we are safe */
3091 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3092 if (IS_ERR(handle)) {
3093 ret = PTR_ERR(handle);
3097 ret = walk_page_buffers(handle, page_bufs, 0,
3098 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3100 err = walk_page_buffers(handle, page_bufs, 0,
3101 PAGE_CACHE_SIZE, NULL, write_end_fn);
3104 err = ext4_journal_stop(handle);
3108 walk_page_buffers(handle, page_bufs, 0,
3109 PAGE_CACHE_SIZE, NULL, bput_one);
3110 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3119 static int ext4_journalled_writepage(struct page *page,
3120 struct writeback_control *wbc)
3122 struct inode *inode = page->mapping->host;
3123 loff_t size = i_size_read(inode);
3126 trace_mark(ext4_journalled_writepage,
3127 "dev %s ino %lu page_index %lu",
3128 inode->i_sb->s_id, inode->i_ino, page->index);
3129 J_ASSERT(PageLocked(page));
3130 if (page->index == size >> PAGE_CACHE_SHIFT)
3131 len = size & ~PAGE_CACHE_MASK;
3133 len = PAGE_CACHE_SIZE;
3135 if (page_has_buffers(page)) {
3136 /* if page has buffers it should all be mapped
3137 * and allocated. If there are not buffers attached
3138 * to the page we know the page is dirty but it lost
3139 * buffers. That means that at some moment in time
3140 * after write_begin() / write_end() has been called
3141 * all buffers have been clean and thus they must have been
3142 * written at least once. So they are all mapped and we can
3143 * happily proceed with mapping them and writing the page.
3145 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3146 ext4_bh_unmapped_or_delay));
3149 if (ext4_journal_current_handle())
3152 if (PageChecked(page)) {
3154 * It's mmapped pagecache. Add buffers and journal it. There
3155 * doesn't seem much point in redirtying the page here.
3157 ClearPageChecked(page);
3158 return __ext4_journalled_writepage(page, wbc);
3161 * It may be a page full of checkpoint-mode buffers. We don't
3162 * really know unless we go poke around in the buffer_heads.
3163 * But block_write_full_page will do the right thing.
3165 return block_write_full_page(page,
3166 ext4_normal_get_block_write,
3170 redirty_page_for_writepage(wbc, page);
3175 static int ext4_readpage(struct file *file, struct page *page)
3177 return mpage_readpage(page, ext4_get_block);
3181 ext4_readpages(struct file *file, struct address_space *mapping,
3182 struct list_head *pages, unsigned nr_pages)
3184 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3187 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3189 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3192 * If it's a full truncate we just forget about the pending dirtying
3195 ClearPageChecked(page);
3198 jbd2_journal_invalidatepage(journal, page, offset);
3200 block_invalidatepage(page, offset);
3203 static int ext4_releasepage(struct page *page, gfp_t wait)
3205 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3207 WARN_ON(PageChecked(page));
3208 if (!page_has_buffers(page))
3211 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3213 return try_to_free_buffers(page);
3217 * If the O_DIRECT write will extend the file then add this inode to the
3218 * orphan list. So recovery will truncate it back to the original size
3219 * if the machine crashes during the write.
3221 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3222 * crashes then stale disk data _may_ be exposed inside the file. But current
3223 * VFS code falls back into buffered path in that case so we are safe.
3225 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3226 const struct iovec *iov, loff_t offset,
3227 unsigned long nr_segs)
3229 struct file *file = iocb->ki_filp;
3230 struct inode *inode = file->f_mapping->host;
3231 struct ext4_inode_info *ei = EXT4_I(inode);
3235 size_t count = iov_length(iov, nr_segs);
3238 loff_t final_size = offset + count;
3240 if (final_size > inode->i_size) {
3241 /* Credits for sb + inode write */
3242 handle = ext4_journal_start(inode, 2);
3243 if (IS_ERR(handle)) {
3244 ret = PTR_ERR(handle);
3247 ret = ext4_orphan_add(handle, inode);
3249 ext4_journal_stop(handle);
3253 ei->i_disksize = inode->i_size;
3254 ext4_journal_stop(handle);
3258 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3260 ext4_get_block, NULL);
3265 /* Credits for sb + inode write */
3266 handle = ext4_journal_start(inode, 2);
3267 if (IS_ERR(handle)) {
3268 /* This is really bad luck. We've written the data
3269 * but cannot extend i_size. Bail out and pretend
3270 * the write failed... */
3271 ret = PTR_ERR(handle);
3275 ext4_orphan_del(handle, inode);
3277 loff_t end = offset + ret;
3278 if (end > inode->i_size) {
3279 ei->i_disksize = end;
3280 i_size_write(inode, end);
3282 * We're going to return a positive `ret'
3283 * here due to non-zero-length I/O, so there's
3284 * no way of reporting error returns from
3285 * ext4_mark_inode_dirty() to userspace. So
3288 ext4_mark_inode_dirty(handle, inode);
3291 err = ext4_journal_stop(handle);
3300 * Pages can be marked dirty completely asynchronously from ext4's journalling
3301 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3302 * much here because ->set_page_dirty is called under VFS locks. The page is
3303 * not necessarily locked.
3305 * We cannot just dirty the page and leave attached buffers clean, because the
3306 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3307 * or jbddirty because all the journalling code will explode.
3309 * So what we do is to mark the page "pending dirty" and next time writepage
3310 * is called, propagate that into the buffers appropriately.
3312 static int ext4_journalled_set_page_dirty(struct page *page)
3314 SetPageChecked(page);
3315 return __set_page_dirty_nobuffers(page);
3318 static const struct address_space_operations ext4_ordered_aops = {
3319 .readpage = ext4_readpage,
3320 .readpages = ext4_readpages,
3321 .writepage = ext4_normal_writepage,
3322 .sync_page = block_sync_page,
3323 .write_begin = ext4_write_begin,
3324 .write_end = ext4_ordered_write_end,
3326 .invalidatepage = ext4_invalidatepage,
3327 .releasepage = ext4_releasepage,
3328 .direct_IO = ext4_direct_IO,
3329 .migratepage = buffer_migrate_page,
3330 .is_partially_uptodate = block_is_partially_uptodate,
3333 static const struct address_space_operations ext4_writeback_aops = {
3334 .readpage = ext4_readpage,
3335 .readpages = ext4_readpages,
3336 .writepage = ext4_normal_writepage,
3337 .sync_page = block_sync_page,
3338 .write_begin = ext4_write_begin,
3339 .write_end = ext4_writeback_write_end,
3341 .invalidatepage = ext4_invalidatepage,
3342 .releasepage = ext4_releasepage,
3343 .direct_IO = ext4_direct_IO,
3344 .migratepage = buffer_migrate_page,
3345 .is_partially_uptodate = block_is_partially_uptodate,
3348 static const struct address_space_operations ext4_journalled_aops = {
3349 .readpage = ext4_readpage,
3350 .readpages = ext4_readpages,
3351 .writepage = ext4_journalled_writepage,
3352 .sync_page = block_sync_page,
3353 .write_begin = ext4_write_begin,
3354 .write_end = ext4_journalled_write_end,
3355 .set_page_dirty = ext4_journalled_set_page_dirty,
3357 .invalidatepage = ext4_invalidatepage,
3358 .releasepage = ext4_releasepage,
3359 .is_partially_uptodate = block_is_partially_uptodate,
3362 static const struct address_space_operations ext4_da_aops = {
3363 .readpage = ext4_readpage,
3364 .readpages = ext4_readpages,
3365 .writepage = ext4_da_writepage,
3366 .writepages = ext4_da_writepages,
3367 .sync_page = block_sync_page,
3368 .write_begin = ext4_da_write_begin,
3369 .write_end = ext4_da_write_end,
3371 .invalidatepage = ext4_da_invalidatepage,
3372 .releasepage = ext4_releasepage,
3373 .direct_IO = ext4_direct_IO,
3374 .migratepage = buffer_migrate_page,
3375 .is_partially_uptodate = block_is_partially_uptodate,
3378 void ext4_set_aops(struct inode *inode)
3380 if (ext4_should_order_data(inode) &&
3381 test_opt(inode->i_sb, DELALLOC))
3382 inode->i_mapping->a_ops = &ext4_da_aops;
3383 else if (ext4_should_order_data(inode))
3384 inode->i_mapping->a_ops = &ext4_ordered_aops;
3385 else if (ext4_should_writeback_data(inode) &&
3386 test_opt(inode->i_sb, DELALLOC))
3387 inode->i_mapping->a_ops = &ext4_da_aops;
3388 else if (ext4_should_writeback_data(inode))
3389 inode->i_mapping->a_ops = &ext4_writeback_aops;
3391 inode->i_mapping->a_ops = &ext4_journalled_aops;
3395 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3396 * up to the end of the block which corresponds to `from'.
3397 * This required during truncate. We need to physically zero the tail end
3398 * of that block so it doesn't yield old data if the file is later grown.
3400 int ext4_block_truncate_page(handle_t *handle,
3401 struct address_space *mapping, loff_t from)
3403 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3404 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3405 unsigned blocksize, length, pos;
3407 struct inode *inode = mapping->host;
3408 struct buffer_head *bh;
3412 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3416 blocksize = inode->i_sb->s_blocksize;
3417 length = blocksize - (offset & (blocksize - 1));
3418 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3421 * For "nobh" option, we can only work if we don't need to
3422 * read-in the page - otherwise we create buffers to do the IO.
3424 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3425 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3426 zero_user(page, offset, length);
3427 set_page_dirty(page);
3431 if (!page_has_buffers(page))
3432 create_empty_buffers(page, blocksize, 0);
3434 /* Find the buffer that contains "offset" */
3435 bh = page_buffers(page);
3437 while (offset >= pos) {
3438 bh = bh->b_this_page;
3444 if (buffer_freed(bh)) {
3445 BUFFER_TRACE(bh, "freed: skip");
3449 if (!buffer_mapped(bh)) {
3450 BUFFER_TRACE(bh, "unmapped");
3451 ext4_get_block(inode, iblock, bh, 0);
3452 /* unmapped? It's a hole - nothing to do */
3453 if (!buffer_mapped(bh)) {
3454 BUFFER_TRACE(bh, "still unmapped");
3459 /* Ok, it's mapped. Make sure it's up-to-date */
3460 if (PageUptodate(page))
3461 set_buffer_uptodate(bh);
3463 if (!buffer_uptodate(bh)) {
3465 ll_rw_block(READ, 1, &bh);
3467 /* Uhhuh. Read error. Complain and punt. */
3468 if (!buffer_uptodate(bh))
3472 if (ext4_should_journal_data(inode)) {
3473 BUFFER_TRACE(bh, "get write access");
3474 err = ext4_journal_get_write_access(handle, bh);
3479 zero_user(page, offset, length);
3481 BUFFER_TRACE(bh, "zeroed end of block");
3484 if (ext4_should_journal_data(inode)) {
3485 err = ext4_handle_dirty_metadata(handle, inode, bh);
3487 if (ext4_should_order_data(inode))
3488 err = ext4_jbd2_file_inode(handle, inode);
3489 mark_buffer_dirty(bh);
3494 page_cache_release(page);
3499 * Probably it should be a library function... search for first non-zero word
3500 * or memcmp with zero_page, whatever is better for particular architecture.
3503 static inline int all_zeroes(__le32 *p, __le32 *q)
3512 * ext4_find_shared - find the indirect blocks for partial truncation.
3513 * @inode: inode in question
3514 * @depth: depth of the affected branch
3515 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3516 * @chain: place to store the pointers to partial indirect blocks
3517 * @top: place to the (detached) top of branch
3519 * This is a helper function used by ext4_truncate().
3521 * When we do truncate() we may have to clean the ends of several
3522 * indirect blocks but leave the blocks themselves alive. Block is
3523 * partially truncated if some data below the new i_size is refered
3524 * from it (and it is on the path to the first completely truncated
3525 * data block, indeed). We have to free the top of that path along
3526 * with everything to the right of the path. Since no allocation
3527 * past the truncation point is possible until ext4_truncate()
3528 * finishes, we may safely do the latter, but top of branch may
3529 * require special attention - pageout below the truncation point
3530 * might try to populate it.
3532 * We atomically detach the top of branch from the tree, store the
3533 * block number of its root in *@top, pointers to buffer_heads of
3534 * partially truncated blocks - in @chain[].bh and pointers to
3535 * their last elements that should not be removed - in
3536 * @chain[].p. Return value is the pointer to last filled element
3539 * The work left to caller to do the actual freeing of subtrees:
3540 * a) free the subtree starting from *@top
3541 * b) free the subtrees whose roots are stored in
3542 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3543 * c) free the subtrees growing from the inode past the @chain[0].
3544 * (no partially truncated stuff there). */
3546 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3547 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3549 Indirect *partial, *p;
3553 /* Make k index the deepest non-null offest + 1 */
3554 for (k = depth; k > 1 && !offsets[k-1]; k--)
3556 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3557 /* Writer: pointers */
3559 partial = chain + k-1;
3561 * If the branch acquired continuation since we've looked at it -
3562 * fine, it should all survive and (new) top doesn't belong to us.
3564 if (!partial->key && *partial->p)
3567 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3570 * OK, we've found the last block that must survive. The rest of our
3571 * branch should be detached before unlocking. However, if that rest
3572 * of branch is all ours and does not grow immediately from the inode
3573 * it's easier to cheat and just decrement partial->p.
3575 if (p == chain + k - 1 && p > chain) {
3579 /* Nope, don't do this in ext4. Must leave the tree intact */
3586 while (partial > p) {
3587 brelse(partial->bh);
3595 * Zero a number of block pointers in either an inode or an indirect block.
3596 * If we restart the transaction we must again get write access to the
3597 * indirect block for further modification.
3599 * We release `count' blocks on disk, but (last - first) may be greater
3600 * than `count' because there can be holes in there.
3602 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3603 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3604 unsigned long count, __le32 *first, __le32 *last)
3607 if (try_to_extend_transaction(handle, inode)) {
3609 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3610 ext4_handle_dirty_metadata(handle, inode, bh);
3612 ext4_mark_inode_dirty(handle, inode);
3613 ext4_journal_test_restart(handle, inode);
3615 BUFFER_TRACE(bh, "retaking write access");
3616 ext4_journal_get_write_access(handle, bh);
3621 * Any buffers which are on the journal will be in memory. We find
3622 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3623 * on them. We've already detached each block from the file, so
3624 * bforget() in jbd2_journal_forget() should be safe.
3626 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3628 for (p = first; p < last; p++) {
3629 u32 nr = le32_to_cpu(*p);
3631 struct buffer_head *tbh;
3634 tbh = sb_find_get_block(inode->i_sb, nr);
3635 ext4_forget(handle, 0, inode, tbh, nr);
3639 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3643 * ext4_free_data - free a list of data blocks
3644 * @handle: handle for this transaction
3645 * @inode: inode we are dealing with
3646 * @this_bh: indirect buffer_head which contains *@first and *@last
3647 * @first: array of block numbers
3648 * @last: points immediately past the end of array
3650 * We are freeing all blocks refered from that array (numbers are stored as
3651 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3653 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3654 * blocks are contiguous then releasing them at one time will only affect one
3655 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3656 * actually use a lot of journal space.
3658 * @this_bh will be %NULL if @first and @last point into the inode's direct
3661 static void ext4_free_data(handle_t *handle, struct inode *inode,
3662 struct buffer_head *this_bh,
3663 __le32 *first, __le32 *last)
3665 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3666 unsigned long count = 0; /* Number of blocks in the run */
3667 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3670 ext4_fsblk_t nr; /* Current block # */
3671 __le32 *p; /* Pointer into inode/ind
3672 for current block */
3675 if (this_bh) { /* For indirect block */
3676 BUFFER_TRACE(this_bh, "get_write_access");
3677 err = ext4_journal_get_write_access(handle, this_bh);
3678 /* Important: if we can't update the indirect pointers
3679 * to the blocks, we can't free them. */
3684 for (p = first; p < last; p++) {
3685 nr = le32_to_cpu(*p);
3687 /* accumulate blocks to free if they're contiguous */
3690 block_to_free_p = p;
3692 } else if (nr == block_to_free + count) {
3695 ext4_clear_blocks(handle, inode, this_bh,
3697 count, block_to_free_p, p);
3699 block_to_free_p = p;
3706 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3707 count, block_to_free_p, p);
3710 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3713 * The buffer head should have an attached journal head at this
3714 * point. However, if the data is corrupted and an indirect
3715 * block pointed to itself, it would have been detached when
3716 * the block was cleared. Check for this instead of OOPSing.
3718 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3719 ext4_handle_dirty_metadata(handle, inode, this_bh);
3721 ext4_error(inode->i_sb, __func__,
3722 "circular indirect block detected, "
3723 "inode=%lu, block=%llu",
3725 (unsigned long long) this_bh->b_blocknr);
3730 * ext4_free_branches - free an array of branches
3731 * @handle: JBD handle for this transaction
3732 * @inode: inode we are dealing with
3733 * @parent_bh: the buffer_head which contains *@first and *@last
3734 * @first: array of block numbers
3735 * @last: pointer immediately past the end of array
3736 * @depth: depth of the branches to free
3738 * We are freeing all blocks refered from these branches (numbers are
3739 * stored as little-endian 32-bit) and updating @inode->i_blocks
3742 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3743 struct buffer_head *parent_bh,
3744 __le32 *first, __le32 *last, int depth)
3749 if (ext4_handle_is_aborted(handle))
3753 struct buffer_head *bh;
3754 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3756 while (--p >= first) {
3757 nr = le32_to_cpu(*p);
3759 continue; /* A hole */
3761 /* Go read the buffer for the next level down */
3762 bh = sb_bread(inode->i_sb, nr);
3765 * A read failure? Report error and clear slot
3769 ext4_error(inode->i_sb, "ext4_free_branches",
3770 "Read failure, inode=%lu, block=%llu",
3775 /* This zaps the entire block. Bottom up. */
3776 BUFFER_TRACE(bh, "free child branches");
3777 ext4_free_branches(handle, inode, bh,
3778 (__le32 *) bh->b_data,
3779 (__le32 *) bh->b_data + addr_per_block,
3783 * We've probably journalled the indirect block several
3784 * times during the truncate. But it's no longer
3785 * needed and we now drop it from the transaction via
3786 * jbd2_journal_revoke().
3788 * That's easy if it's exclusively part of this
3789 * transaction. But if it's part of the committing
3790 * transaction then jbd2_journal_forget() will simply
3791 * brelse() it. That means that if the underlying
3792 * block is reallocated in ext4_get_block(),
3793 * unmap_underlying_metadata() will find this block
3794 * and will try to get rid of it. damn, damn.
3796 * If this block has already been committed to the
3797 * journal, a revoke record will be written. And
3798 * revoke records must be emitted *before* clearing
3799 * this block's bit in the bitmaps.
3801 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3804 * Everything below this this pointer has been
3805 * released. Now let this top-of-subtree go.
3807 * We want the freeing of this indirect block to be
3808 * atomic in the journal with the updating of the
3809 * bitmap block which owns it. So make some room in
3812 * We zero the parent pointer *after* freeing its
3813 * pointee in the bitmaps, so if extend_transaction()
3814 * for some reason fails to put the bitmap changes and
3815 * the release into the same transaction, recovery
3816 * will merely complain about releasing a free block,
3817 * rather than leaking blocks.
3819 if (ext4_handle_is_aborted(handle))
3821 if (try_to_extend_transaction(handle, inode)) {
3822 ext4_mark_inode_dirty(handle, inode);
3823 ext4_journal_test_restart(handle, inode);
3826 ext4_free_blocks(handle, inode, nr, 1, 1);
3830 * The block which we have just freed is
3831 * pointed to by an indirect block: journal it
3833 BUFFER_TRACE(parent_bh, "get_write_access");
3834 if (!ext4_journal_get_write_access(handle,
3837 BUFFER_TRACE(parent_bh,
3838 "call ext4_handle_dirty_metadata");
3839 ext4_handle_dirty_metadata(handle,
3846 /* We have reached the bottom of the tree. */
3847 BUFFER_TRACE(parent_bh, "free data blocks");
3848 ext4_free_data(handle, inode, parent_bh, first, last);
3852 int ext4_can_truncate(struct inode *inode)
3854 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3856 if (S_ISREG(inode->i_mode))
3858 if (S_ISDIR(inode->i_mode))
3860 if (S_ISLNK(inode->i_mode))
3861 return !ext4_inode_is_fast_symlink(inode);
3868 * We block out ext4_get_block() block instantiations across the entire
3869 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3870 * simultaneously on behalf of the same inode.
3872 * As we work through the truncate and commmit bits of it to the journal there
3873 * is one core, guiding principle: the file's tree must always be consistent on
3874 * disk. We must be able to restart the truncate after a crash.
3876 * The file's tree may be transiently inconsistent in memory (although it
3877 * probably isn't), but whenever we close off and commit a journal transaction,
3878 * the contents of (the filesystem + the journal) must be consistent and
3879 * restartable. It's pretty simple, really: bottom up, right to left (although
3880 * left-to-right works OK too).
3882 * Note that at recovery time, journal replay occurs *before* the restart of
3883 * truncate against the orphan inode list.
3885 * The committed inode has the new, desired i_size (which is the same as
3886 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3887 * that this inode's truncate did not complete and it will again call
3888 * ext4_truncate() to have another go. So there will be instantiated blocks
3889 * to the right of the truncation point in a crashed ext4 filesystem. But
3890 * that's fine - as long as they are linked from the inode, the post-crash
3891 * ext4_truncate() run will find them and release them.
3893 void ext4_truncate(struct inode *inode)
3896 struct ext4_inode_info *ei = EXT4_I(inode);
3897 __le32 *i_data = ei->i_data;
3898 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3899 struct address_space *mapping = inode->i_mapping;
3900 ext4_lblk_t offsets[4];
3905 ext4_lblk_t last_block;
3906 unsigned blocksize = inode->i_sb->s_blocksize;
3908 if (!ext4_can_truncate(inode))
3911 if (inode->i_size == 0)
3912 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3914 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3915 ext4_ext_truncate(inode);
3919 handle = start_transaction(inode);
3921 return; /* AKPM: return what? */
3923 last_block = (inode->i_size + blocksize-1)
3924 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3926 if (inode->i_size & (blocksize - 1))
3927 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3930 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3932 goto out_stop; /* error */
3935 * OK. This truncate is going to happen. We add the inode to the
3936 * orphan list, so that if this truncate spans multiple transactions,
3937 * and we crash, we will resume the truncate when the filesystem
3938 * recovers. It also marks the inode dirty, to catch the new size.
3940 * Implication: the file must always be in a sane, consistent
3941 * truncatable state while each transaction commits.
3943 if (ext4_orphan_add(handle, inode))
3947 * From here we block out all ext4_get_block() callers who want to
3948 * modify the block allocation tree.
3950 down_write(&ei->i_data_sem);
3952 ext4_discard_preallocations(inode);
3955 * The orphan list entry will now protect us from any crash which
3956 * occurs before the truncate completes, so it is now safe to propagate
3957 * the new, shorter inode size (held for now in i_size) into the
3958 * on-disk inode. We do this via i_disksize, which is the value which
3959 * ext4 *really* writes onto the disk inode.
3961 ei->i_disksize = inode->i_size;
3963 if (n == 1) { /* direct blocks */
3964 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3965 i_data + EXT4_NDIR_BLOCKS);
3969 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3970 /* Kill the top of shared branch (not detached) */
3972 if (partial == chain) {
3973 /* Shared branch grows from the inode */
3974 ext4_free_branches(handle, inode, NULL,
3975 &nr, &nr+1, (chain+n-1) - partial);
3978 * We mark the inode dirty prior to restart,
3979 * and prior to stop. No need for it here.
3982 /* Shared branch grows from an indirect block */
3983 BUFFER_TRACE(partial->bh, "get_write_access");
3984 ext4_free_branches(handle, inode, partial->bh,
3986 partial->p+1, (chain+n-1) - partial);
3989 /* Clear the ends of indirect blocks on the shared branch */
3990 while (partial > chain) {
3991 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3992 (__le32*)partial->bh->b_data+addr_per_block,
3993 (chain+n-1) - partial);
3994 BUFFER_TRACE(partial->bh, "call brelse");
3995 brelse (partial->bh);
3999 /* Kill the remaining (whole) subtrees */
4000 switch (offsets[0]) {
4002 nr = i_data[EXT4_IND_BLOCK];
4004 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4005 i_data[EXT4_IND_BLOCK] = 0;
4007 case EXT4_IND_BLOCK:
4008 nr = i_data[EXT4_DIND_BLOCK];
4010 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4011 i_data[EXT4_DIND_BLOCK] = 0;
4013 case EXT4_DIND_BLOCK:
4014 nr = i_data[EXT4_TIND_BLOCK];
4016 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4017 i_data[EXT4_TIND_BLOCK] = 0;
4019 case EXT4_TIND_BLOCK:
4023 up_write(&ei->i_data_sem);
4024 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4025 ext4_mark_inode_dirty(handle, inode);
4028 * In a multi-transaction truncate, we only make the final transaction
4032 ext4_handle_sync(handle);
4035 * If this was a simple ftruncate(), and the file will remain alive
4036 * then we need to clear up the orphan record which we created above.
4037 * However, if this was a real unlink then we were called by
4038 * ext4_delete_inode(), and we allow that function to clean up the
4039 * orphan info for us.
4042 ext4_orphan_del(handle, inode);
4044 ext4_journal_stop(handle);
4048 * ext4_get_inode_loc returns with an extra refcount against the inode's
4049 * underlying buffer_head on success. If 'in_mem' is true, we have all
4050 * data in memory that is needed to recreate the on-disk version of this
4053 static int __ext4_get_inode_loc(struct inode *inode,
4054 struct ext4_iloc *iloc, int in_mem)
4056 struct ext4_group_desc *gdp;
4057 struct buffer_head *bh;
4058 struct super_block *sb = inode->i_sb;
4060 int inodes_per_block, inode_offset;
4063 if (!ext4_valid_inum(sb, inode->i_ino))
4066 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4067 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4072 * Figure out the offset within the block group inode table
4074 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4075 inode_offset = ((inode->i_ino - 1) %
4076 EXT4_INODES_PER_GROUP(sb));
4077 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4078 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4080 bh = sb_getblk(sb, block);
4082 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4083 "inode block - inode=%lu, block=%llu",
4084 inode->i_ino, block);
4087 if (!buffer_uptodate(bh)) {
4091 * If the buffer has the write error flag, we have failed
4092 * to write out another inode in the same block. In this
4093 * case, we don't have to read the block because we may
4094 * read the old inode data successfully.
4096 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4097 set_buffer_uptodate(bh);
4099 if (buffer_uptodate(bh)) {
4100 /* someone brought it uptodate while we waited */
4106 * If we have all information of the inode in memory and this
4107 * is the only valid inode in the block, we need not read the
4111 struct buffer_head *bitmap_bh;
4114 start = inode_offset & ~(inodes_per_block - 1);
4116 /* Is the inode bitmap in cache? */
4117 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4122 * If the inode bitmap isn't in cache then the
4123 * optimisation may end up performing two reads instead
4124 * of one, so skip it.
4126 if (!buffer_uptodate(bitmap_bh)) {
4130 for (i = start; i < start + inodes_per_block; i++) {
4131 if (i == inode_offset)
4133 if (ext4_test_bit(i, bitmap_bh->b_data))
4137 if (i == start + inodes_per_block) {
4138 /* all other inodes are free, so skip I/O */
4139 memset(bh->b_data, 0, bh->b_size);
4140 set_buffer_uptodate(bh);
4148 * If we need to do any I/O, try to pre-readahead extra
4149 * blocks from the inode table.
4151 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4152 ext4_fsblk_t b, end, table;
4155 table = ext4_inode_table(sb, gdp);
4156 /* s_inode_readahead_blks is always a power of 2 */
4157 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4160 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4161 num = EXT4_INODES_PER_GROUP(sb);
4162 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4163 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4164 num -= ext4_itable_unused_count(sb, gdp);
4165 table += num / inodes_per_block;
4169 sb_breadahead(sb, b++);
4173 * There are other valid inodes in the buffer, this inode
4174 * has in-inode xattrs, or we don't have this inode in memory.
4175 * Read the block from disk.
4178 bh->b_end_io = end_buffer_read_sync;
4179 submit_bh(READ_META, bh);
4181 if (!buffer_uptodate(bh)) {
4182 ext4_error(sb, __func__,
4183 "unable to read inode block - inode=%lu, "
4184 "block=%llu", inode->i_ino, block);
4194 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4196 /* We have all inode data except xattrs in memory here. */
4197 return __ext4_get_inode_loc(inode, iloc,
4198 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4201 void ext4_set_inode_flags(struct inode *inode)
4203 unsigned int flags = EXT4_I(inode)->i_flags;
4205 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4206 if (flags & EXT4_SYNC_FL)
4207 inode->i_flags |= S_SYNC;
4208 if (flags & EXT4_APPEND_FL)
4209 inode->i_flags |= S_APPEND;
4210 if (flags & EXT4_IMMUTABLE_FL)
4211 inode->i_flags |= S_IMMUTABLE;
4212 if (flags & EXT4_NOATIME_FL)
4213 inode->i_flags |= S_NOATIME;
4214 if (flags & EXT4_DIRSYNC_FL)
4215 inode->i_flags |= S_DIRSYNC;
4218 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4219 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4221 unsigned int flags = ei->vfs_inode.i_flags;
4223 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4224 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4226 ei->i_flags |= EXT4_SYNC_FL;
4227 if (flags & S_APPEND)
4228 ei->i_flags |= EXT4_APPEND_FL;
4229 if (flags & S_IMMUTABLE)
4230 ei->i_flags |= EXT4_IMMUTABLE_FL;
4231 if (flags & S_NOATIME)
4232 ei->i_flags |= EXT4_NOATIME_FL;
4233 if (flags & S_DIRSYNC)
4234 ei->i_flags |= EXT4_DIRSYNC_FL;
4236 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4237 struct ext4_inode_info *ei)
4240 struct inode *inode = &(ei->vfs_inode);
4241 struct super_block *sb = inode->i_sb;
4243 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4244 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4245 /* we are using combined 48 bit field */
4246 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4247 le32_to_cpu(raw_inode->i_blocks_lo);
4248 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4249 /* i_blocks represent file system block size */
4250 return i_blocks << (inode->i_blkbits - 9);
4255 return le32_to_cpu(raw_inode->i_blocks_lo);
4259 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4261 struct ext4_iloc iloc;
4262 struct ext4_inode *raw_inode;
4263 struct ext4_inode_info *ei;
4264 struct buffer_head *bh;
4265 struct inode *inode;
4269 inode = iget_locked(sb, ino);
4271 return ERR_PTR(-ENOMEM);
4272 if (!(inode->i_state & I_NEW))
4276 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4277 ei->i_acl = EXT4_ACL_NOT_CACHED;
4278 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4281 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4285 raw_inode = ext4_raw_inode(&iloc);
4286 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4287 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4288 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4289 if (!(test_opt(inode->i_sb, NO_UID32))) {
4290 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4291 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4293 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4296 ei->i_dir_start_lookup = 0;
4297 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4298 /* We now have enough fields to check if the inode was active or not.
4299 * This is needed because nfsd might try to access dead inodes
4300 * the test is that same one that e2fsck uses
4301 * NeilBrown 1999oct15
4303 if (inode->i_nlink == 0) {
4304 if (inode->i_mode == 0 ||
4305 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4306 /* this inode is deleted */
4311 /* The only unlinked inodes we let through here have
4312 * valid i_mode and are being read by the orphan
4313 * recovery code: that's fine, we're about to complete
4314 * the process of deleting those. */
4316 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4317 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4318 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4319 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4320 cpu_to_le32(EXT4_OS_HURD)) {
4322 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4324 inode->i_size = ext4_isize(raw_inode);
4325 ei->i_disksize = inode->i_size;
4326 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4327 ei->i_block_group = iloc.block_group;
4328 ei->i_last_alloc_group = ~0;
4330 * NOTE! The in-memory inode i_data array is in little-endian order
4331 * even on big-endian machines: we do NOT byteswap the block numbers!
4333 for (block = 0; block < EXT4_N_BLOCKS; block++)
4334 ei->i_data[block] = raw_inode->i_block[block];
4335 INIT_LIST_HEAD(&ei->i_orphan);
4337 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4338 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4339 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4340 EXT4_INODE_SIZE(inode->i_sb)) {
4345 if (ei->i_extra_isize == 0) {
4346 /* The extra space is currently unused. Use it. */
4347 ei->i_extra_isize = sizeof(struct ext4_inode) -
4348 EXT4_GOOD_OLD_INODE_SIZE;
4350 __le32 *magic = (void *)raw_inode +
4351 EXT4_GOOD_OLD_INODE_SIZE +
4353 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4354 ei->i_state |= EXT4_STATE_XATTR;
4357 ei->i_extra_isize = 0;
4359 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4360 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4361 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4362 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4364 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4365 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4366 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4368 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4371 if (ei->i_flags & EXT4_EXTENTS_FL) {
4372 /* Validate extent which is part of inode */
4373 ret = ext4_ext_check_inode(inode);
4381 if (S_ISREG(inode->i_mode)) {
4382 inode->i_op = &ext4_file_inode_operations;
4383 inode->i_fop = &ext4_file_operations;
4384 ext4_set_aops(inode);
4385 } else if (S_ISDIR(inode->i_mode)) {
4386 inode->i_op = &ext4_dir_inode_operations;
4387 inode->i_fop = &ext4_dir_operations;
4388 } else if (S_ISLNK(inode->i_mode)) {
4389 if (ext4_inode_is_fast_symlink(inode)) {
4390 inode->i_op = &ext4_fast_symlink_inode_operations;
4391 nd_terminate_link(ei->i_data, inode->i_size,
4392 sizeof(ei->i_data) - 1);
4394 inode->i_op = &ext4_symlink_inode_operations;
4395 ext4_set_aops(inode);
4398 inode->i_op = &ext4_special_inode_operations;
4399 if (raw_inode->i_block[0])
4400 init_special_inode(inode, inode->i_mode,
4401 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4403 init_special_inode(inode, inode->i_mode,
4404 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4407 ext4_set_inode_flags(inode);
4408 unlock_new_inode(inode);
4413 return ERR_PTR(ret);
4416 static int ext4_inode_blocks_set(handle_t *handle,
4417 struct ext4_inode *raw_inode,
4418 struct ext4_inode_info *ei)
4420 struct inode *inode = &(ei->vfs_inode);
4421 u64 i_blocks = inode->i_blocks;
4422 struct super_block *sb = inode->i_sb;
4424 if (i_blocks <= ~0U) {
4426 * i_blocks can be represnted in a 32 bit variable
4427 * as multiple of 512 bytes
4429 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4430 raw_inode->i_blocks_high = 0;
4431 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4434 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4437 if (i_blocks <= 0xffffffffffffULL) {
4439 * i_blocks can be represented in a 48 bit variable
4440 * as multiple of 512 bytes
4442 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4443 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4444 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4446 ei->i_flags |= EXT4_HUGE_FILE_FL;
4447 /* i_block is stored in file system block size */
4448 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4449 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4450 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4456 * Post the struct inode info into an on-disk inode location in the
4457 * buffer-cache. This gobbles the caller's reference to the
4458 * buffer_head in the inode location struct.
4460 * The caller must have write access to iloc->bh.
4462 static int ext4_do_update_inode(handle_t *handle,
4463 struct inode *inode,
4464 struct ext4_iloc *iloc)
4466 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4467 struct ext4_inode_info *ei = EXT4_I(inode);
4468 struct buffer_head *bh = iloc->bh;
4469 int err = 0, rc, block;
4471 /* For fields not not tracking in the in-memory inode,
4472 * initialise them to zero for new inodes. */
4473 if (ei->i_state & EXT4_STATE_NEW)
4474 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4476 ext4_get_inode_flags(ei);
4477 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4478 if (!(test_opt(inode->i_sb, NO_UID32))) {
4479 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4480 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4482 * Fix up interoperability with old kernels. Otherwise, old inodes get
4483 * re-used with the upper 16 bits of the uid/gid intact
4486 raw_inode->i_uid_high =
4487 cpu_to_le16(high_16_bits(inode->i_uid));
4488 raw_inode->i_gid_high =
4489 cpu_to_le16(high_16_bits(inode->i_gid));
4491 raw_inode->i_uid_high = 0;
4492 raw_inode->i_gid_high = 0;
4495 raw_inode->i_uid_low =
4496 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4497 raw_inode->i_gid_low =
4498 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4499 raw_inode->i_uid_high = 0;
4500 raw_inode->i_gid_high = 0;
4502 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4504 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4505 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4506 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4507 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4509 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4511 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4512 /* clear the migrate flag in the raw_inode */
4513 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4514 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4515 cpu_to_le32(EXT4_OS_HURD))
4516 raw_inode->i_file_acl_high =
4517 cpu_to_le16(ei->i_file_acl >> 32);
4518 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4519 ext4_isize_set(raw_inode, ei->i_disksize);
4520 if (ei->i_disksize > 0x7fffffffULL) {
4521 struct super_block *sb = inode->i_sb;
4522 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4523 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4524 EXT4_SB(sb)->s_es->s_rev_level ==
4525 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4526 /* If this is the first large file
4527 * created, add a flag to the superblock.
4529 err = ext4_journal_get_write_access(handle,
4530 EXT4_SB(sb)->s_sbh);
4533 ext4_update_dynamic_rev(sb);
4534 EXT4_SET_RO_COMPAT_FEATURE(sb,
4535 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4537 ext4_handle_sync(handle);
4538 err = ext4_handle_dirty_metadata(handle, inode,
4539 EXT4_SB(sb)->s_sbh);
4542 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4543 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4544 if (old_valid_dev(inode->i_rdev)) {
4545 raw_inode->i_block[0] =
4546 cpu_to_le32(old_encode_dev(inode->i_rdev));
4547 raw_inode->i_block[1] = 0;
4549 raw_inode->i_block[0] = 0;
4550 raw_inode->i_block[1] =
4551 cpu_to_le32(new_encode_dev(inode->i_rdev));
4552 raw_inode->i_block[2] = 0;
4554 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4555 raw_inode->i_block[block] = ei->i_data[block];
4557 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4558 if (ei->i_extra_isize) {
4559 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4560 raw_inode->i_version_hi =
4561 cpu_to_le32(inode->i_version >> 32);
4562 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4565 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4566 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4569 ei->i_state &= ~EXT4_STATE_NEW;
4573 ext4_std_error(inode->i_sb, err);
4578 * ext4_write_inode()
4580 * We are called from a few places:
4582 * - Within generic_file_write() for O_SYNC files.
4583 * Here, there will be no transaction running. We wait for any running
4584 * trasnaction to commit.
4586 * - Within sys_sync(), kupdate and such.
4587 * We wait on commit, if tol to.
4589 * - Within prune_icache() (PF_MEMALLOC == true)
4590 * Here we simply return. We can't afford to block kswapd on the
4593 * In all cases it is actually safe for us to return without doing anything,
4594 * because the inode has been copied into a raw inode buffer in
4595 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4598 * Note that we are absolutely dependent upon all inode dirtiers doing the
4599 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4600 * which we are interested.
4602 * It would be a bug for them to not do this. The code:
4604 * mark_inode_dirty(inode)
4606 * inode->i_size = expr;
4608 * is in error because a kswapd-driven write_inode() could occur while
4609 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4610 * will no longer be on the superblock's dirty inode list.
4612 int ext4_write_inode(struct inode *inode, int wait)
4614 if (current->flags & PF_MEMALLOC)
4617 if (ext4_journal_current_handle()) {
4618 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4626 return ext4_force_commit(inode->i_sb);
4629 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4633 mark_buffer_dirty(bh);
4634 if (inode && inode_needs_sync(inode)) {
4635 sync_dirty_buffer(bh);
4636 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4637 ext4_error(inode->i_sb, __func__,
4638 "IO error syncing inode, "
4639 "inode=%lu, block=%llu",
4641 (unsigned long long)bh->b_blocknr);
4651 * Called from notify_change.
4653 * We want to trap VFS attempts to truncate the file as soon as
4654 * possible. In particular, we want to make sure that when the VFS
4655 * shrinks i_size, we put the inode on the orphan list and modify
4656 * i_disksize immediately, so that during the subsequent flushing of
4657 * dirty pages and freeing of disk blocks, we can guarantee that any
4658 * commit will leave the blocks being flushed in an unused state on
4659 * disk. (On recovery, the inode will get truncated and the blocks will
4660 * be freed, so we have a strong guarantee that no future commit will
4661 * leave these blocks visible to the user.)
4663 * Another thing we have to assure is that if we are in ordered mode
4664 * and inode is still attached to the committing transaction, we must
4665 * we start writeout of all the dirty pages which are being truncated.
4666 * This way we are sure that all the data written in the previous
4667 * transaction are already on disk (truncate waits for pages under
4670 * Called with inode->i_mutex down.
4672 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4674 struct inode *inode = dentry->d_inode;
4676 const unsigned int ia_valid = attr->ia_valid;
4678 error = inode_change_ok(inode, attr);
4682 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4683 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4686 /* (user+group)*(old+new) structure, inode write (sb,
4687 * inode block, ? - but truncate inode update has it) */
4688 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4689 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4690 if (IS_ERR(handle)) {
4691 error = PTR_ERR(handle);
4694 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4696 ext4_journal_stop(handle);
4699 /* Update corresponding info in inode so that everything is in
4700 * one transaction */
4701 if (attr->ia_valid & ATTR_UID)
4702 inode->i_uid = attr->ia_uid;
4703 if (attr->ia_valid & ATTR_GID)
4704 inode->i_gid = attr->ia_gid;
4705 error = ext4_mark_inode_dirty(handle, inode);
4706 ext4_journal_stop(handle);
4709 if (attr->ia_valid & ATTR_SIZE) {
4710 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4711 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4713 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4720 if (S_ISREG(inode->i_mode) &&
4721 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4724 handle = ext4_journal_start(inode, 3);
4725 if (IS_ERR(handle)) {
4726 error = PTR_ERR(handle);
4730 error = ext4_orphan_add(handle, inode);
4731 EXT4_I(inode)->i_disksize = attr->ia_size;
4732 rc = ext4_mark_inode_dirty(handle, inode);
4735 ext4_journal_stop(handle);
4737 if (ext4_should_order_data(inode)) {
4738 error = ext4_begin_ordered_truncate(inode,
4741 /* Do as much error cleanup as possible */
4742 handle = ext4_journal_start(inode, 3);
4743 if (IS_ERR(handle)) {
4744 ext4_orphan_del(NULL, inode);
4747 ext4_orphan_del(handle, inode);
4748 ext4_journal_stop(handle);
4754 rc = inode_setattr(inode, attr);
4756 /* If inode_setattr's call to ext4_truncate failed to get a
4757 * transaction handle at all, we need to clean up the in-core
4758 * orphan list manually. */
4760 ext4_orphan_del(NULL, inode);
4762 if (!rc && (ia_valid & ATTR_MODE))
4763 rc = ext4_acl_chmod(inode);
4766 ext4_std_error(inode->i_sb, error);
4772 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4775 struct inode *inode;
4776 unsigned long delalloc_blocks;
4778 inode = dentry->d_inode;
4779 generic_fillattr(inode, stat);
4782 * We can't update i_blocks if the block allocation is delayed
4783 * otherwise in the case of system crash before the real block
4784 * allocation is done, we will have i_blocks inconsistent with
4785 * on-disk file blocks.
4786 * We always keep i_blocks updated together with real
4787 * allocation. But to not confuse with user, stat
4788 * will return the blocks that include the delayed allocation
4789 * blocks for this file.
4791 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4792 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4793 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4795 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4799 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4804 /* if nrblocks are contiguous */
4807 * With N contiguous data blocks, it need at most
4808 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4809 * 2 dindirect blocks
4812 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4813 return indirects + 3;
4816 * if nrblocks are not contiguous, worse case, each block touch
4817 * a indirect block, and each indirect block touch a double indirect
4818 * block, plus a triple indirect block
4820 indirects = nrblocks * 2 + 1;
4824 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4826 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4827 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4828 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4832 * Account for index blocks, block groups bitmaps and block group
4833 * descriptor blocks if modify datablocks and index blocks
4834 * worse case, the indexs blocks spread over different block groups
4836 * If datablocks are discontiguous, they are possible to spread over
4837 * different block groups too. If they are contiugous, with flexbg,
4838 * they could still across block group boundary.
4840 * Also account for superblock, inode, quota and xattr blocks
4842 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4844 int groups, gdpblocks;
4849 * How many index blocks need to touch to modify nrblocks?
4850 * The "Chunk" flag indicating whether the nrblocks is
4851 * physically contiguous on disk
4853 * For Direct IO and fallocate, they calls get_block to allocate
4854 * one single extent at a time, so they could set the "Chunk" flag
4856 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4861 * Now let's see how many group bitmaps and group descriptors need
4871 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4872 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4873 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4874 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4876 /* bitmaps and block group descriptor blocks */
4877 ret += groups + gdpblocks;
4879 /* Blocks for super block, inode, quota and xattr blocks */
4880 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4886 * Calulate the total number of credits to reserve to fit
4887 * the modification of a single pages into a single transaction,
4888 * which may include multiple chunks of block allocations.
4890 * This could be called via ext4_write_begin()
4892 * We need to consider the worse case, when
4893 * one new block per extent.
4895 int ext4_writepage_trans_blocks(struct inode *inode)
4897 int bpp = ext4_journal_blocks_per_page(inode);
4900 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4902 /* Account for data blocks for journalled mode */
4903 if (ext4_should_journal_data(inode))
4909 * Calculate the journal credits for a chunk of data modification.
4911 * This is called from DIO, fallocate or whoever calling
4912 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4914 * journal buffers for data blocks are not included here, as DIO
4915 * and fallocate do no need to journal data buffers.
4917 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4919 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4923 * The caller must have previously called ext4_reserve_inode_write().
4924 * Give this, we know that the caller already has write access to iloc->bh.
4926 int ext4_mark_iloc_dirty(handle_t *handle,
4927 struct inode *inode, struct ext4_iloc *iloc)
4931 if (test_opt(inode->i_sb, I_VERSION))
4932 inode_inc_iversion(inode);
4934 /* the do_update_inode consumes one bh->b_count */
4937 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4938 err = ext4_do_update_inode(handle, inode, iloc);
4944 * On success, We end up with an outstanding reference count against
4945 * iloc->bh. This _must_ be cleaned up later.
4949 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4950 struct ext4_iloc *iloc)
4954 err = ext4_get_inode_loc(inode, iloc);
4956 BUFFER_TRACE(iloc->bh, "get_write_access");
4957 err = ext4_journal_get_write_access(handle, iloc->bh);
4963 ext4_std_error(inode->i_sb, err);
4968 * Expand an inode by new_extra_isize bytes.
4969 * Returns 0 on success or negative error number on failure.
4971 static int ext4_expand_extra_isize(struct inode *inode,
4972 unsigned int new_extra_isize,
4973 struct ext4_iloc iloc,
4976 struct ext4_inode *raw_inode;
4977 struct ext4_xattr_ibody_header *header;
4978 struct ext4_xattr_entry *entry;
4980 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4983 raw_inode = ext4_raw_inode(&iloc);
4985 header = IHDR(inode, raw_inode);
4986 entry = IFIRST(header);
4988 /* No extended attributes present */
4989 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4990 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4991 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4993 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4997 /* try to expand with EAs present */
4998 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5003 * What we do here is to mark the in-core inode as clean with respect to inode
5004 * dirtiness (it may still be data-dirty).
5005 * This means that the in-core inode may be reaped by prune_icache
5006 * without having to perform any I/O. This is a very good thing,
5007 * because *any* task may call prune_icache - even ones which
5008 * have a transaction open against a different journal.
5010 * Is this cheating? Not really. Sure, we haven't written the
5011 * inode out, but prune_icache isn't a user-visible syncing function.
5012 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5013 * we start and wait on commits.
5015 * Is this efficient/effective? Well, we're being nice to the system
5016 * by cleaning up our inodes proactively so they can be reaped
5017 * without I/O. But we are potentially leaving up to five seconds'
5018 * worth of inodes floating about which prune_icache wants us to
5019 * write out. One way to fix that would be to get prune_icache()
5020 * to do a write_super() to free up some memory. It has the desired
5023 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5025 struct ext4_iloc iloc;
5026 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5027 static unsigned int mnt_count;
5031 err = ext4_reserve_inode_write(handle, inode, &iloc);
5032 if (ext4_handle_valid(handle) &&
5033 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5034 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5036 * We need extra buffer credits since we may write into EA block
5037 * with this same handle. If journal_extend fails, then it will
5038 * only result in a minor loss of functionality for that inode.
5039 * If this is felt to be critical, then e2fsck should be run to
5040 * force a large enough s_min_extra_isize.
5042 if ((jbd2_journal_extend(handle,
5043 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5044 ret = ext4_expand_extra_isize(inode,
5045 sbi->s_want_extra_isize,
5048 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5050 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5051 ext4_warning(inode->i_sb, __func__,
5052 "Unable to expand inode %lu. Delete"
5053 " some EAs or run e2fsck.",
5056 le16_to_cpu(sbi->s_es->s_mnt_count);
5062 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5067 * ext4_dirty_inode() is called from __mark_inode_dirty()
5069 * We're really interested in the case where a file is being extended.
5070 * i_size has been changed by generic_commit_write() and we thus need
5071 * to include the updated inode in the current transaction.
5073 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5074 * are allocated to the file.
5076 * If the inode is marked synchronous, we don't honour that here - doing
5077 * so would cause a commit on atime updates, which we don't bother doing.
5078 * We handle synchronous inodes at the highest possible level.
5080 void ext4_dirty_inode(struct inode *inode)
5082 handle_t *current_handle = ext4_journal_current_handle();
5085 if (!ext4_handle_valid(current_handle)) {
5086 ext4_mark_inode_dirty(current_handle, inode);
5090 handle = ext4_journal_start(inode, 2);
5093 if (current_handle &&
5094 current_handle->h_transaction != handle->h_transaction) {
5095 /* This task has a transaction open against a different fs */
5096 printk(KERN_EMERG "%s: transactions do not match!\n",
5099 jbd_debug(5, "marking dirty. outer handle=%p\n",
5101 ext4_mark_inode_dirty(handle, inode);
5103 ext4_journal_stop(handle);
5110 * Bind an inode's backing buffer_head into this transaction, to prevent
5111 * it from being flushed to disk early. Unlike
5112 * ext4_reserve_inode_write, this leaves behind no bh reference and
5113 * returns no iloc structure, so the caller needs to repeat the iloc
5114 * lookup to mark the inode dirty later.
5116 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5118 struct ext4_iloc iloc;
5122 err = ext4_get_inode_loc(inode, &iloc);
5124 BUFFER_TRACE(iloc.bh, "get_write_access");
5125 err = jbd2_journal_get_write_access(handle, iloc.bh);
5127 err = ext4_handle_dirty_metadata(handle,
5133 ext4_std_error(inode->i_sb, err);
5138 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5145 * We have to be very careful here: changing a data block's
5146 * journaling status dynamically is dangerous. If we write a
5147 * data block to the journal, change the status and then delete
5148 * that block, we risk forgetting to revoke the old log record
5149 * from the journal and so a subsequent replay can corrupt data.
5150 * So, first we make sure that the journal is empty and that
5151 * nobody is changing anything.
5154 journal = EXT4_JOURNAL(inode);
5157 if (is_journal_aborted(journal))
5160 jbd2_journal_lock_updates(journal);
5161 jbd2_journal_flush(journal);
5164 * OK, there are no updates running now, and all cached data is
5165 * synced to disk. We are now in a completely consistent state
5166 * which doesn't have anything in the journal, and we know that
5167 * no filesystem updates are running, so it is safe to modify
5168 * the inode's in-core data-journaling state flag now.
5172 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5174 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5175 ext4_set_aops(inode);
5177 jbd2_journal_unlock_updates(journal);
5179 /* Finally we can mark the inode as dirty. */
5181 handle = ext4_journal_start(inode, 1);
5183 return PTR_ERR(handle);
5185 err = ext4_mark_inode_dirty(handle, inode);
5186 ext4_handle_sync(handle);
5187 ext4_journal_stop(handle);
5188 ext4_std_error(inode->i_sb, err);
5193 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5195 return !buffer_mapped(bh);
5198 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
5204 struct file *file = vma->vm_file;
5205 struct inode *inode = file->f_path.dentry->d_inode;
5206 struct address_space *mapping = inode->i_mapping;
5209 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5210 * get i_mutex because we are already holding mmap_sem.
5212 down_read(&inode->i_alloc_sem);
5213 size = i_size_read(inode);
5214 if (page->mapping != mapping || size <= page_offset(page)
5215 || !PageUptodate(page)) {
5216 /* page got truncated from under us? */
5220 if (PageMappedToDisk(page))
5223 if (page->index == size >> PAGE_CACHE_SHIFT)
5224 len = size & ~PAGE_CACHE_MASK;
5226 len = PAGE_CACHE_SIZE;
5228 if (page_has_buffers(page)) {
5229 /* return if we have all the buffers mapped */
5230 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5235 * OK, we need to fill the hole... Do write_begin write_end
5236 * to do block allocation/reservation.We are not holding
5237 * inode.i__mutex here. That allow * parallel write_begin,
5238 * write_end call. lock_page prevent this from happening
5239 * on the same page though
5241 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5242 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5245 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5246 len, len, page, fsdata);
5251 up_read(&inode->i_alloc_sem);