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);
63 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
64 struct buffer_head *bh_result, int create);
65 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
66 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
67 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
68 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
71 * Test whether an inode is a fast symlink.
73 static int ext4_inode_is_fast_symlink(struct inode *inode)
75 int ea_blocks = EXT4_I(inode)->i_file_acl ?
76 (inode->i_sb->s_blocksize >> 9) : 0;
78 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
82 * Work out how many blocks we need to proceed with the next chunk of a
83 * truncate transaction.
85 static unsigned long blocks_for_truncate(struct inode *inode)
89 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
91 /* Give ourselves just enough room to cope with inodes in which
92 * i_blocks is corrupt: we've seen disk corruptions in the past
93 * which resulted in random data in an inode which looked enough
94 * like a regular file for ext4 to try to delete it. Things
95 * will go a bit crazy if that happens, but at least we should
96 * try not to panic the whole kernel. */
100 /* But we need to bound the transaction so we don't overflow the
102 if (needed > EXT4_MAX_TRANS_DATA)
103 needed = EXT4_MAX_TRANS_DATA;
105 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
109 * Truncate transactions can be complex and absolutely huge. So we need to
110 * be able to restart the transaction at a conventient checkpoint to make
111 * sure we don't overflow the journal.
113 * start_transaction gets us a new handle for a truncate transaction,
114 * and extend_transaction tries to extend the existing one a bit. If
115 * extend fails, we need to propagate the failure up and restart the
116 * transaction in the top-level truncate loop. --sct
118 static handle_t *start_transaction(struct inode *inode)
122 result = ext4_journal_start(inode, blocks_for_truncate(inode));
126 ext4_std_error(inode->i_sb, PTR_ERR(result));
131 * Try to extend this transaction for the purposes of truncation.
133 * Returns 0 if we managed to create more room. If we can't create more
134 * room, and the transaction must be restarted we return 1.
136 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
138 if (!ext4_handle_valid(handle))
140 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
142 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
148 * Restart the transaction associated with *handle. This does a commit,
149 * so before we call here everything must be consistently dirtied against
152 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
158 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
159 * moment, get_block can be called only for blocks inside i_size since
160 * page cache has been already dropped and writes are blocked by
161 * i_mutex. So we can safely drop the i_data_sem here.
163 BUG_ON(EXT4_JOURNAL(inode) == NULL);
164 jbd_debug(2, "restarting handle %p\n", handle);
165 up_write(&EXT4_I(inode)->i_data_sem);
166 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
167 down_write(&EXT4_I(inode)->i_data_sem);
168 ext4_discard_preallocations(inode);
174 * Called at the last iput() if i_nlink is zero.
176 void ext4_evict_inode(struct inode *inode)
181 if (inode->i_nlink) {
182 truncate_inode_pages(&inode->i_data, 0);
186 if (!is_bad_inode(inode))
187 dquot_initialize(inode);
189 if (ext4_should_order_data(inode))
190 ext4_begin_ordered_truncate(inode, 0);
191 truncate_inode_pages(&inode->i_data, 0);
193 if (is_bad_inode(inode))
196 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
197 if (IS_ERR(handle)) {
198 ext4_std_error(inode->i_sb, PTR_ERR(handle));
200 * If we're going to skip the normal cleanup, we still need to
201 * make sure that the in-core orphan linked list is properly
204 ext4_orphan_del(NULL, inode);
209 ext4_handle_sync(handle);
211 err = ext4_mark_inode_dirty(handle, inode);
213 ext4_warning(inode->i_sb,
214 "couldn't mark inode dirty (err %d)", err);
218 ext4_truncate(inode);
221 * ext4_ext_truncate() doesn't reserve any slop when it
222 * restarts journal transactions; therefore there may not be
223 * enough credits left in the handle to remove the inode from
224 * the orphan list and set the dtime field.
226 if (!ext4_handle_has_enough_credits(handle, 3)) {
227 err = ext4_journal_extend(handle, 3);
229 err = ext4_journal_restart(handle, 3);
231 ext4_warning(inode->i_sb,
232 "couldn't extend journal (err %d)", err);
234 ext4_journal_stop(handle);
235 ext4_orphan_del(NULL, inode);
241 * Kill off the orphan record which ext4_truncate created.
242 * AKPM: I think this can be inside the above `if'.
243 * Note that ext4_orphan_del() has to be able to cope with the
244 * deletion of a non-existent orphan - this is because we don't
245 * know if ext4_truncate() actually created an orphan record.
246 * (Well, we could do this if we need to, but heck - it works)
248 ext4_orphan_del(handle, inode);
249 EXT4_I(inode)->i_dtime = get_seconds();
252 * One subtle ordering requirement: if anything has gone wrong
253 * (transaction abort, IO errors, whatever), then we can still
254 * do these next steps (the fs will already have been marked as
255 * having errors), but we can't free the inode if the mark_dirty
258 if (ext4_mark_inode_dirty(handle, inode))
259 /* If that failed, just do the required in-core inode clear. */
260 ext4_clear_inode(inode);
262 ext4_free_inode(handle, inode);
263 ext4_journal_stop(handle);
266 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
272 struct buffer_head *bh;
275 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
277 p->key = *(p->p = v);
282 * ext4_block_to_path - parse the block number into array of offsets
283 * @inode: inode in question (we are only interested in its superblock)
284 * @i_block: block number to be parsed
285 * @offsets: array to store the offsets in
286 * @boundary: set this non-zero if the referred-to block is likely to be
287 * followed (on disk) by an indirect block.
289 * To store the locations of file's data ext4 uses a data structure common
290 * for UNIX filesystems - tree of pointers anchored in the inode, with
291 * data blocks at leaves and indirect blocks in intermediate nodes.
292 * This function translates the block number into path in that tree -
293 * return value is the path length and @offsets[n] is the offset of
294 * pointer to (n+1)th node in the nth one. If @block is out of range
295 * (negative or too large) warning is printed and zero returned.
297 * Note: function doesn't find node addresses, so no IO is needed. All
298 * we need to know is the capacity of indirect blocks (taken from the
303 * Portability note: the last comparison (check that we fit into triple
304 * indirect block) is spelled differently, because otherwise on an
305 * architecture with 32-bit longs and 8Kb pages we might get into trouble
306 * if our filesystem had 8Kb blocks. We might use long long, but that would
307 * kill us on x86. Oh, well, at least the sign propagation does not matter -
308 * i_block would have to be negative in the very beginning, so we would not
312 static int ext4_block_to_path(struct inode *inode,
314 ext4_lblk_t offsets[4], int *boundary)
316 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
317 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
318 const long direct_blocks = EXT4_NDIR_BLOCKS,
319 indirect_blocks = ptrs,
320 double_blocks = (1 << (ptrs_bits * 2));
324 if (i_block < direct_blocks) {
325 offsets[n++] = i_block;
326 final = direct_blocks;
327 } else if ((i_block -= direct_blocks) < indirect_blocks) {
328 offsets[n++] = EXT4_IND_BLOCK;
329 offsets[n++] = i_block;
331 } else if ((i_block -= indirect_blocks) < double_blocks) {
332 offsets[n++] = EXT4_DIND_BLOCK;
333 offsets[n++] = i_block >> ptrs_bits;
334 offsets[n++] = i_block & (ptrs - 1);
336 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
337 offsets[n++] = EXT4_TIND_BLOCK;
338 offsets[n++] = i_block >> (ptrs_bits * 2);
339 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
340 offsets[n++] = i_block & (ptrs - 1);
343 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
344 i_block + direct_blocks +
345 indirect_blocks + double_blocks, inode->i_ino);
348 *boundary = final - 1 - (i_block & (ptrs - 1));
352 static int __ext4_check_blockref(const char *function, unsigned int line,
354 __le32 *p, unsigned int max)
356 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
360 while (bref < p+max) {
361 blk = le32_to_cpu(*bref++);
363 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
365 es->s_last_error_block = cpu_to_le64(blk);
366 ext4_error_inode(inode, function, line, blk,
375 #define ext4_check_indirect_blockref(inode, bh) \
376 __ext4_check_blockref(__func__, __LINE__, inode, \
377 (__le32 *)(bh)->b_data, \
378 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
380 #define ext4_check_inode_blockref(inode) \
381 __ext4_check_blockref(__func__, __LINE__, inode, \
382 EXT4_I(inode)->i_data, \
386 * ext4_get_branch - read the chain of indirect blocks leading to data
387 * @inode: inode in question
388 * @depth: depth of the chain (1 - direct pointer, etc.)
389 * @offsets: offsets of pointers in inode/indirect blocks
390 * @chain: place to store the result
391 * @err: here we store the error value
393 * Function fills the array of triples <key, p, bh> and returns %NULL
394 * if everything went OK or the pointer to the last filled triple
395 * (incomplete one) otherwise. Upon the return chain[i].key contains
396 * the number of (i+1)-th block in the chain (as it is stored in memory,
397 * i.e. little-endian 32-bit), chain[i].p contains the address of that
398 * number (it points into struct inode for i==0 and into the bh->b_data
399 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
400 * block for i>0 and NULL for i==0. In other words, it holds the block
401 * numbers of the chain, addresses they were taken from (and where we can
402 * verify that chain did not change) and buffer_heads hosting these
405 * Function stops when it stumbles upon zero pointer (absent block)
406 * (pointer to last triple returned, *@err == 0)
407 * or when it gets an IO error reading an indirect block
408 * (ditto, *@err == -EIO)
409 * or when it reads all @depth-1 indirect blocks successfully and finds
410 * the whole chain, all way to the data (returns %NULL, *err == 0).
412 * Need to be called with
413 * down_read(&EXT4_I(inode)->i_data_sem)
415 static Indirect *ext4_get_branch(struct inode *inode, int depth,
416 ext4_lblk_t *offsets,
417 Indirect chain[4], int *err)
419 struct super_block *sb = inode->i_sb;
421 struct buffer_head *bh;
424 /* i_data is not going away, no lock needed */
425 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
429 bh = sb_getblk(sb, le32_to_cpu(p->key));
433 if (!bh_uptodate_or_lock(bh)) {
434 if (bh_submit_read(bh) < 0) {
438 /* validate block references */
439 if (ext4_check_indirect_blockref(inode, bh)) {
445 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
459 * ext4_find_near - find a place for allocation with sufficient locality
461 * @ind: descriptor of indirect block.
463 * This function returns the preferred place for block allocation.
464 * It is used when heuristic for sequential allocation fails.
466 * + if there is a block to the left of our position - allocate near it.
467 * + if pointer will live in indirect block - allocate near that block.
468 * + if pointer will live in inode - allocate in the same
471 * In the latter case we colour the starting block by the callers PID to
472 * prevent it from clashing with concurrent allocations for a different inode
473 * in the same block group. The PID is used here so that functionally related
474 * files will be close-by on-disk.
476 * Caller must make sure that @ind is valid and will stay that way.
478 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
480 struct ext4_inode_info *ei = EXT4_I(inode);
481 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
483 ext4_fsblk_t bg_start;
484 ext4_fsblk_t last_block;
485 ext4_grpblk_t colour;
486 ext4_group_t block_group;
487 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
489 /* Try to find previous block */
490 for (p = ind->p - 1; p >= start; p--) {
492 return le32_to_cpu(*p);
495 /* No such thing, so let's try location of indirect block */
497 return ind->bh->b_blocknr;
500 * It is going to be referred to from the inode itself? OK, just put it
501 * into the same cylinder group then.
503 block_group = ei->i_block_group;
504 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
505 block_group &= ~(flex_size-1);
506 if (S_ISREG(inode->i_mode))
509 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
510 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
513 * If we are doing delayed allocation, we don't need take
514 * colour into account.
516 if (test_opt(inode->i_sb, DELALLOC))
519 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
520 colour = (current->pid % 16) *
521 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
523 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
524 return bg_start + colour;
528 * ext4_find_goal - find a preferred place for allocation.
530 * @block: block we want
531 * @partial: pointer to the last triple within a chain
533 * Normally this function find the preferred place for block allocation,
535 * Because this is only used for non-extent files, we limit the block nr
538 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
544 * XXX need to get goal block from mballoc's data structures
547 goal = ext4_find_near(inode, partial);
548 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
553 * ext4_blks_to_allocate: Look up the block map and count the number
554 * of direct blocks need to be allocated for the given branch.
556 * @branch: chain of indirect blocks
557 * @k: number of blocks need for indirect blocks
558 * @blks: number of data blocks to be mapped.
559 * @blocks_to_boundary: the offset in the indirect block
561 * return the total number of blocks to be allocate, including the
562 * direct and indirect blocks.
564 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
565 int blocks_to_boundary)
567 unsigned int count = 0;
570 * Simple case, [t,d]Indirect block(s) has not allocated yet
571 * then it's clear blocks on that path have not allocated
574 /* right now we don't handle cross boundary allocation */
575 if (blks < blocks_to_boundary + 1)
578 count += blocks_to_boundary + 1;
583 while (count < blks && count <= blocks_to_boundary &&
584 le32_to_cpu(*(branch[0].p + count)) == 0) {
591 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
592 * @indirect_blks: the number of blocks need to allocate for indirect
595 * @new_blocks: on return it will store the new block numbers for
596 * the indirect blocks(if needed) and the first direct block,
597 * @blks: on return it will store the total number of allocated
600 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
601 ext4_lblk_t iblock, ext4_fsblk_t goal,
602 int indirect_blks, int blks,
603 ext4_fsblk_t new_blocks[4], int *err)
605 struct ext4_allocation_request ar;
607 unsigned long count = 0, blk_allocated = 0;
609 ext4_fsblk_t current_block = 0;
613 * Here we try to allocate the requested multiple blocks at once,
614 * on a best-effort basis.
615 * To build a branch, we should allocate blocks for
616 * the indirect blocks(if not allocated yet), and at least
617 * the first direct block of this branch. That's the
618 * minimum number of blocks need to allocate(required)
620 /* first we try to allocate the indirect blocks */
621 target = indirect_blks;
624 /* allocating blocks for indirect blocks and direct blocks */
625 current_block = ext4_new_meta_blocks(handle, inode,
630 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
631 EXT4_ERROR_INODE(inode,
632 "current_block %llu + count %lu > %d!",
633 current_block, count,
634 EXT4_MAX_BLOCK_FILE_PHYS);
640 /* allocate blocks for indirect blocks */
641 while (index < indirect_blks && count) {
642 new_blocks[index++] = current_block++;
647 * save the new block number
648 * for the first direct block
650 new_blocks[index] = current_block;
651 printk(KERN_INFO "%s returned more blocks than "
652 "requested\n", __func__);
658 target = blks - count ;
659 blk_allocated = count;
662 /* Now allocate data blocks */
663 memset(&ar, 0, sizeof(ar));
668 if (S_ISREG(inode->i_mode))
669 /* enable in-core preallocation only for regular files */
670 ar.flags = EXT4_MB_HINT_DATA;
672 current_block = ext4_mb_new_blocks(handle, &ar, err);
673 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
674 EXT4_ERROR_INODE(inode,
675 "current_block %llu + ar.len %d > %d!",
676 current_block, ar.len,
677 EXT4_MAX_BLOCK_FILE_PHYS);
682 if (*err && (target == blks)) {
684 * if the allocation failed and we didn't allocate
690 if (target == blks) {
692 * save the new block number
693 * for the first direct block
695 new_blocks[index] = current_block;
697 blk_allocated += ar.len;
700 /* total number of blocks allocated for direct blocks */
705 for (i = 0; i < index; i++)
706 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
711 * ext4_alloc_branch - allocate and set up a chain of blocks.
713 * @indirect_blks: number of allocated indirect blocks
714 * @blks: number of allocated direct blocks
715 * @offsets: offsets (in the blocks) to store the pointers to next.
716 * @branch: place to store the chain in.
718 * This function allocates blocks, zeroes out all but the last one,
719 * links them into chain and (if we are synchronous) writes them to disk.
720 * In other words, it prepares a branch that can be spliced onto the
721 * inode. It stores the information about that chain in the branch[], in
722 * the same format as ext4_get_branch() would do. We are calling it after
723 * we had read the existing part of chain and partial points to the last
724 * triple of that (one with zero ->key). Upon the exit we have the same
725 * picture as after the successful ext4_get_block(), except that in one
726 * place chain is disconnected - *branch->p is still zero (we did not
727 * set the last link), but branch->key contains the number that should
728 * be placed into *branch->p to fill that gap.
730 * If allocation fails we free all blocks we've allocated (and forget
731 * their buffer_heads) and return the error value the from failed
732 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
733 * as described above and return 0.
735 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
736 ext4_lblk_t iblock, int indirect_blks,
737 int *blks, ext4_fsblk_t goal,
738 ext4_lblk_t *offsets, Indirect *branch)
740 int blocksize = inode->i_sb->s_blocksize;
743 struct buffer_head *bh;
745 ext4_fsblk_t new_blocks[4];
746 ext4_fsblk_t current_block;
748 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
749 *blks, new_blocks, &err);
753 branch[0].key = cpu_to_le32(new_blocks[0]);
755 * metadata blocks and data blocks are allocated.
757 for (n = 1; n <= indirect_blks; n++) {
759 * Get buffer_head for parent block, zero it out
760 * and set the pointer to new one, then send
763 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
766 BUFFER_TRACE(bh, "call get_create_access");
767 err = ext4_journal_get_create_access(handle, bh);
769 /* Don't brelse(bh) here; it's done in
770 * ext4_journal_forget() below */
775 memset(bh->b_data, 0, blocksize);
776 branch[n].p = (__le32 *) bh->b_data + offsets[n];
777 branch[n].key = cpu_to_le32(new_blocks[n]);
778 *branch[n].p = branch[n].key;
779 if (n == indirect_blks) {
780 current_block = new_blocks[n];
782 * End of chain, update the last new metablock of
783 * the chain to point to the new allocated
784 * data blocks numbers
786 for (i = 1; i < num; i++)
787 *(branch[n].p + i) = cpu_to_le32(++current_block);
789 BUFFER_TRACE(bh, "marking uptodate");
790 set_buffer_uptodate(bh);
793 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
794 err = ext4_handle_dirty_metadata(handle, inode, bh);
801 /* Allocation failed, free what we already allocated */
802 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
803 for (i = 1; i <= n ; i++) {
805 * branch[i].bh is newly allocated, so there is no
806 * need to revoke the block, which is why we don't
807 * need to set EXT4_FREE_BLOCKS_METADATA.
809 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
810 EXT4_FREE_BLOCKS_FORGET);
812 for (i = n+1; i < indirect_blks; i++)
813 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
815 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
821 * ext4_splice_branch - splice the allocated branch onto inode.
823 * @block: (logical) number of block we are adding
824 * @chain: chain of indirect blocks (with a missing link - see
826 * @where: location of missing link
827 * @num: number of indirect blocks we are adding
828 * @blks: number of direct blocks we are adding
830 * This function fills the missing link and does all housekeeping needed in
831 * inode (->i_blocks, etc.). In case of success we end up with the full
832 * chain to new block and return 0.
834 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
835 ext4_lblk_t block, Indirect *where, int num,
840 ext4_fsblk_t current_block;
843 * If we're splicing into a [td]indirect block (as opposed to the
844 * inode) then we need to get write access to the [td]indirect block
848 BUFFER_TRACE(where->bh, "get_write_access");
849 err = ext4_journal_get_write_access(handle, where->bh);
855 *where->p = where->key;
858 * Update the host buffer_head or inode to point to more just allocated
859 * direct blocks blocks
861 if (num == 0 && blks > 1) {
862 current_block = le32_to_cpu(where->key) + 1;
863 for (i = 1; i < blks; i++)
864 *(where->p + i) = cpu_to_le32(current_block++);
867 /* We are done with atomic stuff, now do the rest of housekeeping */
868 /* had we spliced it onto indirect block? */
871 * If we spliced it onto an indirect block, we haven't
872 * altered the inode. Note however that if it is being spliced
873 * onto an indirect block at the very end of the file (the
874 * file is growing) then we *will* alter the inode to reflect
875 * the new i_size. But that is not done here - it is done in
876 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
878 jbd_debug(5, "splicing indirect only\n");
879 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
880 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
885 * OK, we spliced it into the inode itself on a direct block.
887 ext4_mark_inode_dirty(handle, inode);
888 jbd_debug(5, "splicing direct\n");
893 for (i = 1; i <= num; i++) {
895 * branch[i].bh is newly allocated, so there is no
896 * need to revoke the block, which is why we don't
897 * need to set EXT4_FREE_BLOCKS_METADATA.
899 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
900 EXT4_FREE_BLOCKS_FORGET);
902 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
909 * The ext4_ind_map_blocks() function handles non-extents inodes
910 * (i.e., using the traditional indirect/double-indirect i_blocks
911 * scheme) for ext4_map_blocks().
913 * Allocation strategy is simple: if we have to allocate something, we will
914 * have to go the whole way to leaf. So let's do it before attaching anything
915 * to tree, set linkage between the newborn blocks, write them if sync is
916 * required, recheck the path, free and repeat if check fails, otherwise
917 * set the last missing link (that will protect us from any truncate-generated
918 * removals - all blocks on the path are immune now) and possibly force the
919 * write on the parent block.
920 * That has a nice additional property: no special recovery from the failed
921 * allocations is needed - we simply release blocks and do not touch anything
922 * reachable from inode.
924 * `handle' can be NULL if create == 0.
926 * return > 0, # of blocks mapped or allocated.
927 * return = 0, if plain lookup failed.
928 * return < 0, error case.
930 * The ext4_ind_get_blocks() function should be called with
931 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
932 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
933 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
936 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
937 struct ext4_map_blocks *map,
941 ext4_lblk_t offsets[4];
946 int blocks_to_boundary = 0;
949 ext4_fsblk_t first_block = 0;
951 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
952 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
953 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
954 &blocks_to_boundary);
959 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
961 /* Simplest case - block found, no allocation needed */
963 first_block = le32_to_cpu(chain[depth - 1].key);
966 while (count < map->m_len && count <= blocks_to_boundary) {
969 blk = le32_to_cpu(*(chain[depth-1].p + count));
971 if (blk == first_block + count)
979 /* Next simple case - plain lookup or failed read of indirect block */
980 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
984 * Okay, we need to do block allocation.
986 goal = ext4_find_goal(inode, map->m_lblk, partial);
988 /* the number of blocks need to allocate for [d,t]indirect blocks */
989 indirect_blks = (chain + depth) - partial - 1;
992 * Next look up the indirect map to count the totoal number of
993 * direct blocks to allocate for this branch.
995 count = ext4_blks_to_allocate(partial, indirect_blks,
996 map->m_len, blocks_to_boundary);
998 * Block out ext4_truncate while we alter the tree
1000 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1002 offsets + (partial - chain), partial);
1005 * The ext4_splice_branch call will free and forget any buffers
1006 * on the new chain if there is a failure, but that risks using
1007 * up transaction credits, especially for bitmaps where the
1008 * credits cannot be returned. Can we handle this somehow? We
1009 * may need to return -EAGAIN upwards in the worst case. --sct
1012 err = ext4_splice_branch(handle, inode, map->m_lblk,
1013 partial, indirect_blks, count);
1017 map->m_flags |= EXT4_MAP_NEW;
1019 ext4_update_inode_fsync_trans(handle, inode, 1);
1021 map->m_flags |= EXT4_MAP_MAPPED;
1022 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1024 if (count > blocks_to_boundary)
1025 map->m_flags |= EXT4_MAP_BOUNDARY;
1027 /* Clean up and exit */
1028 partial = chain + depth - 1; /* the whole chain */
1030 while (partial > chain) {
1031 BUFFER_TRACE(partial->bh, "call brelse");
1032 brelse(partial->bh);
1040 qsize_t *ext4_get_reserved_space(struct inode *inode)
1042 return &EXT4_I(inode)->i_reserved_quota;
1047 * Calculate the number of metadata blocks need to reserve
1048 * to allocate a new block at @lblocks for non extent file based file
1050 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1053 struct ext4_inode_info *ei = EXT4_I(inode);
1054 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1057 if (lblock < EXT4_NDIR_BLOCKS)
1060 lblock -= EXT4_NDIR_BLOCKS;
1062 if (ei->i_da_metadata_calc_len &&
1063 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1064 ei->i_da_metadata_calc_len++;
1067 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1068 ei->i_da_metadata_calc_len = 1;
1069 blk_bits = order_base_2(lblock);
1070 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1074 * Calculate the number of metadata blocks need to reserve
1075 * to allocate a block located at @lblock
1077 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1079 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1080 return ext4_ext_calc_metadata_amount(inode, lblock);
1082 return ext4_indirect_calc_metadata_amount(inode, lblock);
1086 * Called with i_data_sem down, which is important since we can call
1087 * ext4_discard_preallocations() from here.
1089 void ext4_da_update_reserve_space(struct inode *inode,
1090 int used, int quota_claim)
1092 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1093 struct ext4_inode_info *ei = EXT4_I(inode);
1095 spin_lock(&ei->i_block_reservation_lock);
1096 trace_ext4_da_update_reserve_space(inode, used);
1097 if (unlikely(used > ei->i_reserved_data_blocks)) {
1098 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1099 "with only %d reserved data blocks\n",
1100 __func__, inode->i_ino, used,
1101 ei->i_reserved_data_blocks);
1103 used = ei->i_reserved_data_blocks;
1106 /* Update per-inode reservations */
1107 ei->i_reserved_data_blocks -= used;
1108 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1109 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1110 used + ei->i_allocated_meta_blocks);
1111 ei->i_allocated_meta_blocks = 0;
1113 if (ei->i_reserved_data_blocks == 0) {
1115 * We can release all of the reserved metadata blocks
1116 * only when we have written all of the delayed
1117 * allocation blocks.
1119 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1120 ei->i_reserved_meta_blocks);
1121 ei->i_reserved_meta_blocks = 0;
1122 ei->i_da_metadata_calc_len = 0;
1124 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1126 /* Update quota subsystem for data blocks */
1128 dquot_claim_block(inode, used);
1131 * We did fallocate with an offset that is already delayed
1132 * allocated. So on delayed allocated writeback we should
1133 * not re-claim the quota for fallocated blocks.
1135 dquot_release_reservation_block(inode, used);
1139 * If we have done all the pending block allocations and if
1140 * there aren't any writers on the inode, we can discard the
1141 * inode's preallocations.
1143 if ((ei->i_reserved_data_blocks == 0) &&
1144 (atomic_read(&inode->i_writecount) == 0))
1145 ext4_discard_preallocations(inode);
1148 static int __check_block_validity(struct inode *inode, const char *func,
1150 struct ext4_map_blocks *map)
1152 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1154 ext4_error_inode(inode, func, line, map->m_pblk,
1155 "lblock %lu mapped to illegal pblock "
1156 "(length %d)", (unsigned long) map->m_lblk,
1163 #define check_block_validity(inode, map) \
1164 __check_block_validity((inode), __func__, __LINE__, (map))
1167 * Return the number of contiguous dirty pages in a given inode
1168 * starting at page frame idx.
1170 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1171 unsigned int max_pages)
1173 struct address_space *mapping = inode->i_mapping;
1175 struct pagevec pvec;
1177 int i, nr_pages, done = 0;
1181 pagevec_init(&pvec, 0);
1184 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1185 PAGECACHE_TAG_DIRTY,
1186 (pgoff_t)PAGEVEC_SIZE);
1189 for (i = 0; i < nr_pages; i++) {
1190 struct page *page = pvec.pages[i];
1191 struct buffer_head *bh, *head;
1194 if (unlikely(page->mapping != mapping) ||
1196 PageWriteback(page) ||
1197 page->index != idx) {
1202 if (page_has_buffers(page)) {
1203 bh = head = page_buffers(page);
1205 if (!buffer_delay(bh) &&
1206 !buffer_unwritten(bh))
1208 bh = bh->b_this_page;
1209 } while (!done && (bh != head));
1216 if (num >= max_pages) {
1221 pagevec_release(&pvec);
1227 * The ext4_map_blocks() function tries to look up the requested blocks,
1228 * and returns if the blocks are already mapped.
1230 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1231 * and store the allocated blocks in the result buffer head and mark it
1234 * If file type is extents based, it will call ext4_ext_map_blocks(),
1235 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1238 * On success, it returns the number of blocks being mapped or allocate.
1239 * if create==0 and the blocks are pre-allocated and uninitialized block,
1240 * the result buffer head is unmapped. If the create ==1, it will make sure
1241 * the buffer head is mapped.
1243 * It returns 0 if plain look up failed (blocks have not been allocated), in
1244 * that casem, buffer head is unmapped
1246 * It returns the error in case of allocation failure.
1248 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1249 struct ext4_map_blocks *map, int flags)
1254 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1255 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1256 (unsigned long) map->m_lblk);
1258 * Try to see if we can get the block without requesting a new
1259 * file system block.
1261 down_read((&EXT4_I(inode)->i_data_sem));
1262 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1263 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1265 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1267 up_read((&EXT4_I(inode)->i_data_sem));
1269 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1270 int ret = check_block_validity(inode, map);
1275 /* If it is only a block(s) look up */
1276 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1280 * Returns if the blocks have already allocated
1282 * Note that if blocks have been preallocated
1283 * ext4_ext_get_block() returns th create = 0
1284 * with buffer head unmapped.
1286 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1290 * When we call get_blocks without the create flag, the
1291 * BH_Unwritten flag could have gotten set if the blocks
1292 * requested were part of a uninitialized extent. We need to
1293 * clear this flag now that we are committed to convert all or
1294 * part of the uninitialized extent to be an initialized
1295 * extent. This is because we need to avoid the combination
1296 * of BH_Unwritten and BH_Mapped flags being simultaneously
1297 * set on the buffer_head.
1299 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1302 * New blocks allocate and/or writing to uninitialized extent
1303 * will possibly result in updating i_data, so we take
1304 * the write lock of i_data_sem, and call get_blocks()
1305 * with create == 1 flag.
1307 down_write((&EXT4_I(inode)->i_data_sem));
1310 * if the caller is from delayed allocation writeout path
1311 * we have already reserved fs blocks for allocation
1312 * let the underlying get_block() function know to
1313 * avoid double accounting
1315 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1316 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1318 * We need to check for EXT4 here because migrate
1319 * could have changed the inode type in between
1321 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1322 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1324 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1326 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1328 * We allocated new blocks which will result in
1329 * i_data's format changing. Force the migrate
1330 * to fail by clearing migrate flags
1332 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1336 * Update reserved blocks/metadata blocks after successful
1337 * block allocation which had been deferred till now. We don't
1338 * support fallocate for non extent files. So we can update
1339 * reserve space here.
1342 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1343 ext4_da_update_reserve_space(inode, retval, 1);
1345 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1346 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1348 up_write((&EXT4_I(inode)->i_data_sem));
1349 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1350 int ret = check_block_validity(inode, map);
1357 /* Maximum number of blocks we map for direct IO at once. */
1358 #define DIO_MAX_BLOCKS 4096
1360 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1361 struct buffer_head *bh, int flags)
1363 handle_t *handle = ext4_journal_current_handle();
1364 struct ext4_map_blocks map;
1365 int ret = 0, started = 0;
1368 map.m_lblk = iblock;
1369 map.m_len = bh->b_size >> inode->i_blkbits;
1371 if (flags && !handle) {
1372 /* Direct IO write... */
1373 if (map.m_len > DIO_MAX_BLOCKS)
1374 map.m_len = DIO_MAX_BLOCKS;
1375 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1376 handle = ext4_journal_start(inode, dio_credits);
1377 if (IS_ERR(handle)) {
1378 ret = PTR_ERR(handle);
1384 ret = ext4_map_blocks(handle, inode, &map, flags);
1386 map_bh(bh, inode->i_sb, map.m_pblk);
1387 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1388 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1392 ext4_journal_stop(handle);
1396 int ext4_get_block(struct inode *inode, sector_t iblock,
1397 struct buffer_head *bh, int create)
1399 return _ext4_get_block(inode, iblock, bh,
1400 create ? EXT4_GET_BLOCKS_CREATE : 0);
1404 * `handle' can be NULL if create is zero
1406 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1407 ext4_lblk_t block, int create, int *errp)
1409 struct ext4_map_blocks map;
1410 struct buffer_head *bh;
1413 J_ASSERT(handle != NULL || create == 0);
1417 err = ext4_map_blocks(handle, inode, &map,
1418 create ? EXT4_GET_BLOCKS_CREATE : 0);
1426 bh = sb_getblk(inode->i_sb, map.m_pblk);
1431 if (map.m_flags & EXT4_MAP_NEW) {
1432 J_ASSERT(create != 0);
1433 J_ASSERT(handle != NULL);
1436 * Now that we do not always journal data, we should
1437 * keep in mind whether this should always journal the
1438 * new buffer as metadata. For now, regular file
1439 * writes use ext4_get_block instead, so it's not a
1443 BUFFER_TRACE(bh, "call get_create_access");
1444 fatal = ext4_journal_get_create_access(handle, bh);
1445 if (!fatal && !buffer_uptodate(bh)) {
1446 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1447 set_buffer_uptodate(bh);
1450 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1451 err = ext4_handle_dirty_metadata(handle, inode, bh);
1455 BUFFER_TRACE(bh, "not a new buffer");
1465 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1466 ext4_lblk_t block, int create, int *err)
1468 struct buffer_head *bh;
1470 bh = ext4_getblk(handle, inode, block, create, err);
1473 if (buffer_uptodate(bh))
1475 ll_rw_block(READ_META, 1, &bh);
1477 if (buffer_uptodate(bh))
1484 static int walk_page_buffers(handle_t *handle,
1485 struct buffer_head *head,
1489 int (*fn)(handle_t *handle,
1490 struct buffer_head *bh))
1492 struct buffer_head *bh;
1493 unsigned block_start, block_end;
1494 unsigned blocksize = head->b_size;
1496 struct buffer_head *next;
1498 for (bh = head, block_start = 0;
1499 ret == 0 && (bh != head || !block_start);
1500 block_start = block_end, bh = next) {
1501 next = bh->b_this_page;
1502 block_end = block_start + blocksize;
1503 if (block_end <= from || block_start >= to) {
1504 if (partial && !buffer_uptodate(bh))
1508 err = (*fn)(handle, bh);
1516 * To preserve ordering, it is essential that the hole instantiation and
1517 * the data write be encapsulated in a single transaction. We cannot
1518 * close off a transaction and start a new one between the ext4_get_block()
1519 * and the commit_write(). So doing the jbd2_journal_start at the start of
1520 * prepare_write() is the right place.
1522 * Also, this function can nest inside ext4_writepage() ->
1523 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1524 * has generated enough buffer credits to do the whole page. So we won't
1525 * block on the journal in that case, which is good, because the caller may
1528 * By accident, ext4 can be reentered when a transaction is open via
1529 * quota file writes. If we were to commit the transaction while thus
1530 * reentered, there can be a deadlock - we would be holding a quota
1531 * lock, and the commit would never complete if another thread had a
1532 * transaction open and was blocking on the quota lock - a ranking
1535 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1536 * will _not_ run commit under these circumstances because handle->h_ref
1537 * is elevated. We'll still have enough credits for the tiny quotafile
1540 static int do_journal_get_write_access(handle_t *handle,
1541 struct buffer_head *bh)
1543 int dirty = buffer_dirty(bh);
1546 if (!buffer_mapped(bh) || buffer_freed(bh))
1549 * __block_prepare_write() could have dirtied some buffers. Clean
1550 * the dirty bit as jbd2_journal_get_write_access() could complain
1551 * otherwise about fs integrity issues. Setting of the dirty bit
1552 * by __block_prepare_write() isn't a real problem here as we clear
1553 * the bit before releasing a page lock and thus writeback cannot
1554 * ever write the buffer.
1557 clear_buffer_dirty(bh);
1558 ret = ext4_journal_get_write_access(handle, bh);
1560 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1565 * Truncate blocks that were not used by write. We have to truncate the
1566 * pagecache as well so that corresponding buffers get properly unmapped.
1568 static void ext4_truncate_failed_write(struct inode *inode)
1570 truncate_inode_pages(inode->i_mapping, inode->i_size);
1571 ext4_truncate(inode);
1574 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1575 struct buffer_head *bh_result, int create);
1576 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1577 loff_t pos, unsigned len, unsigned flags,
1578 struct page **pagep, void **fsdata)
1580 struct inode *inode = mapping->host;
1581 int ret, needed_blocks;
1588 trace_ext4_write_begin(inode, pos, len, flags);
1590 * Reserve one block more for addition to orphan list in case
1591 * we allocate blocks but write fails for some reason
1593 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1594 index = pos >> PAGE_CACHE_SHIFT;
1595 from = pos & (PAGE_CACHE_SIZE - 1);
1599 handle = ext4_journal_start(inode, needed_blocks);
1600 if (IS_ERR(handle)) {
1601 ret = PTR_ERR(handle);
1605 /* We cannot recurse into the filesystem as the transaction is already
1607 flags |= AOP_FLAG_NOFS;
1609 page = grab_cache_page_write_begin(mapping, index, flags);
1611 ext4_journal_stop(handle);
1617 if (ext4_should_dioread_nolock(inode))
1618 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1620 ret = __block_write_begin(page, pos, len, ext4_get_block);
1622 if (!ret && ext4_should_journal_data(inode)) {
1623 ret = walk_page_buffers(handle, page_buffers(page),
1624 from, to, NULL, do_journal_get_write_access);
1629 page_cache_release(page);
1631 * __block_write_begin may have instantiated a few blocks
1632 * outside i_size. Trim these off again. Don't need
1633 * i_size_read because we hold i_mutex.
1635 * Add inode to orphan list in case we crash before
1638 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1639 ext4_orphan_add(handle, inode);
1641 ext4_journal_stop(handle);
1642 if (pos + len > inode->i_size) {
1643 ext4_truncate_failed_write(inode);
1645 * If truncate failed early the inode might
1646 * still be on the orphan list; we need to
1647 * make sure the inode is removed from the
1648 * orphan list in that case.
1651 ext4_orphan_del(NULL, inode);
1655 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1661 /* For write_end() in data=journal mode */
1662 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1664 if (!buffer_mapped(bh) || buffer_freed(bh))
1666 set_buffer_uptodate(bh);
1667 return ext4_handle_dirty_metadata(handle, NULL, bh);
1670 static int ext4_generic_write_end(struct file *file,
1671 struct address_space *mapping,
1672 loff_t pos, unsigned len, unsigned copied,
1673 struct page *page, void *fsdata)
1675 int i_size_changed = 0;
1676 struct inode *inode = mapping->host;
1677 handle_t *handle = ext4_journal_current_handle();
1679 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1682 * No need to use i_size_read() here, the i_size
1683 * cannot change under us because we hold i_mutex.
1685 * But it's important to update i_size while still holding page lock:
1686 * page writeout could otherwise come in and zero beyond i_size.
1688 if (pos + copied > inode->i_size) {
1689 i_size_write(inode, pos + copied);
1693 if (pos + copied > EXT4_I(inode)->i_disksize) {
1694 /* We need to mark inode dirty even if
1695 * new_i_size is less that inode->i_size
1696 * bu greater than i_disksize.(hint delalloc)
1698 ext4_update_i_disksize(inode, (pos + copied));
1702 page_cache_release(page);
1705 * Don't mark the inode dirty under page lock. First, it unnecessarily
1706 * makes the holding time of page lock longer. Second, it forces lock
1707 * ordering of page lock and transaction start for journaling
1711 ext4_mark_inode_dirty(handle, inode);
1717 * We need to pick up the new inode size which generic_commit_write gave us
1718 * `file' can be NULL - eg, when called from page_symlink().
1720 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1721 * buffers are managed internally.
1723 static int ext4_ordered_write_end(struct file *file,
1724 struct address_space *mapping,
1725 loff_t pos, unsigned len, unsigned copied,
1726 struct page *page, void *fsdata)
1728 handle_t *handle = ext4_journal_current_handle();
1729 struct inode *inode = mapping->host;
1732 trace_ext4_ordered_write_end(inode, pos, len, copied);
1733 ret = ext4_jbd2_file_inode(handle, inode);
1736 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1739 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1740 /* if we have allocated more blocks and copied
1741 * less. We will have blocks allocated outside
1742 * inode->i_size. So truncate them
1744 ext4_orphan_add(handle, inode);
1748 ret2 = ext4_journal_stop(handle);
1752 if (pos + len > inode->i_size) {
1753 ext4_truncate_failed_write(inode);
1755 * If truncate failed early the inode might still be
1756 * on the orphan list; we need to make sure the inode
1757 * is removed from the orphan list in that case.
1760 ext4_orphan_del(NULL, inode);
1764 return ret ? ret : copied;
1767 static int ext4_writeback_write_end(struct file *file,
1768 struct address_space *mapping,
1769 loff_t pos, unsigned len, unsigned copied,
1770 struct page *page, void *fsdata)
1772 handle_t *handle = ext4_journal_current_handle();
1773 struct inode *inode = mapping->host;
1776 trace_ext4_writeback_write_end(inode, pos, len, copied);
1777 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1780 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1781 /* if we have allocated more blocks and copied
1782 * less. We will have blocks allocated outside
1783 * inode->i_size. So truncate them
1785 ext4_orphan_add(handle, inode);
1790 ret2 = ext4_journal_stop(handle);
1794 if (pos + len > inode->i_size) {
1795 ext4_truncate_failed_write(inode);
1797 * If truncate failed early the inode might still be
1798 * on the orphan list; we need to make sure the inode
1799 * is removed from the orphan list in that case.
1802 ext4_orphan_del(NULL, inode);
1805 return ret ? ret : copied;
1808 static int ext4_journalled_write_end(struct file *file,
1809 struct address_space *mapping,
1810 loff_t pos, unsigned len, unsigned copied,
1811 struct page *page, void *fsdata)
1813 handle_t *handle = ext4_journal_current_handle();
1814 struct inode *inode = mapping->host;
1820 trace_ext4_journalled_write_end(inode, pos, len, copied);
1821 from = pos & (PAGE_CACHE_SIZE - 1);
1825 if (!PageUptodate(page))
1827 page_zero_new_buffers(page, from+copied, to);
1830 ret = walk_page_buffers(handle, page_buffers(page), from,
1831 to, &partial, write_end_fn);
1833 SetPageUptodate(page);
1834 new_i_size = pos + copied;
1835 if (new_i_size > inode->i_size)
1836 i_size_write(inode, pos+copied);
1837 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1838 if (new_i_size > EXT4_I(inode)->i_disksize) {
1839 ext4_update_i_disksize(inode, new_i_size);
1840 ret2 = ext4_mark_inode_dirty(handle, inode);
1846 page_cache_release(page);
1847 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1848 /* if we have allocated more blocks and copied
1849 * less. We will have blocks allocated outside
1850 * inode->i_size. So truncate them
1852 ext4_orphan_add(handle, inode);
1854 ret2 = ext4_journal_stop(handle);
1857 if (pos + len > inode->i_size) {
1858 ext4_truncate_failed_write(inode);
1860 * If truncate failed early the inode might still be
1861 * on the orphan list; we need to make sure the inode
1862 * is removed from the orphan list in that case.
1865 ext4_orphan_del(NULL, inode);
1868 return ret ? ret : copied;
1872 * Reserve a single block located at lblock
1874 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1877 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1878 struct ext4_inode_info *ei = EXT4_I(inode);
1879 unsigned long md_needed;
1883 * recalculate the amount of metadata blocks to reserve
1884 * in order to allocate nrblocks
1885 * worse case is one extent per block
1888 spin_lock(&ei->i_block_reservation_lock);
1889 md_needed = ext4_calc_metadata_amount(inode, lblock);
1890 trace_ext4_da_reserve_space(inode, md_needed);
1891 spin_unlock(&ei->i_block_reservation_lock);
1894 * We will charge metadata quota at writeout time; this saves
1895 * us from metadata over-estimation, though we may go over by
1896 * a small amount in the end. Here we just reserve for data.
1898 ret = dquot_reserve_block(inode, 1);
1902 * We do still charge estimated metadata to the sb though;
1903 * we cannot afford to run out of free blocks.
1905 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1906 dquot_release_reservation_block(inode, 1);
1907 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1913 spin_lock(&ei->i_block_reservation_lock);
1914 ei->i_reserved_data_blocks++;
1915 ei->i_reserved_meta_blocks += md_needed;
1916 spin_unlock(&ei->i_block_reservation_lock);
1918 return 0; /* success */
1921 static void ext4_da_release_space(struct inode *inode, int to_free)
1923 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1924 struct ext4_inode_info *ei = EXT4_I(inode);
1927 return; /* Nothing to release, exit */
1929 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1931 trace_ext4_da_release_space(inode, to_free);
1932 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1934 * if there aren't enough reserved blocks, then the
1935 * counter is messed up somewhere. Since this
1936 * function is called from invalidate page, it's
1937 * harmless to return without any action.
1939 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1940 "ino %lu, to_free %d with only %d reserved "
1941 "data blocks\n", inode->i_ino, to_free,
1942 ei->i_reserved_data_blocks);
1944 to_free = ei->i_reserved_data_blocks;
1946 ei->i_reserved_data_blocks -= to_free;
1948 if (ei->i_reserved_data_blocks == 0) {
1950 * We can release all of the reserved metadata blocks
1951 * only when we have written all of the delayed
1952 * allocation blocks.
1954 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1955 ei->i_reserved_meta_blocks);
1956 ei->i_reserved_meta_blocks = 0;
1957 ei->i_da_metadata_calc_len = 0;
1960 /* update fs dirty data blocks counter */
1961 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1963 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1965 dquot_release_reservation_block(inode, to_free);
1968 static void ext4_da_page_release_reservation(struct page *page,
1969 unsigned long offset)
1972 struct buffer_head *head, *bh;
1973 unsigned int curr_off = 0;
1975 head = page_buffers(page);
1978 unsigned int next_off = curr_off + bh->b_size;
1980 if ((offset <= curr_off) && (buffer_delay(bh))) {
1982 clear_buffer_delay(bh);
1984 curr_off = next_off;
1985 } while ((bh = bh->b_this_page) != head);
1986 ext4_da_release_space(page->mapping->host, to_release);
1990 * Delayed allocation stuff
1994 * mpage_da_submit_io - walks through extent of pages and try to write
1995 * them with writepage() call back
1997 * @mpd->inode: inode
1998 * @mpd->first_page: first page of the extent
1999 * @mpd->next_page: page after the last page of the extent
2001 * By the time mpage_da_submit_io() is called we expect all blocks
2002 * to be allocated. this may be wrong if allocation failed.
2004 * As pages are already locked by write_cache_pages(), we can't use it
2006 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2007 struct ext4_map_blocks *map)
2009 struct pagevec pvec;
2010 unsigned long index, end;
2011 int ret = 0, err, nr_pages, i;
2012 struct inode *inode = mpd->inode;
2013 struct address_space *mapping = inode->i_mapping;
2014 loff_t size = i_size_read(inode);
2015 unsigned int len, block_start;
2016 struct buffer_head *bh, *page_bufs = NULL;
2017 int journal_data = ext4_should_journal_data(inode);
2018 sector_t pblock = 0, cur_logical = 0;
2020 BUG_ON(mpd->next_page <= mpd->first_page);
2022 * We need to start from the first_page to the next_page - 1
2023 * to make sure we also write the mapped dirty buffer_heads.
2024 * If we look at mpd->b_blocknr we would only be looking
2025 * at the currently mapped buffer_heads.
2027 index = mpd->first_page;
2028 end = mpd->next_page - 1;
2030 pagevec_init(&pvec, 0);
2031 while (index <= end) {
2032 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2035 for (i = 0; i < nr_pages; i++) {
2036 int commit_write = 0, redirty_page = 0;
2037 struct page *page = pvec.pages[i];
2039 index = page->index;
2043 if (index == size >> PAGE_CACHE_SHIFT)
2044 len = size & ~PAGE_CACHE_MASK;
2046 len = PAGE_CACHE_SIZE;
2048 cur_logical = index << (PAGE_CACHE_SHIFT -
2050 pblock = map->m_pblk + (cur_logical -
2055 BUG_ON(!PageLocked(page));
2056 BUG_ON(PageWriteback(page));
2059 * If the page does not have buffers (for
2060 * whatever reason), try to create them using
2061 * block_prepare_write. If this fails,
2062 * redirty the page and move on.
2064 if (!page_has_buffers(page)) {
2065 if (block_prepare_write(page, 0, len,
2066 noalloc_get_block_write)) {
2068 redirty_page_for_writepage(mpd->wbc,
2076 bh = page_bufs = page_buffers(page);
2081 if (map && (cur_logical >= map->m_lblk) &&
2082 (cur_logical <= (map->m_lblk +
2083 (map->m_len - 1)))) {
2084 if (buffer_delay(bh)) {
2085 clear_buffer_delay(bh);
2086 bh->b_blocknr = pblock;
2088 if (buffer_unwritten(bh) ||
2090 BUG_ON(bh->b_blocknr != pblock);
2091 if (map->m_flags & EXT4_MAP_UNINIT)
2092 set_buffer_uninit(bh);
2093 clear_buffer_unwritten(bh);
2096 /* redirty page if block allocation undone */
2097 if (buffer_delay(bh) || buffer_unwritten(bh))
2099 bh = bh->b_this_page;
2100 block_start += bh->b_size;
2103 } while (bh != page_bufs);
2109 /* mark the buffer_heads as dirty & uptodate */
2110 block_commit_write(page, 0, len);
2112 if (journal_data && PageChecked(page))
2113 err = __ext4_journalled_writepage(page, len);
2114 else if (buffer_uninit(page_bufs)) {
2115 ext4_set_bh_endio(page_bufs, inode);
2116 err = block_write_full_page_endio(page,
2117 noalloc_get_block_write,
2118 mpd->wbc, ext4_end_io_buffer_write);
2120 err = block_write_full_page(page,
2121 noalloc_get_block_write, mpd->wbc);
2124 mpd->pages_written++;
2126 * In error case, we have to continue because
2127 * remaining pages are still locked
2132 pagevec_release(&pvec);
2137 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2138 sector_t logical, long blk_cnt)
2142 struct pagevec pvec;
2143 struct inode *inode = mpd->inode;
2144 struct address_space *mapping = inode->i_mapping;
2146 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2147 end = (logical + blk_cnt - 1) >>
2148 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2149 while (index <= end) {
2150 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2153 for (i = 0; i < nr_pages; i++) {
2154 struct page *page = pvec.pages[i];
2155 if (page->index > end)
2157 BUG_ON(!PageLocked(page));
2158 BUG_ON(PageWriteback(page));
2159 block_invalidatepage(page, 0);
2160 ClearPageUptodate(page);
2163 index = pvec.pages[nr_pages - 1]->index + 1;
2164 pagevec_release(&pvec);
2169 static void ext4_print_free_blocks(struct inode *inode)
2171 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2172 printk(KERN_CRIT "Total free blocks count %lld\n",
2173 ext4_count_free_blocks(inode->i_sb));
2174 printk(KERN_CRIT "Free/Dirty block details\n");
2175 printk(KERN_CRIT "free_blocks=%lld\n",
2176 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2177 printk(KERN_CRIT "dirty_blocks=%lld\n",
2178 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2179 printk(KERN_CRIT "Block reservation details\n");
2180 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2181 EXT4_I(inode)->i_reserved_data_blocks);
2182 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2183 EXT4_I(inode)->i_reserved_meta_blocks);
2188 * mpage_da_map_and_submit - go through given space, map them
2189 * if necessary, and then submit them for I/O
2191 * @mpd - bh describing space
2193 * The function skips space we know is already mapped to disk blocks.
2196 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2198 int err, blks, get_blocks_flags;
2199 struct ext4_map_blocks map, *mapp = NULL;
2200 sector_t next = mpd->b_blocknr;
2201 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2202 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2203 handle_t *handle = NULL;
2206 * If the blocks are mapped already, or we couldn't accumulate
2207 * any blocks, then proceed immediately to the submission stage.
2209 if ((mpd->b_size == 0) ||
2210 ((mpd->b_state & (1 << BH_Mapped)) &&
2211 !(mpd->b_state & (1 << BH_Delay)) &&
2212 !(mpd->b_state & (1 << BH_Unwritten))))
2215 handle = ext4_journal_current_handle();
2219 * Call ext4_map_blocks() to allocate any delayed allocation
2220 * blocks, or to convert an uninitialized extent to be
2221 * initialized (in the case where we have written into
2222 * one or more preallocated blocks).
2224 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2225 * indicate that we are on the delayed allocation path. This
2226 * affects functions in many different parts of the allocation
2227 * call path. This flag exists primarily because we don't
2228 * want to change *many* call functions, so ext4_map_blocks()
2229 * will set the magic i_delalloc_reserved_flag once the
2230 * inode's allocation semaphore is taken.
2232 * If the blocks in questions were delalloc blocks, set
2233 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2234 * variables are updated after the blocks have been allocated.
2237 map.m_len = max_blocks;
2238 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2239 if (ext4_should_dioread_nolock(mpd->inode))
2240 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2241 if (mpd->b_state & (1 << BH_Delay))
2242 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2244 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2246 struct super_block *sb = mpd->inode->i_sb;
2250 * If get block returns EAGAIN or ENOSPC and there
2251 * appears to be free blocks we will call
2252 * ext4_writepage() for all of the pages which will
2253 * just redirty the pages.
2258 if (err == -ENOSPC &&
2259 ext4_count_free_blocks(sb)) {
2265 * get block failure will cause us to loop in
2266 * writepages, because a_ops->writepage won't be able
2267 * to make progress. The page will be redirtied by
2268 * writepage and writepages will again try to write
2271 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2272 ext4_msg(sb, KERN_CRIT,
2273 "delayed block allocation failed for inode %lu "
2274 "at logical offset %llu with max blocks %zd "
2275 "with error %d", mpd->inode->i_ino,
2276 (unsigned long long) next,
2277 mpd->b_size >> mpd->inode->i_blkbits, err);
2278 ext4_msg(sb, KERN_CRIT,
2279 "This should not happen!! Data will be lost\n");
2281 ext4_print_free_blocks(mpd->inode);
2283 /* invalidate all the pages */
2284 ext4_da_block_invalidatepages(mpd, next,
2285 mpd->b_size >> mpd->inode->i_blkbits);
2291 if (map.m_flags & EXT4_MAP_NEW) {
2292 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2295 for (i = 0; i < map.m_len; i++)
2296 unmap_underlying_metadata(bdev, map.m_pblk + i);
2299 if (ext4_should_order_data(mpd->inode)) {
2300 err = ext4_jbd2_file_inode(handle, mpd->inode);
2302 /* This only happens if the journal is aborted */
2307 * Update on-disk size along with block allocation.
2309 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2310 if (disksize > i_size_read(mpd->inode))
2311 disksize = i_size_read(mpd->inode);
2312 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2313 ext4_update_i_disksize(mpd->inode, disksize);
2314 err = ext4_mark_inode_dirty(handle, mpd->inode);
2316 ext4_error(mpd->inode->i_sb,
2317 "Failed to mark inode %lu dirty",
2322 mpage_da_submit_io(mpd, mapp);
2326 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2327 (1 << BH_Delay) | (1 << BH_Unwritten))
2330 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2332 * @mpd->lbh - extent of blocks
2333 * @logical - logical number of the block in the file
2334 * @bh - bh of the block (used to access block's state)
2336 * the function is used to collect contig. blocks in same state
2338 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2339 sector_t logical, size_t b_size,
2340 unsigned long b_state)
2343 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2346 * XXX Don't go larger than mballoc is willing to allocate
2347 * This is a stopgap solution. We eventually need to fold
2348 * mpage_da_submit_io() into this function and then call
2349 * ext4_map_blocks() multiple times in a loop
2351 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2354 /* check if thereserved journal credits might overflow */
2355 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2356 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2358 * With non-extent format we are limited by the journal
2359 * credit available. Total credit needed to insert
2360 * nrblocks contiguous blocks is dependent on the
2361 * nrblocks. So limit nrblocks.
2364 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2365 EXT4_MAX_TRANS_DATA) {
2367 * Adding the new buffer_head would make it cross the
2368 * allowed limit for which we have journal credit
2369 * reserved. So limit the new bh->b_size
2371 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2372 mpd->inode->i_blkbits;
2373 /* we will do mpage_da_submit_io in the next loop */
2377 * First block in the extent
2379 if (mpd->b_size == 0) {
2380 mpd->b_blocknr = logical;
2381 mpd->b_size = b_size;
2382 mpd->b_state = b_state & BH_FLAGS;
2386 next = mpd->b_blocknr + nrblocks;
2388 * Can we merge the block to our big extent?
2390 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2391 mpd->b_size += b_size;
2397 * We couldn't merge the block to our extent, so we
2398 * need to flush current extent and start new one
2400 mpage_da_map_and_submit(mpd);
2404 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2406 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2410 * __mpage_da_writepage - finds extent of pages and blocks
2412 * @page: page to consider
2413 * @wbc: not used, we just follow rules
2416 * The function finds extents of pages and scan them for all blocks.
2418 static int __mpage_da_writepage(struct page *page,
2419 struct writeback_control *wbc, void *data)
2421 struct mpage_da_data *mpd = data;
2422 struct inode *inode = mpd->inode;
2423 struct buffer_head *bh, *head;
2427 * Can we merge this page to current extent?
2429 if (mpd->next_page != page->index) {
2431 * Nope, we can't. So, we map non-allocated blocks
2432 * and start IO on them
2434 if (mpd->next_page != mpd->first_page) {
2435 mpage_da_map_and_submit(mpd);
2437 * skip rest of the page in the page_vec
2439 redirty_page_for_writepage(wbc, page);
2441 return MPAGE_DA_EXTENT_TAIL;
2445 * Start next extent of pages ...
2447 mpd->first_page = page->index;
2457 mpd->next_page = page->index + 1;
2458 logical = (sector_t) page->index <<
2459 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2461 if (!page_has_buffers(page)) {
2462 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2463 (1 << BH_Dirty) | (1 << BH_Uptodate));
2465 return MPAGE_DA_EXTENT_TAIL;
2468 * Page with regular buffer heads, just add all dirty ones
2470 head = page_buffers(page);
2473 BUG_ON(buffer_locked(bh));
2475 * We need to try to allocate
2476 * unmapped blocks in the same page.
2477 * Otherwise we won't make progress
2478 * with the page in ext4_writepage
2480 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2481 mpage_add_bh_to_extent(mpd, logical,
2485 return MPAGE_DA_EXTENT_TAIL;
2486 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2488 * mapped dirty buffer. We need to update
2489 * the b_state because we look at
2490 * b_state in mpage_da_map_blocks. We don't
2491 * update b_size because if we find an
2492 * unmapped buffer_head later we need to
2493 * use the b_state flag of that buffer_head.
2495 if (mpd->b_size == 0)
2496 mpd->b_state = bh->b_state & BH_FLAGS;
2499 } while ((bh = bh->b_this_page) != head);
2506 * This is a special get_blocks_t callback which is used by
2507 * ext4_da_write_begin(). It will either return mapped block or
2508 * reserve space for a single block.
2510 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2511 * We also have b_blocknr = -1 and b_bdev initialized properly
2513 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2514 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2515 * initialized properly.
2517 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2518 struct buffer_head *bh, int create)
2520 struct ext4_map_blocks map;
2522 sector_t invalid_block = ~((sector_t) 0xffff);
2524 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2527 BUG_ON(create == 0);
2528 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2530 map.m_lblk = iblock;
2534 * first, we need to know whether the block is allocated already
2535 * preallocated blocks are unmapped but should treated
2536 * the same as allocated blocks.
2538 ret = ext4_map_blocks(NULL, inode, &map, 0);
2542 if (buffer_delay(bh))
2543 return 0; /* Not sure this could or should happen */
2545 * XXX: __block_prepare_write() unmaps passed block,
2548 ret = ext4_da_reserve_space(inode, iblock);
2550 /* not enough space to reserve */
2553 map_bh(bh, inode->i_sb, invalid_block);
2555 set_buffer_delay(bh);
2559 map_bh(bh, inode->i_sb, map.m_pblk);
2560 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2562 if (buffer_unwritten(bh)) {
2563 /* A delayed write to unwritten bh should be marked
2564 * new and mapped. Mapped ensures that we don't do
2565 * get_block multiple times when we write to the same
2566 * offset and new ensures that we do proper zero out
2567 * for partial write.
2570 set_buffer_mapped(bh);
2576 * This function is used as a standard get_block_t calback function
2577 * when there is no desire to allocate any blocks. It is used as a
2578 * callback function for block_prepare_write() and block_write_full_page().
2579 * These functions should only try to map a single block at a time.
2581 * Since this function doesn't do block allocations even if the caller
2582 * requests it by passing in create=1, it is critically important that
2583 * any caller checks to make sure that any buffer heads are returned
2584 * by this function are either all already mapped or marked for
2585 * delayed allocation before calling block_write_full_page(). Otherwise,
2586 * b_blocknr could be left unitialized, and the page write functions will
2587 * be taken by surprise.
2589 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2590 struct buffer_head *bh_result, int create)
2592 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2593 return _ext4_get_block(inode, iblock, bh_result, 0);
2596 static int bget_one(handle_t *handle, struct buffer_head *bh)
2602 static int bput_one(handle_t *handle, struct buffer_head *bh)
2608 static int __ext4_journalled_writepage(struct page *page,
2611 struct address_space *mapping = page->mapping;
2612 struct inode *inode = mapping->host;
2613 struct buffer_head *page_bufs;
2614 handle_t *handle = NULL;
2618 ClearPageChecked(page);
2619 page_bufs = page_buffers(page);
2621 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2622 /* As soon as we unlock the page, it can go away, but we have
2623 * references to buffers so we are safe */
2626 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2627 if (IS_ERR(handle)) {
2628 ret = PTR_ERR(handle);
2632 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2633 do_journal_get_write_access);
2635 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2639 err = ext4_journal_stop(handle);
2643 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2644 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2649 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2650 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2653 * Note that we don't need to start a transaction unless we're journaling data
2654 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2655 * need to file the inode to the transaction's list in ordered mode because if
2656 * we are writing back data added by write(), the inode is already there and if
2657 * we are writing back data modified via mmap(), noone guarantees in which
2658 * transaction the data will hit the disk. In case we are journaling data, we
2659 * cannot start transaction directly because transaction start ranks above page
2660 * lock so we have to do some magic.
2662 * This function can get called via...
2663 * - ext4_da_writepages after taking page lock (have journal handle)
2664 * - journal_submit_inode_data_buffers (no journal handle)
2665 * - shrink_page_list via pdflush (no journal handle)
2666 * - grab_page_cache when doing write_begin (have journal handle)
2668 * We don't do any block allocation in this function. If we have page with
2669 * multiple blocks we need to write those buffer_heads that are mapped. This
2670 * is important for mmaped based write. So if we do with blocksize 1K
2671 * truncate(f, 1024);
2672 * a = mmap(f, 0, 4096);
2674 * truncate(f, 4096);
2675 * we have in the page first buffer_head mapped via page_mkwrite call back
2676 * but other bufer_heads would be unmapped but dirty(dirty done via the
2677 * do_wp_page). So writepage should write the first block. If we modify
2678 * the mmap area beyond 1024 we will again get a page_fault and the
2679 * page_mkwrite callback will do the block allocation and mark the
2680 * buffer_heads mapped.
2682 * We redirty the page if we have any buffer_heads that is either delay or
2683 * unwritten in the page.
2685 * We can get recursively called as show below.
2687 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2690 * But since we don't do any block allocation we should not deadlock.
2691 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2693 static int ext4_writepage(struct page *page,
2694 struct writeback_control *wbc)
2696 int ret = 0, commit_write = 0;
2699 struct buffer_head *page_bufs = NULL;
2700 struct inode *inode = page->mapping->host;
2702 trace_ext4_writepage(inode, page);
2703 size = i_size_read(inode);
2704 if (page->index == size >> PAGE_CACHE_SHIFT)
2705 len = size & ~PAGE_CACHE_MASK;
2707 len = PAGE_CACHE_SIZE;
2710 * If the page does not have buffers (for whatever reason),
2711 * try to create them using block_prepare_write. If this
2712 * fails, redirty the page and move on.
2714 if (!page_buffers(page)) {
2715 if (block_prepare_write(page, 0, len,
2716 noalloc_get_block_write)) {
2718 redirty_page_for_writepage(wbc, page);
2724 page_bufs = page_buffers(page);
2725 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2726 ext4_bh_delay_or_unwritten)) {
2728 * We don't want to do block allocation So redirty the
2729 * page and return We may reach here when we do a
2730 * journal commit via
2731 * journal_submit_inode_data_buffers. If we don't
2732 * have mapping block we just ignore them. We can also
2733 * reach here via shrink_page_list
2738 /* now mark the buffer_heads as dirty and uptodate */
2739 block_commit_write(page, 0, len);
2741 if (PageChecked(page) && ext4_should_journal_data(inode))
2743 * It's mmapped pagecache. Add buffers and journal it. There
2744 * doesn't seem much point in redirtying the page here.
2746 return __ext4_journalled_writepage(page, len);
2748 if (buffer_uninit(page_bufs)) {
2749 ext4_set_bh_endio(page_bufs, inode);
2750 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2751 wbc, ext4_end_io_buffer_write);
2753 ret = block_write_full_page(page, noalloc_get_block_write,
2760 * This is called via ext4_da_writepages() to
2761 * calulate the total number of credits to reserve to fit
2762 * a single extent allocation into a single transaction,
2763 * ext4_da_writpeages() will loop calling this before
2764 * the block allocation.
2767 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2769 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2772 * With non-extent format the journal credit needed to
2773 * insert nrblocks contiguous block is dependent on
2774 * number of contiguous block. So we will limit
2775 * number of contiguous block to a sane value
2777 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2778 (max_blocks > EXT4_MAX_TRANS_DATA))
2779 max_blocks = EXT4_MAX_TRANS_DATA;
2781 return ext4_chunk_trans_blocks(inode, max_blocks);
2785 * write_cache_pages_da - walk the list of dirty pages of the given
2786 * address space and call the callback function (which usually writes
2789 * This is a forked version of write_cache_pages(). Differences:
2790 * Range cyclic is ignored.
2791 * no_nrwrite_index_update is always presumed true
2793 static int write_cache_pages_da(struct address_space *mapping,
2794 struct writeback_control *wbc,
2795 struct mpage_da_data *mpd)
2799 struct pagevec pvec;
2802 pgoff_t end; /* Inclusive */
2803 long nr_to_write = wbc->nr_to_write;
2805 pagevec_init(&pvec, 0);
2806 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2807 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2809 while (!done && (index <= end)) {
2812 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
2813 PAGECACHE_TAG_DIRTY,
2814 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2818 for (i = 0; i < nr_pages; i++) {
2819 struct page *page = pvec.pages[i];
2822 * At this point, the page may be truncated or
2823 * invalidated (changing page->mapping to NULL), or
2824 * even swizzled back from swapper_space to tmpfs file
2825 * mapping. However, page->index will not change
2826 * because we have a reference on the page.
2828 if (page->index > end) {
2836 * Page truncated or invalidated. We can freely skip it
2837 * then, even for data integrity operations: the page
2838 * has disappeared concurrently, so there could be no
2839 * real expectation of this data interity operation
2840 * even if there is now a new, dirty page at the same
2841 * pagecache address.
2843 if (unlikely(page->mapping != mapping)) {
2849 if (!PageDirty(page)) {
2850 /* someone wrote it for us */
2851 goto continue_unlock;
2854 if (PageWriteback(page)) {
2855 if (wbc->sync_mode != WB_SYNC_NONE)
2856 wait_on_page_writeback(page);
2858 goto continue_unlock;
2861 BUG_ON(PageWriteback(page));
2862 if (!clear_page_dirty_for_io(page))
2863 goto continue_unlock;
2865 ret = __mpage_da_writepage(page, wbc, mpd);
2866 if (unlikely(ret)) {
2867 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2876 if (nr_to_write > 0) {
2878 if (nr_to_write == 0 &&
2879 wbc->sync_mode == WB_SYNC_NONE) {
2881 * We stop writing back only if we are
2882 * not doing integrity sync. In case of
2883 * integrity sync we have to keep going
2884 * because someone may be concurrently
2885 * dirtying pages, and we might have
2886 * synced a lot of newly appeared dirty
2887 * pages, but have not synced all of the
2895 pagevec_release(&pvec);
2902 static int ext4_da_writepages(struct address_space *mapping,
2903 struct writeback_control *wbc)
2906 int range_whole = 0;
2907 handle_t *handle = NULL;
2908 struct mpage_da_data mpd;
2909 struct inode *inode = mapping->host;
2910 int pages_written = 0;
2912 unsigned int max_pages;
2913 int range_cyclic, cycled = 1, io_done = 0;
2914 int needed_blocks, ret = 0;
2915 long desired_nr_to_write, nr_to_writebump = 0;
2916 loff_t range_start = wbc->range_start;
2917 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2919 trace_ext4_da_writepages(inode, wbc);
2922 * No pages to write? This is mainly a kludge to avoid starting
2923 * a transaction for special inodes like journal inode on last iput()
2924 * because that could violate lock ordering on umount
2926 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2930 * If the filesystem has aborted, it is read-only, so return
2931 * right away instead of dumping stack traces later on that
2932 * will obscure the real source of the problem. We test
2933 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2934 * the latter could be true if the filesystem is mounted
2935 * read-only, and in that case, ext4_da_writepages should
2936 * *never* be called, so if that ever happens, we would want
2939 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2942 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2945 range_cyclic = wbc->range_cyclic;
2946 if (wbc->range_cyclic) {
2947 index = mapping->writeback_index;
2950 wbc->range_start = index << PAGE_CACHE_SHIFT;
2951 wbc->range_end = LLONG_MAX;
2952 wbc->range_cyclic = 0;
2954 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2957 * This works around two forms of stupidity. The first is in
2958 * the writeback code, which caps the maximum number of pages
2959 * written to be 1024 pages. This is wrong on multiple
2960 * levels; different architectues have a different page size,
2961 * which changes the maximum amount of data which gets
2962 * written. Secondly, 4 megabytes is way too small. XFS
2963 * forces this value to be 16 megabytes by multiplying
2964 * nr_to_write parameter by four, and then relies on its
2965 * allocator to allocate larger extents to make them
2966 * contiguous. Unfortunately this brings us to the second
2967 * stupidity, which is that ext4's mballoc code only allocates
2968 * at most 2048 blocks. So we force contiguous writes up to
2969 * the number of dirty blocks in the inode, or
2970 * sbi->max_writeback_mb_bump whichever is smaller.
2972 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2973 if (!range_cyclic && range_whole) {
2974 if (wbc->nr_to_write == LONG_MAX)
2975 desired_nr_to_write = wbc->nr_to_write;
2977 desired_nr_to_write = wbc->nr_to_write * 8;
2979 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2981 if (desired_nr_to_write > max_pages)
2982 desired_nr_to_write = max_pages;
2984 if (wbc->nr_to_write < desired_nr_to_write) {
2985 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2986 wbc->nr_to_write = desired_nr_to_write;
2990 mpd.inode = mapping->host;
2992 pages_skipped = wbc->pages_skipped;
2995 while (!ret && wbc->nr_to_write > 0) {
2998 * we insert one extent at a time. So we need
2999 * credit needed for single extent allocation.
3000 * journalled mode is currently not supported
3003 BUG_ON(ext4_should_journal_data(inode));
3004 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3006 /* start a new transaction*/
3007 handle = ext4_journal_start(inode, needed_blocks);
3008 if (IS_ERR(handle)) {
3009 ret = PTR_ERR(handle);
3010 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3011 "%ld pages, ino %lu; err %d", __func__,
3012 wbc->nr_to_write, inode->i_ino, ret);
3013 goto out_writepages;
3017 * Now call __mpage_da_writepage to find the next
3018 * contiguous region of logical blocks that need
3019 * blocks to be allocated by ext4. We don't actually
3020 * submit the blocks for I/O here, even though
3021 * write_cache_pages thinks it will, and will set the
3022 * pages as clean for write before calling
3023 * __mpage_da_writepage().
3031 mpd.pages_written = 0;
3033 ret = write_cache_pages_da(mapping, wbc, &mpd);
3035 * If we have a contiguous extent of pages and we
3036 * haven't done the I/O yet, map the blocks and submit
3039 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3040 mpage_da_map_and_submit(&mpd);
3041 ret = MPAGE_DA_EXTENT_TAIL;
3043 trace_ext4_da_write_pages(inode, &mpd);
3044 wbc->nr_to_write -= mpd.pages_written;
3046 ext4_journal_stop(handle);
3048 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3049 /* commit the transaction which would
3050 * free blocks released in the transaction
3053 jbd2_journal_force_commit_nested(sbi->s_journal);
3054 wbc->pages_skipped = pages_skipped;
3056 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3058 * got one extent now try with
3061 pages_written += mpd.pages_written;
3062 wbc->pages_skipped = pages_skipped;
3065 } else if (wbc->nr_to_write)
3067 * There is no more writeout needed
3068 * or we requested for a noblocking writeout
3069 * and we found the device congested
3073 if (!io_done && !cycled) {
3076 wbc->range_start = index << PAGE_CACHE_SHIFT;
3077 wbc->range_end = mapping->writeback_index - 1;
3080 if (pages_skipped != wbc->pages_skipped)
3081 ext4_msg(inode->i_sb, KERN_CRIT,
3082 "This should not happen leaving %s "
3083 "with nr_to_write = %ld ret = %d",
3084 __func__, wbc->nr_to_write, ret);
3087 index += pages_written;
3088 wbc->range_cyclic = range_cyclic;
3089 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3091 * set the writeback_index so that range_cyclic
3092 * mode will write it back later
3094 mapping->writeback_index = index;
3097 wbc->nr_to_write -= nr_to_writebump;
3098 wbc->range_start = range_start;
3099 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3103 #define FALL_BACK_TO_NONDELALLOC 1
3104 static int ext4_nonda_switch(struct super_block *sb)
3106 s64 free_blocks, dirty_blocks;
3107 struct ext4_sb_info *sbi = EXT4_SB(sb);
3110 * switch to non delalloc mode if we are running low
3111 * on free block. The free block accounting via percpu
3112 * counters can get slightly wrong with percpu_counter_batch getting
3113 * accumulated on each CPU without updating global counters
3114 * Delalloc need an accurate free block accounting. So switch
3115 * to non delalloc when we are near to error range.
3117 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3118 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3119 if (2 * free_blocks < 3 * dirty_blocks ||
3120 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3122 * free block count is less than 150% of dirty blocks
3123 * or free blocks is less than watermark
3128 * Even if we don't switch but are nearing capacity,
3129 * start pushing delalloc when 1/2 of free blocks are dirty.
3131 if (free_blocks < 2 * dirty_blocks)
3132 writeback_inodes_sb_if_idle(sb);
3137 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3138 loff_t pos, unsigned len, unsigned flags,
3139 struct page **pagep, void **fsdata)
3141 int ret, retries = 0;
3144 struct inode *inode = mapping->host;
3147 index = pos >> PAGE_CACHE_SHIFT;
3149 if (ext4_nonda_switch(inode->i_sb)) {
3150 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3151 return ext4_write_begin(file, mapping, pos,
3152 len, flags, pagep, fsdata);
3154 *fsdata = (void *)0;
3155 trace_ext4_da_write_begin(inode, pos, len, flags);
3158 * With delayed allocation, we don't log the i_disksize update
3159 * if there is delayed block allocation. But we still need
3160 * to journalling the i_disksize update if writes to the end
3161 * of file which has an already mapped buffer.
3163 handle = ext4_journal_start(inode, 1);
3164 if (IS_ERR(handle)) {
3165 ret = PTR_ERR(handle);
3168 /* We cannot recurse into the filesystem as the transaction is already
3170 flags |= AOP_FLAG_NOFS;
3172 page = grab_cache_page_write_begin(mapping, index, flags);
3174 ext4_journal_stop(handle);
3180 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3183 ext4_journal_stop(handle);
3184 page_cache_release(page);
3186 * block_write_begin may have instantiated a few blocks
3187 * outside i_size. Trim these off again. Don't need
3188 * i_size_read because we hold i_mutex.
3190 if (pos + len > inode->i_size)
3191 ext4_truncate_failed_write(inode);
3194 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3201 * Check if we should update i_disksize
3202 * when write to the end of file but not require block allocation
3204 static int ext4_da_should_update_i_disksize(struct page *page,
3205 unsigned long offset)
3207 struct buffer_head *bh;
3208 struct inode *inode = page->mapping->host;
3212 bh = page_buffers(page);
3213 idx = offset >> inode->i_blkbits;
3215 for (i = 0; i < idx; i++)
3216 bh = bh->b_this_page;
3218 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3223 static int ext4_da_write_end(struct file *file,
3224 struct address_space *mapping,
3225 loff_t pos, unsigned len, unsigned copied,
3226 struct page *page, void *fsdata)
3228 struct inode *inode = mapping->host;
3230 handle_t *handle = ext4_journal_current_handle();
3232 unsigned long start, end;
3233 int write_mode = (int)(unsigned long)fsdata;
3235 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3236 if (ext4_should_order_data(inode)) {
3237 return ext4_ordered_write_end(file, mapping, pos,
3238 len, copied, page, fsdata);
3239 } else if (ext4_should_writeback_data(inode)) {
3240 return ext4_writeback_write_end(file, mapping, pos,
3241 len, copied, page, fsdata);
3247 trace_ext4_da_write_end(inode, pos, len, copied);
3248 start = pos & (PAGE_CACHE_SIZE - 1);
3249 end = start + copied - 1;
3252 * generic_write_end() will run mark_inode_dirty() if i_size
3253 * changes. So let's piggyback the i_disksize mark_inode_dirty
3257 new_i_size = pos + copied;
3258 if (new_i_size > EXT4_I(inode)->i_disksize) {
3259 if (ext4_da_should_update_i_disksize(page, end)) {
3260 down_write(&EXT4_I(inode)->i_data_sem);
3261 if (new_i_size > EXT4_I(inode)->i_disksize) {
3263 * Updating i_disksize when extending file
3264 * without needing block allocation
3266 if (ext4_should_order_data(inode))
3267 ret = ext4_jbd2_file_inode(handle,
3270 EXT4_I(inode)->i_disksize = new_i_size;
3272 up_write(&EXT4_I(inode)->i_data_sem);
3273 /* We need to mark inode dirty even if
3274 * new_i_size is less that inode->i_size
3275 * bu greater than i_disksize.(hint delalloc)
3277 ext4_mark_inode_dirty(handle, inode);
3280 ret2 = generic_write_end(file, mapping, pos, len, copied,
3285 ret2 = ext4_journal_stop(handle);
3289 return ret ? ret : copied;
3292 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3295 * Drop reserved blocks
3297 BUG_ON(!PageLocked(page));
3298 if (!page_has_buffers(page))
3301 ext4_da_page_release_reservation(page, offset);
3304 ext4_invalidatepage(page, offset);
3310 * Force all delayed allocation blocks to be allocated for a given inode.
3312 int ext4_alloc_da_blocks(struct inode *inode)
3314 trace_ext4_alloc_da_blocks(inode);
3316 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3317 !EXT4_I(inode)->i_reserved_meta_blocks)
3321 * We do something simple for now. The filemap_flush() will
3322 * also start triggering a write of the data blocks, which is
3323 * not strictly speaking necessary (and for users of
3324 * laptop_mode, not even desirable). However, to do otherwise
3325 * would require replicating code paths in:
3327 * ext4_da_writepages() ->
3328 * write_cache_pages() ---> (via passed in callback function)
3329 * __mpage_da_writepage() -->
3330 * mpage_add_bh_to_extent()
3331 * mpage_da_map_blocks()
3333 * The problem is that write_cache_pages(), located in
3334 * mm/page-writeback.c, marks pages clean in preparation for
3335 * doing I/O, which is not desirable if we're not planning on
3338 * We could call write_cache_pages(), and then redirty all of
3339 * the pages by calling redirty_page_for_writeback() but that
3340 * would be ugly in the extreme. So instead we would need to
3341 * replicate parts of the code in the above functions,
3342 * simplifying them becuase we wouldn't actually intend to
3343 * write out the pages, but rather only collect contiguous
3344 * logical block extents, call the multi-block allocator, and
3345 * then update the buffer heads with the block allocations.
3347 * For now, though, we'll cheat by calling filemap_flush(),
3348 * which will map the blocks, and start the I/O, but not
3349 * actually wait for the I/O to complete.
3351 return filemap_flush(inode->i_mapping);
3355 * bmap() is special. It gets used by applications such as lilo and by
3356 * the swapper to find the on-disk block of a specific piece of data.
3358 * Naturally, this is dangerous if the block concerned is still in the
3359 * journal. If somebody makes a swapfile on an ext4 data-journaling
3360 * filesystem and enables swap, then they may get a nasty shock when the
3361 * data getting swapped to that swapfile suddenly gets overwritten by
3362 * the original zero's written out previously to the journal and
3363 * awaiting writeback in the kernel's buffer cache.
3365 * So, if we see any bmap calls here on a modified, data-journaled file,
3366 * take extra steps to flush any blocks which might be in the cache.
3368 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3370 struct inode *inode = mapping->host;
3374 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3375 test_opt(inode->i_sb, DELALLOC)) {
3377 * With delalloc we want to sync the file
3378 * so that we can make sure we allocate
3381 filemap_write_and_wait(mapping);
3384 if (EXT4_JOURNAL(inode) &&
3385 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3387 * This is a REALLY heavyweight approach, but the use of
3388 * bmap on dirty files is expected to be extremely rare:
3389 * only if we run lilo or swapon on a freshly made file
3390 * do we expect this to happen.
3392 * (bmap requires CAP_SYS_RAWIO so this does not
3393 * represent an unprivileged user DOS attack --- we'd be
3394 * in trouble if mortal users could trigger this path at
3397 * NB. EXT4_STATE_JDATA is not set on files other than
3398 * regular files. If somebody wants to bmap a directory
3399 * or symlink and gets confused because the buffer
3400 * hasn't yet been flushed to disk, they deserve
3401 * everything they get.
3404 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3405 journal = EXT4_JOURNAL(inode);
3406 jbd2_journal_lock_updates(journal);
3407 err = jbd2_journal_flush(journal);
3408 jbd2_journal_unlock_updates(journal);
3414 return generic_block_bmap(mapping, block, ext4_get_block);
3417 static int ext4_readpage(struct file *file, struct page *page)
3419 return mpage_readpage(page, ext4_get_block);
3423 ext4_readpages(struct file *file, struct address_space *mapping,
3424 struct list_head *pages, unsigned nr_pages)
3426 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3429 static void ext4_free_io_end(ext4_io_end_t *io)
3438 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3440 struct buffer_head *head, *bh;
3441 unsigned int curr_off = 0;
3443 if (!page_has_buffers(page))
3445 head = bh = page_buffers(page);
3447 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3449 ext4_free_io_end(bh->b_private);
3450 bh->b_private = NULL;
3451 bh->b_end_io = NULL;
3453 curr_off = curr_off + bh->b_size;
3454 bh = bh->b_this_page;
3455 } while (bh != head);
3458 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3460 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3463 * free any io_end structure allocated for buffers to be discarded
3465 if (ext4_should_dioread_nolock(page->mapping->host))
3466 ext4_invalidatepage_free_endio(page, offset);
3468 * If it's a full truncate we just forget about the pending dirtying
3471 ClearPageChecked(page);
3474 jbd2_journal_invalidatepage(journal, page, offset);
3476 block_invalidatepage(page, offset);
3479 static int ext4_releasepage(struct page *page, gfp_t wait)
3481 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3483 WARN_ON(PageChecked(page));
3484 if (!page_has_buffers(page))
3487 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3489 return try_to_free_buffers(page);
3493 * O_DIRECT for ext3 (or indirect map) based files
3495 * If the O_DIRECT write will extend the file then add this inode to the
3496 * orphan list. So recovery will truncate it back to the original size
3497 * if the machine crashes during the write.
3499 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3500 * crashes then stale disk data _may_ be exposed inside the file. But current
3501 * VFS code falls back into buffered path in that case so we are safe.
3503 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3504 const struct iovec *iov, loff_t offset,
3505 unsigned long nr_segs)
3507 struct file *file = iocb->ki_filp;
3508 struct inode *inode = file->f_mapping->host;
3509 struct ext4_inode_info *ei = EXT4_I(inode);
3513 size_t count = iov_length(iov, nr_segs);
3517 loff_t final_size = offset + count;
3519 if (final_size > inode->i_size) {
3520 /* Credits for sb + inode write */
3521 handle = ext4_journal_start(inode, 2);
3522 if (IS_ERR(handle)) {
3523 ret = PTR_ERR(handle);
3526 ret = ext4_orphan_add(handle, inode);
3528 ext4_journal_stop(handle);
3532 ei->i_disksize = inode->i_size;
3533 ext4_journal_stop(handle);
3538 if (rw == READ && ext4_should_dioread_nolock(inode))
3539 ret = __blockdev_direct_IO(rw, iocb, inode,
3540 inode->i_sb->s_bdev, iov,
3542 ext4_get_block, NULL, NULL, 0);
3544 ret = blockdev_direct_IO(rw, iocb, inode,
3545 inode->i_sb->s_bdev, iov,
3547 ext4_get_block, NULL);
3549 if (unlikely((rw & WRITE) && ret < 0)) {
3550 loff_t isize = i_size_read(inode);
3551 loff_t end = offset + iov_length(iov, nr_segs);
3554 vmtruncate(inode, isize);
3557 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3563 /* Credits for sb + inode write */
3564 handle = ext4_journal_start(inode, 2);
3565 if (IS_ERR(handle)) {
3566 /* This is really bad luck. We've written the data
3567 * but cannot extend i_size. Bail out and pretend
3568 * the write failed... */
3569 ret = PTR_ERR(handle);
3571 ext4_orphan_del(NULL, inode);
3576 ext4_orphan_del(handle, inode);
3578 loff_t end = offset + ret;
3579 if (end > inode->i_size) {
3580 ei->i_disksize = end;
3581 i_size_write(inode, end);
3583 * We're going to return a positive `ret'
3584 * here due to non-zero-length I/O, so there's
3585 * no way of reporting error returns from
3586 * ext4_mark_inode_dirty() to userspace. So
3589 ext4_mark_inode_dirty(handle, inode);
3592 err = ext4_journal_stop(handle);
3601 * ext4_get_block used when preparing for a DIO write or buffer write.
3602 * We allocate an uinitialized extent if blocks haven't been allocated.
3603 * The extent will be converted to initialized after the IO is complete.
3605 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3606 struct buffer_head *bh_result, int create)
3608 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3609 inode->i_ino, create);
3610 return _ext4_get_block(inode, iblock, bh_result,
3611 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3614 static void dump_completed_IO(struct inode * inode)
3617 struct list_head *cur, *before, *after;
3618 ext4_io_end_t *io, *io0, *io1;
3619 unsigned long flags;
3621 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
3622 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
3626 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
3627 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3628 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
3631 io0 = container_of(before, ext4_io_end_t, list);
3633 io1 = container_of(after, ext4_io_end_t, list);
3635 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
3636 io, inode->i_ino, io0, io1);
3638 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3643 * check a range of space and convert unwritten extents to written.
3645 static int ext4_end_io_nolock(ext4_io_end_t *io)
3647 struct inode *inode = io->inode;
3648 loff_t offset = io->offset;
3649 ssize_t size = io->size;
3652 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
3653 "list->prev 0x%p\n",
3654 io, inode->i_ino, io->list.next, io->list.prev);
3656 if (list_empty(&io->list))
3659 if (io->flag != EXT4_IO_UNWRITTEN)
3662 ret = ext4_convert_unwritten_extents(inode, offset, size);
3664 printk(KERN_EMERG "%s: failed to convert unwritten"
3665 "extents to written extents, error is %d"
3666 " io is still on inode %lu aio dio list\n",
3667 __func__, ret, inode->i_ino);
3672 aio_complete(io->iocb, io->result, 0);
3673 /* clear the DIO AIO unwritten flag */
3679 * work on completed aio dio IO, to convert unwritten extents to extents
3681 static void ext4_end_io_work(struct work_struct *work)
3683 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
3684 struct inode *inode = io->inode;
3685 struct ext4_inode_info *ei = EXT4_I(inode);
3686 unsigned long flags;
3689 mutex_lock(&inode->i_mutex);
3690 ret = ext4_end_io_nolock(io);
3692 mutex_unlock(&inode->i_mutex);
3696 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3697 if (!list_empty(&io->list))
3698 list_del_init(&io->list);
3699 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3700 mutex_unlock(&inode->i_mutex);
3701 ext4_free_io_end(io);
3705 * This function is called from ext4_sync_file().
3707 * When IO is completed, the work to convert unwritten extents to
3708 * written is queued on workqueue but may not get immediately
3709 * scheduled. When fsync is called, we need to ensure the
3710 * conversion is complete before fsync returns.
3711 * The inode keeps track of a list of pending/completed IO that
3712 * might needs to do the conversion. This function walks through
3713 * the list and convert the related unwritten extents for completed IO
3715 * The function return the number of pending IOs on success.
3717 int flush_completed_IO(struct inode *inode)
3720 struct ext4_inode_info *ei = EXT4_I(inode);
3721 unsigned long flags;
3725 if (list_empty(&ei->i_completed_io_list))
3728 dump_completed_IO(inode);
3729 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3730 while (!list_empty(&ei->i_completed_io_list)){
3731 io = list_entry(ei->i_completed_io_list.next,
3732 ext4_io_end_t, list);
3734 * Calling ext4_end_io_nolock() to convert completed
3737 * When ext4_sync_file() is called, run_queue() may already
3738 * about to flush the work corresponding to this io structure.
3739 * It will be upset if it founds the io structure related
3740 * to the work-to-be schedule is freed.
3742 * Thus we need to keep the io structure still valid here after
3743 * convertion finished. The io structure has a flag to
3744 * avoid double converting from both fsync and background work
3747 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3748 ret = ext4_end_io_nolock(io);
3749 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3753 list_del_init(&io->list);
3755 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3756 return (ret2 < 0) ? ret2 : 0;
3759 static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
3761 ext4_io_end_t *io = NULL;
3763 io = kmalloc(sizeof(*io), flags);
3774 INIT_WORK(&io->work, ext4_end_io_work);
3775 INIT_LIST_HEAD(&io->list);
3781 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3782 ssize_t size, void *private, int ret,
3785 ext4_io_end_t *io_end = iocb->private;
3786 struct workqueue_struct *wq;
3787 unsigned long flags;
3788 struct ext4_inode_info *ei;
3790 /* if not async direct IO or dio with 0 bytes write, just return */
3791 if (!io_end || !size)
3794 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3795 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3796 iocb->private, io_end->inode->i_ino, iocb, offset,
3799 /* if not aio dio with unwritten extents, just free io and return */
3800 if (io_end->flag != EXT4_IO_UNWRITTEN){
3801 ext4_free_io_end(io_end);
3802 iocb->private = NULL;
3805 aio_complete(iocb, ret, 0);
3809 io_end->offset = offset;
3810 io_end->size = size;
3812 io_end->iocb = iocb;
3813 io_end->result = ret;
3815 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3817 /* Add the io_end to per-inode completed aio dio list*/
3818 ei = EXT4_I(io_end->inode);
3819 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3820 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3821 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3823 /* queue the work to convert unwritten extents to written */
3824 queue_work(wq, &io_end->work);
3825 iocb->private = NULL;
3828 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3830 ext4_io_end_t *io_end = bh->b_private;
3831 struct workqueue_struct *wq;
3832 struct inode *inode;
3833 unsigned long flags;
3835 if (!test_clear_buffer_uninit(bh) || !io_end)
3838 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3839 printk("sb umounted, discard end_io request for inode %lu\n",
3840 io_end->inode->i_ino);
3841 ext4_free_io_end(io_end);
3845 io_end->flag = EXT4_IO_UNWRITTEN;
3846 inode = io_end->inode;
3848 /* Add the io_end to per-inode completed io list*/
3849 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3850 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3851 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3853 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3854 /* queue the work to convert unwritten extents to written */
3855 queue_work(wq, &io_end->work);
3857 bh->b_private = NULL;
3858 bh->b_end_io = NULL;
3859 clear_buffer_uninit(bh);
3860 end_buffer_async_write(bh, uptodate);
3863 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3865 ext4_io_end_t *io_end;
3866 struct page *page = bh->b_page;
3867 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3868 size_t size = bh->b_size;
3871 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3873 if (printk_ratelimit())
3874 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3878 io_end->offset = offset;
3879 io_end->size = size;
3881 * We need to hold a reference to the page to make sure it
3882 * doesn't get evicted before ext4_end_io_work() has a chance
3883 * to convert the extent from written to unwritten.
3885 io_end->page = page;
3886 get_page(io_end->page);
3888 bh->b_private = io_end;
3889 bh->b_end_io = ext4_end_io_buffer_write;
3894 * For ext4 extent files, ext4 will do direct-io write to holes,
3895 * preallocated extents, and those write extend the file, no need to
3896 * fall back to buffered IO.
3898 * For holes, we fallocate those blocks, mark them as unintialized
3899 * If those blocks were preallocated, we mark sure they are splited, but
3900 * still keep the range to write as unintialized.
3902 * The unwrritten extents will be converted to written when DIO is completed.
3903 * For async direct IO, since the IO may still pending when return, we
3904 * set up an end_io call back function, which will do the convertion
3905 * when async direct IO completed.
3907 * If the O_DIRECT write will extend the file then add this inode to the
3908 * orphan list. So recovery will truncate it back to the original size
3909 * if the machine crashes during the write.
3912 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3913 const struct iovec *iov, loff_t offset,
3914 unsigned long nr_segs)
3916 struct file *file = iocb->ki_filp;
3917 struct inode *inode = file->f_mapping->host;
3919 size_t count = iov_length(iov, nr_segs);
3921 loff_t final_size = offset + count;
3922 if (rw == WRITE && final_size <= inode->i_size) {
3924 * We could direct write to holes and fallocate.
3926 * Allocated blocks to fill the hole are marked as uninitialized
3927 * to prevent paralel buffered read to expose the stale data
3928 * before DIO complete the data IO.
3930 * As to previously fallocated extents, ext4 get_block
3931 * will just simply mark the buffer mapped but still
3932 * keep the extents uninitialized.
3934 * for non AIO case, we will convert those unwritten extents
3935 * to written after return back from blockdev_direct_IO.
3937 * for async DIO, the conversion needs to be defered when
3938 * the IO is completed. The ext4 end_io callback function
3939 * will be called to take care of the conversion work.
3940 * Here for async case, we allocate an io_end structure to
3943 iocb->private = NULL;
3944 EXT4_I(inode)->cur_aio_dio = NULL;
3945 if (!is_sync_kiocb(iocb)) {
3946 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3950 * we save the io structure for current async
3951 * direct IO, so that later ext4_map_blocks()
3952 * could flag the io structure whether there
3953 * is a unwritten extents needs to be converted
3954 * when IO is completed.
3956 EXT4_I(inode)->cur_aio_dio = iocb->private;
3959 ret = blockdev_direct_IO(rw, iocb, inode,
3960 inode->i_sb->s_bdev, iov,
3962 ext4_get_block_write,
3965 EXT4_I(inode)->cur_aio_dio = NULL;
3967 * The io_end structure takes a reference to the inode,
3968 * that structure needs to be destroyed and the
3969 * reference to the inode need to be dropped, when IO is
3970 * complete, even with 0 byte write, or failed.
3972 * In the successful AIO DIO case, the io_end structure will be
3973 * desctroyed and the reference to the inode will be dropped
3974 * after the end_io call back function is called.
3976 * In the case there is 0 byte write, or error case, since
3977 * VFS direct IO won't invoke the end_io call back function,
3978 * we need to free the end_io structure here.
3980 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3981 ext4_free_io_end(iocb->private);
3982 iocb->private = NULL;
3983 } else if (ret > 0 && ext4_test_inode_state(inode,
3984 EXT4_STATE_DIO_UNWRITTEN)) {
3987 * for non AIO case, since the IO is already
3988 * completed, we could do the convertion right here
3990 err = ext4_convert_unwritten_extents(inode,
3994 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3999 /* for write the the end of file case, we fall back to old way */
4000 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4003 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
4004 const struct iovec *iov, loff_t offset,
4005 unsigned long nr_segs)
4007 struct file *file = iocb->ki_filp;
4008 struct inode *inode = file->f_mapping->host;
4010 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4011 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
4013 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
4017 * Pages can be marked dirty completely asynchronously from ext4's journalling
4018 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
4019 * much here because ->set_page_dirty is called under VFS locks. The page is
4020 * not necessarily locked.
4022 * We cannot just dirty the page and leave attached buffers clean, because the
4023 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
4024 * or jbddirty because all the journalling code will explode.
4026 * So what we do is to mark the page "pending dirty" and next time writepage
4027 * is called, propagate that into the buffers appropriately.
4029 static int ext4_journalled_set_page_dirty(struct page *page)
4031 SetPageChecked(page);
4032 return __set_page_dirty_nobuffers(page);
4035 static const struct address_space_operations ext4_ordered_aops = {
4036 .readpage = ext4_readpage,
4037 .readpages = ext4_readpages,
4038 .writepage = ext4_writepage,
4039 .sync_page = block_sync_page,
4040 .write_begin = ext4_write_begin,
4041 .write_end = ext4_ordered_write_end,
4043 .invalidatepage = ext4_invalidatepage,
4044 .releasepage = ext4_releasepage,
4045 .direct_IO = ext4_direct_IO,
4046 .migratepage = buffer_migrate_page,
4047 .is_partially_uptodate = block_is_partially_uptodate,
4048 .error_remove_page = generic_error_remove_page,
4051 static const struct address_space_operations ext4_writeback_aops = {
4052 .readpage = ext4_readpage,
4053 .readpages = ext4_readpages,
4054 .writepage = ext4_writepage,
4055 .sync_page = block_sync_page,
4056 .write_begin = ext4_write_begin,
4057 .write_end = ext4_writeback_write_end,
4059 .invalidatepage = ext4_invalidatepage,
4060 .releasepage = ext4_releasepage,
4061 .direct_IO = ext4_direct_IO,
4062 .migratepage = buffer_migrate_page,
4063 .is_partially_uptodate = block_is_partially_uptodate,
4064 .error_remove_page = generic_error_remove_page,
4067 static const struct address_space_operations ext4_journalled_aops = {
4068 .readpage = ext4_readpage,
4069 .readpages = ext4_readpages,
4070 .writepage = ext4_writepage,
4071 .sync_page = block_sync_page,
4072 .write_begin = ext4_write_begin,
4073 .write_end = ext4_journalled_write_end,
4074 .set_page_dirty = ext4_journalled_set_page_dirty,
4076 .invalidatepage = ext4_invalidatepage,
4077 .releasepage = ext4_releasepage,
4078 .is_partially_uptodate = block_is_partially_uptodate,
4079 .error_remove_page = generic_error_remove_page,
4082 static const struct address_space_operations ext4_da_aops = {
4083 .readpage = ext4_readpage,
4084 .readpages = ext4_readpages,
4085 .writepage = ext4_writepage,
4086 .writepages = ext4_da_writepages,
4087 .sync_page = block_sync_page,
4088 .write_begin = ext4_da_write_begin,
4089 .write_end = ext4_da_write_end,
4091 .invalidatepage = ext4_da_invalidatepage,
4092 .releasepage = ext4_releasepage,
4093 .direct_IO = ext4_direct_IO,
4094 .migratepage = buffer_migrate_page,
4095 .is_partially_uptodate = block_is_partially_uptodate,
4096 .error_remove_page = generic_error_remove_page,
4099 void ext4_set_aops(struct inode *inode)
4101 if (ext4_should_order_data(inode) &&
4102 test_opt(inode->i_sb, DELALLOC))
4103 inode->i_mapping->a_ops = &ext4_da_aops;
4104 else if (ext4_should_order_data(inode))
4105 inode->i_mapping->a_ops = &ext4_ordered_aops;
4106 else if (ext4_should_writeback_data(inode) &&
4107 test_opt(inode->i_sb, DELALLOC))
4108 inode->i_mapping->a_ops = &ext4_da_aops;
4109 else if (ext4_should_writeback_data(inode))
4110 inode->i_mapping->a_ops = &ext4_writeback_aops;
4112 inode->i_mapping->a_ops = &ext4_journalled_aops;
4116 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4117 * up to the end of the block which corresponds to `from'.
4118 * This required during truncate. We need to physically zero the tail end
4119 * of that block so it doesn't yield old data if the file is later grown.
4121 int ext4_block_truncate_page(handle_t *handle,
4122 struct address_space *mapping, loff_t from)
4124 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
4125 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4126 unsigned blocksize, length, pos;
4128 struct inode *inode = mapping->host;
4129 struct buffer_head *bh;
4133 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
4134 mapping_gfp_mask(mapping) & ~__GFP_FS);
4138 blocksize = inode->i_sb->s_blocksize;
4139 length = blocksize - (offset & (blocksize - 1));
4140 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
4142 if (!page_has_buffers(page))
4143 create_empty_buffers(page, blocksize, 0);
4145 /* Find the buffer that contains "offset" */
4146 bh = page_buffers(page);
4148 while (offset >= pos) {
4149 bh = bh->b_this_page;
4155 if (buffer_freed(bh)) {
4156 BUFFER_TRACE(bh, "freed: skip");
4160 if (!buffer_mapped(bh)) {
4161 BUFFER_TRACE(bh, "unmapped");
4162 ext4_get_block(inode, iblock, bh, 0);
4163 /* unmapped? It's a hole - nothing to do */
4164 if (!buffer_mapped(bh)) {
4165 BUFFER_TRACE(bh, "still unmapped");
4170 /* Ok, it's mapped. Make sure it's up-to-date */
4171 if (PageUptodate(page))
4172 set_buffer_uptodate(bh);
4174 if (!buffer_uptodate(bh)) {
4176 ll_rw_block(READ, 1, &bh);
4178 /* Uhhuh. Read error. Complain and punt. */
4179 if (!buffer_uptodate(bh))
4183 if (ext4_should_journal_data(inode)) {
4184 BUFFER_TRACE(bh, "get write access");
4185 err = ext4_journal_get_write_access(handle, bh);
4190 zero_user(page, offset, length);
4192 BUFFER_TRACE(bh, "zeroed end of block");
4195 if (ext4_should_journal_data(inode)) {
4196 err = ext4_handle_dirty_metadata(handle, inode, bh);
4198 if (ext4_should_order_data(inode))
4199 err = ext4_jbd2_file_inode(handle, inode);
4200 mark_buffer_dirty(bh);
4205 page_cache_release(page);
4210 * Probably it should be a library function... search for first non-zero word
4211 * or memcmp with zero_page, whatever is better for particular architecture.
4214 static inline int all_zeroes(__le32 *p, __le32 *q)
4223 * ext4_find_shared - find the indirect blocks for partial truncation.
4224 * @inode: inode in question
4225 * @depth: depth of the affected branch
4226 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4227 * @chain: place to store the pointers to partial indirect blocks
4228 * @top: place to the (detached) top of branch
4230 * This is a helper function used by ext4_truncate().
4232 * When we do truncate() we may have to clean the ends of several
4233 * indirect blocks but leave the blocks themselves alive. Block is
4234 * partially truncated if some data below the new i_size is refered
4235 * from it (and it is on the path to the first completely truncated
4236 * data block, indeed). We have to free the top of that path along
4237 * with everything to the right of the path. Since no allocation
4238 * past the truncation point is possible until ext4_truncate()
4239 * finishes, we may safely do the latter, but top of branch may
4240 * require special attention - pageout below the truncation point
4241 * might try to populate it.
4243 * We atomically detach the top of branch from the tree, store the
4244 * block number of its root in *@top, pointers to buffer_heads of
4245 * partially truncated blocks - in @chain[].bh and pointers to
4246 * their last elements that should not be removed - in
4247 * @chain[].p. Return value is the pointer to last filled element
4250 * The work left to caller to do the actual freeing of subtrees:
4251 * a) free the subtree starting from *@top
4252 * b) free the subtrees whose roots are stored in
4253 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4254 * c) free the subtrees growing from the inode past the @chain[0].
4255 * (no partially truncated stuff there). */
4257 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4258 ext4_lblk_t offsets[4], Indirect chain[4],
4261 Indirect *partial, *p;
4265 /* Make k index the deepest non-null offset + 1 */
4266 for (k = depth; k > 1 && !offsets[k-1]; k--)
4268 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4269 /* Writer: pointers */
4271 partial = chain + k-1;
4273 * If the branch acquired continuation since we've looked at it -
4274 * fine, it should all survive and (new) top doesn't belong to us.
4276 if (!partial->key && *partial->p)
4279 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4282 * OK, we've found the last block that must survive. The rest of our
4283 * branch should be detached before unlocking. However, if that rest
4284 * of branch is all ours and does not grow immediately from the inode
4285 * it's easier to cheat and just decrement partial->p.
4287 if (p == chain + k - 1 && p > chain) {
4291 /* Nope, don't do this in ext4. Must leave the tree intact */
4298 while (partial > p) {
4299 brelse(partial->bh);
4307 * Zero a number of block pointers in either an inode or an indirect block.
4308 * If we restart the transaction we must again get write access to the
4309 * indirect block for further modification.
4311 * We release `count' blocks on disk, but (last - first) may be greater
4312 * than `count' because there can be holes in there.
4314 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4315 struct buffer_head *bh,
4316 ext4_fsblk_t block_to_free,
4317 unsigned long count, __le32 *first,
4321 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4323 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4324 flags |= EXT4_FREE_BLOCKS_METADATA;
4326 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4328 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4329 "blocks %llu len %lu",
4330 (unsigned long long) block_to_free, count);
4334 if (try_to_extend_transaction(handle, inode)) {
4336 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4337 ext4_handle_dirty_metadata(handle, inode, bh);
4339 ext4_mark_inode_dirty(handle, inode);
4340 ext4_truncate_restart_trans(handle, inode,
4341 blocks_for_truncate(inode));
4343 BUFFER_TRACE(bh, "retaking write access");
4344 ext4_journal_get_write_access(handle, bh);
4348 for (p = first; p < last; p++)
4351 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4356 * ext4_free_data - free a list of data blocks
4357 * @handle: handle for this transaction
4358 * @inode: inode we are dealing with
4359 * @this_bh: indirect buffer_head which contains *@first and *@last
4360 * @first: array of block numbers
4361 * @last: points immediately past the end of array
4363 * We are freeing all blocks refered from that array (numbers are stored as
4364 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4366 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4367 * blocks are contiguous then releasing them at one time will only affect one
4368 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4369 * actually use a lot of journal space.
4371 * @this_bh will be %NULL if @first and @last point into the inode's direct
4374 static void ext4_free_data(handle_t *handle, struct inode *inode,
4375 struct buffer_head *this_bh,
4376 __le32 *first, __le32 *last)
4378 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4379 unsigned long count = 0; /* Number of blocks in the run */
4380 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4383 ext4_fsblk_t nr; /* Current block # */
4384 __le32 *p; /* Pointer into inode/ind
4385 for current block */
4388 if (this_bh) { /* For indirect block */
4389 BUFFER_TRACE(this_bh, "get_write_access");
4390 err = ext4_journal_get_write_access(handle, this_bh);
4391 /* Important: if we can't update the indirect pointers
4392 * to the blocks, we can't free them. */
4397 for (p = first; p < last; p++) {
4398 nr = le32_to_cpu(*p);
4400 /* accumulate blocks to free if they're contiguous */
4403 block_to_free_p = p;
4405 } else if (nr == block_to_free + count) {
4408 if (ext4_clear_blocks(handle, inode, this_bh,
4409 block_to_free, count,
4410 block_to_free_p, p))
4413 block_to_free_p = p;
4420 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4421 count, block_to_free_p, p);
4424 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4427 * The buffer head should have an attached journal head at this
4428 * point. However, if the data is corrupted and an indirect
4429 * block pointed to itself, it would have been detached when
4430 * the block was cleared. Check for this instead of OOPSing.
4432 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4433 ext4_handle_dirty_metadata(handle, inode, this_bh);
4435 EXT4_ERROR_INODE(inode,
4436 "circular indirect block detected at "
4438 (unsigned long long) this_bh->b_blocknr);
4443 * ext4_free_branches - free an array of branches
4444 * @handle: JBD handle for this transaction
4445 * @inode: inode we are dealing with
4446 * @parent_bh: the buffer_head which contains *@first and *@last
4447 * @first: array of block numbers
4448 * @last: pointer immediately past the end of array
4449 * @depth: depth of the branches to free
4451 * We are freeing all blocks refered from these branches (numbers are
4452 * stored as little-endian 32-bit) and updating @inode->i_blocks
4455 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4456 struct buffer_head *parent_bh,
4457 __le32 *first, __le32 *last, int depth)
4462 if (ext4_handle_is_aborted(handle))
4466 struct buffer_head *bh;
4467 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4469 while (--p >= first) {
4470 nr = le32_to_cpu(*p);
4472 continue; /* A hole */
4474 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4476 EXT4_ERROR_INODE(inode,
4477 "invalid indirect mapped "
4478 "block %lu (level %d)",
4479 (unsigned long) nr, depth);
4483 /* Go read the buffer for the next level down */
4484 bh = sb_bread(inode->i_sb, nr);
4487 * A read failure? Report error and clear slot
4491 EXT4_ERROR_INODE_BLOCK(inode, nr,
4496 /* This zaps the entire block. Bottom up. */
4497 BUFFER_TRACE(bh, "free child branches");
4498 ext4_free_branches(handle, inode, bh,
4499 (__le32 *) bh->b_data,
4500 (__le32 *) bh->b_data + addr_per_block,
4504 * Everything below this this pointer has been
4505 * released. Now let this top-of-subtree go.
4507 * We want the freeing of this indirect block to be
4508 * atomic in the journal with the updating of the
4509 * bitmap block which owns it. So make some room in
4512 * We zero the parent pointer *after* freeing its
4513 * pointee in the bitmaps, so if extend_transaction()
4514 * for some reason fails to put the bitmap changes and
4515 * the release into the same transaction, recovery
4516 * will merely complain about releasing a free block,
4517 * rather than leaking blocks.
4519 if (ext4_handle_is_aborted(handle))
4521 if (try_to_extend_transaction(handle, inode)) {
4522 ext4_mark_inode_dirty(handle, inode);
4523 ext4_truncate_restart_trans(handle, inode,
4524 blocks_for_truncate(inode));
4528 * The forget flag here is critical because if
4529 * we are journaling (and not doing data
4530 * journaling), we have to make sure a revoke
4531 * record is written to prevent the journal
4532 * replay from overwriting the (former)
4533 * indirect block if it gets reallocated as a
4534 * data block. This must happen in the same
4535 * transaction where the data blocks are
4538 ext4_free_blocks(handle, inode, 0, nr, 1,
4539 EXT4_FREE_BLOCKS_METADATA|
4540 EXT4_FREE_BLOCKS_FORGET);
4544 * The block which we have just freed is
4545 * pointed to by an indirect block: journal it
4547 BUFFER_TRACE(parent_bh, "get_write_access");
4548 if (!ext4_journal_get_write_access(handle,
4551 BUFFER_TRACE(parent_bh,
4552 "call ext4_handle_dirty_metadata");
4553 ext4_handle_dirty_metadata(handle,
4560 /* We have reached the bottom of the tree. */
4561 BUFFER_TRACE(parent_bh, "free data blocks");
4562 ext4_free_data(handle, inode, parent_bh, first, last);
4566 int ext4_can_truncate(struct inode *inode)
4568 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4570 if (S_ISREG(inode->i_mode))
4572 if (S_ISDIR(inode->i_mode))
4574 if (S_ISLNK(inode->i_mode))
4575 return !ext4_inode_is_fast_symlink(inode);
4582 * We block out ext4_get_block() block instantiations across the entire
4583 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4584 * simultaneously on behalf of the same inode.
4586 * As we work through the truncate and commmit bits of it to the journal there
4587 * is one core, guiding principle: the file's tree must always be consistent on
4588 * disk. We must be able to restart the truncate after a crash.
4590 * The file's tree may be transiently inconsistent in memory (although it
4591 * probably isn't), but whenever we close off and commit a journal transaction,
4592 * the contents of (the filesystem + the journal) must be consistent and
4593 * restartable. It's pretty simple, really: bottom up, right to left (although
4594 * left-to-right works OK too).
4596 * Note that at recovery time, journal replay occurs *before* the restart of
4597 * truncate against the orphan inode list.
4599 * The committed inode has the new, desired i_size (which is the same as
4600 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4601 * that this inode's truncate did not complete and it will again call
4602 * ext4_truncate() to have another go. So there will be instantiated blocks
4603 * to the right of the truncation point in a crashed ext4 filesystem. But
4604 * that's fine - as long as they are linked from the inode, the post-crash
4605 * ext4_truncate() run will find them and release them.
4607 void ext4_truncate(struct inode *inode)
4610 struct ext4_inode_info *ei = EXT4_I(inode);
4611 __le32 *i_data = ei->i_data;
4612 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4613 struct address_space *mapping = inode->i_mapping;
4614 ext4_lblk_t offsets[4];
4619 ext4_lblk_t last_block;
4620 unsigned blocksize = inode->i_sb->s_blocksize;
4622 if (!ext4_can_truncate(inode))
4625 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4627 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4628 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4630 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4631 ext4_ext_truncate(inode);
4635 handle = start_transaction(inode);
4637 return; /* AKPM: return what? */
4639 last_block = (inode->i_size + blocksize-1)
4640 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4642 if (inode->i_size & (blocksize - 1))
4643 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4646 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4648 goto out_stop; /* error */
4651 * OK. This truncate is going to happen. We add the inode to the
4652 * orphan list, so that if this truncate spans multiple transactions,
4653 * and we crash, we will resume the truncate when the filesystem
4654 * recovers. It also marks the inode dirty, to catch the new size.
4656 * Implication: the file must always be in a sane, consistent
4657 * truncatable state while each transaction commits.
4659 if (ext4_orphan_add(handle, inode))
4663 * From here we block out all ext4_get_block() callers who want to
4664 * modify the block allocation tree.
4666 down_write(&ei->i_data_sem);
4668 ext4_discard_preallocations(inode);
4671 * The orphan list entry will now protect us from any crash which
4672 * occurs before the truncate completes, so it is now safe to propagate
4673 * the new, shorter inode size (held for now in i_size) into the
4674 * on-disk inode. We do this via i_disksize, which is the value which
4675 * ext4 *really* writes onto the disk inode.
4677 ei->i_disksize = inode->i_size;
4679 if (n == 1) { /* direct blocks */
4680 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4681 i_data + EXT4_NDIR_BLOCKS);
4685 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4686 /* Kill the top of shared branch (not detached) */
4688 if (partial == chain) {
4689 /* Shared branch grows from the inode */
4690 ext4_free_branches(handle, inode, NULL,
4691 &nr, &nr+1, (chain+n-1) - partial);
4694 * We mark the inode dirty prior to restart,
4695 * and prior to stop. No need for it here.
4698 /* Shared branch grows from an indirect block */
4699 BUFFER_TRACE(partial->bh, "get_write_access");
4700 ext4_free_branches(handle, inode, partial->bh,
4702 partial->p+1, (chain+n-1) - partial);
4705 /* Clear the ends of indirect blocks on the shared branch */
4706 while (partial > chain) {
4707 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4708 (__le32*)partial->bh->b_data+addr_per_block,
4709 (chain+n-1) - partial);
4710 BUFFER_TRACE(partial->bh, "call brelse");
4711 brelse(partial->bh);
4715 /* Kill the remaining (whole) subtrees */
4716 switch (offsets[0]) {
4718 nr = i_data[EXT4_IND_BLOCK];
4720 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4721 i_data[EXT4_IND_BLOCK] = 0;
4723 case EXT4_IND_BLOCK:
4724 nr = i_data[EXT4_DIND_BLOCK];
4726 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4727 i_data[EXT4_DIND_BLOCK] = 0;
4729 case EXT4_DIND_BLOCK:
4730 nr = i_data[EXT4_TIND_BLOCK];
4732 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4733 i_data[EXT4_TIND_BLOCK] = 0;
4735 case EXT4_TIND_BLOCK:
4739 up_write(&ei->i_data_sem);
4740 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4741 ext4_mark_inode_dirty(handle, inode);
4744 * In a multi-transaction truncate, we only make the final transaction
4748 ext4_handle_sync(handle);
4751 * If this was a simple ftruncate(), and the file will remain alive
4752 * then we need to clear up the orphan record which we created above.
4753 * However, if this was a real unlink then we were called by
4754 * ext4_delete_inode(), and we allow that function to clean up the
4755 * orphan info for us.
4758 ext4_orphan_del(handle, inode);
4760 ext4_journal_stop(handle);
4764 * ext4_get_inode_loc returns with an extra refcount against the inode's
4765 * underlying buffer_head on success. If 'in_mem' is true, we have all
4766 * data in memory that is needed to recreate the on-disk version of this
4769 static int __ext4_get_inode_loc(struct inode *inode,
4770 struct ext4_iloc *iloc, int in_mem)
4772 struct ext4_group_desc *gdp;
4773 struct buffer_head *bh;
4774 struct super_block *sb = inode->i_sb;
4776 int inodes_per_block, inode_offset;
4779 if (!ext4_valid_inum(sb, inode->i_ino))
4782 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4783 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4788 * Figure out the offset within the block group inode table
4790 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4791 inode_offset = ((inode->i_ino - 1) %
4792 EXT4_INODES_PER_GROUP(sb));
4793 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4794 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4796 bh = sb_getblk(sb, block);
4798 EXT4_ERROR_INODE_BLOCK(inode, block,
4799 "unable to read itable block");
4802 if (!buffer_uptodate(bh)) {
4806 * If the buffer has the write error flag, we have failed
4807 * to write out another inode in the same block. In this
4808 * case, we don't have to read the block because we may
4809 * read the old inode data successfully.
4811 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4812 set_buffer_uptodate(bh);
4814 if (buffer_uptodate(bh)) {
4815 /* someone brought it uptodate while we waited */
4821 * If we have all information of the inode in memory and this
4822 * is the only valid inode in the block, we need not read the
4826 struct buffer_head *bitmap_bh;
4829 start = inode_offset & ~(inodes_per_block - 1);
4831 /* Is the inode bitmap in cache? */
4832 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4837 * If the inode bitmap isn't in cache then the
4838 * optimisation may end up performing two reads instead
4839 * of one, so skip it.
4841 if (!buffer_uptodate(bitmap_bh)) {
4845 for (i = start; i < start + inodes_per_block; i++) {
4846 if (i == inode_offset)
4848 if (ext4_test_bit(i, bitmap_bh->b_data))
4852 if (i == start + inodes_per_block) {
4853 /* all other inodes are free, so skip I/O */
4854 memset(bh->b_data, 0, bh->b_size);
4855 set_buffer_uptodate(bh);
4863 * If we need to do any I/O, try to pre-readahead extra
4864 * blocks from the inode table.
4866 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4867 ext4_fsblk_t b, end, table;
4870 table = ext4_inode_table(sb, gdp);
4871 /* s_inode_readahead_blks is always a power of 2 */
4872 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4875 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4876 num = EXT4_INODES_PER_GROUP(sb);
4877 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4878 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4879 num -= ext4_itable_unused_count(sb, gdp);
4880 table += num / inodes_per_block;
4884 sb_breadahead(sb, b++);
4888 * There are other valid inodes in the buffer, this inode
4889 * has in-inode xattrs, or we don't have this inode in memory.
4890 * Read the block from disk.
4893 bh->b_end_io = end_buffer_read_sync;
4894 submit_bh(READ_META, bh);
4896 if (!buffer_uptodate(bh)) {
4897 EXT4_ERROR_INODE_BLOCK(inode, block,
4898 "unable to read itable block");
4908 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4910 /* We have all inode data except xattrs in memory here. */
4911 return __ext4_get_inode_loc(inode, iloc,
4912 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4915 void ext4_set_inode_flags(struct inode *inode)
4917 unsigned int flags = EXT4_I(inode)->i_flags;
4919 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4920 if (flags & EXT4_SYNC_FL)
4921 inode->i_flags |= S_SYNC;
4922 if (flags & EXT4_APPEND_FL)
4923 inode->i_flags |= S_APPEND;
4924 if (flags & EXT4_IMMUTABLE_FL)
4925 inode->i_flags |= S_IMMUTABLE;
4926 if (flags & EXT4_NOATIME_FL)
4927 inode->i_flags |= S_NOATIME;
4928 if (flags & EXT4_DIRSYNC_FL)
4929 inode->i_flags |= S_DIRSYNC;
4932 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4933 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4935 unsigned int vfs_fl;
4936 unsigned long old_fl, new_fl;
4939 vfs_fl = ei->vfs_inode.i_flags;
4940 old_fl = ei->i_flags;
4941 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4942 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4944 if (vfs_fl & S_SYNC)
4945 new_fl |= EXT4_SYNC_FL;
4946 if (vfs_fl & S_APPEND)
4947 new_fl |= EXT4_APPEND_FL;
4948 if (vfs_fl & S_IMMUTABLE)
4949 new_fl |= EXT4_IMMUTABLE_FL;
4950 if (vfs_fl & S_NOATIME)
4951 new_fl |= EXT4_NOATIME_FL;
4952 if (vfs_fl & S_DIRSYNC)
4953 new_fl |= EXT4_DIRSYNC_FL;
4954 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4957 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4958 struct ext4_inode_info *ei)
4961 struct inode *inode = &(ei->vfs_inode);
4962 struct super_block *sb = inode->i_sb;
4964 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4965 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4966 /* we are using combined 48 bit field */
4967 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4968 le32_to_cpu(raw_inode->i_blocks_lo);
4969 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4970 /* i_blocks represent file system block size */
4971 return i_blocks << (inode->i_blkbits - 9);
4976 return le32_to_cpu(raw_inode->i_blocks_lo);
4980 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4982 struct ext4_iloc iloc;
4983 struct ext4_inode *raw_inode;
4984 struct ext4_inode_info *ei;
4985 struct inode *inode;
4986 journal_t *journal = EXT4_SB(sb)->s_journal;
4990 inode = iget_locked(sb, ino);
4992 return ERR_PTR(-ENOMEM);
4993 if (!(inode->i_state & I_NEW))
4999 ret = __ext4_get_inode_loc(inode, &iloc, 0);
5002 raw_inode = ext4_raw_inode(&iloc);
5003 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
5004 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
5005 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
5006 if (!(test_opt(inode->i_sb, NO_UID32))) {
5007 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
5008 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
5010 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
5012 ei->i_state_flags = 0;
5013 ei->i_dir_start_lookup = 0;
5014 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
5015 /* We now have enough fields to check if the inode was active or not.
5016 * This is needed because nfsd might try to access dead inodes
5017 * the test is that same one that e2fsck uses
5018 * NeilBrown 1999oct15
5020 if (inode->i_nlink == 0) {
5021 if (inode->i_mode == 0 ||
5022 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
5023 /* this inode is deleted */
5027 /* The only unlinked inodes we let through here have
5028 * valid i_mode and are being read by the orphan
5029 * recovery code: that's fine, we're about to complete
5030 * the process of deleting those. */
5032 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
5033 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
5034 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
5035 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
5037 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
5038 inode->i_size = ext4_isize(raw_inode);
5039 ei->i_disksize = inode->i_size;
5041 ei->i_reserved_quota = 0;
5043 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
5044 ei->i_block_group = iloc.block_group;
5045 ei->i_last_alloc_group = ~0;
5047 * NOTE! The in-memory inode i_data array is in little-endian order
5048 * even on big-endian machines: we do NOT byteswap the block numbers!
5050 for (block = 0; block < EXT4_N_BLOCKS; block++)
5051 ei->i_data[block] = raw_inode->i_block[block];
5052 INIT_LIST_HEAD(&ei->i_orphan);
5055 * Set transaction id's of transactions that have to be committed
5056 * to finish f[data]sync. We set them to currently running transaction
5057 * as we cannot be sure that the inode or some of its metadata isn't
5058 * part of the transaction - the inode could have been reclaimed and
5059 * now it is reread from disk.
5062 transaction_t *transaction;
5065 read_lock(&journal->j_state_lock);
5066 if (journal->j_running_transaction)
5067 transaction = journal->j_running_transaction;
5069 transaction = journal->j_committing_transaction;
5071 tid = transaction->t_tid;
5073 tid = journal->j_commit_sequence;
5074 read_unlock(&journal->j_state_lock);
5075 ei->i_sync_tid = tid;
5076 ei->i_datasync_tid = tid;
5079 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5080 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
5081 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
5082 EXT4_INODE_SIZE(inode->i_sb)) {
5086 if (ei->i_extra_isize == 0) {
5087 /* The extra space is currently unused. Use it. */
5088 ei->i_extra_isize = sizeof(struct ext4_inode) -
5089 EXT4_GOOD_OLD_INODE_SIZE;
5091 __le32 *magic = (void *)raw_inode +
5092 EXT4_GOOD_OLD_INODE_SIZE +
5094 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
5095 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5098 ei->i_extra_isize = 0;
5100 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5101 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5102 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5103 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5105 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5106 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5107 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5109 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5113 if (ei->i_file_acl &&
5114 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5115 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5119 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5120 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5121 (S_ISLNK(inode->i_mode) &&
5122 !ext4_inode_is_fast_symlink(inode)))
5123 /* Validate extent which is part of inode */
5124 ret = ext4_ext_check_inode(inode);
5125 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5126 (S_ISLNK(inode->i_mode) &&
5127 !ext4_inode_is_fast_symlink(inode))) {
5128 /* Validate block references which are part of inode */
5129 ret = ext4_check_inode_blockref(inode);
5134 if (S_ISREG(inode->i_mode)) {
5135 inode->i_op = &ext4_file_inode_operations;
5136 inode->i_fop = &ext4_file_operations;
5137 ext4_set_aops(inode);
5138 } else if (S_ISDIR(inode->i_mode)) {
5139 inode->i_op = &ext4_dir_inode_operations;
5140 inode->i_fop = &ext4_dir_operations;
5141 } else if (S_ISLNK(inode->i_mode)) {
5142 if (ext4_inode_is_fast_symlink(inode)) {
5143 inode->i_op = &ext4_fast_symlink_inode_operations;
5144 nd_terminate_link(ei->i_data, inode->i_size,
5145 sizeof(ei->i_data) - 1);
5147 inode->i_op = &ext4_symlink_inode_operations;
5148 ext4_set_aops(inode);
5150 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5151 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5152 inode->i_op = &ext4_special_inode_operations;
5153 if (raw_inode->i_block[0])
5154 init_special_inode(inode, inode->i_mode,
5155 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5157 init_special_inode(inode, inode->i_mode,
5158 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5161 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5165 ext4_set_inode_flags(inode);
5166 unlock_new_inode(inode);
5172 return ERR_PTR(ret);
5175 static int ext4_inode_blocks_set(handle_t *handle,
5176 struct ext4_inode *raw_inode,
5177 struct ext4_inode_info *ei)
5179 struct inode *inode = &(ei->vfs_inode);
5180 u64 i_blocks = inode->i_blocks;
5181 struct super_block *sb = inode->i_sb;
5183 if (i_blocks <= ~0U) {
5185 * i_blocks can be represnted in a 32 bit variable
5186 * as multiple of 512 bytes
5188 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5189 raw_inode->i_blocks_high = 0;
5190 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5193 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5196 if (i_blocks <= 0xffffffffffffULL) {
5198 * i_blocks can be represented in a 48 bit variable
5199 * as multiple of 512 bytes
5201 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5202 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5203 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5205 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5206 /* i_block is stored in file system block size */
5207 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5208 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5209 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5215 * Post the struct inode info into an on-disk inode location in the
5216 * buffer-cache. This gobbles the caller's reference to the
5217 * buffer_head in the inode location struct.
5219 * The caller must have write access to iloc->bh.
5221 static int ext4_do_update_inode(handle_t *handle,
5222 struct inode *inode,
5223 struct ext4_iloc *iloc)
5225 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5226 struct ext4_inode_info *ei = EXT4_I(inode);
5227 struct buffer_head *bh = iloc->bh;
5228 int err = 0, rc, block;
5230 /* For fields not not tracking in the in-memory inode,
5231 * initialise them to zero for new inodes. */
5232 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5233 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5235 ext4_get_inode_flags(ei);
5236 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5237 if (!(test_opt(inode->i_sb, NO_UID32))) {
5238 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5239 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5241 * Fix up interoperability with old kernels. Otherwise, old inodes get
5242 * re-used with the upper 16 bits of the uid/gid intact
5245 raw_inode->i_uid_high =
5246 cpu_to_le16(high_16_bits(inode->i_uid));
5247 raw_inode->i_gid_high =
5248 cpu_to_le16(high_16_bits(inode->i_gid));
5250 raw_inode->i_uid_high = 0;
5251 raw_inode->i_gid_high = 0;
5254 raw_inode->i_uid_low =
5255 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5256 raw_inode->i_gid_low =
5257 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5258 raw_inode->i_uid_high = 0;
5259 raw_inode->i_gid_high = 0;
5261 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5263 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5264 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5265 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5266 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5268 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5270 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5271 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5272 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5273 cpu_to_le32(EXT4_OS_HURD))
5274 raw_inode->i_file_acl_high =
5275 cpu_to_le16(ei->i_file_acl >> 32);
5276 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5277 ext4_isize_set(raw_inode, ei->i_disksize);
5278 if (ei->i_disksize > 0x7fffffffULL) {
5279 struct super_block *sb = inode->i_sb;
5280 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5281 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5282 EXT4_SB(sb)->s_es->s_rev_level ==
5283 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5284 /* If this is the first large file
5285 * created, add a flag to the superblock.
5287 err = ext4_journal_get_write_access(handle,
5288 EXT4_SB(sb)->s_sbh);
5291 ext4_update_dynamic_rev(sb);
5292 EXT4_SET_RO_COMPAT_FEATURE(sb,
5293 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5295 ext4_handle_sync(handle);
5296 err = ext4_handle_dirty_metadata(handle, NULL,
5297 EXT4_SB(sb)->s_sbh);
5300 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5301 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5302 if (old_valid_dev(inode->i_rdev)) {
5303 raw_inode->i_block[0] =
5304 cpu_to_le32(old_encode_dev(inode->i_rdev));
5305 raw_inode->i_block[1] = 0;
5307 raw_inode->i_block[0] = 0;
5308 raw_inode->i_block[1] =
5309 cpu_to_le32(new_encode_dev(inode->i_rdev));
5310 raw_inode->i_block[2] = 0;
5313 for (block = 0; block < EXT4_N_BLOCKS; block++)
5314 raw_inode->i_block[block] = ei->i_data[block];
5316 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5317 if (ei->i_extra_isize) {
5318 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5319 raw_inode->i_version_hi =
5320 cpu_to_le32(inode->i_version >> 32);
5321 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5324 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5325 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5328 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5330 ext4_update_inode_fsync_trans(handle, inode, 0);
5333 ext4_std_error(inode->i_sb, err);
5338 * ext4_write_inode()
5340 * We are called from a few places:
5342 * - Within generic_file_write() for O_SYNC files.
5343 * Here, there will be no transaction running. We wait for any running
5344 * trasnaction to commit.
5346 * - Within sys_sync(), kupdate and such.
5347 * We wait on commit, if tol to.
5349 * - Within prune_icache() (PF_MEMALLOC == true)
5350 * Here we simply return. We can't afford to block kswapd on the
5353 * In all cases it is actually safe for us to return without doing anything,
5354 * because the inode has been copied into a raw inode buffer in
5355 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5358 * Note that we are absolutely dependent upon all inode dirtiers doing the
5359 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5360 * which we are interested.
5362 * It would be a bug for them to not do this. The code:
5364 * mark_inode_dirty(inode)
5366 * inode->i_size = expr;
5368 * is in error because a kswapd-driven write_inode() could occur while
5369 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5370 * will no longer be on the superblock's dirty inode list.
5372 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5376 if (current->flags & PF_MEMALLOC)
5379 if (EXT4_SB(inode->i_sb)->s_journal) {
5380 if (ext4_journal_current_handle()) {
5381 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5386 if (wbc->sync_mode != WB_SYNC_ALL)
5389 err = ext4_force_commit(inode->i_sb);
5391 struct ext4_iloc iloc;
5393 err = __ext4_get_inode_loc(inode, &iloc, 0);
5396 if (wbc->sync_mode == WB_SYNC_ALL)
5397 sync_dirty_buffer(iloc.bh);
5398 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5399 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5400 "IO error syncing inode");
5411 * Called from notify_change.
5413 * We want to trap VFS attempts to truncate the file as soon as
5414 * possible. In particular, we want to make sure that when the VFS
5415 * shrinks i_size, we put the inode on the orphan list and modify
5416 * i_disksize immediately, so that during the subsequent flushing of
5417 * dirty pages and freeing of disk blocks, we can guarantee that any
5418 * commit will leave the blocks being flushed in an unused state on
5419 * disk. (On recovery, the inode will get truncated and the blocks will
5420 * be freed, so we have a strong guarantee that no future commit will
5421 * leave these blocks visible to the user.)
5423 * Another thing we have to assure is that if we are in ordered mode
5424 * and inode is still attached to the committing transaction, we must
5425 * we start writeout of all the dirty pages which are being truncated.
5426 * This way we are sure that all the data written in the previous
5427 * transaction are already on disk (truncate waits for pages under
5430 * Called with inode->i_mutex down.
5432 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5434 struct inode *inode = dentry->d_inode;
5436 const unsigned int ia_valid = attr->ia_valid;
5438 error = inode_change_ok(inode, attr);
5442 if (is_quota_modification(inode, attr))
5443 dquot_initialize(inode);
5444 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5445 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5448 /* (user+group)*(old+new) structure, inode write (sb,
5449 * inode block, ? - but truncate inode update has it) */
5450 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5451 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5452 if (IS_ERR(handle)) {
5453 error = PTR_ERR(handle);
5456 error = dquot_transfer(inode, attr);
5458 ext4_journal_stop(handle);
5461 /* Update corresponding info in inode so that everything is in
5462 * one transaction */
5463 if (attr->ia_valid & ATTR_UID)
5464 inode->i_uid = attr->ia_uid;
5465 if (attr->ia_valid & ATTR_GID)
5466 inode->i_gid = attr->ia_gid;
5467 error = ext4_mark_inode_dirty(handle, inode);
5468 ext4_journal_stop(handle);
5471 if (attr->ia_valid & ATTR_SIZE) {
5472 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5475 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5480 if (S_ISREG(inode->i_mode) &&
5481 attr->ia_valid & ATTR_SIZE &&
5482 (attr->ia_size < inode->i_size ||
5483 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5486 handle = ext4_journal_start(inode, 3);
5487 if (IS_ERR(handle)) {
5488 error = PTR_ERR(handle);
5492 error = ext4_orphan_add(handle, inode);
5493 EXT4_I(inode)->i_disksize = attr->ia_size;
5494 rc = ext4_mark_inode_dirty(handle, inode);
5497 ext4_journal_stop(handle);
5499 if (ext4_should_order_data(inode)) {
5500 error = ext4_begin_ordered_truncate(inode,
5503 /* Do as much error cleanup as possible */
5504 handle = ext4_journal_start(inode, 3);
5505 if (IS_ERR(handle)) {
5506 ext4_orphan_del(NULL, inode);
5509 ext4_orphan_del(handle, inode);
5510 ext4_journal_stop(handle);
5514 /* ext4_truncate will clear the flag */
5515 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5516 ext4_truncate(inode);
5519 if ((attr->ia_valid & ATTR_SIZE) &&
5520 attr->ia_size != i_size_read(inode))
5521 rc = vmtruncate(inode, attr->ia_size);
5524 setattr_copy(inode, attr);
5525 mark_inode_dirty(inode);
5529 * If the call to ext4_truncate failed to get a transaction handle at
5530 * all, we need to clean up the in-core orphan list manually.
5533 ext4_orphan_del(NULL, inode);
5535 if (!rc && (ia_valid & ATTR_MODE))
5536 rc = ext4_acl_chmod(inode);
5539 ext4_std_error(inode->i_sb, error);
5545 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5548 struct inode *inode;
5549 unsigned long delalloc_blocks;
5551 inode = dentry->d_inode;
5552 generic_fillattr(inode, stat);
5555 * We can't update i_blocks if the block allocation is delayed
5556 * otherwise in the case of system crash before the real block
5557 * allocation is done, we will have i_blocks inconsistent with
5558 * on-disk file blocks.
5559 * We always keep i_blocks updated together with real
5560 * allocation. But to not confuse with user, stat
5561 * will return the blocks that include the delayed allocation
5562 * blocks for this file.
5564 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5565 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5566 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5568 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5572 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5577 /* if nrblocks are contiguous */
5580 * With N contiguous data blocks, it need at most
5581 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5582 * 2 dindirect blocks
5585 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5586 return indirects + 3;
5589 * if nrblocks are not contiguous, worse case, each block touch
5590 * a indirect block, and each indirect block touch a double indirect
5591 * block, plus a triple indirect block
5593 indirects = nrblocks * 2 + 1;
5597 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5599 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5600 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5601 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5605 * Account for index blocks, block groups bitmaps and block group
5606 * descriptor blocks if modify datablocks and index blocks
5607 * worse case, the indexs blocks spread over different block groups
5609 * If datablocks are discontiguous, they are possible to spread over
5610 * different block groups too. If they are contiuguous, with flexbg,
5611 * they could still across block group boundary.
5613 * Also account for superblock, inode, quota and xattr blocks
5615 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5617 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5623 * How many index blocks need to touch to modify nrblocks?
5624 * The "Chunk" flag indicating whether the nrblocks is
5625 * physically contiguous on disk
5627 * For Direct IO and fallocate, they calls get_block to allocate
5628 * one single extent at a time, so they could set the "Chunk" flag
5630 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5635 * Now let's see how many group bitmaps and group descriptors need
5645 if (groups > ngroups)
5647 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5648 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5650 /* bitmaps and block group descriptor blocks */
5651 ret += groups + gdpblocks;
5653 /* Blocks for super block, inode, quota and xattr blocks */
5654 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5660 * Calulate the total number of credits to reserve to fit
5661 * the modification of a single pages into a single transaction,
5662 * which may include multiple chunks of block allocations.
5664 * This could be called via ext4_write_begin()
5666 * We need to consider the worse case, when
5667 * one new block per extent.
5669 int ext4_writepage_trans_blocks(struct inode *inode)
5671 int bpp = ext4_journal_blocks_per_page(inode);
5674 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5676 /* Account for data blocks for journalled mode */
5677 if (ext4_should_journal_data(inode))
5683 * Calculate the journal credits for a chunk of data modification.
5685 * This is called from DIO, fallocate or whoever calling
5686 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5688 * journal buffers for data blocks are not included here, as DIO
5689 * and fallocate do no need to journal data buffers.
5691 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5693 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5697 * The caller must have previously called ext4_reserve_inode_write().
5698 * Give this, we know that the caller already has write access to iloc->bh.
5700 int ext4_mark_iloc_dirty(handle_t *handle,
5701 struct inode *inode, struct ext4_iloc *iloc)
5705 if (test_opt(inode->i_sb, I_VERSION))
5706 inode_inc_iversion(inode);
5708 /* the do_update_inode consumes one bh->b_count */
5711 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5712 err = ext4_do_update_inode(handle, inode, iloc);
5718 * On success, We end up with an outstanding reference count against
5719 * iloc->bh. This _must_ be cleaned up later.
5723 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5724 struct ext4_iloc *iloc)
5728 err = ext4_get_inode_loc(inode, iloc);
5730 BUFFER_TRACE(iloc->bh, "get_write_access");
5731 err = ext4_journal_get_write_access(handle, iloc->bh);
5737 ext4_std_error(inode->i_sb, err);
5742 * Expand an inode by new_extra_isize bytes.
5743 * Returns 0 on success or negative error number on failure.
5745 static int ext4_expand_extra_isize(struct inode *inode,
5746 unsigned int new_extra_isize,
5747 struct ext4_iloc iloc,
5750 struct ext4_inode *raw_inode;
5751 struct ext4_xattr_ibody_header *header;
5753 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5756 raw_inode = ext4_raw_inode(&iloc);
5758 header = IHDR(inode, raw_inode);
5760 /* No extended attributes present */
5761 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5762 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5763 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5765 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5769 /* try to expand with EAs present */
5770 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5775 * What we do here is to mark the in-core inode as clean with respect to inode
5776 * dirtiness (it may still be data-dirty).
5777 * This means that the in-core inode may be reaped by prune_icache
5778 * without having to perform any I/O. This is a very good thing,
5779 * because *any* task may call prune_icache - even ones which
5780 * have a transaction open against a different journal.
5782 * Is this cheating? Not really. Sure, we haven't written the
5783 * inode out, but prune_icache isn't a user-visible syncing function.
5784 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5785 * we start and wait on commits.
5787 * Is this efficient/effective? Well, we're being nice to the system
5788 * by cleaning up our inodes proactively so they can be reaped
5789 * without I/O. But we are potentially leaving up to five seconds'
5790 * worth of inodes floating about which prune_icache wants us to
5791 * write out. One way to fix that would be to get prune_icache()
5792 * to do a write_super() to free up some memory. It has the desired
5795 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5797 struct ext4_iloc iloc;
5798 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5799 static unsigned int mnt_count;
5803 err = ext4_reserve_inode_write(handle, inode, &iloc);
5804 if (ext4_handle_valid(handle) &&
5805 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5806 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5808 * We need extra buffer credits since we may write into EA block
5809 * with this same handle. If journal_extend fails, then it will
5810 * only result in a minor loss of functionality for that inode.
5811 * If this is felt to be critical, then e2fsck should be run to
5812 * force a large enough s_min_extra_isize.
5814 if ((jbd2_journal_extend(handle,
5815 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5816 ret = ext4_expand_extra_isize(inode,
5817 sbi->s_want_extra_isize,
5820 ext4_set_inode_state(inode,
5821 EXT4_STATE_NO_EXPAND);
5823 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5824 ext4_warning(inode->i_sb,
5825 "Unable to expand inode %lu. Delete"
5826 " some EAs or run e2fsck.",
5829 le16_to_cpu(sbi->s_es->s_mnt_count);
5835 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5840 * ext4_dirty_inode() is called from __mark_inode_dirty()
5842 * We're really interested in the case where a file is being extended.
5843 * i_size has been changed by generic_commit_write() and we thus need
5844 * to include the updated inode in the current transaction.
5846 * Also, dquot_alloc_block() will always dirty the inode when blocks
5847 * are allocated to the file.
5849 * If the inode is marked synchronous, we don't honour that here - doing
5850 * so would cause a commit on atime updates, which we don't bother doing.
5851 * We handle synchronous inodes at the highest possible level.
5853 void ext4_dirty_inode(struct inode *inode)
5857 handle = ext4_journal_start(inode, 2);
5861 ext4_mark_inode_dirty(handle, inode);
5863 ext4_journal_stop(handle);
5870 * Bind an inode's backing buffer_head into this transaction, to prevent
5871 * it from being flushed to disk early. Unlike
5872 * ext4_reserve_inode_write, this leaves behind no bh reference and
5873 * returns no iloc structure, so the caller needs to repeat the iloc
5874 * lookup to mark the inode dirty later.
5876 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5878 struct ext4_iloc iloc;
5882 err = ext4_get_inode_loc(inode, &iloc);
5884 BUFFER_TRACE(iloc.bh, "get_write_access");
5885 err = jbd2_journal_get_write_access(handle, iloc.bh);
5887 err = ext4_handle_dirty_metadata(handle,
5893 ext4_std_error(inode->i_sb, err);
5898 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5905 * We have to be very careful here: changing a data block's
5906 * journaling status dynamically is dangerous. If we write a
5907 * data block to the journal, change the status and then delete
5908 * that block, we risk forgetting to revoke the old log record
5909 * from the journal and so a subsequent replay can corrupt data.
5910 * So, first we make sure that the journal is empty and that
5911 * nobody is changing anything.
5914 journal = EXT4_JOURNAL(inode);
5917 if (is_journal_aborted(journal))
5920 jbd2_journal_lock_updates(journal);
5921 jbd2_journal_flush(journal);
5924 * OK, there are no updates running now, and all cached data is
5925 * synced to disk. We are now in a completely consistent state
5926 * which doesn't have anything in the journal, and we know that
5927 * no filesystem updates are running, so it is safe to modify
5928 * the inode's in-core data-journaling state flag now.
5932 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5934 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5935 ext4_set_aops(inode);
5937 jbd2_journal_unlock_updates(journal);
5939 /* Finally we can mark the inode as dirty. */
5941 handle = ext4_journal_start(inode, 1);
5943 return PTR_ERR(handle);
5945 err = ext4_mark_inode_dirty(handle, inode);
5946 ext4_handle_sync(handle);
5947 ext4_journal_stop(handle);
5948 ext4_std_error(inode->i_sb, err);
5953 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5955 return !buffer_mapped(bh);
5958 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5960 struct page *page = vmf->page;
5965 struct file *file = vma->vm_file;
5966 struct inode *inode = file->f_path.dentry->d_inode;
5967 struct address_space *mapping = inode->i_mapping;
5970 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5971 * get i_mutex because we are already holding mmap_sem.
5973 down_read(&inode->i_alloc_sem);
5974 size = i_size_read(inode);
5975 if (page->mapping != mapping || size <= page_offset(page)
5976 || !PageUptodate(page)) {
5977 /* page got truncated from under us? */
5981 if (PageMappedToDisk(page))
5984 if (page->index == size >> PAGE_CACHE_SHIFT)
5985 len = size & ~PAGE_CACHE_MASK;
5987 len = PAGE_CACHE_SIZE;
5991 * return if we have all the buffers mapped. This avoid
5992 * the need to call write_begin/write_end which does a
5993 * journal_start/journal_stop which can block and take
5996 if (page_has_buffers(page)) {
5997 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5998 ext4_bh_unmapped)) {
6005 * OK, we need to fill the hole... Do write_begin write_end
6006 * to do block allocation/reservation.We are not holding
6007 * inode.i__mutex here. That allow * parallel write_begin,
6008 * write_end call. lock_page prevent this from happening
6009 * on the same page though
6011 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
6012 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
6015 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
6016 len, len, page, fsdata);
6022 ret = VM_FAULT_SIGBUS;
6023 up_read(&inode->i_alloc_sem);