2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode *inode,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode,
62 static void ext4_invalidatepage(struct page *page, unsigned long offset);
65 * Test whether an inode is a fast symlink.
67 static int ext4_inode_is_fast_symlink(struct inode *inode)
69 int ea_blocks = EXT4_I(inode)->i_file_acl ?
70 (inode->i_sb->s_blocksize >> 9) : 0;
72 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
76 * Work out how many blocks we need to proceed with the next chunk of a
77 * truncate transaction.
79 static unsigned long blocks_for_truncate(struct inode *inode)
83 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
85 /* Give ourselves just enough room to cope with inodes in which
86 * i_blocks is corrupt: we've seen disk corruptions in the past
87 * which resulted in random data in an inode which looked enough
88 * like a regular file for ext4 to try to delete it. Things
89 * will go a bit crazy if that happens, but at least we should
90 * try not to panic the whole kernel. */
94 /* But we need to bound the transaction so we don't overflow the
96 if (needed > EXT4_MAX_TRANS_DATA)
97 needed = EXT4_MAX_TRANS_DATA;
99 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
103 * Truncate transactions can be complex and absolutely huge. So we need to
104 * be able to restart the transaction at a conventient checkpoint to make
105 * sure we don't overflow the journal.
107 * start_transaction gets us a new handle for a truncate transaction,
108 * and extend_transaction tries to extend the existing one a bit. If
109 * extend fails, we need to propagate the failure up and restart the
110 * transaction in the top-level truncate loop. --sct
112 static handle_t *start_transaction(struct inode *inode)
116 result = ext4_journal_start(inode, blocks_for_truncate(inode));
120 ext4_std_error(inode->i_sb, PTR_ERR(result));
125 * Try to extend this transaction for the purposes of truncation.
127 * Returns 0 if we managed to create more room. If we can't create more
128 * room, and the transaction must be restarted we return 1.
130 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
132 if (!ext4_handle_valid(handle))
134 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
136 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
142 * Restart the transaction associated with *handle. This does a commit,
143 * so before we call here everything must be consistently dirtied against
146 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
152 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
153 * moment, get_block can be called only for blocks inside i_size since
154 * page cache has been already dropped and writes are blocked by
155 * i_mutex. So we can safely drop the i_data_sem here.
157 BUG_ON(EXT4_JOURNAL(inode) == NULL);
158 jbd_debug(2, "restarting handle %p\n", handle);
159 up_write(&EXT4_I(inode)->i_data_sem);
160 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
161 down_write(&EXT4_I(inode)->i_data_sem);
162 ext4_discard_preallocations(inode);
168 * Called at the last iput() if i_nlink is zero.
170 void ext4_evict_inode(struct inode *inode)
175 if (inode->i_nlink) {
176 truncate_inode_pages(&inode->i_data, 0);
180 if (!is_bad_inode(inode))
181 dquot_initialize(inode);
183 if (ext4_should_order_data(inode))
184 ext4_begin_ordered_truncate(inode, 0);
185 truncate_inode_pages(&inode->i_data, 0);
187 if (is_bad_inode(inode))
190 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
191 if (IS_ERR(handle)) {
192 ext4_std_error(inode->i_sb, PTR_ERR(handle));
194 * If we're going to skip the normal cleanup, we still need to
195 * make sure that the in-core orphan linked list is properly
198 ext4_orphan_del(NULL, inode);
203 ext4_handle_sync(handle);
205 err = ext4_mark_inode_dirty(handle, inode);
207 ext4_warning(inode->i_sb,
208 "couldn't mark inode dirty (err %d)", err);
212 ext4_truncate(inode);
215 * ext4_ext_truncate() doesn't reserve any slop when it
216 * restarts journal transactions; therefore there may not be
217 * enough credits left in the handle to remove the inode from
218 * the orphan list and set the dtime field.
220 if (!ext4_handle_has_enough_credits(handle, 3)) {
221 err = ext4_journal_extend(handle, 3);
223 err = ext4_journal_restart(handle, 3);
225 ext4_warning(inode->i_sb,
226 "couldn't extend journal (err %d)", err);
228 ext4_journal_stop(handle);
229 ext4_orphan_del(NULL, inode);
235 * Kill off the orphan record which ext4_truncate created.
236 * AKPM: I think this can be inside the above `if'.
237 * Note that ext4_orphan_del() has to be able to cope with the
238 * deletion of a non-existent orphan - this is because we don't
239 * know if ext4_truncate() actually created an orphan record.
240 * (Well, we could do this if we need to, but heck - it works)
242 ext4_orphan_del(handle, inode);
243 EXT4_I(inode)->i_dtime = get_seconds();
246 * One subtle ordering requirement: if anything has gone wrong
247 * (transaction abort, IO errors, whatever), then we can still
248 * do these next steps (the fs will already have been marked as
249 * having errors), but we can't free the inode if the mark_dirty
252 if (ext4_mark_inode_dirty(handle, inode))
253 /* If that failed, just do the required in-core inode clear. */
254 ext4_clear_inode(inode);
256 ext4_free_inode(handle, inode);
257 ext4_journal_stop(handle);
260 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
266 struct buffer_head *bh;
269 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
271 p->key = *(p->p = v);
276 * ext4_block_to_path - parse the block number into array of offsets
277 * @inode: inode in question (we are only interested in its superblock)
278 * @i_block: block number to be parsed
279 * @offsets: array to store the offsets in
280 * @boundary: set this non-zero if the referred-to block is likely to be
281 * followed (on disk) by an indirect block.
283 * To store the locations of file's data ext4 uses a data structure common
284 * for UNIX filesystems - tree of pointers anchored in the inode, with
285 * data blocks at leaves and indirect blocks in intermediate nodes.
286 * This function translates the block number into path in that tree -
287 * return value is the path length and @offsets[n] is the offset of
288 * pointer to (n+1)th node in the nth one. If @block is out of range
289 * (negative or too large) warning is printed and zero returned.
291 * Note: function doesn't find node addresses, so no IO is needed. All
292 * we need to know is the capacity of indirect blocks (taken from the
297 * Portability note: the last comparison (check that we fit into triple
298 * indirect block) is spelled differently, because otherwise on an
299 * architecture with 32-bit longs and 8Kb pages we might get into trouble
300 * if our filesystem had 8Kb blocks. We might use long long, but that would
301 * kill us on x86. Oh, well, at least the sign propagation does not matter -
302 * i_block would have to be negative in the very beginning, so we would not
306 static int ext4_block_to_path(struct inode *inode,
308 ext4_lblk_t offsets[4], int *boundary)
310 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
311 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
312 const long direct_blocks = EXT4_NDIR_BLOCKS,
313 indirect_blocks = ptrs,
314 double_blocks = (1 << (ptrs_bits * 2));
318 if (i_block < direct_blocks) {
319 offsets[n++] = i_block;
320 final = direct_blocks;
321 } else if ((i_block -= direct_blocks) < indirect_blocks) {
322 offsets[n++] = EXT4_IND_BLOCK;
323 offsets[n++] = i_block;
325 } else if ((i_block -= indirect_blocks) < double_blocks) {
326 offsets[n++] = EXT4_DIND_BLOCK;
327 offsets[n++] = i_block >> ptrs_bits;
328 offsets[n++] = i_block & (ptrs - 1);
330 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
331 offsets[n++] = EXT4_TIND_BLOCK;
332 offsets[n++] = i_block >> (ptrs_bits * 2);
333 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
334 offsets[n++] = i_block & (ptrs - 1);
337 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
338 i_block + direct_blocks +
339 indirect_blocks + double_blocks, inode->i_ino);
342 *boundary = final - 1 - (i_block & (ptrs - 1));
346 static int __ext4_check_blockref(const char *function, unsigned int line,
348 __le32 *p, unsigned int max)
350 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
354 while (bref < p+max) {
355 blk = le32_to_cpu(*bref++);
357 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
359 es->s_last_error_block = cpu_to_le64(blk);
360 ext4_error_inode(inode, function, line, blk,
369 #define ext4_check_indirect_blockref(inode, bh) \
370 __ext4_check_blockref(__func__, __LINE__, inode, \
371 (__le32 *)(bh)->b_data, \
372 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
374 #define ext4_check_inode_blockref(inode) \
375 __ext4_check_blockref(__func__, __LINE__, inode, \
376 EXT4_I(inode)->i_data, \
380 * ext4_get_branch - read the chain of indirect blocks leading to data
381 * @inode: inode in question
382 * @depth: depth of the chain (1 - direct pointer, etc.)
383 * @offsets: offsets of pointers in inode/indirect blocks
384 * @chain: place to store the result
385 * @err: here we store the error value
387 * Function fills the array of triples <key, p, bh> and returns %NULL
388 * if everything went OK or the pointer to the last filled triple
389 * (incomplete one) otherwise. Upon the return chain[i].key contains
390 * the number of (i+1)-th block in the chain (as it is stored in memory,
391 * i.e. little-endian 32-bit), chain[i].p contains the address of that
392 * number (it points into struct inode for i==0 and into the bh->b_data
393 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
394 * block for i>0 and NULL for i==0. In other words, it holds the block
395 * numbers of the chain, addresses they were taken from (and where we can
396 * verify that chain did not change) and buffer_heads hosting these
399 * Function stops when it stumbles upon zero pointer (absent block)
400 * (pointer to last triple returned, *@err == 0)
401 * or when it gets an IO error reading an indirect block
402 * (ditto, *@err == -EIO)
403 * or when it reads all @depth-1 indirect blocks successfully and finds
404 * the whole chain, all way to the data (returns %NULL, *err == 0).
406 * Need to be called with
407 * down_read(&EXT4_I(inode)->i_data_sem)
409 static Indirect *ext4_get_branch(struct inode *inode, int depth,
410 ext4_lblk_t *offsets,
411 Indirect chain[4], int *err)
413 struct super_block *sb = inode->i_sb;
415 struct buffer_head *bh;
418 /* i_data is not going away, no lock needed */
419 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
423 bh = sb_getblk(sb, le32_to_cpu(p->key));
427 if (!bh_uptodate_or_lock(bh)) {
428 if (bh_submit_read(bh) < 0) {
432 /* validate block references */
433 if (ext4_check_indirect_blockref(inode, bh)) {
439 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
453 * ext4_find_near - find a place for allocation with sufficient locality
455 * @ind: descriptor of indirect block.
457 * This function returns the preferred place for block allocation.
458 * It is used when heuristic for sequential allocation fails.
460 * + if there is a block to the left of our position - allocate near it.
461 * + if pointer will live in indirect block - allocate near that block.
462 * + if pointer will live in inode - allocate in the same
465 * In the latter case we colour the starting block by the callers PID to
466 * prevent it from clashing with concurrent allocations for a different inode
467 * in the same block group. The PID is used here so that functionally related
468 * files will be close-by on-disk.
470 * Caller must make sure that @ind is valid and will stay that way.
472 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
474 struct ext4_inode_info *ei = EXT4_I(inode);
475 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
477 ext4_fsblk_t bg_start;
478 ext4_fsblk_t last_block;
479 ext4_grpblk_t colour;
480 ext4_group_t block_group;
481 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
483 /* Try to find previous block */
484 for (p = ind->p - 1; p >= start; p--) {
486 return le32_to_cpu(*p);
489 /* No such thing, so let's try location of indirect block */
491 return ind->bh->b_blocknr;
494 * It is going to be referred to from the inode itself? OK, just put it
495 * into the same cylinder group then.
497 block_group = ei->i_block_group;
498 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
499 block_group &= ~(flex_size-1);
500 if (S_ISREG(inode->i_mode))
503 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
504 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
507 * If we are doing delayed allocation, we don't need take
508 * colour into account.
510 if (test_opt(inode->i_sb, DELALLOC))
513 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
514 colour = (current->pid % 16) *
515 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
517 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
518 return bg_start + colour;
522 * ext4_find_goal - find a preferred place for allocation.
524 * @block: block we want
525 * @partial: pointer to the last triple within a chain
527 * Normally this function find the preferred place for block allocation,
529 * Because this is only used for non-extent files, we limit the block nr
532 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
538 * XXX need to get goal block from mballoc's data structures
541 goal = ext4_find_near(inode, partial);
542 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
547 * ext4_blks_to_allocate: Look up the block map and count the number
548 * of direct blocks need to be allocated for the given branch.
550 * @branch: chain of indirect blocks
551 * @k: number of blocks need for indirect blocks
552 * @blks: number of data blocks to be mapped.
553 * @blocks_to_boundary: the offset in the indirect block
555 * return the total number of blocks to be allocate, including the
556 * direct and indirect blocks.
558 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
559 int blocks_to_boundary)
561 unsigned int count = 0;
564 * Simple case, [t,d]Indirect block(s) has not allocated yet
565 * then it's clear blocks on that path have not allocated
568 /* right now we don't handle cross boundary allocation */
569 if (blks < blocks_to_boundary + 1)
572 count += blocks_to_boundary + 1;
577 while (count < blks && count <= blocks_to_boundary &&
578 le32_to_cpu(*(branch[0].p + count)) == 0) {
585 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
586 * @indirect_blks: the number of blocks need to allocate for indirect
589 * @new_blocks: on return it will store the new block numbers for
590 * the indirect blocks(if needed) and the first direct block,
591 * @blks: on return it will store the total number of allocated
594 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
595 ext4_lblk_t iblock, ext4_fsblk_t goal,
596 int indirect_blks, int blks,
597 ext4_fsblk_t new_blocks[4], int *err)
599 struct ext4_allocation_request ar;
601 unsigned long count = 0, blk_allocated = 0;
603 ext4_fsblk_t current_block = 0;
607 * Here we try to allocate the requested multiple blocks at once,
608 * on a best-effort basis.
609 * To build a branch, we should allocate blocks for
610 * the indirect blocks(if not allocated yet), and at least
611 * the first direct block of this branch. That's the
612 * minimum number of blocks need to allocate(required)
614 /* first we try to allocate the indirect blocks */
615 target = indirect_blks;
618 /* allocating blocks for indirect blocks and direct blocks */
619 current_block = ext4_new_meta_blocks(handle, inode,
624 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
625 EXT4_ERROR_INODE(inode,
626 "current_block %llu + count %lu > %d!",
627 current_block, count,
628 EXT4_MAX_BLOCK_FILE_PHYS);
634 /* allocate blocks for indirect blocks */
635 while (index < indirect_blks && count) {
636 new_blocks[index++] = current_block++;
641 * save the new block number
642 * for the first direct block
644 new_blocks[index] = current_block;
645 printk(KERN_INFO "%s returned more blocks than "
646 "requested\n", __func__);
652 target = blks - count ;
653 blk_allocated = count;
656 /* Now allocate data blocks */
657 memset(&ar, 0, sizeof(ar));
662 if (S_ISREG(inode->i_mode))
663 /* enable in-core preallocation only for regular files */
664 ar.flags = EXT4_MB_HINT_DATA;
666 current_block = ext4_mb_new_blocks(handle, &ar, err);
667 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
668 EXT4_ERROR_INODE(inode,
669 "current_block %llu + ar.len %d > %d!",
670 current_block, ar.len,
671 EXT4_MAX_BLOCK_FILE_PHYS);
676 if (*err && (target == blks)) {
678 * if the allocation failed and we didn't allocate
684 if (target == blks) {
686 * save the new block number
687 * for the first direct block
689 new_blocks[index] = current_block;
691 blk_allocated += ar.len;
694 /* total number of blocks allocated for direct blocks */
699 for (i = 0; i < index; i++)
700 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
705 * ext4_alloc_branch - allocate and set up a chain of blocks.
707 * @indirect_blks: number of allocated indirect blocks
708 * @blks: number of allocated direct blocks
709 * @offsets: offsets (in the blocks) to store the pointers to next.
710 * @branch: place to store the chain in.
712 * This function allocates blocks, zeroes out all but the last one,
713 * links them into chain and (if we are synchronous) writes them to disk.
714 * In other words, it prepares a branch that can be spliced onto the
715 * inode. It stores the information about that chain in the branch[], in
716 * the same format as ext4_get_branch() would do. We are calling it after
717 * we had read the existing part of chain and partial points to the last
718 * triple of that (one with zero ->key). Upon the exit we have the same
719 * picture as after the successful ext4_get_block(), except that in one
720 * place chain is disconnected - *branch->p is still zero (we did not
721 * set the last link), but branch->key contains the number that should
722 * be placed into *branch->p to fill that gap.
724 * If allocation fails we free all blocks we've allocated (and forget
725 * their buffer_heads) and return the error value the from failed
726 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
727 * as described above and return 0.
729 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
730 ext4_lblk_t iblock, int indirect_blks,
731 int *blks, ext4_fsblk_t goal,
732 ext4_lblk_t *offsets, Indirect *branch)
734 int blocksize = inode->i_sb->s_blocksize;
737 struct buffer_head *bh;
739 ext4_fsblk_t new_blocks[4];
740 ext4_fsblk_t current_block;
742 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
743 *blks, new_blocks, &err);
747 branch[0].key = cpu_to_le32(new_blocks[0]);
749 * metadata blocks and data blocks are allocated.
751 for (n = 1; n <= indirect_blks; n++) {
753 * Get buffer_head for parent block, zero it out
754 * and set the pointer to new one, then send
757 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
760 BUFFER_TRACE(bh, "call get_create_access");
761 err = ext4_journal_get_create_access(handle, bh);
763 /* Don't brelse(bh) here; it's done in
764 * ext4_journal_forget() below */
769 memset(bh->b_data, 0, blocksize);
770 branch[n].p = (__le32 *) bh->b_data + offsets[n];
771 branch[n].key = cpu_to_le32(new_blocks[n]);
772 *branch[n].p = branch[n].key;
773 if (n == indirect_blks) {
774 current_block = new_blocks[n];
776 * End of chain, update the last new metablock of
777 * the chain to point to the new allocated
778 * data blocks numbers
780 for (i = 1; i < num; i++)
781 *(branch[n].p + i) = cpu_to_le32(++current_block);
783 BUFFER_TRACE(bh, "marking uptodate");
784 set_buffer_uptodate(bh);
787 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
788 err = ext4_handle_dirty_metadata(handle, inode, bh);
795 /* Allocation failed, free what we already allocated */
796 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
797 for (i = 1; i <= n ; i++) {
799 * branch[i].bh is newly allocated, so there is no
800 * need to revoke the block, which is why we don't
801 * need to set EXT4_FREE_BLOCKS_METADATA.
803 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
804 EXT4_FREE_BLOCKS_FORGET);
806 for (i = n+1; i < indirect_blks; i++)
807 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
809 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
815 * ext4_splice_branch - splice the allocated branch onto inode.
817 * @block: (logical) number of block we are adding
818 * @chain: chain of indirect blocks (with a missing link - see
820 * @where: location of missing link
821 * @num: number of indirect blocks we are adding
822 * @blks: number of direct blocks we are adding
824 * This function fills the missing link and does all housekeeping needed in
825 * inode (->i_blocks, etc.). In case of success we end up with the full
826 * chain to new block and return 0.
828 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
829 ext4_lblk_t block, Indirect *where, int num,
834 ext4_fsblk_t current_block;
837 * If we're splicing into a [td]indirect block (as opposed to the
838 * inode) then we need to get write access to the [td]indirect block
842 BUFFER_TRACE(where->bh, "get_write_access");
843 err = ext4_journal_get_write_access(handle, where->bh);
849 *where->p = where->key;
852 * Update the host buffer_head or inode to point to more just allocated
853 * direct blocks blocks
855 if (num == 0 && blks > 1) {
856 current_block = le32_to_cpu(where->key) + 1;
857 for (i = 1; i < blks; i++)
858 *(where->p + i) = cpu_to_le32(current_block++);
861 /* We are done with atomic stuff, now do the rest of housekeeping */
862 /* had we spliced it onto indirect block? */
865 * If we spliced it onto an indirect block, we haven't
866 * altered the inode. Note however that if it is being spliced
867 * onto an indirect block at the very end of the file (the
868 * file is growing) then we *will* alter the inode to reflect
869 * the new i_size. But that is not done here - it is done in
870 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
872 jbd_debug(5, "splicing indirect only\n");
873 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
874 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
879 * OK, we spliced it into the inode itself on a direct block.
881 ext4_mark_inode_dirty(handle, inode);
882 jbd_debug(5, "splicing direct\n");
887 for (i = 1; i <= num; i++) {
889 * branch[i].bh is newly allocated, so there is no
890 * need to revoke the block, which is why we don't
891 * need to set EXT4_FREE_BLOCKS_METADATA.
893 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
894 EXT4_FREE_BLOCKS_FORGET);
896 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
903 * The ext4_ind_map_blocks() function handles non-extents inodes
904 * (i.e., using the traditional indirect/double-indirect i_blocks
905 * scheme) for ext4_map_blocks().
907 * Allocation strategy is simple: if we have to allocate something, we will
908 * have to go the whole way to leaf. So let's do it before attaching anything
909 * to tree, set linkage between the newborn blocks, write them if sync is
910 * required, recheck the path, free and repeat if check fails, otherwise
911 * set the last missing link (that will protect us from any truncate-generated
912 * removals - all blocks on the path are immune now) and possibly force the
913 * write on the parent block.
914 * That has a nice additional property: no special recovery from the failed
915 * allocations is needed - we simply release blocks and do not touch anything
916 * reachable from inode.
918 * `handle' can be NULL if create == 0.
920 * return > 0, # of blocks mapped or allocated.
921 * return = 0, if plain lookup failed.
922 * return < 0, error case.
924 * The ext4_ind_get_blocks() function should be called with
925 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
926 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
927 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
930 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
931 struct ext4_map_blocks *map,
935 ext4_lblk_t offsets[4];
940 int blocks_to_boundary = 0;
943 ext4_fsblk_t first_block = 0;
945 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
946 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
947 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
948 &blocks_to_boundary);
953 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
955 /* Simplest case - block found, no allocation needed */
957 first_block = le32_to_cpu(chain[depth - 1].key);
960 while (count < map->m_len && count <= blocks_to_boundary) {
963 blk = le32_to_cpu(*(chain[depth-1].p + count));
965 if (blk == first_block + count)
973 /* Next simple case - plain lookup or failed read of indirect block */
974 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
978 * Okay, we need to do block allocation.
980 goal = ext4_find_goal(inode, map->m_lblk, partial);
982 /* the number of blocks need to allocate for [d,t]indirect blocks */
983 indirect_blks = (chain + depth) - partial - 1;
986 * Next look up the indirect map to count the totoal number of
987 * direct blocks to allocate for this branch.
989 count = ext4_blks_to_allocate(partial, indirect_blks,
990 map->m_len, blocks_to_boundary);
992 * Block out ext4_truncate while we alter the tree
994 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
996 offsets + (partial - chain), partial);
999 * The ext4_splice_branch call will free and forget any buffers
1000 * on the new chain if there is a failure, but that risks using
1001 * up transaction credits, especially for bitmaps where the
1002 * credits cannot be returned. Can we handle this somehow? We
1003 * may need to return -EAGAIN upwards in the worst case. --sct
1006 err = ext4_splice_branch(handle, inode, map->m_lblk,
1007 partial, indirect_blks, count);
1011 map->m_flags |= EXT4_MAP_NEW;
1013 ext4_update_inode_fsync_trans(handle, inode, 1);
1015 map->m_flags |= EXT4_MAP_MAPPED;
1016 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1018 if (count > blocks_to_boundary)
1019 map->m_flags |= EXT4_MAP_BOUNDARY;
1021 /* Clean up and exit */
1022 partial = chain + depth - 1; /* the whole chain */
1024 while (partial > chain) {
1025 BUFFER_TRACE(partial->bh, "call brelse");
1026 brelse(partial->bh);
1034 qsize_t *ext4_get_reserved_space(struct inode *inode)
1036 return &EXT4_I(inode)->i_reserved_quota;
1041 * Calculate the number of metadata blocks need to reserve
1042 * to allocate a new block at @lblocks for non extent file based file
1044 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1047 struct ext4_inode_info *ei = EXT4_I(inode);
1048 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1051 if (lblock < EXT4_NDIR_BLOCKS)
1054 lblock -= EXT4_NDIR_BLOCKS;
1056 if (ei->i_da_metadata_calc_len &&
1057 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1058 ei->i_da_metadata_calc_len++;
1061 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1062 ei->i_da_metadata_calc_len = 1;
1063 blk_bits = order_base_2(lblock);
1064 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1068 * Calculate the number of metadata blocks need to reserve
1069 * to allocate a block located at @lblock
1071 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1073 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1074 return ext4_ext_calc_metadata_amount(inode, lblock);
1076 return ext4_indirect_calc_metadata_amount(inode, lblock);
1080 * Called with i_data_sem down, which is important since we can call
1081 * ext4_discard_preallocations() from here.
1083 void ext4_da_update_reserve_space(struct inode *inode,
1084 int used, int quota_claim)
1086 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1087 struct ext4_inode_info *ei = EXT4_I(inode);
1089 spin_lock(&ei->i_block_reservation_lock);
1090 trace_ext4_da_update_reserve_space(inode, used);
1091 if (unlikely(used > ei->i_reserved_data_blocks)) {
1092 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1093 "with only %d reserved data blocks\n",
1094 __func__, inode->i_ino, used,
1095 ei->i_reserved_data_blocks);
1097 used = ei->i_reserved_data_blocks;
1100 /* Update per-inode reservations */
1101 ei->i_reserved_data_blocks -= used;
1102 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1103 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1104 used + ei->i_allocated_meta_blocks);
1105 ei->i_allocated_meta_blocks = 0;
1107 if (ei->i_reserved_data_blocks == 0) {
1109 * We can release all of the reserved metadata blocks
1110 * only when we have written all of the delayed
1111 * allocation blocks.
1113 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1114 ei->i_reserved_meta_blocks);
1115 ei->i_reserved_meta_blocks = 0;
1116 ei->i_da_metadata_calc_len = 0;
1118 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1120 /* Update quota subsystem for data blocks */
1122 dquot_claim_block(inode, used);
1125 * We did fallocate with an offset that is already delayed
1126 * allocated. So on delayed allocated writeback we should
1127 * not re-claim the quota for fallocated blocks.
1129 dquot_release_reservation_block(inode, used);
1133 * If we have done all the pending block allocations and if
1134 * there aren't any writers on the inode, we can discard the
1135 * inode's preallocations.
1137 if ((ei->i_reserved_data_blocks == 0) &&
1138 (atomic_read(&inode->i_writecount) == 0))
1139 ext4_discard_preallocations(inode);
1142 static int __check_block_validity(struct inode *inode, const char *func,
1144 struct ext4_map_blocks *map)
1146 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1148 ext4_error_inode(inode, func, line, map->m_pblk,
1149 "lblock %lu mapped to illegal pblock "
1150 "(length %d)", (unsigned long) map->m_lblk,
1157 #define check_block_validity(inode, map) \
1158 __check_block_validity((inode), __func__, __LINE__, (map))
1161 * Return the number of contiguous dirty pages in a given inode
1162 * starting at page frame idx.
1164 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1165 unsigned int max_pages)
1167 struct address_space *mapping = inode->i_mapping;
1169 struct pagevec pvec;
1171 int i, nr_pages, done = 0;
1175 pagevec_init(&pvec, 0);
1178 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1179 PAGECACHE_TAG_DIRTY,
1180 (pgoff_t)PAGEVEC_SIZE);
1183 for (i = 0; i < nr_pages; i++) {
1184 struct page *page = pvec.pages[i];
1185 struct buffer_head *bh, *head;
1188 if (unlikely(page->mapping != mapping) ||
1190 PageWriteback(page) ||
1191 page->index != idx) {
1196 if (page_has_buffers(page)) {
1197 bh = head = page_buffers(page);
1199 if (!buffer_delay(bh) &&
1200 !buffer_unwritten(bh))
1202 bh = bh->b_this_page;
1203 } while (!done && (bh != head));
1210 if (num >= max_pages) {
1215 pagevec_release(&pvec);
1221 * The ext4_map_blocks() function tries to look up the requested blocks,
1222 * and returns if the blocks are already mapped.
1224 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1225 * and store the allocated blocks in the result buffer head and mark it
1228 * If file type is extents based, it will call ext4_ext_map_blocks(),
1229 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1232 * On success, it returns the number of blocks being mapped or allocate.
1233 * if create==0 and the blocks are pre-allocated and uninitialized block,
1234 * the result buffer head is unmapped. If the create ==1, it will make sure
1235 * the buffer head is mapped.
1237 * It returns 0 if plain look up failed (blocks have not been allocated), in
1238 * that casem, buffer head is unmapped
1240 * It returns the error in case of allocation failure.
1242 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1243 struct ext4_map_blocks *map, int flags)
1248 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1249 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1250 (unsigned long) map->m_lblk);
1252 * Try to see if we can get the block without requesting a new
1253 * file system block.
1255 down_read((&EXT4_I(inode)->i_data_sem));
1256 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1257 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1259 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1261 up_read((&EXT4_I(inode)->i_data_sem));
1263 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1264 int ret = check_block_validity(inode, map);
1269 /* If it is only a block(s) look up */
1270 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1274 * Returns if the blocks have already allocated
1276 * Note that if blocks have been preallocated
1277 * ext4_ext_get_block() returns th create = 0
1278 * with buffer head unmapped.
1280 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1284 * When we call get_blocks without the create flag, the
1285 * BH_Unwritten flag could have gotten set if the blocks
1286 * requested were part of a uninitialized extent. We need to
1287 * clear this flag now that we are committed to convert all or
1288 * part of the uninitialized extent to be an initialized
1289 * extent. This is because we need to avoid the combination
1290 * of BH_Unwritten and BH_Mapped flags being simultaneously
1291 * set on the buffer_head.
1293 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1296 * New blocks allocate and/or writing to uninitialized extent
1297 * will possibly result in updating i_data, so we take
1298 * the write lock of i_data_sem, and call get_blocks()
1299 * with create == 1 flag.
1301 down_write((&EXT4_I(inode)->i_data_sem));
1304 * if the caller is from delayed allocation writeout path
1305 * we have already reserved fs blocks for allocation
1306 * let the underlying get_block() function know to
1307 * avoid double accounting
1309 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1310 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1312 * We need to check for EXT4 here because migrate
1313 * could have changed the inode type in between
1315 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1316 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1318 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1320 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1322 * We allocated new blocks which will result in
1323 * i_data's format changing. Force the migrate
1324 * to fail by clearing migrate flags
1326 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1330 * Update reserved blocks/metadata blocks after successful
1331 * block allocation which had been deferred till now. We don't
1332 * support fallocate for non extent files. So we can update
1333 * reserve space here.
1336 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1337 ext4_da_update_reserve_space(inode, retval, 1);
1339 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1340 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1342 up_write((&EXT4_I(inode)->i_data_sem));
1343 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1344 int ret = check_block_validity(inode, map);
1351 /* Maximum number of blocks we map for direct IO at once. */
1352 #define DIO_MAX_BLOCKS 4096
1354 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1355 struct buffer_head *bh, int flags)
1357 handle_t *handle = ext4_journal_current_handle();
1358 struct ext4_map_blocks map;
1359 int ret = 0, started = 0;
1362 map.m_lblk = iblock;
1363 map.m_len = bh->b_size >> inode->i_blkbits;
1365 if (flags && !handle) {
1366 /* Direct IO write... */
1367 if (map.m_len > DIO_MAX_BLOCKS)
1368 map.m_len = DIO_MAX_BLOCKS;
1369 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1370 handle = ext4_journal_start(inode, dio_credits);
1371 if (IS_ERR(handle)) {
1372 ret = PTR_ERR(handle);
1378 ret = ext4_map_blocks(handle, inode, &map, flags);
1380 map_bh(bh, inode->i_sb, map.m_pblk);
1381 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1382 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1386 ext4_journal_stop(handle);
1390 int ext4_get_block(struct inode *inode, sector_t iblock,
1391 struct buffer_head *bh, int create)
1393 return _ext4_get_block(inode, iblock, bh,
1394 create ? EXT4_GET_BLOCKS_CREATE : 0);
1398 * `handle' can be NULL if create is zero
1400 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1401 ext4_lblk_t block, int create, int *errp)
1403 struct ext4_map_blocks map;
1404 struct buffer_head *bh;
1407 J_ASSERT(handle != NULL || create == 0);
1411 err = ext4_map_blocks(handle, inode, &map,
1412 create ? EXT4_GET_BLOCKS_CREATE : 0);
1420 bh = sb_getblk(inode->i_sb, map.m_pblk);
1425 if (map.m_flags & EXT4_MAP_NEW) {
1426 J_ASSERT(create != 0);
1427 J_ASSERT(handle != NULL);
1430 * Now that we do not always journal data, we should
1431 * keep in mind whether this should always journal the
1432 * new buffer as metadata. For now, regular file
1433 * writes use ext4_get_block instead, so it's not a
1437 BUFFER_TRACE(bh, "call get_create_access");
1438 fatal = ext4_journal_get_create_access(handle, bh);
1439 if (!fatal && !buffer_uptodate(bh)) {
1440 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1441 set_buffer_uptodate(bh);
1444 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1445 err = ext4_handle_dirty_metadata(handle, inode, bh);
1449 BUFFER_TRACE(bh, "not a new buffer");
1459 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1460 ext4_lblk_t block, int create, int *err)
1462 struct buffer_head *bh;
1464 bh = ext4_getblk(handle, inode, block, create, err);
1467 if (buffer_uptodate(bh))
1469 ll_rw_block(READ_META, 1, &bh);
1471 if (buffer_uptodate(bh))
1478 static int walk_page_buffers(handle_t *handle,
1479 struct buffer_head *head,
1483 int (*fn)(handle_t *handle,
1484 struct buffer_head *bh))
1486 struct buffer_head *bh;
1487 unsigned block_start, block_end;
1488 unsigned blocksize = head->b_size;
1490 struct buffer_head *next;
1492 for (bh = head, block_start = 0;
1493 ret == 0 && (bh != head || !block_start);
1494 block_start = block_end, bh = next) {
1495 next = bh->b_this_page;
1496 block_end = block_start + blocksize;
1497 if (block_end <= from || block_start >= to) {
1498 if (partial && !buffer_uptodate(bh))
1502 err = (*fn)(handle, bh);
1510 * To preserve ordering, it is essential that the hole instantiation and
1511 * the data write be encapsulated in a single transaction. We cannot
1512 * close off a transaction and start a new one between the ext4_get_block()
1513 * and the commit_write(). So doing the jbd2_journal_start at the start of
1514 * prepare_write() is the right place.
1516 * Also, this function can nest inside ext4_writepage() ->
1517 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1518 * has generated enough buffer credits to do the whole page. So we won't
1519 * block on the journal in that case, which is good, because the caller may
1522 * By accident, ext4 can be reentered when a transaction is open via
1523 * quota file writes. If we were to commit the transaction while thus
1524 * reentered, there can be a deadlock - we would be holding a quota
1525 * lock, and the commit would never complete if another thread had a
1526 * transaction open and was blocking on the quota lock - a ranking
1529 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1530 * will _not_ run commit under these circumstances because handle->h_ref
1531 * is elevated. We'll still have enough credits for the tiny quotafile
1534 static int do_journal_get_write_access(handle_t *handle,
1535 struct buffer_head *bh)
1537 int dirty = buffer_dirty(bh);
1540 if (!buffer_mapped(bh) || buffer_freed(bh))
1543 * __block_prepare_write() could have dirtied some buffers. Clean
1544 * the dirty bit as jbd2_journal_get_write_access() could complain
1545 * otherwise about fs integrity issues. Setting of the dirty bit
1546 * by __block_prepare_write() isn't a real problem here as we clear
1547 * the bit before releasing a page lock and thus writeback cannot
1548 * ever write the buffer.
1551 clear_buffer_dirty(bh);
1552 ret = ext4_journal_get_write_access(handle, bh);
1554 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1559 * Truncate blocks that were not used by write. We have to truncate the
1560 * pagecache as well so that corresponding buffers get properly unmapped.
1562 static void ext4_truncate_failed_write(struct inode *inode)
1564 truncate_inode_pages(inode->i_mapping, inode->i_size);
1565 ext4_truncate(inode);
1568 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1569 struct buffer_head *bh_result, int create);
1570 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1571 loff_t pos, unsigned len, unsigned flags,
1572 struct page **pagep, void **fsdata)
1574 struct inode *inode = mapping->host;
1575 int ret, needed_blocks;
1582 trace_ext4_write_begin(inode, pos, len, flags);
1584 * Reserve one block more for addition to orphan list in case
1585 * we allocate blocks but write fails for some reason
1587 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1588 index = pos >> PAGE_CACHE_SHIFT;
1589 from = pos & (PAGE_CACHE_SIZE - 1);
1593 handle = ext4_journal_start(inode, needed_blocks);
1594 if (IS_ERR(handle)) {
1595 ret = PTR_ERR(handle);
1599 /* We cannot recurse into the filesystem as the transaction is already
1601 flags |= AOP_FLAG_NOFS;
1603 page = grab_cache_page_write_begin(mapping, index, flags);
1605 ext4_journal_stop(handle);
1611 if (ext4_should_dioread_nolock(inode))
1612 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1614 ret = __block_write_begin(page, pos, len, ext4_get_block);
1616 if (!ret && ext4_should_journal_data(inode)) {
1617 ret = walk_page_buffers(handle, page_buffers(page),
1618 from, to, NULL, do_journal_get_write_access);
1623 page_cache_release(page);
1625 * __block_write_begin may have instantiated a few blocks
1626 * outside i_size. Trim these off again. Don't need
1627 * i_size_read because we hold i_mutex.
1629 * Add inode to orphan list in case we crash before
1632 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1633 ext4_orphan_add(handle, inode);
1635 ext4_journal_stop(handle);
1636 if (pos + len > inode->i_size) {
1637 ext4_truncate_failed_write(inode);
1639 * If truncate failed early the inode might
1640 * still be on the orphan list; we need to
1641 * make sure the inode is removed from the
1642 * orphan list in that case.
1645 ext4_orphan_del(NULL, inode);
1649 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1655 /* For write_end() in data=journal mode */
1656 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1658 if (!buffer_mapped(bh) || buffer_freed(bh))
1660 set_buffer_uptodate(bh);
1661 return ext4_handle_dirty_metadata(handle, NULL, bh);
1664 static int ext4_generic_write_end(struct file *file,
1665 struct address_space *mapping,
1666 loff_t pos, unsigned len, unsigned copied,
1667 struct page *page, void *fsdata)
1669 int i_size_changed = 0;
1670 struct inode *inode = mapping->host;
1671 handle_t *handle = ext4_journal_current_handle();
1673 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1676 * No need to use i_size_read() here, the i_size
1677 * cannot change under us because we hold i_mutex.
1679 * But it's important to update i_size while still holding page lock:
1680 * page writeout could otherwise come in and zero beyond i_size.
1682 if (pos + copied > inode->i_size) {
1683 i_size_write(inode, pos + copied);
1687 if (pos + copied > EXT4_I(inode)->i_disksize) {
1688 /* We need to mark inode dirty even if
1689 * new_i_size is less that inode->i_size
1690 * bu greater than i_disksize.(hint delalloc)
1692 ext4_update_i_disksize(inode, (pos + copied));
1696 page_cache_release(page);
1699 * Don't mark the inode dirty under page lock. First, it unnecessarily
1700 * makes the holding time of page lock longer. Second, it forces lock
1701 * ordering of page lock and transaction start for journaling
1705 ext4_mark_inode_dirty(handle, inode);
1711 * We need to pick up the new inode size which generic_commit_write gave us
1712 * `file' can be NULL - eg, when called from page_symlink().
1714 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1715 * buffers are managed internally.
1717 static int ext4_ordered_write_end(struct file *file,
1718 struct address_space *mapping,
1719 loff_t pos, unsigned len, unsigned copied,
1720 struct page *page, void *fsdata)
1722 handle_t *handle = ext4_journal_current_handle();
1723 struct inode *inode = mapping->host;
1726 trace_ext4_ordered_write_end(inode, pos, len, copied);
1727 ret = ext4_jbd2_file_inode(handle, inode);
1730 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1733 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1734 /* if we have allocated more blocks and copied
1735 * less. We will have blocks allocated outside
1736 * inode->i_size. So truncate them
1738 ext4_orphan_add(handle, inode);
1742 ret2 = ext4_journal_stop(handle);
1746 if (pos + len > inode->i_size) {
1747 ext4_truncate_failed_write(inode);
1749 * If truncate failed early the inode might still be
1750 * on the orphan list; we need to make sure the inode
1751 * is removed from the orphan list in that case.
1754 ext4_orphan_del(NULL, inode);
1758 return ret ? ret : copied;
1761 static int ext4_writeback_write_end(struct file *file,
1762 struct address_space *mapping,
1763 loff_t pos, unsigned len, unsigned copied,
1764 struct page *page, void *fsdata)
1766 handle_t *handle = ext4_journal_current_handle();
1767 struct inode *inode = mapping->host;
1770 trace_ext4_writeback_write_end(inode, pos, len, copied);
1771 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1774 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1775 /* if we have allocated more blocks and copied
1776 * less. We will have blocks allocated outside
1777 * inode->i_size. So truncate them
1779 ext4_orphan_add(handle, inode);
1784 ret2 = ext4_journal_stop(handle);
1788 if (pos + len > inode->i_size) {
1789 ext4_truncate_failed_write(inode);
1791 * If truncate failed early the inode might still be
1792 * on the orphan list; we need to make sure the inode
1793 * is removed from the orphan list in that case.
1796 ext4_orphan_del(NULL, inode);
1799 return ret ? ret : copied;
1802 static int ext4_journalled_write_end(struct file *file,
1803 struct address_space *mapping,
1804 loff_t pos, unsigned len, unsigned copied,
1805 struct page *page, void *fsdata)
1807 handle_t *handle = ext4_journal_current_handle();
1808 struct inode *inode = mapping->host;
1814 trace_ext4_journalled_write_end(inode, pos, len, copied);
1815 from = pos & (PAGE_CACHE_SIZE - 1);
1819 if (!PageUptodate(page))
1821 page_zero_new_buffers(page, from+copied, to);
1824 ret = walk_page_buffers(handle, page_buffers(page), from,
1825 to, &partial, write_end_fn);
1827 SetPageUptodate(page);
1828 new_i_size = pos + copied;
1829 if (new_i_size > inode->i_size)
1830 i_size_write(inode, pos+copied);
1831 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1832 if (new_i_size > EXT4_I(inode)->i_disksize) {
1833 ext4_update_i_disksize(inode, new_i_size);
1834 ret2 = ext4_mark_inode_dirty(handle, inode);
1840 page_cache_release(page);
1841 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1842 /* if we have allocated more blocks and copied
1843 * less. We will have blocks allocated outside
1844 * inode->i_size. So truncate them
1846 ext4_orphan_add(handle, inode);
1848 ret2 = ext4_journal_stop(handle);
1851 if (pos + len > inode->i_size) {
1852 ext4_truncate_failed_write(inode);
1854 * If truncate failed early the inode might still be
1855 * on the orphan list; we need to make sure the inode
1856 * is removed from the orphan list in that case.
1859 ext4_orphan_del(NULL, inode);
1862 return ret ? ret : copied;
1866 * Reserve a single block located at lblock
1868 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1871 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1872 struct ext4_inode_info *ei = EXT4_I(inode);
1873 unsigned long md_needed;
1877 * recalculate the amount of metadata blocks to reserve
1878 * in order to allocate nrblocks
1879 * worse case is one extent per block
1882 spin_lock(&ei->i_block_reservation_lock);
1883 md_needed = ext4_calc_metadata_amount(inode, lblock);
1884 trace_ext4_da_reserve_space(inode, md_needed);
1885 spin_unlock(&ei->i_block_reservation_lock);
1888 * We will charge metadata quota at writeout time; this saves
1889 * us from metadata over-estimation, though we may go over by
1890 * a small amount in the end. Here we just reserve for data.
1892 ret = dquot_reserve_block(inode, 1);
1896 * We do still charge estimated metadata to the sb though;
1897 * we cannot afford to run out of free blocks.
1899 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1900 dquot_release_reservation_block(inode, 1);
1901 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1907 spin_lock(&ei->i_block_reservation_lock);
1908 ei->i_reserved_data_blocks++;
1909 ei->i_reserved_meta_blocks += md_needed;
1910 spin_unlock(&ei->i_block_reservation_lock);
1912 return 0; /* success */
1915 static void ext4_da_release_space(struct inode *inode, int to_free)
1917 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1918 struct ext4_inode_info *ei = EXT4_I(inode);
1921 return; /* Nothing to release, exit */
1923 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1925 trace_ext4_da_release_space(inode, to_free);
1926 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1928 * if there aren't enough reserved blocks, then the
1929 * counter is messed up somewhere. Since this
1930 * function is called from invalidate page, it's
1931 * harmless to return without any action.
1933 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1934 "ino %lu, to_free %d with only %d reserved "
1935 "data blocks\n", inode->i_ino, to_free,
1936 ei->i_reserved_data_blocks);
1938 to_free = ei->i_reserved_data_blocks;
1940 ei->i_reserved_data_blocks -= to_free;
1942 if (ei->i_reserved_data_blocks == 0) {
1944 * We can release all of the reserved metadata blocks
1945 * only when we have written all of the delayed
1946 * allocation blocks.
1948 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1949 ei->i_reserved_meta_blocks);
1950 ei->i_reserved_meta_blocks = 0;
1951 ei->i_da_metadata_calc_len = 0;
1954 /* update fs dirty data blocks counter */
1955 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1957 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1959 dquot_release_reservation_block(inode, to_free);
1962 static void ext4_da_page_release_reservation(struct page *page,
1963 unsigned long offset)
1966 struct buffer_head *head, *bh;
1967 unsigned int curr_off = 0;
1969 head = page_buffers(page);
1972 unsigned int next_off = curr_off + bh->b_size;
1974 if ((offset <= curr_off) && (buffer_delay(bh))) {
1976 clear_buffer_delay(bh);
1978 curr_off = next_off;
1979 } while ((bh = bh->b_this_page) != head);
1980 ext4_da_release_space(page->mapping->host, to_release);
1984 * Delayed allocation stuff
1988 * mpage_da_submit_io - walks through extent of pages and try to write
1989 * them with writepage() call back
1991 * @mpd->inode: inode
1992 * @mpd->first_page: first page of the extent
1993 * @mpd->next_page: page after the last page of the extent
1995 * By the time mpage_da_submit_io() is called we expect all blocks
1996 * to be allocated. this may be wrong if allocation failed.
1998 * As pages are already locked by write_cache_pages(), we can't use it
2000 static int mpage_da_submit_io(struct mpage_da_data *mpd)
2003 struct pagevec pvec;
2004 unsigned long index, end;
2005 int ret = 0, err, nr_pages, i;
2006 struct inode *inode = mpd->inode;
2007 struct address_space *mapping = inode->i_mapping;
2009 BUG_ON(mpd->next_page <= mpd->first_page);
2011 * We need to start from the first_page to the next_page - 1
2012 * to make sure we also write the mapped dirty buffer_heads.
2013 * If we look at mpd->b_blocknr we would only be looking
2014 * at the currently mapped buffer_heads.
2016 index = mpd->first_page;
2017 end = mpd->next_page - 1;
2019 pagevec_init(&pvec, 0);
2020 while (index <= end) {
2021 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2024 for (i = 0; i < nr_pages; i++) {
2025 struct page *page = pvec.pages[i];
2027 index = page->index;
2032 BUG_ON(!PageLocked(page));
2033 BUG_ON(PageWriteback(page));
2035 pages_skipped = mpd->wbc->pages_skipped;
2036 err = mapping->a_ops->writepage(page, mpd->wbc);
2037 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
2039 * have successfully written the page
2040 * without skipping the same
2042 mpd->pages_written++;
2044 * In error case, we have to continue because
2045 * remaining pages are still locked
2046 * XXX: unlock and re-dirty them?
2051 pagevec_release(&pvec);
2057 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
2059 * the function goes through all passed space and put actual disk
2060 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
2062 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd,
2063 struct ext4_map_blocks *map)
2065 struct inode *inode = mpd->inode;
2066 struct address_space *mapping = inode->i_mapping;
2067 int blocks = map->m_len;
2068 sector_t pblock = map->m_pblk, cur_logical;
2069 struct buffer_head *head, *bh;
2071 struct pagevec pvec;
2074 index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2075 end = (map->m_lblk + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2076 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2078 pagevec_init(&pvec, 0);
2080 while (index <= end) {
2081 /* XXX: optimize tail */
2082 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2085 for (i = 0; i < nr_pages; i++) {
2086 struct page *page = pvec.pages[i];
2088 index = page->index;
2093 BUG_ON(!PageLocked(page));
2094 BUG_ON(PageWriteback(page));
2095 BUG_ON(!page_has_buffers(page));
2097 bh = page_buffers(page);
2100 /* skip blocks out of the range */
2102 if (cur_logical >= map->m_lblk)
2105 } while ((bh = bh->b_this_page) != head);
2108 if (cur_logical >= map->m_lblk + blocks)
2111 if (buffer_delay(bh) || buffer_unwritten(bh)) {
2113 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
2115 if (buffer_delay(bh)) {
2116 clear_buffer_delay(bh);
2117 bh->b_blocknr = pblock;
2120 * unwritten already should have
2121 * blocknr assigned. Verify that
2123 clear_buffer_unwritten(bh);
2124 BUG_ON(bh->b_blocknr != pblock);
2127 } else if (buffer_mapped(bh))
2128 BUG_ON(bh->b_blocknr != pblock);
2130 if (map->m_flags & EXT4_MAP_UNINIT)
2131 set_buffer_uninit(bh);
2134 } while ((bh = bh->b_this_page) != head);
2136 pagevec_release(&pvec);
2141 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2142 sector_t logical, long blk_cnt)
2146 struct pagevec pvec;
2147 struct inode *inode = mpd->inode;
2148 struct address_space *mapping = inode->i_mapping;
2150 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2151 end = (logical + blk_cnt - 1) >>
2152 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2153 while (index <= end) {
2154 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2157 for (i = 0; i < nr_pages; i++) {
2158 struct page *page = pvec.pages[i];
2159 if (page->index > end)
2161 BUG_ON(!PageLocked(page));
2162 BUG_ON(PageWriteback(page));
2163 block_invalidatepage(page, 0);
2164 ClearPageUptodate(page);
2167 index = pvec.pages[nr_pages - 1]->index + 1;
2168 pagevec_release(&pvec);
2173 static void ext4_print_free_blocks(struct inode *inode)
2175 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2176 printk(KERN_CRIT "Total free blocks count %lld\n",
2177 ext4_count_free_blocks(inode->i_sb));
2178 printk(KERN_CRIT "Free/Dirty block details\n");
2179 printk(KERN_CRIT "free_blocks=%lld\n",
2180 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2181 printk(KERN_CRIT "dirty_blocks=%lld\n",
2182 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2183 printk(KERN_CRIT "Block reservation details\n");
2184 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2185 EXT4_I(inode)->i_reserved_data_blocks);
2186 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2187 EXT4_I(inode)->i_reserved_meta_blocks);
2192 * mpage_da_map_blocks - go through given space
2194 * @mpd - bh describing space
2196 * The function skips space we know is already mapped to disk blocks.
2199 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2201 int err, blks, get_blocks_flags;
2202 struct ext4_map_blocks map;
2203 sector_t next = mpd->b_blocknr;
2204 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2205 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2206 handle_t *handle = NULL;
2209 * We consider only non-mapped and non-allocated blocks
2211 if ((mpd->b_state & (1 << BH_Mapped)) &&
2212 !(mpd->b_state & (1 << BH_Delay)) &&
2213 !(mpd->b_state & (1 << BH_Unwritten)))
2217 * If we didn't accumulate anything to write simply return
2222 handle = ext4_journal_current_handle();
2226 * Call ext4_map_blocks() to allocate any delayed allocation
2227 * blocks, or to convert an uninitialized extent to be
2228 * initialized (in the case where we have written into
2229 * one or more preallocated blocks).
2231 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2232 * indicate that we are on the delayed allocation path. This
2233 * affects functions in many different parts of the allocation
2234 * call path. This flag exists primarily because we don't
2235 * want to change *many* call functions, so ext4_map_blocks()
2236 * will set the magic i_delalloc_reserved_flag once the
2237 * inode's allocation semaphore is taken.
2239 * If the blocks in questions were delalloc blocks, set
2240 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2241 * variables are updated after the blocks have been allocated.
2244 map.m_len = max_blocks;
2245 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2246 if (ext4_should_dioread_nolock(mpd->inode))
2247 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2248 if (mpd->b_state & (1 << BH_Delay))
2249 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2251 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2253 struct super_block *sb = mpd->inode->i_sb;
2257 * If get block returns with error we simply
2258 * return. Later writepage will redirty the page and
2259 * writepages will find the dirty page again
2264 if (err == -ENOSPC &&
2265 ext4_count_free_blocks(sb)) {
2271 * get block failure will cause us to loop in
2272 * writepages, because a_ops->writepage won't be able
2273 * to make progress. The page will be redirtied by
2274 * writepage and writepages will again try to write
2277 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2278 ext4_msg(sb, KERN_CRIT,
2279 "delayed block allocation failed for inode %lu "
2280 "at logical offset %llu with max blocks %zd "
2281 "with error %d", mpd->inode->i_ino,
2282 (unsigned long long) next,
2283 mpd->b_size >> mpd->inode->i_blkbits, err);
2284 ext4_msg(sb, KERN_CRIT,
2285 "This should not happen!! Data will be lost\n");
2287 ext4_print_free_blocks(mpd->inode);
2289 /* invalidate all the pages */
2290 ext4_da_block_invalidatepages(mpd, next,
2291 mpd->b_size >> mpd->inode->i_blkbits);
2296 if (map.m_flags & EXT4_MAP_NEW) {
2297 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2300 for (i = 0; i < map.m_len; i++)
2301 unmap_underlying_metadata(bdev, map.m_pblk + i);
2305 * If blocks are delayed marked, we need to
2306 * put actual blocknr and drop delayed bit
2308 if ((mpd->b_state & (1 << BH_Delay)) ||
2309 (mpd->b_state & (1 << BH_Unwritten)))
2310 mpage_put_bnr_to_bhs(mpd, &map);
2312 if (ext4_should_order_data(mpd->inode)) {
2313 err = ext4_jbd2_file_inode(handle, mpd->inode);
2319 * Update on-disk size along with block allocation.
2321 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2322 if (disksize > i_size_read(mpd->inode))
2323 disksize = i_size_read(mpd->inode);
2324 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2325 ext4_update_i_disksize(mpd->inode, disksize);
2326 return ext4_mark_inode_dirty(handle, mpd->inode);
2332 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2333 (1 << BH_Delay) | (1 << BH_Unwritten))
2336 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2338 * @mpd->lbh - extent of blocks
2339 * @logical - logical number of the block in the file
2340 * @bh - bh of the block (used to access block's state)
2342 * the function is used to collect contig. blocks in same state
2344 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2345 sector_t logical, size_t b_size,
2346 unsigned long b_state)
2349 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2352 * XXX Don't go larger than mballoc is willing to allocate
2353 * This is a stopgap solution. We eventually need to fold
2354 * mpage_da_submit_io() into this function and then call
2355 * ext4_map_blocks() multiple times in a loop
2357 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2360 /* check if thereserved journal credits might overflow */
2361 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2362 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2364 * With non-extent format we are limited by the journal
2365 * credit available. Total credit needed to insert
2366 * nrblocks contiguous blocks is dependent on the
2367 * nrblocks. So limit nrblocks.
2370 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2371 EXT4_MAX_TRANS_DATA) {
2373 * Adding the new buffer_head would make it cross the
2374 * allowed limit for which we have journal credit
2375 * reserved. So limit the new bh->b_size
2377 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2378 mpd->inode->i_blkbits;
2379 /* we will do mpage_da_submit_io in the next loop */
2383 * First block in the extent
2385 if (mpd->b_size == 0) {
2386 mpd->b_blocknr = logical;
2387 mpd->b_size = b_size;
2388 mpd->b_state = b_state & BH_FLAGS;
2392 next = mpd->b_blocknr + nrblocks;
2394 * Can we merge the block to our big extent?
2396 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2397 mpd->b_size += b_size;
2403 * We couldn't merge the block to our extent, so we
2404 * need to flush current extent and start new one
2406 if (mpage_da_map_blocks(mpd) == 0)
2407 mpage_da_submit_io(mpd);
2412 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2414 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2418 * __mpage_da_writepage - finds extent of pages and blocks
2420 * @page: page to consider
2421 * @wbc: not used, we just follow rules
2424 * The function finds extents of pages and scan them for all blocks.
2426 static int __mpage_da_writepage(struct page *page,
2427 struct writeback_control *wbc, void *data)
2429 struct mpage_da_data *mpd = data;
2430 struct inode *inode = mpd->inode;
2431 struct buffer_head *bh, *head;
2435 * Can we merge this page to current extent?
2437 if (mpd->next_page != page->index) {
2439 * Nope, we can't. So, we map non-allocated blocks
2440 * and start IO on them using writepage()
2442 if (mpd->next_page != mpd->first_page) {
2443 if (mpage_da_map_blocks(mpd) == 0)
2444 mpage_da_submit_io(mpd);
2446 * skip rest of the page in the page_vec
2449 redirty_page_for_writepage(wbc, page);
2451 return MPAGE_DA_EXTENT_TAIL;
2455 * Start next extent of pages ...
2457 mpd->first_page = page->index;
2467 mpd->next_page = page->index + 1;
2468 logical = (sector_t) page->index <<
2469 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2471 if (!page_has_buffers(page)) {
2472 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2473 (1 << BH_Dirty) | (1 << BH_Uptodate));
2475 return MPAGE_DA_EXTENT_TAIL;
2478 * Page with regular buffer heads, just add all dirty ones
2480 head = page_buffers(page);
2483 BUG_ON(buffer_locked(bh));
2485 * We need to try to allocate
2486 * unmapped blocks in the same page.
2487 * Otherwise we won't make progress
2488 * with the page in ext4_writepage
2490 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2491 mpage_add_bh_to_extent(mpd, logical,
2495 return MPAGE_DA_EXTENT_TAIL;
2496 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2498 * mapped dirty buffer. We need to update
2499 * the b_state because we look at
2500 * b_state in mpage_da_map_blocks. We don't
2501 * update b_size because if we find an
2502 * unmapped buffer_head later we need to
2503 * use the b_state flag of that buffer_head.
2505 if (mpd->b_size == 0)
2506 mpd->b_state = bh->b_state & BH_FLAGS;
2509 } while ((bh = bh->b_this_page) != head);
2516 * This is a special get_blocks_t callback which is used by
2517 * ext4_da_write_begin(). It will either return mapped block or
2518 * reserve space for a single block.
2520 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2521 * We also have b_blocknr = -1 and b_bdev initialized properly
2523 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2524 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2525 * initialized properly.
2527 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2528 struct buffer_head *bh, int create)
2530 struct ext4_map_blocks map;
2532 sector_t invalid_block = ~((sector_t) 0xffff);
2534 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2537 BUG_ON(create == 0);
2538 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2540 map.m_lblk = iblock;
2544 * first, we need to know whether the block is allocated already
2545 * preallocated blocks are unmapped but should treated
2546 * the same as allocated blocks.
2548 ret = ext4_map_blocks(NULL, inode, &map, 0);
2552 if (buffer_delay(bh))
2553 return 0; /* Not sure this could or should happen */
2555 * XXX: __block_prepare_write() unmaps passed block,
2558 ret = ext4_da_reserve_space(inode, iblock);
2560 /* not enough space to reserve */
2563 map_bh(bh, inode->i_sb, invalid_block);
2565 set_buffer_delay(bh);
2569 map_bh(bh, inode->i_sb, map.m_pblk);
2570 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2572 if (buffer_unwritten(bh)) {
2573 /* A delayed write to unwritten bh should be marked
2574 * new and mapped. Mapped ensures that we don't do
2575 * get_block multiple times when we write to the same
2576 * offset and new ensures that we do proper zero out
2577 * for partial write.
2580 set_buffer_mapped(bh);
2586 * This function is used as a standard get_block_t calback function
2587 * when there is no desire to allocate any blocks. It is used as a
2588 * callback function for block_prepare_write() and block_write_full_page().
2589 * These functions should only try to map a single block at a time.
2591 * Since this function doesn't do block allocations even if the caller
2592 * requests it by passing in create=1, it is critically important that
2593 * any caller checks to make sure that any buffer heads are returned
2594 * by this function are either all already mapped or marked for
2595 * delayed allocation before calling block_write_full_page(). Otherwise,
2596 * b_blocknr could be left unitialized, and the page write functions will
2597 * be taken by surprise.
2599 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2600 struct buffer_head *bh_result, int create)
2602 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2603 return _ext4_get_block(inode, iblock, bh_result, 0);
2606 static int bget_one(handle_t *handle, struct buffer_head *bh)
2612 static int bput_one(handle_t *handle, struct buffer_head *bh)
2618 static int __ext4_journalled_writepage(struct page *page,
2621 struct address_space *mapping = page->mapping;
2622 struct inode *inode = mapping->host;
2623 struct buffer_head *page_bufs;
2624 handle_t *handle = NULL;
2628 page_bufs = page_buffers(page);
2630 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2631 /* As soon as we unlock the page, it can go away, but we have
2632 * references to buffers so we are safe */
2635 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2636 if (IS_ERR(handle)) {
2637 ret = PTR_ERR(handle);
2641 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2642 do_journal_get_write_access);
2644 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2648 err = ext4_journal_stop(handle);
2652 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2653 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2658 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2659 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2662 * Note that we don't need to start a transaction unless we're journaling data
2663 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2664 * need to file the inode to the transaction's list in ordered mode because if
2665 * we are writing back data added by write(), the inode is already there and if
2666 * we are writing back data modified via mmap(), noone guarantees in which
2667 * transaction the data will hit the disk. In case we are journaling data, we
2668 * cannot start transaction directly because transaction start ranks above page
2669 * lock so we have to do some magic.
2671 * This function can get called via...
2672 * - ext4_da_writepages after taking page lock (have journal handle)
2673 * - journal_submit_inode_data_buffers (no journal handle)
2674 * - shrink_page_list via pdflush (no journal handle)
2675 * - grab_page_cache when doing write_begin (have journal handle)
2677 * We don't do any block allocation in this function. If we have page with
2678 * multiple blocks we need to write those buffer_heads that are mapped. This
2679 * is important for mmaped based write. So if we do with blocksize 1K
2680 * truncate(f, 1024);
2681 * a = mmap(f, 0, 4096);
2683 * truncate(f, 4096);
2684 * we have in the page first buffer_head mapped via page_mkwrite call back
2685 * but other bufer_heads would be unmapped but dirty(dirty done via the
2686 * do_wp_page). So writepage should write the first block. If we modify
2687 * the mmap area beyond 1024 we will again get a page_fault and the
2688 * page_mkwrite callback will do the block allocation and mark the
2689 * buffer_heads mapped.
2691 * We redirty the page if we have any buffer_heads that is either delay or
2692 * unwritten in the page.
2694 * We can get recursively called as show below.
2696 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2699 * But since we don't do any block allocation we should not deadlock.
2700 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2702 static int ext4_writepage(struct page *page,
2703 struct writeback_control *wbc)
2708 struct buffer_head *page_bufs = NULL;
2709 struct inode *inode = page->mapping->host;
2711 trace_ext4_writepage(inode, page);
2712 size = i_size_read(inode);
2713 if (page->index == size >> PAGE_CACHE_SHIFT)
2714 len = size & ~PAGE_CACHE_MASK;
2716 len = PAGE_CACHE_SIZE;
2718 if (page_has_buffers(page)) {
2719 page_bufs = page_buffers(page);
2720 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2721 ext4_bh_delay_or_unwritten)) {
2723 * We don't want to do block allocation
2724 * So redirty the page and return
2725 * We may reach here when we do a journal commit
2726 * via journal_submit_inode_data_buffers.
2727 * If we don't have mapping block we just ignore
2728 * them. We can also reach here via shrink_page_list
2730 redirty_page_for_writepage(wbc, page);
2736 * The test for page_has_buffers() is subtle:
2737 * We know the page is dirty but it lost buffers. That means
2738 * that at some moment in time after write_begin()/write_end()
2739 * has been called all buffers have been clean and thus they
2740 * must have been written at least once. So they are all
2741 * mapped and we can happily proceed with mapping them
2742 * and writing the page.
2744 * Try to initialize the buffer_heads and check whether
2745 * all are mapped and non delay. We don't want to
2746 * do block allocation here.
2748 ret = block_prepare_write(page, 0, len,
2749 noalloc_get_block_write);
2751 page_bufs = page_buffers(page);
2752 /* check whether all are mapped and non delay */
2753 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2754 ext4_bh_delay_or_unwritten)) {
2755 redirty_page_for_writepage(wbc, page);
2761 * We can't do block allocation here
2762 * so just redity the page and unlock
2765 redirty_page_for_writepage(wbc, page);
2769 /* now mark the buffer_heads as dirty and uptodate */
2770 block_commit_write(page, 0, len);
2773 if (PageChecked(page) && ext4_should_journal_data(inode)) {
2775 * It's mmapped pagecache. Add buffers and journal it. There
2776 * doesn't seem much point in redirtying the page here.
2778 ClearPageChecked(page);
2779 return __ext4_journalled_writepage(page, len);
2782 if (page_bufs && buffer_uninit(page_bufs)) {
2783 ext4_set_bh_endio(page_bufs, inode);
2784 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2785 wbc, ext4_end_io_buffer_write);
2787 ret = block_write_full_page(page, noalloc_get_block_write,
2794 * This is called via ext4_da_writepages() to
2795 * calulate the total number of credits to reserve to fit
2796 * a single extent allocation into a single transaction,
2797 * ext4_da_writpeages() will loop calling this before
2798 * the block allocation.
2801 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2803 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2806 * With non-extent format the journal credit needed to
2807 * insert nrblocks contiguous block is dependent on
2808 * number of contiguous block. So we will limit
2809 * number of contiguous block to a sane value
2811 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2812 (max_blocks > EXT4_MAX_TRANS_DATA))
2813 max_blocks = EXT4_MAX_TRANS_DATA;
2815 return ext4_chunk_trans_blocks(inode, max_blocks);
2819 * write_cache_pages_da - walk the list of dirty pages of the given
2820 * address space and call the callback function (which usually writes
2823 * This is a forked version of write_cache_pages(). Differences:
2824 * Range cyclic is ignored.
2825 * no_nrwrite_index_update is always presumed true
2827 static int write_cache_pages_da(struct address_space *mapping,
2828 struct writeback_control *wbc,
2829 struct mpage_da_data *mpd)
2833 struct pagevec pvec;
2836 pgoff_t end; /* Inclusive */
2837 long nr_to_write = wbc->nr_to_write;
2839 pagevec_init(&pvec, 0);
2840 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2841 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2843 while (!done && (index <= end)) {
2846 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2847 PAGECACHE_TAG_DIRTY,
2848 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2852 for (i = 0; i < nr_pages; i++) {
2853 struct page *page = pvec.pages[i];
2856 * At this point, the page may be truncated or
2857 * invalidated (changing page->mapping to NULL), or
2858 * even swizzled back from swapper_space to tmpfs file
2859 * mapping. However, page->index will not change
2860 * because we have a reference on the page.
2862 if (page->index > end) {
2870 * Page truncated or invalidated. We can freely skip it
2871 * then, even for data integrity operations: the page
2872 * has disappeared concurrently, so there could be no
2873 * real expectation of this data interity operation
2874 * even if there is now a new, dirty page at the same
2875 * pagecache address.
2877 if (unlikely(page->mapping != mapping)) {
2883 if (!PageDirty(page)) {
2884 /* someone wrote it for us */
2885 goto continue_unlock;
2888 if (PageWriteback(page)) {
2889 if (wbc->sync_mode != WB_SYNC_NONE)
2890 wait_on_page_writeback(page);
2892 goto continue_unlock;
2895 BUG_ON(PageWriteback(page));
2896 if (!clear_page_dirty_for_io(page))
2897 goto continue_unlock;
2899 ret = __mpage_da_writepage(page, wbc, mpd);
2900 if (unlikely(ret)) {
2901 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2910 if (nr_to_write > 0) {
2912 if (nr_to_write == 0 &&
2913 wbc->sync_mode == WB_SYNC_NONE) {
2915 * We stop writing back only if we are
2916 * not doing integrity sync. In case of
2917 * integrity sync we have to keep going
2918 * because someone may be concurrently
2919 * dirtying pages, and we might have
2920 * synced a lot of newly appeared dirty
2921 * pages, but have not synced all of the
2929 pagevec_release(&pvec);
2936 static int ext4_da_writepages(struct address_space *mapping,
2937 struct writeback_control *wbc)
2940 int range_whole = 0;
2941 handle_t *handle = NULL;
2942 struct mpage_da_data mpd;
2943 struct inode *inode = mapping->host;
2944 int pages_written = 0;
2946 unsigned int max_pages;
2947 int range_cyclic, cycled = 1, io_done = 0;
2948 int needed_blocks, ret = 0;
2949 long desired_nr_to_write, nr_to_writebump = 0;
2950 loff_t range_start = wbc->range_start;
2951 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2953 trace_ext4_da_writepages(inode, wbc);
2956 * No pages to write? This is mainly a kludge to avoid starting
2957 * a transaction for special inodes like journal inode on last iput()
2958 * because that could violate lock ordering on umount
2960 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2964 * If the filesystem has aborted, it is read-only, so return
2965 * right away instead of dumping stack traces later on that
2966 * will obscure the real source of the problem. We test
2967 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2968 * the latter could be true if the filesystem is mounted
2969 * read-only, and in that case, ext4_da_writepages should
2970 * *never* be called, so if that ever happens, we would want
2973 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2976 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2979 range_cyclic = wbc->range_cyclic;
2980 if (wbc->range_cyclic) {
2981 index = mapping->writeback_index;
2984 wbc->range_start = index << PAGE_CACHE_SHIFT;
2985 wbc->range_end = LLONG_MAX;
2986 wbc->range_cyclic = 0;
2988 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2991 * This works around two forms of stupidity. The first is in
2992 * the writeback code, which caps the maximum number of pages
2993 * written to be 1024 pages. This is wrong on multiple
2994 * levels; different architectues have a different page size,
2995 * which changes the maximum amount of data which gets
2996 * written. Secondly, 4 megabytes is way too small. XFS
2997 * forces this value to be 16 megabytes by multiplying
2998 * nr_to_write parameter by four, and then relies on its
2999 * allocator to allocate larger extents to make them
3000 * contiguous. Unfortunately this brings us to the second
3001 * stupidity, which is that ext4's mballoc code only allocates
3002 * at most 2048 blocks. So we force contiguous writes up to
3003 * the number of dirty blocks in the inode, or
3004 * sbi->max_writeback_mb_bump whichever is smaller.
3006 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
3007 if (!range_cyclic && range_whole) {
3008 if (wbc->nr_to_write == LONG_MAX)
3009 desired_nr_to_write = wbc->nr_to_write;
3011 desired_nr_to_write = wbc->nr_to_write * 8;
3013 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3015 if (desired_nr_to_write > max_pages)
3016 desired_nr_to_write = max_pages;
3018 if (wbc->nr_to_write < desired_nr_to_write) {
3019 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
3020 wbc->nr_to_write = desired_nr_to_write;
3024 mpd.inode = mapping->host;
3026 pages_skipped = wbc->pages_skipped;
3029 while (!ret && wbc->nr_to_write > 0) {
3032 * we insert one extent at a time. So we need
3033 * credit needed for single extent allocation.
3034 * journalled mode is currently not supported
3037 BUG_ON(ext4_should_journal_data(inode));
3038 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3040 /* start a new transaction*/
3041 handle = ext4_journal_start(inode, needed_blocks);
3042 if (IS_ERR(handle)) {
3043 ret = PTR_ERR(handle);
3044 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3045 "%ld pages, ino %lu; err %d", __func__,
3046 wbc->nr_to_write, inode->i_ino, ret);
3047 goto out_writepages;
3051 * Now call __mpage_da_writepage to find the next
3052 * contiguous region of logical blocks that need
3053 * blocks to be allocated by ext4. We don't actually
3054 * submit the blocks for I/O here, even though
3055 * write_cache_pages thinks it will, and will set the
3056 * pages as clean for write before calling
3057 * __mpage_da_writepage().
3065 mpd.pages_written = 0;
3067 ret = write_cache_pages_da(mapping, wbc, &mpd);
3069 * If we have a contiguous extent of pages and we
3070 * haven't done the I/O yet, map the blocks and submit
3073 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3074 if (mpage_da_map_blocks(&mpd) == 0)
3075 mpage_da_submit_io(&mpd);
3077 ret = MPAGE_DA_EXTENT_TAIL;
3079 trace_ext4_da_write_pages(inode, &mpd);
3080 wbc->nr_to_write -= mpd.pages_written;
3082 ext4_journal_stop(handle);
3084 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3085 /* commit the transaction which would
3086 * free blocks released in the transaction
3089 jbd2_journal_force_commit_nested(sbi->s_journal);
3090 wbc->pages_skipped = pages_skipped;
3092 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3094 * got one extent now try with
3097 pages_written += mpd.pages_written;
3098 wbc->pages_skipped = pages_skipped;
3101 } else if (wbc->nr_to_write)
3103 * There is no more writeout needed
3104 * or we requested for a noblocking writeout
3105 * and we found the device congested
3109 if (!io_done && !cycled) {
3112 wbc->range_start = index << PAGE_CACHE_SHIFT;
3113 wbc->range_end = mapping->writeback_index - 1;
3116 if (pages_skipped != wbc->pages_skipped)
3117 ext4_msg(inode->i_sb, KERN_CRIT,
3118 "This should not happen leaving %s "
3119 "with nr_to_write = %ld ret = %d",
3120 __func__, wbc->nr_to_write, ret);
3123 index += pages_written;
3124 wbc->range_cyclic = range_cyclic;
3125 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3127 * set the writeback_index so that range_cyclic
3128 * mode will write it back later
3130 mapping->writeback_index = index;
3133 wbc->nr_to_write -= nr_to_writebump;
3134 wbc->range_start = range_start;
3135 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3139 #define FALL_BACK_TO_NONDELALLOC 1
3140 static int ext4_nonda_switch(struct super_block *sb)
3142 s64 free_blocks, dirty_blocks;
3143 struct ext4_sb_info *sbi = EXT4_SB(sb);
3146 * switch to non delalloc mode if we are running low
3147 * on free block. The free block accounting via percpu
3148 * counters can get slightly wrong with percpu_counter_batch getting
3149 * accumulated on each CPU without updating global counters
3150 * Delalloc need an accurate free block accounting. So switch
3151 * to non delalloc when we are near to error range.
3153 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3154 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3155 if (2 * free_blocks < 3 * dirty_blocks ||
3156 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3158 * free block count is less than 150% of dirty blocks
3159 * or free blocks is less than watermark
3164 * Even if we don't switch but are nearing capacity,
3165 * start pushing delalloc when 1/2 of free blocks are dirty.
3167 if (free_blocks < 2 * dirty_blocks)
3168 writeback_inodes_sb_if_idle(sb);
3173 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3174 loff_t pos, unsigned len, unsigned flags,
3175 struct page **pagep, void **fsdata)
3177 int ret, retries = 0;
3180 struct inode *inode = mapping->host;
3183 index = pos >> PAGE_CACHE_SHIFT;
3185 if (ext4_nonda_switch(inode->i_sb)) {
3186 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3187 return ext4_write_begin(file, mapping, pos,
3188 len, flags, pagep, fsdata);
3190 *fsdata = (void *)0;
3191 trace_ext4_da_write_begin(inode, pos, len, flags);
3194 * With delayed allocation, we don't log the i_disksize update
3195 * if there is delayed block allocation. But we still need
3196 * to journalling the i_disksize update if writes to the end
3197 * of file which has an already mapped buffer.
3199 handle = ext4_journal_start(inode, 1);
3200 if (IS_ERR(handle)) {
3201 ret = PTR_ERR(handle);
3204 /* We cannot recurse into the filesystem as the transaction is already
3206 flags |= AOP_FLAG_NOFS;
3208 page = grab_cache_page_write_begin(mapping, index, flags);
3210 ext4_journal_stop(handle);
3216 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3219 ext4_journal_stop(handle);
3220 page_cache_release(page);
3222 * block_write_begin may have instantiated a few blocks
3223 * outside i_size. Trim these off again. Don't need
3224 * i_size_read because we hold i_mutex.
3226 if (pos + len > inode->i_size)
3227 ext4_truncate_failed_write(inode);
3230 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3237 * Check if we should update i_disksize
3238 * when write to the end of file but not require block allocation
3240 static int ext4_da_should_update_i_disksize(struct page *page,
3241 unsigned long offset)
3243 struct buffer_head *bh;
3244 struct inode *inode = page->mapping->host;
3248 bh = page_buffers(page);
3249 idx = offset >> inode->i_blkbits;
3251 for (i = 0; i < idx; i++)
3252 bh = bh->b_this_page;
3254 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3259 static int ext4_da_write_end(struct file *file,
3260 struct address_space *mapping,
3261 loff_t pos, unsigned len, unsigned copied,
3262 struct page *page, void *fsdata)
3264 struct inode *inode = mapping->host;
3266 handle_t *handle = ext4_journal_current_handle();
3268 unsigned long start, end;
3269 int write_mode = (int)(unsigned long)fsdata;
3271 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3272 if (ext4_should_order_data(inode)) {
3273 return ext4_ordered_write_end(file, mapping, pos,
3274 len, copied, page, fsdata);
3275 } else if (ext4_should_writeback_data(inode)) {
3276 return ext4_writeback_write_end(file, mapping, pos,
3277 len, copied, page, fsdata);
3283 trace_ext4_da_write_end(inode, pos, len, copied);
3284 start = pos & (PAGE_CACHE_SIZE - 1);
3285 end = start + copied - 1;
3288 * generic_write_end() will run mark_inode_dirty() if i_size
3289 * changes. So let's piggyback the i_disksize mark_inode_dirty
3293 new_i_size = pos + copied;
3294 if (new_i_size > EXT4_I(inode)->i_disksize) {
3295 if (ext4_da_should_update_i_disksize(page, end)) {
3296 down_write(&EXT4_I(inode)->i_data_sem);
3297 if (new_i_size > EXT4_I(inode)->i_disksize) {
3299 * Updating i_disksize when extending file
3300 * without needing block allocation
3302 if (ext4_should_order_data(inode))
3303 ret = ext4_jbd2_file_inode(handle,
3306 EXT4_I(inode)->i_disksize = new_i_size;
3308 up_write(&EXT4_I(inode)->i_data_sem);
3309 /* We need to mark inode dirty even if
3310 * new_i_size is less that inode->i_size
3311 * bu greater than i_disksize.(hint delalloc)
3313 ext4_mark_inode_dirty(handle, inode);
3316 ret2 = generic_write_end(file, mapping, pos, len, copied,
3321 ret2 = ext4_journal_stop(handle);
3325 return ret ? ret : copied;
3328 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3331 * Drop reserved blocks
3333 BUG_ON(!PageLocked(page));
3334 if (!page_has_buffers(page))
3337 ext4_da_page_release_reservation(page, offset);
3340 ext4_invalidatepage(page, offset);
3346 * Force all delayed allocation blocks to be allocated for a given inode.
3348 int ext4_alloc_da_blocks(struct inode *inode)
3350 trace_ext4_alloc_da_blocks(inode);
3352 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3353 !EXT4_I(inode)->i_reserved_meta_blocks)
3357 * We do something simple for now. The filemap_flush() will
3358 * also start triggering a write of the data blocks, which is
3359 * not strictly speaking necessary (and for users of
3360 * laptop_mode, not even desirable). However, to do otherwise
3361 * would require replicating code paths in:
3363 * ext4_da_writepages() ->
3364 * write_cache_pages() ---> (via passed in callback function)
3365 * __mpage_da_writepage() -->
3366 * mpage_add_bh_to_extent()
3367 * mpage_da_map_blocks()
3369 * The problem is that write_cache_pages(), located in
3370 * mm/page-writeback.c, marks pages clean in preparation for
3371 * doing I/O, which is not desirable if we're not planning on
3374 * We could call write_cache_pages(), and then redirty all of
3375 * the pages by calling redirty_page_for_writeback() but that
3376 * would be ugly in the extreme. So instead we would need to
3377 * replicate parts of the code in the above functions,
3378 * simplifying them becuase we wouldn't actually intend to
3379 * write out the pages, but rather only collect contiguous
3380 * logical block extents, call the multi-block allocator, and
3381 * then update the buffer heads with the block allocations.
3383 * For now, though, we'll cheat by calling filemap_flush(),
3384 * which will map the blocks, and start the I/O, but not
3385 * actually wait for the I/O to complete.
3387 return filemap_flush(inode->i_mapping);
3391 * bmap() is special. It gets used by applications such as lilo and by
3392 * the swapper to find the on-disk block of a specific piece of data.
3394 * Naturally, this is dangerous if the block concerned is still in the
3395 * journal. If somebody makes a swapfile on an ext4 data-journaling
3396 * filesystem and enables swap, then they may get a nasty shock when the
3397 * data getting swapped to that swapfile suddenly gets overwritten by
3398 * the original zero's written out previously to the journal and
3399 * awaiting writeback in the kernel's buffer cache.
3401 * So, if we see any bmap calls here on a modified, data-journaled file,
3402 * take extra steps to flush any blocks which might be in the cache.
3404 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3406 struct inode *inode = mapping->host;
3410 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3411 test_opt(inode->i_sb, DELALLOC)) {
3413 * With delalloc we want to sync the file
3414 * so that we can make sure we allocate
3417 filemap_write_and_wait(mapping);
3420 if (EXT4_JOURNAL(inode) &&
3421 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3423 * This is a REALLY heavyweight approach, but the use of
3424 * bmap on dirty files is expected to be extremely rare:
3425 * only if we run lilo or swapon on a freshly made file
3426 * do we expect this to happen.
3428 * (bmap requires CAP_SYS_RAWIO so this does not
3429 * represent an unprivileged user DOS attack --- we'd be
3430 * in trouble if mortal users could trigger this path at
3433 * NB. EXT4_STATE_JDATA is not set on files other than
3434 * regular files. If somebody wants to bmap a directory
3435 * or symlink and gets confused because the buffer
3436 * hasn't yet been flushed to disk, they deserve
3437 * everything they get.
3440 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3441 journal = EXT4_JOURNAL(inode);
3442 jbd2_journal_lock_updates(journal);
3443 err = jbd2_journal_flush(journal);
3444 jbd2_journal_unlock_updates(journal);
3450 return generic_block_bmap(mapping, block, ext4_get_block);
3453 static int ext4_readpage(struct file *file, struct page *page)
3455 return mpage_readpage(page, ext4_get_block);
3459 ext4_readpages(struct file *file, struct address_space *mapping,
3460 struct list_head *pages, unsigned nr_pages)
3462 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3465 static void ext4_free_io_end(ext4_io_end_t *io)
3474 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3476 struct buffer_head *head, *bh;
3477 unsigned int curr_off = 0;
3479 if (!page_has_buffers(page))
3481 head = bh = page_buffers(page);
3483 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3485 ext4_free_io_end(bh->b_private);
3486 bh->b_private = NULL;
3487 bh->b_end_io = NULL;
3489 curr_off = curr_off + bh->b_size;
3490 bh = bh->b_this_page;
3491 } while (bh != head);
3494 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3496 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3499 * free any io_end structure allocated for buffers to be discarded
3501 if (ext4_should_dioread_nolock(page->mapping->host))
3502 ext4_invalidatepage_free_endio(page, offset);
3504 * If it's a full truncate we just forget about the pending dirtying
3507 ClearPageChecked(page);
3510 jbd2_journal_invalidatepage(journal, page, offset);
3512 block_invalidatepage(page, offset);
3515 static int ext4_releasepage(struct page *page, gfp_t wait)
3517 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3519 WARN_ON(PageChecked(page));
3520 if (!page_has_buffers(page))
3523 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3525 return try_to_free_buffers(page);
3529 * O_DIRECT for ext3 (or indirect map) based files
3531 * If the O_DIRECT write will extend the file then add this inode to the
3532 * orphan list. So recovery will truncate it back to the original size
3533 * if the machine crashes during the write.
3535 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3536 * crashes then stale disk data _may_ be exposed inside the file. But current
3537 * VFS code falls back into buffered path in that case so we are safe.
3539 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3540 const struct iovec *iov, loff_t offset,
3541 unsigned long nr_segs)
3543 struct file *file = iocb->ki_filp;
3544 struct inode *inode = file->f_mapping->host;
3545 struct ext4_inode_info *ei = EXT4_I(inode);
3549 size_t count = iov_length(iov, nr_segs);
3553 loff_t final_size = offset + count;
3555 if (final_size > inode->i_size) {
3556 /* Credits for sb + inode write */
3557 handle = ext4_journal_start(inode, 2);
3558 if (IS_ERR(handle)) {
3559 ret = PTR_ERR(handle);
3562 ret = ext4_orphan_add(handle, inode);
3564 ext4_journal_stop(handle);
3568 ei->i_disksize = inode->i_size;
3569 ext4_journal_stop(handle);
3574 if (rw == READ && ext4_should_dioread_nolock(inode))
3575 ret = __blockdev_direct_IO(rw, iocb, inode,
3576 inode->i_sb->s_bdev, iov,
3578 ext4_get_block, NULL, NULL, 0);
3580 ret = blockdev_direct_IO(rw, iocb, inode,
3581 inode->i_sb->s_bdev, iov,
3583 ext4_get_block, NULL);
3585 if (unlikely((rw & WRITE) && ret < 0)) {
3586 loff_t isize = i_size_read(inode);
3587 loff_t end = offset + iov_length(iov, nr_segs);
3590 vmtruncate(inode, isize);
3593 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3599 /* Credits for sb + inode write */
3600 handle = ext4_journal_start(inode, 2);
3601 if (IS_ERR(handle)) {
3602 /* This is really bad luck. We've written the data
3603 * but cannot extend i_size. Bail out and pretend
3604 * the write failed... */
3605 ret = PTR_ERR(handle);
3607 ext4_orphan_del(NULL, inode);
3612 ext4_orphan_del(handle, inode);
3614 loff_t end = offset + ret;
3615 if (end > inode->i_size) {
3616 ei->i_disksize = end;
3617 i_size_write(inode, end);
3619 * We're going to return a positive `ret'
3620 * here due to non-zero-length I/O, so there's
3621 * no way of reporting error returns from
3622 * ext4_mark_inode_dirty() to userspace. So
3625 ext4_mark_inode_dirty(handle, inode);
3628 err = ext4_journal_stop(handle);
3637 * ext4_get_block used when preparing for a DIO write or buffer write.
3638 * We allocate an uinitialized extent if blocks haven't been allocated.
3639 * The extent will be converted to initialized after the IO is complete.
3641 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3642 struct buffer_head *bh_result, int create)
3644 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3645 inode->i_ino, create);
3646 return _ext4_get_block(inode, iblock, bh_result,
3647 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3650 static void dump_completed_IO(struct inode * inode)
3653 struct list_head *cur, *before, *after;
3654 ext4_io_end_t *io, *io0, *io1;
3655 unsigned long flags;
3657 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3658 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3662 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3663 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3664 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3667 io0 = container_of(before, ext4_io_end_t, list);
3669 io1 = container_of(after, ext4_io_end_t, list);
3671 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3672 io, inode->i_ino, io0, io1);
3674 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3679 * check a range of space and convert unwritten extents to written.
3681 static int ext4_end_io_nolock(ext4_io_end_t *io)
3683 struct inode *inode = io->inode;
3684 loff_t offset = io->offset;
3685 ssize_t size = io->size;
3688 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3689 "list->prev 0x%p\n",
3690 io, inode->i_ino, io->list.next, io->list.prev);
3692 if (list_empty(&io->list))
3695 if (io->flag != EXT4_IO_UNWRITTEN)
3698 ret = ext4_convert_unwritten_extents(inode, offset, size);
3700 printk(KERN_EMERG "%s: failed to convert unwritten"
3701 "extents to written extents, error is %d"
3702 " io is still on inode %lu aio dio list\n",
3703 __func__, ret, inode->i_ino);
3708 aio_complete(io->iocb, io->result, 0);
3709 /* clear the DIO AIO unwritten flag */
3715 * work on completed aio dio IO, to convert unwritten extents to extents
3717 static void ext4_end_io_work(struct work_struct *work)
3719 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3720 struct inode *inode = io->inode;
3721 struct ext4_inode_info *ei = EXT4_I(inode);
3722 unsigned long flags;
3725 mutex_lock(&inode->i_mutex);
3726 ret = ext4_end_io_nolock(io);
3728 mutex_unlock(&inode->i_mutex);
3732 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3733 if (!list_empty(&io->list))
3734 list_del_init(&io->list);
3735 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3736 mutex_unlock(&inode->i_mutex);
3737 ext4_free_io_end(io);
3741 * This function is called from ext4_sync_file().
3743 * When IO is completed, the work to convert unwritten extents to
3744 * written is queued on workqueue but may not get immediately
3745 * scheduled. When fsync is called, we need to ensure the
3746 * conversion is complete before fsync returns.
3747 * The inode keeps track of a list of pending/completed IO that
3748 * might needs to do the conversion. This function walks through
3749 * the list and convert the related unwritten extents for completed IO
3751 * The function return the number of pending IOs on success.
3753 int flush_completed_IO(struct inode *inode)
3756 struct ext4_inode_info *ei = EXT4_I(inode);
3757 unsigned long flags;
3761 if (list_empty(&ei->i_completed_io_list))
3764 dump_completed_IO(inode);
3765 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3766 while (!list_empty(&ei->i_completed_io_list)){
3767 io = list_entry(ei->i_completed_io_list.next,
3768 ext4_io_end_t, list);
3770 * Calling ext4_end_io_nolock() to convert completed
3773 * When ext4_sync_file() is called, run_queue() may already
3774 * about to flush the work corresponding to this io structure.
3775 * It will be upset if it founds the io structure related
3776 * to the work-to-be schedule is freed.
3778 * Thus we need to keep the io structure still valid here after
3779 * convertion finished. The io structure has a flag to
3780 * avoid double converting from both fsync and background work
3783 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3784 ret = ext4_end_io_nolock(io);
3785 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3789 list_del_init(&io->list);
3791 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3792 return (ret2 < 0) ? ret2 : 0;
3795 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3797 ext4_io_end_t *io = NULL;
3799 io = kmalloc(sizeof(*io), flags);
3810 INIT_WORK(&io->work, ext4_end_io_work);
3811 INIT_LIST_HEAD(&io->list);
3817 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3818 ssize_t size, void *private, int ret,
3821 ext4_io_end_t *io_end = iocb->private;
3822 struct workqueue_struct *wq;
3823 unsigned long flags;
3824 struct ext4_inode_info *ei;
3826 /* if not async direct IO or dio with 0 bytes write, just return */
3827 if (!io_end || !size)
3830 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3831 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3832 iocb->private, io_end->inode->i_ino, iocb, offset,
3835 /* if not aio dio with unwritten extents, just free io and return */
3836 if (io_end->flag != EXT4_IO_UNWRITTEN){
3837 ext4_free_io_end(io_end);
3838 iocb->private = NULL;
3841 aio_complete(iocb, ret, 0);
3845 io_end->offset = offset;
3846 io_end->size = size;
3848 io_end->iocb = iocb;
3849 io_end->result = ret;
3851 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3853 /* queue the work to convert unwritten extents to written */
3854 queue_work(wq, &io_end->work);
3856 /* Add the io_end to per-inode completed aio dio list*/
3857 ei = EXT4_I(io_end->inode);
3858 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3859 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3860 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3861 iocb->private = NULL;
3864 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3866 ext4_io_end_t *io_end = bh->b_private;
3867 struct workqueue_struct *wq;
3868 struct inode *inode;
3869 unsigned long flags;
3871 if (!test_clear_buffer_uninit(bh) || !io_end)
3874 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3875 printk("sb umounted, discard end_io request for inode %lu\n",
3876 io_end->inode->i_ino);
3877 ext4_free_io_end(io_end);
3881 io_end->flag = EXT4_IO_UNWRITTEN;
3882 inode = io_end->inode;
3884 /* Add the io_end to per-inode completed io list*/
3885 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3886 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3887 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3889 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3890 /* queue the work to convert unwritten extents to written */
3891 queue_work(wq, &io_end->work);
3893 bh->b_private = NULL;
3894 bh->b_end_io = NULL;
3895 clear_buffer_uninit(bh);
3896 end_buffer_async_write(bh, uptodate);
3899 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3901 ext4_io_end_t *io_end;
3902 struct page *page = bh->b_page;
3903 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3904 size_t size = bh->b_size;
3907 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3909 if (printk_ratelimit())
3910 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3914 io_end->offset = offset;
3915 io_end->size = size;
3917 * We need to hold a reference to the page to make sure it
3918 * doesn't get evicted before ext4_end_io_work() has a chance
3919 * to convert the extent from written to unwritten.
3921 io_end->page = page;
3922 get_page(io_end->page);
3924 bh->b_private = io_end;
3925 bh->b_end_io = ext4_end_io_buffer_write;
3930 * For ext4 extent files, ext4 will do direct-io write to holes,
3931 * preallocated extents, and those write extend the file, no need to
3932 * fall back to buffered IO.
3934 * For holes, we fallocate those blocks, mark them as unintialized
3935 * If those blocks were preallocated, we mark sure they are splited, but
3936 * still keep the range to write as unintialized.
3938 * The unwrritten extents will be converted to written when DIO is completed.
3939 * For async direct IO, since the IO may still pending when return, we
3940 * set up an end_io call back function, which will do the convertion
3941 * when async direct IO completed.
3943 * If the O_DIRECT write will extend the file then add this inode to the
3944 * orphan list. So recovery will truncate it back to the original size
3945 * if the machine crashes during the write.
3948 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3949 const struct iovec *iov, loff_t offset,
3950 unsigned long nr_segs)
3952 struct file *file = iocb->ki_filp;
3953 struct inode *inode = file->f_mapping->host;
3955 size_t count = iov_length(iov, nr_segs);
3957 loff_t final_size = offset + count;
3958 if (rw == WRITE && final_size <= inode->i_size) {
3960 * We could direct write to holes and fallocate.
3962 * Allocated blocks to fill the hole are marked as uninitialized
3963 * to prevent paralel buffered read to expose the stale data
3964 * before DIO complete the data IO.
3966 * As to previously fallocated extents, ext4 get_block
3967 * will just simply mark the buffer mapped but still
3968 * keep the extents uninitialized.
3970 * for non AIO case, we will convert those unwritten extents
3971 * to written after return back from blockdev_direct_IO.
3973 * for async DIO, the conversion needs to be defered when
3974 * the IO is completed. The ext4 end_io callback function
3975 * will be called to take care of the conversion work.
3976 * Here for async case, we allocate an io_end structure to
3979 iocb->private = NULL;
3980 EXT4_I(inode)->cur_aio_dio = NULL;
3981 if (!is_sync_kiocb(iocb)) {
3982 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3986 * we save the io structure for current async
3987 * direct IO, so that later ext4_map_blocks()
3988 * could flag the io structure whether there
3989 * is a unwritten extents needs to be converted
3990 * when IO is completed.
3992 EXT4_I(inode)->cur_aio_dio = iocb->private;
3995 ret = blockdev_direct_IO(rw, iocb, inode,
3996 inode->i_sb->s_bdev, iov,
3998 ext4_get_block_write,
4001 EXT4_I(inode)->cur_aio_dio = NULL;
4003 * The io_end structure takes a reference to the inode,
4004 * that structure needs to be destroyed and the
4005 * reference to the inode need to be dropped, when IO is
4006 * complete, even with 0 byte write, or failed.
4008 * In the successful AIO DIO case, the io_end structure will be
4009 * desctroyed and the reference to the inode will be dropped
4010 * after the end_io call back function is called.
4012 * In the case there is 0 byte write, or error case, since
4013 * VFS direct IO won't invoke the end_io call back function,
4014 * we need to free the end_io structure here.
4016 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
4017 ext4_free_io_end(iocb->private);
4018 iocb->private = NULL;
4019 } else if (ret > 0 && ext4_test_inode_state(inode,
4020 EXT4_STATE_DIO_UNWRITTEN)) {
4023 * for non AIO case, since the IO is already
4024 * completed, we could do the convertion right here
4026 err = ext4_convert_unwritten_extents(inode,
4030 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
4035 /* for write the the end of file case, we fall back to old way */
4036 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4039 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
4040 const struct iovec *iov, loff_t offset,
4041 unsigned long nr_segs)
4043 struct file *file = iocb->ki_filp;
4044 struct inode *inode = file->f_mapping->host;
4046 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4047 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4049 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4053 * Pages can be marked dirty completely asynchronously from ext4's journalling
4054 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4055 * much here because ->set_page_dirty is called under VFS locks. The page is
4056 * not necessarily locked.
4058 * We cannot just dirty the page and leave attached buffers clean, because the
4059 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4060 * or jbddirty because all the journalling code will explode.
4062 * So what we do is to mark the page "pending dirty" and next time writepage
4063 * is called, propagate that into the buffers appropriately.
4065 static int ext4_journalled_set_page_dirty(struct page *page)
4067 SetPageChecked(page);
4068 return __set_page_dirty_nobuffers(page);
4071 static const struct address_space_operations ext4_ordered_aops = {
4072 .readpage = ext4_readpage,
4073 .readpages = ext4_readpages,
4074 .writepage = ext4_writepage,
4075 .sync_page = block_sync_page,
4076 .write_begin = ext4_write_begin,
4077 .write_end = ext4_ordered_write_end,
4079 .invalidatepage = ext4_invalidatepage,
4080 .releasepage = ext4_releasepage,
4081 .direct_IO = ext4_direct_IO,
4082 .migratepage = buffer_migrate_page,
4083 .is_partially_uptodate = block_is_partially_uptodate,
4084 .error_remove_page = generic_error_remove_page,
4087 static const struct address_space_operations ext4_writeback_aops = {
4088 .readpage = ext4_readpage,
4089 .readpages = ext4_readpages,
4090 .writepage = ext4_writepage,
4091 .sync_page = block_sync_page,
4092 .write_begin = ext4_write_begin,
4093 .write_end = ext4_writeback_write_end,
4095 .invalidatepage = ext4_invalidatepage,
4096 .releasepage = ext4_releasepage,
4097 .direct_IO = ext4_direct_IO,
4098 .migratepage = buffer_migrate_page,
4099 .is_partially_uptodate = block_is_partially_uptodate,
4100 .error_remove_page = generic_error_remove_page,
4103 static const struct address_space_operations ext4_journalled_aops = {
4104 .readpage = ext4_readpage,
4105 .readpages = ext4_readpages,
4106 .writepage = ext4_writepage,
4107 .sync_page = block_sync_page,
4108 .write_begin = ext4_write_begin,
4109 .write_end = ext4_journalled_write_end,
4110 .set_page_dirty = ext4_journalled_set_page_dirty,
4112 .invalidatepage = ext4_invalidatepage,
4113 .releasepage = ext4_releasepage,
4114 .is_partially_uptodate = block_is_partially_uptodate,
4115 .error_remove_page = generic_error_remove_page,
4118 static const struct address_space_operations ext4_da_aops = {
4119 .readpage = ext4_readpage,
4120 .readpages = ext4_readpages,
4121 .writepage = ext4_writepage,
4122 .writepages = ext4_da_writepages,
4123 .sync_page = block_sync_page,
4124 .write_begin = ext4_da_write_begin,
4125 .write_end = ext4_da_write_end,
4127 .invalidatepage = ext4_da_invalidatepage,
4128 .releasepage = ext4_releasepage,
4129 .direct_IO = ext4_direct_IO,
4130 .migratepage = buffer_migrate_page,
4131 .is_partially_uptodate = block_is_partially_uptodate,
4132 .error_remove_page = generic_error_remove_page,
4135 void ext4_set_aops(struct inode *inode)
4137 if (ext4_should_order_data(inode) &&
4138 test_opt(inode->i_sb, DELALLOC))
4139 inode->i_mapping->a_ops = &ext4_da_aops;
4140 else if (ext4_should_order_data(inode))
4141 inode->i_mapping->a_ops = &ext4_ordered_aops;
4142 else if (ext4_should_writeback_data(inode) &&
4143 test_opt(inode->i_sb, DELALLOC))
4144 inode->i_mapping->a_ops = &ext4_da_aops;
4145 else if (ext4_should_writeback_data(inode))
4146 inode->i_mapping->a_ops = &ext4_writeback_aops;
4148 inode->i_mapping->a_ops = &ext4_journalled_aops;
4152 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4153 * up to the end of the block which corresponds to `from'.
4154 * This required during truncate. We need to physically zero the tail end
4155 * of that block so it doesn't yield old data if the file is later grown.
4157 int ext4_block_truncate_page(handle_t *handle,
4158 struct address_space *mapping, loff_t from)
4160 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4161 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4162 unsigned blocksize, length, pos;
4164 struct inode *inode = mapping->host;
4165 struct buffer_head *bh;
4169 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4170 mapping_gfp_mask(mapping) & ~__GFP_FS);
4174 blocksize = inode->i_sb->s_blocksize;
4175 length = blocksize - (offset & (blocksize - 1));
4176 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4178 if (!page_has_buffers(page))
4179 create_empty_buffers(page, blocksize, 0);
4181 /* Find the buffer that contains "offset" */
4182 bh = page_buffers(page);
4184 while (offset >= pos) {
4185 bh = bh->b_this_page;
4191 if (buffer_freed(bh)) {
4192 BUFFER_TRACE(bh, "freed: skip");
4196 if (!buffer_mapped(bh)) {
4197 BUFFER_TRACE(bh, "unmapped");
4198 ext4_get_block(inode, iblock, bh, 0);
4199 /* unmapped? It's a hole - nothing to do */
4200 if (!buffer_mapped(bh)) {
4201 BUFFER_TRACE(bh, "still unmapped");
4206 /* Ok, it's mapped. Make sure it's up-to-date */
4207 if (PageUptodate(page))
4208 set_buffer_uptodate(bh);
4210 if (!buffer_uptodate(bh)) {
4212 ll_rw_block(READ, 1, &bh);
4214 /* Uhhuh. Read error. Complain and punt. */
4215 if (!buffer_uptodate(bh))
4219 if (ext4_should_journal_data(inode)) {
4220 BUFFER_TRACE(bh, "get write access");
4221 err = ext4_journal_get_write_access(handle, bh);
4226 zero_user(page, offset, length);
4228 BUFFER_TRACE(bh, "zeroed end of block");
4231 if (ext4_should_journal_data(inode)) {
4232 err = ext4_handle_dirty_metadata(handle, inode, bh);
4234 if (ext4_should_order_data(inode))
4235 err = ext4_jbd2_file_inode(handle, inode);
4236 mark_buffer_dirty(bh);
4241 page_cache_release(page);
4246 * Probably it should be a library function... search for first non-zero word
4247 * or memcmp with zero_page, whatever is better for particular architecture.
4250 static inline int all_zeroes(__le32 *p, __le32 *q)
4259 * ext4_find_shared - find the indirect blocks for partial truncation.
4260 * @inode: inode in question
4261 * @depth: depth of the affected branch
4262 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4263 * @chain: place to store the pointers to partial indirect blocks
4264 * @top: place to the (detached) top of branch
4266 * This is a helper function used by ext4_truncate().
4268 * When we do truncate() we may have to clean the ends of several
4269 * indirect blocks but leave the blocks themselves alive. Block is
4270 * partially truncated if some data below the new i_size is refered
4271 * from it (and it is on the path to the first completely truncated
4272 * data block, indeed). We have to free the top of that path along
4273 * with everything to the right of the path. Since no allocation
4274 * past the truncation point is possible until ext4_truncate()
4275 * finishes, we may safely do the latter, but top of branch may
4276 * require special attention - pageout below the truncation point
4277 * might try to populate it.
4279 * We atomically detach the top of branch from the tree, store the
4280 * block number of its root in *@top, pointers to buffer_heads of
4281 * partially truncated blocks - in @chain[].bh and pointers to
4282 * their last elements that should not be removed - in
4283 * @chain[].p. Return value is the pointer to last filled element
4286 * The work left to caller to do the actual freeing of subtrees:
4287 * a) free the subtree starting from *@top
4288 * b) free the subtrees whose roots are stored in
4289 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4290 * c) free the subtrees growing from the inode past the @chain[0].
4291 * (no partially truncated stuff there). */
4293 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4294 ext4_lblk_t offsets[4], Indirect chain[4],
4297 Indirect *partial, *p;
4301 /* Make k index the deepest non-null offset + 1 */
4302 for (k = depth; k > 1 && !offsets[k-1]; k--)
4304 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4305 /* Writer: pointers */
4307 partial = chain + k-1;
4309 * If the branch acquired continuation since we've looked at it -
4310 * fine, it should all survive and (new) top doesn't belong to us.
4312 if (!partial->key && *partial->p)
4315 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4318 * OK, we've found the last block that must survive. The rest of our
4319 * branch should be detached before unlocking. However, if that rest
4320 * of branch is all ours and does not grow immediately from the inode
4321 * it's easier to cheat and just decrement partial->p.
4323 if (p == chain + k - 1 && p > chain) {
4327 /* Nope, don't do this in ext4. Must leave the tree intact */
4334 while (partial > p) {
4335 brelse(partial->bh);
4343 * Zero a number of block pointers in either an inode or an indirect block.
4344 * If we restart the transaction we must again get write access to the
4345 * indirect block for further modification.
4347 * We release `count' blocks on disk, but (last - first) may be greater
4348 * than `count' because there can be holes in there.
4350 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4351 struct buffer_head *bh,
4352 ext4_fsblk_t block_to_free,
4353 unsigned long count, __le32 *first,
4357 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4359 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4360 flags |= EXT4_FREE_BLOCKS_METADATA;
4362 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4364 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4365 "blocks %llu len %lu",
4366 (unsigned long long) block_to_free, count);
4370 if (try_to_extend_transaction(handle, inode)) {
4372 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4373 ext4_handle_dirty_metadata(handle, inode, bh);
4375 ext4_mark_inode_dirty(handle, inode);
4376 ext4_truncate_restart_trans(handle, inode,
4377 blocks_for_truncate(inode));
4379 BUFFER_TRACE(bh, "retaking write access");
4380 ext4_journal_get_write_access(handle, bh);
4384 for (p = first; p < last; p++)
4387 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4392 * ext4_free_data - free a list of data blocks
4393 * @handle: handle for this transaction
4394 * @inode: inode we are dealing with
4395 * @this_bh: indirect buffer_head which contains *@first and *@last
4396 * @first: array of block numbers
4397 * @last: points immediately past the end of array
4399 * We are freeing all blocks refered from that array (numbers are stored as
4400 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4402 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4403 * blocks are contiguous then releasing them at one time will only affect one
4404 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4405 * actually use a lot of journal space.
4407 * @this_bh will be %NULL if @first and @last point into the inode's direct
4410 static void ext4_free_data(handle_t *handle, struct inode *inode,
4411 struct buffer_head *this_bh,
4412 __le32 *first, __le32 *last)
4414 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4415 unsigned long count = 0; /* Number of blocks in the run */
4416 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4419 ext4_fsblk_t nr; /* Current block # */
4420 __le32 *p; /* Pointer into inode/ind
4421 for current block */
4424 if (this_bh) { /* For indirect block */
4425 BUFFER_TRACE(this_bh, "get_write_access");
4426 err = ext4_journal_get_write_access(handle, this_bh);
4427 /* Important: if we can't update the indirect pointers
4428 * to the blocks, we can't free them. */
4433 for (p = first; p < last; p++) {
4434 nr = le32_to_cpu(*p);
4436 /* accumulate blocks to free if they're contiguous */
4439 block_to_free_p = p;
4441 } else if (nr == block_to_free + count) {
4444 if (ext4_clear_blocks(handle, inode, this_bh,
4445 block_to_free, count,
4446 block_to_free_p, p))
4449 block_to_free_p = p;
4456 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4457 count, block_to_free_p, p);
4460 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4463 * The buffer head should have an attached journal head at this
4464 * point. However, if the data is corrupted and an indirect
4465 * block pointed to itself, it would have been detached when
4466 * the block was cleared. Check for this instead of OOPSing.
4468 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4469 ext4_handle_dirty_metadata(handle, inode, this_bh);
4471 EXT4_ERROR_INODE(inode,
4472 "circular indirect block detected at "
4474 (unsigned long long) this_bh->b_blocknr);
4479 * ext4_free_branches - free an array of branches
4480 * @handle: JBD handle for this transaction
4481 * @inode: inode we are dealing with
4482 * @parent_bh: the buffer_head which contains *@first and *@last
4483 * @first: array of block numbers
4484 * @last: pointer immediately past the end of array
4485 * @depth: depth of the branches to free
4487 * We are freeing all blocks refered from these branches (numbers are
4488 * stored as little-endian 32-bit) and updating @inode->i_blocks
4491 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4492 struct buffer_head *parent_bh,
4493 __le32 *first, __le32 *last, int depth)
4498 if (ext4_handle_is_aborted(handle))
4502 struct buffer_head *bh;
4503 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4505 while (--p >= first) {
4506 nr = le32_to_cpu(*p);
4508 continue; /* A hole */
4510 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4512 EXT4_ERROR_INODE(inode,
4513 "invalid indirect mapped "
4514 "block %lu (level %d)",
4515 (unsigned long) nr, depth);
4519 /* Go read the buffer for the next level down */
4520 bh = sb_bread(inode->i_sb, nr);
4523 * A read failure? Report error and clear slot
4527 EXT4_ERROR_INODE_BLOCK(inode, nr,
4532 /* This zaps the entire block. Bottom up. */
4533 BUFFER_TRACE(bh, "free child branches");
4534 ext4_free_branches(handle, inode, bh,
4535 (__le32 *) bh->b_data,
4536 (__le32 *) bh->b_data + addr_per_block,
4540 * Everything below this this pointer has been
4541 * released. Now let this top-of-subtree go.
4543 * We want the freeing of this indirect block to be
4544 * atomic in the journal with the updating of the
4545 * bitmap block which owns it. So make some room in
4548 * We zero the parent pointer *after* freeing its
4549 * pointee in the bitmaps, so if extend_transaction()
4550 * for some reason fails to put the bitmap changes and
4551 * the release into the same transaction, recovery
4552 * will merely complain about releasing a free block,
4553 * rather than leaking blocks.
4555 if (ext4_handle_is_aborted(handle))
4557 if (try_to_extend_transaction(handle, inode)) {
4558 ext4_mark_inode_dirty(handle, inode);
4559 ext4_truncate_restart_trans(handle, inode,
4560 blocks_for_truncate(inode));
4564 * The forget flag here is critical because if
4565 * we are journaling (and not doing data
4566 * journaling), we have to make sure a revoke
4567 * record is written to prevent the journal
4568 * replay from overwriting the (former)
4569 * indirect block if it gets reallocated as a
4570 * data block. This must happen in the same
4571 * transaction where the data blocks are
4574 ext4_free_blocks(handle, inode, 0, nr, 1,
4575 EXT4_FREE_BLOCKS_METADATA|
4576 EXT4_FREE_BLOCKS_FORGET);
4580 * The block which we have just freed is
4581 * pointed to by an indirect block: journal it
4583 BUFFER_TRACE(parent_bh, "get_write_access");
4584 if (!ext4_journal_get_write_access(handle,
4587 BUFFER_TRACE(parent_bh,
4588 "call ext4_handle_dirty_metadata");
4589 ext4_handle_dirty_metadata(handle,
4596 /* We have reached the bottom of the tree. */
4597 BUFFER_TRACE(parent_bh, "free data blocks");
4598 ext4_free_data(handle, inode, parent_bh, first, last);
4602 int ext4_can_truncate(struct inode *inode)
4604 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4606 if (S_ISREG(inode->i_mode))
4608 if (S_ISDIR(inode->i_mode))
4610 if (S_ISLNK(inode->i_mode))
4611 return !ext4_inode_is_fast_symlink(inode);
4618 * We block out ext4_get_block() block instantiations across the entire
4619 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4620 * simultaneously on behalf of the same inode.
4622 * As we work through the truncate and commmit bits of it to the journal there
4623 * is one core, guiding principle: the file's tree must always be consistent on
4624 * disk. We must be able to restart the truncate after a crash.
4626 * The file's tree may be transiently inconsistent in memory (although it
4627 * probably isn't), but whenever we close off and commit a journal transaction,
4628 * the contents of (the filesystem + the journal) must be consistent and
4629 * restartable. It's pretty simple, really: bottom up, right to left (although
4630 * left-to-right works OK too).
4632 * Note that at recovery time, journal replay occurs *before* the restart of
4633 * truncate against the orphan inode list.
4635 * The committed inode has the new, desired i_size (which is the same as
4636 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4637 * that this inode's truncate did not complete and it will again call
4638 * ext4_truncate() to have another go. So there will be instantiated blocks
4639 * to the right of the truncation point in a crashed ext4 filesystem. But
4640 * that's fine - as long as they are linked from the inode, the post-crash
4641 * ext4_truncate() run will find them and release them.
4643 void ext4_truncate(struct inode *inode)
4646 struct ext4_inode_info *ei = EXT4_I(inode);
4647 __le32 *i_data = ei->i_data;
4648 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4649 struct address_space *mapping = inode->i_mapping;
4650 ext4_lblk_t offsets[4];
4655 ext4_lblk_t last_block;
4656 unsigned blocksize = inode->i_sb->s_blocksize;
4658 if (!ext4_can_truncate(inode))
4661 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4663 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4664 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4666 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4667 ext4_ext_truncate(inode);
4671 handle = start_transaction(inode);
4673 return; /* AKPM: return what? */
4675 last_block = (inode->i_size + blocksize-1)
4676 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4678 if (inode->i_size & (blocksize - 1))
4679 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4682 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4684 goto out_stop; /* error */
4687 * OK. This truncate is going to happen. We add the inode to the
4688 * orphan list, so that if this truncate spans multiple transactions,
4689 * and we crash, we will resume the truncate when the filesystem
4690 * recovers. It also marks the inode dirty, to catch the new size.
4692 * Implication: the file must always be in a sane, consistent
4693 * truncatable state while each transaction commits.
4695 if (ext4_orphan_add(handle, inode))
4699 * From here we block out all ext4_get_block() callers who want to
4700 * modify the block allocation tree.
4702 down_write(&ei->i_data_sem);
4704 ext4_discard_preallocations(inode);
4707 * The orphan list entry will now protect us from any crash which
4708 * occurs before the truncate completes, so it is now safe to propagate
4709 * the new, shorter inode size (held for now in i_size) into the
4710 * on-disk inode. We do this via i_disksize, which is the value which
4711 * ext4 *really* writes onto the disk inode.
4713 ei->i_disksize = inode->i_size;
4715 if (n == 1) { /* direct blocks */
4716 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4717 i_data + EXT4_NDIR_BLOCKS);
4721 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4722 /* Kill the top of shared branch (not detached) */
4724 if (partial == chain) {
4725 /* Shared branch grows from the inode */
4726 ext4_free_branches(handle, inode, NULL,
4727 &nr, &nr+1, (chain+n-1) - partial);
4730 * We mark the inode dirty prior to restart,
4731 * and prior to stop. No need for it here.
4734 /* Shared branch grows from an indirect block */
4735 BUFFER_TRACE(partial->bh, "get_write_access");
4736 ext4_free_branches(handle, inode, partial->bh,
4738 partial->p+1, (chain+n-1) - partial);
4741 /* Clear the ends of indirect blocks on the shared branch */
4742 while (partial > chain) {
4743 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4744 (__le32*)partial->bh->b_data+addr_per_block,
4745 (chain+n-1) - partial);
4746 BUFFER_TRACE(partial->bh, "call brelse");
4747 brelse(partial->bh);
4751 /* Kill the remaining (whole) subtrees */
4752 switch (offsets[0]) {
4754 nr = i_data[EXT4_IND_BLOCK];
4756 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4757 i_data[EXT4_IND_BLOCK] = 0;
4759 case EXT4_IND_BLOCK:
4760 nr = i_data[EXT4_DIND_BLOCK];
4762 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4763 i_data[EXT4_DIND_BLOCK] = 0;
4765 case EXT4_DIND_BLOCK:
4766 nr = i_data[EXT4_TIND_BLOCK];
4768 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4769 i_data[EXT4_TIND_BLOCK] = 0;
4771 case EXT4_TIND_BLOCK:
4775 up_write(&ei->i_data_sem);
4776 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4777 ext4_mark_inode_dirty(handle, inode);
4780 * In a multi-transaction truncate, we only make the final transaction
4784 ext4_handle_sync(handle);
4787 * If this was a simple ftruncate(), and the file will remain alive
4788 * then we need to clear up the orphan record which we created above.
4789 * However, if this was a real unlink then we were called by
4790 * ext4_delete_inode(), and we allow that function to clean up the
4791 * orphan info for us.
4794 ext4_orphan_del(handle, inode);
4796 ext4_journal_stop(handle);
4800 * ext4_get_inode_loc returns with an extra refcount against the inode's
4801 * underlying buffer_head on success. If 'in_mem' is true, we have all
4802 * data in memory that is needed to recreate the on-disk version of this
4805 static int __ext4_get_inode_loc(struct inode *inode,
4806 struct ext4_iloc *iloc, int in_mem)
4808 struct ext4_group_desc *gdp;
4809 struct buffer_head *bh;
4810 struct super_block *sb = inode->i_sb;
4812 int inodes_per_block, inode_offset;
4815 if (!ext4_valid_inum(sb, inode->i_ino))
4818 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4819 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4824 * Figure out the offset within the block group inode table
4826 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4827 inode_offset = ((inode->i_ino - 1) %
4828 EXT4_INODES_PER_GROUP(sb));
4829 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4830 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4832 bh = sb_getblk(sb, block);
4834 EXT4_ERROR_INODE_BLOCK(inode, block,
4835 "unable to read itable block");
4838 if (!buffer_uptodate(bh)) {
4842 * If the buffer has the write error flag, we have failed
4843 * to write out another inode in the same block. In this
4844 * case, we don't have to read the block because we may
4845 * read the old inode data successfully.
4847 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4848 set_buffer_uptodate(bh);
4850 if (buffer_uptodate(bh)) {
4851 /* someone brought it uptodate while we waited */
4857 * If we have all information of the inode in memory and this
4858 * is the only valid inode in the block, we need not read the
4862 struct buffer_head *bitmap_bh;
4865 start = inode_offset & ~(inodes_per_block - 1);
4867 /* Is the inode bitmap in cache? */
4868 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4873 * If the inode bitmap isn't in cache then the
4874 * optimisation may end up performing two reads instead
4875 * of one, so skip it.
4877 if (!buffer_uptodate(bitmap_bh)) {
4881 for (i = start; i < start + inodes_per_block; i++) {
4882 if (i == inode_offset)
4884 if (ext4_test_bit(i, bitmap_bh->b_data))
4888 if (i == start + inodes_per_block) {
4889 /* all other inodes are free, so skip I/O */
4890 memset(bh->b_data, 0, bh->b_size);
4891 set_buffer_uptodate(bh);
4899 * If we need to do any I/O, try to pre-readahead extra
4900 * blocks from the inode table.
4902 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4903 ext4_fsblk_t b, end, table;
4906 table = ext4_inode_table(sb, gdp);
4907 /* s_inode_readahead_blks is always a power of 2 */
4908 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4911 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4912 num = EXT4_INODES_PER_GROUP(sb);
4913 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4914 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4915 num -= ext4_itable_unused_count(sb, gdp);
4916 table += num / inodes_per_block;
4920 sb_breadahead(sb, b++);
4924 * There are other valid inodes in the buffer, this inode
4925 * has in-inode xattrs, or we don't have this inode in memory.
4926 * Read the block from disk.
4929 bh->b_end_io = end_buffer_read_sync;
4930 submit_bh(READ_META, bh);
4932 if (!buffer_uptodate(bh)) {
4933 EXT4_ERROR_INODE_BLOCK(inode, block,
4934 "unable to read itable block");
4944 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4946 /* We have all inode data except xattrs in memory here. */
4947 return __ext4_get_inode_loc(inode, iloc,
4948 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4951 void ext4_set_inode_flags(struct inode *inode)
4953 unsigned int flags = EXT4_I(inode)->i_flags;
4955 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4956 if (flags & EXT4_SYNC_FL)
4957 inode->i_flags |= S_SYNC;
4958 if (flags & EXT4_APPEND_FL)
4959 inode->i_flags |= S_APPEND;
4960 if (flags & EXT4_IMMUTABLE_FL)
4961 inode->i_flags |= S_IMMUTABLE;
4962 if (flags & EXT4_NOATIME_FL)
4963 inode->i_flags |= S_NOATIME;
4964 if (flags & EXT4_DIRSYNC_FL)
4965 inode->i_flags |= S_DIRSYNC;
4968 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4969 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4971 unsigned int vfs_fl;
4972 unsigned long old_fl, new_fl;
4975 vfs_fl = ei->vfs_inode.i_flags;
4976 old_fl = ei->i_flags;
4977 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4978 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4980 if (vfs_fl & S_SYNC)
4981 new_fl |= EXT4_SYNC_FL;
4982 if (vfs_fl & S_APPEND)
4983 new_fl |= EXT4_APPEND_FL;
4984 if (vfs_fl & S_IMMUTABLE)
4985 new_fl |= EXT4_IMMUTABLE_FL;
4986 if (vfs_fl & S_NOATIME)
4987 new_fl |= EXT4_NOATIME_FL;
4988 if (vfs_fl & S_DIRSYNC)
4989 new_fl |= EXT4_DIRSYNC_FL;
4990 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4993 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4994 struct ext4_inode_info *ei)
4997 struct inode *inode = &(ei->vfs_inode);
4998 struct super_block *sb = inode->i_sb;
5000 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5001 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
5002 /* we are using combined 48 bit field */
5003 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
5004 le32_to_cpu(raw_inode->i_blocks_lo);
5005 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
5006 /* i_blocks represent file system block size */
5007 return i_blocks << (inode->i_blkbits - 9);
5012 return le32_to_cpu(raw_inode->i_blocks_lo);
5016 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
5018 struct ext4_iloc iloc;
5019 struct ext4_inode *raw_inode;
5020 struct ext4_inode_info *ei;
5021 struct inode *inode;
5022 journal_t *journal = EXT4_SB(sb)->s_journal;
5026 inode = iget_locked(sb, ino);
5028 return ERR_PTR(-ENOMEM);
5029 if (!(inode->i_state & I_NEW))
5035 ret = __ext4_get_inode_loc(inode, &iloc, 0);
5038 raw_inode = ext4_raw_inode(&iloc);
5039 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5040 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5041 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5042 if (!(test_opt(inode->i_sb, NO_UID32))) {
5043 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5044 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5046 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5048 ei->i_state_flags = 0;
5049 ei->i_dir_start_lookup = 0;
5050 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5051 /* We now have enough fields to check if the inode was active or not.
5052 * This is needed because nfsd might try to access dead inodes
5053 * the test is that same one that e2fsck uses
5054 * NeilBrown 1999oct15
5056 if (inode->i_nlink == 0) {
5057 if (inode->i_mode == 0 ||
5058 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5059 /* this inode is deleted */
5063 /* The only unlinked inodes we let through here have
5064 * valid i_mode and are being read by the orphan
5065 * recovery code: that's fine, we're about to complete
5066 * the process of deleting those. */
5068 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5069 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5070 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5071 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5073 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5074 inode->i_size = ext4_isize(raw_inode);
5075 ei->i_disksize = inode->i_size;
5077 ei->i_reserved_quota = 0;
5079 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5080 ei->i_block_group = iloc.block_group;
5081 ei->i_last_alloc_group = ~0;
5083 * NOTE! The in-memory inode i_data array is in little-endian order
5084 * even on big-endian machines: we do NOT byteswap the block numbers!
5086 for (block = 0; block < EXT4_N_BLOCKS; block++)
5087 ei->i_data[block] = raw_inode->i_block[block];
5088 INIT_LIST_HEAD(&ei->i_orphan);
5091 * Set transaction id's of transactions that have to be committed
5092 * to finish f[data]sync. We set them to currently running transaction
5093 * as we cannot be sure that the inode or some of its metadata isn't
5094 * part of the transaction - the inode could have been reclaimed and
5095 * now it is reread from disk.
5098 transaction_t *transaction;
5101 read_lock(&journal->j_state_lock);
5102 if (journal->j_running_transaction)
5103 transaction = journal->j_running_transaction;
5105 transaction = journal->j_committing_transaction;
5107 tid = transaction->t_tid;
5109 tid = journal->j_commit_sequence;
5110 read_unlock(&journal->j_state_lock);
5111 ei->i_sync_tid = tid;
5112 ei->i_datasync_tid = tid;
5115 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5116 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5117 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5118 EXT4_INODE_SIZE(inode->i_sb)) {
5122 if (ei->i_extra_isize == 0) {
5123 /* The extra space is currently unused. Use it. */
5124 ei->i_extra_isize = sizeof(struct ext4_inode) -
5125 EXT4_GOOD_OLD_INODE_SIZE;
5127 __le32 *magic = (void *)raw_inode +
5128 EXT4_GOOD_OLD_INODE_SIZE +
5130 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5131 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5134 ei->i_extra_isize = 0;
5136 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5137 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5138 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5139 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5141 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5142 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5143 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5145 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5149 if (ei->i_file_acl &&
5150 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5151 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5155 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5156 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5157 (S_ISLNK(inode->i_mode) &&
5158 !ext4_inode_is_fast_symlink(inode)))
5159 /* Validate extent which is part of inode */
5160 ret = ext4_ext_check_inode(inode);
5161 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5162 (S_ISLNK(inode->i_mode) &&
5163 !ext4_inode_is_fast_symlink(inode))) {
5164 /* Validate block references which are part of inode */
5165 ret = ext4_check_inode_blockref(inode);
5170 if (S_ISREG(inode->i_mode)) {
5171 inode->i_op = &ext4_file_inode_operations;
5172 inode->i_fop = &ext4_file_operations;
5173 ext4_set_aops(inode);
5174 } else if (S_ISDIR(inode->i_mode)) {
5175 inode->i_op = &ext4_dir_inode_operations;
5176 inode->i_fop = &ext4_dir_operations;
5177 } else if (S_ISLNK(inode->i_mode)) {
5178 if (ext4_inode_is_fast_symlink(inode)) {
5179 inode->i_op = &ext4_fast_symlink_inode_operations;
5180 nd_terminate_link(ei->i_data, inode->i_size,
5181 sizeof(ei->i_data) - 1);
5183 inode->i_op = &ext4_symlink_inode_operations;
5184 ext4_set_aops(inode);
5186 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5187 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5188 inode->i_op = &ext4_special_inode_operations;
5189 if (raw_inode->i_block[0])
5190 init_special_inode(inode, inode->i_mode,
5191 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5193 init_special_inode(inode, inode->i_mode,
5194 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5197 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5201 ext4_set_inode_flags(inode);
5202 unlock_new_inode(inode);
5208 return ERR_PTR(ret);
5211 static int ext4_inode_blocks_set(handle_t *handle,
5212 struct ext4_inode *raw_inode,
5213 struct ext4_inode_info *ei)
5215 struct inode *inode = &(ei->vfs_inode);
5216 u64 i_blocks = inode->i_blocks;
5217 struct super_block *sb = inode->i_sb;
5219 if (i_blocks <= ~0U) {
5221 * i_blocks can be represnted in a 32 bit variable
5222 * as multiple of 512 bytes
5224 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5225 raw_inode->i_blocks_high = 0;
5226 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5229 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5232 if (i_blocks <= 0xffffffffffffULL) {
5234 * i_blocks can be represented in a 48 bit variable
5235 * as multiple of 512 bytes
5237 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5238 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5239 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5241 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5242 /* i_block is stored in file system block size */
5243 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5244 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5245 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5251 * Post the struct inode info into an on-disk inode location in the
5252 * buffer-cache. This gobbles the caller's reference to the
5253 * buffer_head in the inode location struct.
5255 * The caller must have write access to iloc->bh.
5257 static int ext4_do_update_inode(handle_t *handle,
5258 struct inode *inode,
5259 struct ext4_iloc *iloc)
5261 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5262 struct ext4_inode_info *ei = EXT4_I(inode);
5263 struct buffer_head *bh = iloc->bh;
5264 int err = 0, rc, block;
5266 /* For fields not not tracking in the in-memory inode,
5267 * initialise them to zero for new inodes. */
5268 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5269 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5271 ext4_get_inode_flags(ei);
5272 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5273 if (!(test_opt(inode->i_sb, NO_UID32))) {
5274 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5275 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5277 * Fix up interoperability with old kernels. Otherwise, old inodes get
5278 * re-used with the upper 16 bits of the uid/gid intact
5281 raw_inode->i_uid_high =
5282 cpu_to_le16(high_16_bits(inode->i_uid));
5283 raw_inode->i_gid_high =
5284 cpu_to_le16(high_16_bits(inode->i_gid));
5286 raw_inode->i_uid_high = 0;
5287 raw_inode->i_gid_high = 0;
5290 raw_inode->i_uid_low =
5291 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5292 raw_inode->i_gid_low =
5293 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5294 raw_inode->i_uid_high = 0;
5295 raw_inode->i_gid_high = 0;
5297 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5299 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5300 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5301 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5302 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5304 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5306 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5307 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5308 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5309 cpu_to_le32(EXT4_OS_HURD))
5310 raw_inode->i_file_acl_high =
5311 cpu_to_le16(ei->i_file_acl >> 32);
5312 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5313 ext4_isize_set(raw_inode, ei->i_disksize);
5314 if (ei->i_disksize > 0x7fffffffULL) {
5315 struct super_block *sb = inode->i_sb;
5316 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5317 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5318 EXT4_SB(sb)->s_es->s_rev_level ==
5319 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5320 /* If this is the first large file
5321 * created, add a flag to the superblock.
5323 err = ext4_journal_get_write_access(handle,
5324 EXT4_SB(sb)->s_sbh);
5327 ext4_update_dynamic_rev(sb);
5328 EXT4_SET_RO_COMPAT_FEATURE(sb,
5329 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5331 ext4_handle_sync(handle);
5332 err = ext4_handle_dirty_metadata(handle, NULL,
5333 EXT4_SB(sb)->s_sbh);
5336 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5337 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5338 if (old_valid_dev(inode->i_rdev)) {
5339 raw_inode->i_block[0] =
5340 cpu_to_le32(old_encode_dev(inode->i_rdev));
5341 raw_inode->i_block[1] = 0;
5343 raw_inode->i_block[0] = 0;
5344 raw_inode->i_block[1] =
5345 cpu_to_le32(new_encode_dev(inode->i_rdev));
5346 raw_inode->i_block[2] = 0;
5349 for (block = 0; block < EXT4_N_BLOCKS; block++)
5350 raw_inode->i_block[block] = ei->i_data[block];
5352 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5353 if (ei->i_extra_isize) {
5354 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5355 raw_inode->i_version_hi =
5356 cpu_to_le32(inode->i_version >> 32);
5357 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5360 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5361 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5364 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5366 ext4_update_inode_fsync_trans(handle, inode, 0);
5369 ext4_std_error(inode->i_sb, err);
5374 * ext4_write_inode()
5376 * We are called from a few places:
5378 * - Within generic_file_write() for O_SYNC files.
5379 * Here, there will be no transaction running. We wait for any running
5380 * trasnaction to commit.
5382 * - Within sys_sync(), kupdate and such.
5383 * We wait on commit, if tol to.
5385 * - Within prune_icache() (PF_MEMALLOC == true)
5386 * Here we simply return. We can't afford to block kswapd on the
5389 * In all cases it is actually safe for us to return without doing anything,
5390 * because the inode has been copied into a raw inode buffer in
5391 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5394 * Note that we are absolutely dependent upon all inode dirtiers doing the
5395 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5396 * which we are interested.
5398 * It would be a bug for them to not do this. The code:
5400 * mark_inode_dirty(inode)
5402 * inode->i_size = expr;
5404 * is in error because a kswapd-driven write_inode() could occur while
5405 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5406 * will no longer be on the superblock's dirty inode list.
5408 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5412 if (current->flags & PF_MEMALLOC)
5415 if (EXT4_SB(inode->i_sb)->s_journal) {
5416 if (ext4_journal_current_handle()) {
5417 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5422 if (wbc->sync_mode != WB_SYNC_ALL)
5425 err = ext4_force_commit(inode->i_sb);
5427 struct ext4_iloc iloc;
5429 err = __ext4_get_inode_loc(inode, &iloc, 0);
5432 if (wbc->sync_mode == WB_SYNC_ALL)
5433 sync_dirty_buffer(iloc.bh);
5434 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5435 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5436 "IO error syncing inode");
5447 * Called from notify_change.
5449 * We want to trap VFS attempts to truncate the file as soon as
5450 * possible. In particular, we want to make sure that when the VFS
5451 * shrinks i_size, we put the inode on the orphan list and modify
5452 * i_disksize immediately, so that during the subsequent flushing of
5453 * dirty pages and freeing of disk blocks, we can guarantee that any
5454 * commit will leave the blocks being flushed in an unused state on
5455 * disk. (On recovery, the inode will get truncated and the blocks will
5456 * be freed, so we have a strong guarantee that no future commit will
5457 * leave these blocks visible to the user.)
5459 * Another thing we have to assure is that if we are in ordered mode
5460 * and inode is still attached to the committing transaction, we must
5461 * we start writeout of all the dirty pages which are being truncated.
5462 * This way we are sure that all the data written in the previous
5463 * transaction are already on disk (truncate waits for pages under
5466 * Called with inode->i_mutex down.
5468 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5470 struct inode *inode = dentry->d_inode;
5472 const unsigned int ia_valid = attr->ia_valid;
5474 error = inode_change_ok(inode, attr);
5478 if (is_quota_modification(inode, attr))
5479 dquot_initialize(inode);
5480 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5481 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5484 /* (user+group)*(old+new) structure, inode write (sb,
5485 * inode block, ? - but truncate inode update has it) */
5486 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5487 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5488 if (IS_ERR(handle)) {
5489 error = PTR_ERR(handle);
5492 error = dquot_transfer(inode, attr);
5494 ext4_journal_stop(handle);
5497 /* Update corresponding info in inode so that everything is in
5498 * one transaction */
5499 if (attr->ia_valid & ATTR_UID)
5500 inode->i_uid = attr->ia_uid;
5501 if (attr->ia_valid & ATTR_GID)
5502 inode->i_gid = attr->ia_gid;
5503 error = ext4_mark_inode_dirty(handle, inode);
5504 ext4_journal_stop(handle);
5507 if (attr->ia_valid & ATTR_SIZE) {
5508 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5509 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5511 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5516 if (S_ISREG(inode->i_mode) &&
5517 attr->ia_valid & ATTR_SIZE &&
5518 (attr->ia_size < inode->i_size ||
5519 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5522 handle = ext4_journal_start(inode, 3);
5523 if (IS_ERR(handle)) {
5524 error = PTR_ERR(handle);
5528 error = ext4_orphan_add(handle, inode);
5529 EXT4_I(inode)->i_disksize = attr->ia_size;
5530 rc = ext4_mark_inode_dirty(handle, inode);
5533 ext4_journal_stop(handle);
5535 if (ext4_should_order_data(inode)) {
5536 error = ext4_begin_ordered_truncate(inode,
5539 /* Do as much error cleanup as possible */
5540 handle = ext4_journal_start(inode, 3);
5541 if (IS_ERR(handle)) {
5542 ext4_orphan_del(NULL, inode);
5545 ext4_orphan_del(handle, inode);
5546 ext4_journal_stop(handle);
5550 /* ext4_truncate will clear the flag */
5551 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5552 ext4_truncate(inode);
5555 if ((attr->ia_valid & ATTR_SIZE) &&
5556 attr->ia_size != i_size_read(inode))
5557 rc = vmtruncate(inode, attr->ia_size);
5560 setattr_copy(inode, attr);
5561 mark_inode_dirty(inode);
5565 * If the call to ext4_truncate failed to get a transaction handle at
5566 * all, we need to clean up the in-core orphan list manually.
5569 ext4_orphan_del(NULL, inode);
5571 if (!rc && (ia_valid & ATTR_MODE))
5572 rc = ext4_acl_chmod(inode);
5575 ext4_std_error(inode->i_sb, error);
5581 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5584 struct inode *inode;
5585 unsigned long delalloc_blocks;
5587 inode = dentry->d_inode;
5588 generic_fillattr(inode, stat);
5591 * We can't update i_blocks if the block allocation is delayed
5592 * otherwise in the case of system crash before the real block
5593 * allocation is done, we will have i_blocks inconsistent with
5594 * on-disk file blocks.
5595 * We always keep i_blocks updated together with real
5596 * allocation. But to not confuse with user, stat
5597 * will return the blocks that include the delayed allocation
5598 * blocks for this file.
5600 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5601 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5602 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5604 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5608 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5613 /* if nrblocks are contiguous */
5616 * With N contiguous data blocks, it need at most
5617 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5618 * 2 dindirect blocks
5621 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5622 return indirects + 3;
5625 * if nrblocks are not contiguous, worse case, each block touch
5626 * a indirect block, and each indirect block touch a double indirect
5627 * block, plus a triple indirect block
5629 indirects = nrblocks * 2 + 1;
5633 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5635 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5636 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5637 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5641 * Account for index blocks, block groups bitmaps and block group
5642 * descriptor blocks if modify datablocks and index blocks
5643 * worse case, the indexs blocks spread over different block groups
5645 * If datablocks are discontiguous, they are possible to spread over
5646 * different block groups too. If they are contiuguous, with flexbg,
5647 * they could still across block group boundary.
5649 * Also account for superblock, inode, quota and xattr blocks
5651 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5653 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5659 * How many index blocks need to touch to modify nrblocks?
5660 * The "Chunk" flag indicating whether the nrblocks is
5661 * physically contiguous on disk
5663 * For Direct IO and fallocate, they calls get_block to allocate
5664 * one single extent at a time, so they could set the "Chunk" flag
5666 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5671 * Now let's see how many group bitmaps and group descriptors need
5681 if (groups > ngroups)
5683 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5684 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5686 /* bitmaps and block group descriptor blocks */
5687 ret += groups + gdpblocks;
5689 /* Blocks for super block, inode, quota and xattr blocks */
5690 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5696 * Calulate the total number of credits to reserve to fit
5697 * the modification of a single pages into a single transaction,
5698 * which may include multiple chunks of block allocations.
5700 * This could be called via ext4_write_begin()
5702 * We need to consider the worse case, when
5703 * one new block per extent.
5705 int ext4_writepage_trans_blocks(struct inode *inode)
5707 int bpp = ext4_journal_blocks_per_page(inode);
5710 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5712 /* Account for data blocks for journalled mode */
5713 if (ext4_should_journal_data(inode))
5719 * Calculate the journal credits for a chunk of data modification.
5721 * This is called from DIO, fallocate or whoever calling
5722 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5724 * journal buffers for data blocks are not included here, as DIO
5725 * and fallocate do no need to journal data buffers.
5727 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5729 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5733 * The caller must have previously called ext4_reserve_inode_write().
5734 * Give this, we know that the caller already has write access to iloc->bh.
5736 int ext4_mark_iloc_dirty(handle_t *handle,
5737 struct inode *inode, struct ext4_iloc *iloc)
5741 if (test_opt(inode->i_sb, I_VERSION))
5742 inode_inc_iversion(inode);
5744 /* the do_update_inode consumes one bh->b_count */
5747 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5748 err = ext4_do_update_inode(handle, inode, iloc);
5754 * On success, We end up with an outstanding reference count against
5755 * iloc->bh. This _must_ be cleaned up later.
5759 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5760 struct ext4_iloc *iloc)
5764 err = ext4_get_inode_loc(inode, iloc);
5766 BUFFER_TRACE(iloc->bh, "get_write_access");
5767 err = ext4_journal_get_write_access(handle, iloc->bh);
5773 ext4_std_error(inode->i_sb, err);
5778 * Expand an inode by new_extra_isize bytes.
5779 * Returns 0 on success or negative error number on failure.
5781 static int ext4_expand_extra_isize(struct inode *inode,
5782 unsigned int new_extra_isize,
5783 struct ext4_iloc iloc,
5786 struct ext4_inode *raw_inode;
5787 struct ext4_xattr_ibody_header *header;
5789 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5792 raw_inode = ext4_raw_inode(&iloc);
5794 header = IHDR(inode, raw_inode);
5796 /* No extended attributes present */
5797 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5798 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5799 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5801 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5805 /* try to expand with EAs present */
5806 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5811 * What we do here is to mark the in-core inode as clean with respect to inode
5812 * dirtiness (it may still be data-dirty).
5813 * This means that the in-core inode may be reaped by prune_icache
5814 * without having to perform any I/O. This is a very good thing,
5815 * because *any* task may call prune_icache - even ones which
5816 * have a transaction open against a different journal.
5818 * Is this cheating? Not really. Sure, we haven't written the
5819 * inode out, but prune_icache isn't a user-visible syncing function.
5820 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5821 * we start and wait on commits.
5823 * Is this efficient/effective? Well, we're being nice to the system
5824 * by cleaning up our inodes proactively so they can be reaped
5825 * without I/O. But we are potentially leaving up to five seconds'
5826 * worth of inodes floating about which prune_icache wants us to
5827 * write out. One way to fix that would be to get prune_icache()
5828 * to do a write_super() to free up some memory. It has the desired
5831 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5833 struct ext4_iloc iloc;
5834 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5835 static unsigned int mnt_count;
5839 err = ext4_reserve_inode_write(handle, inode, &iloc);
5840 if (ext4_handle_valid(handle) &&
5841 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5842 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5844 * We need extra buffer credits since we may write into EA block
5845 * with this same handle. If journal_extend fails, then it will
5846 * only result in a minor loss of functionality for that inode.
5847 * If this is felt to be critical, then e2fsck should be run to
5848 * force a large enough s_min_extra_isize.
5850 if ((jbd2_journal_extend(handle,
5851 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5852 ret = ext4_expand_extra_isize(inode,
5853 sbi->s_want_extra_isize,
5856 ext4_set_inode_state(inode,
5857 EXT4_STATE_NO_EXPAND);
5859 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5860 ext4_warning(inode->i_sb,
5861 "Unable to expand inode %lu. Delete"
5862 " some EAs or run e2fsck.",
5865 le16_to_cpu(sbi->s_es->s_mnt_count);
5871 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5876 * ext4_dirty_inode() is called from __mark_inode_dirty()
5878 * We're really interested in the case where a file is being extended.
5879 * i_size has been changed by generic_commit_write() and we thus need
5880 * to include the updated inode in the current transaction.
5882 * Also, dquot_alloc_block() will always dirty the inode when blocks
5883 * are allocated to the file.
5885 * If the inode is marked synchronous, we don't honour that here - doing
5886 * so would cause a commit on atime updates, which we don't bother doing.
5887 * We handle synchronous inodes at the highest possible level.
5889 void ext4_dirty_inode(struct inode *inode)
5893 handle = ext4_journal_start(inode, 2);
5897 ext4_mark_inode_dirty(handle, inode);
5899 ext4_journal_stop(handle);
5906 * Bind an inode's backing buffer_head into this transaction, to prevent
5907 * it from being flushed to disk early. Unlike
5908 * ext4_reserve_inode_write, this leaves behind no bh reference and
5909 * returns no iloc structure, so the caller needs to repeat the iloc
5910 * lookup to mark the inode dirty later.
5912 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5914 struct ext4_iloc iloc;
5918 err = ext4_get_inode_loc(inode, &iloc);
5920 BUFFER_TRACE(iloc.bh, "get_write_access");
5921 err = jbd2_journal_get_write_access(handle, iloc.bh);
5923 err = ext4_handle_dirty_metadata(handle,
5929 ext4_std_error(inode->i_sb, err);
5934 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5941 * We have to be very careful here: changing a data block's
5942 * journaling status dynamically is dangerous. If we write a
5943 * data block to the journal, change the status and then delete
5944 * that block, we risk forgetting to revoke the old log record
5945 * from the journal and so a subsequent replay can corrupt data.
5946 * So, first we make sure that the journal is empty and that
5947 * nobody is changing anything.
5950 journal = EXT4_JOURNAL(inode);
5953 if (is_journal_aborted(journal))
5956 jbd2_journal_lock_updates(journal);
5957 jbd2_journal_flush(journal);
5960 * OK, there are no updates running now, and all cached data is
5961 * synced to disk. We are now in a completely consistent state
5962 * which doesn't have anything in the journal, and we know that
5963 * no filesystem updates are running, so it is safe to modify
5964 * the inode's in-core data-journaling state flag now.
5968 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5970 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5971 ext4_set_aops(inode);
5973 jbd2_journal_unlock_updates(journal);
5975 /* Finally we can mark the inode as dirty. */
5977 handle = ext4_journal_start(inode, 1);
5979 return PTR_ERR(handle);
5981 err = ext4_mark_inode_dirty(handle, inode);
5982 ext4_handle_sync(handle);
5983 ext4_journal_stop(handle);
5984 ext4_std_error(inode->i_sb, err);
5989 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5991 return !buffer_mapped(bh);
5994 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5996 struct page *page = vmf->page;
6001 struct file *file = vma->vm_file;
6002 struct inode *inode = file->f_path.dentry->d_inode;
6003 struct address_space *mapping = inode->i_mapping;
6006 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
6007 * get i_mutex because we are already holding mmap_sem.
6009 down_read(&inode->i_alloc_sem);
6010 size = i_size_read(inode);
6011 if (page->mapping != mapping || size <= page_offset(page)
6012 || !PageUptodate(page)) {
6013 /* page got truncated from under us? */
6017 if (PageMappedToDisk(page))
6020 if (page->index == size >> PAGE_CACHE_SHIFT)
6021 len = size & ~PAGE_CACHE_MASK;
6023 len = PAGE_CACHE_SIZE;
6027 * return if we have all the buffers mapped. This avoid
6028 * the need to call write_begin/write_end which does a
6029 * journal_start/journal_stop which can block and take
6032 if (page_has_buffers(page)) {
6033 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
6034 ext4_bh_unmapped)) {
6041 * OK, we need to fill the hole... Do write_begin write_end
6042 * to do block allocation/reservation.We are not holding
6043 * inode.i__mutex here. That allow * parallel write_begin,
6044 * write_end call. lock_page prevent this from happening
6045 * on the same page though
6047 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6048 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6051 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6052 len, len, page, fsdata);
6058 ret = VM_FAULT_SIGBUS;
6059 up_read(&inode->i_alloc_sem);