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 trace_ext4_begin_ordered_truncate(inode, new_size);
57 return jbd2_journal_begin_ordered_truncate(
58 EXT4_SB(inode->i_sb)->s_journal,
59 &EXT4_I(inode)->jinode,
63 static void ext4_invalidatepage(struct page *page, unsigned long offset);
64 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
65 struct buffer_head *bh_result, int create);
66 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
67 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
68 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
69 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
72 * Test whether an inode is a fast symlink.
74 static int ext4_inode_is_fast_symlink(struct inode *inode)
76 int ea_blocks = EXT4_I(inode)->i_file_acl ?
77 (inode->i_sb->s_blocksize >> 9) : 0;
79 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
83 * Work out how many blocks we need to proceed with the next chunk of a
84 * truncate transaction.
86 static unsigned long blocks_for_truncate(struct inode *inode)
90 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
92 /* Give ourselves just enough room to cope with inodes in which
93 * i_blocks is corrupt: we've seen disk corruptions in the past
94 * which resulted in random data in an inode which looked enough
95 * like a regular file for ext4 to try to delete it. Things
96 * will go a bit crazy if that happens, but at least we should
97 * try not to panic the whole kernel. */
101 /* But we need to bound the transaction so we don't overflow the
103 if (needed > EXT4_MAX_TRANS_DATA)
104 needed = EXT4_MAX_TRANS_DATA;
106 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
110 * Truncate transactions can be complex and absolutely huge. So we need to
111 * be able to restart the transaction at a conventient checkpoint to make
112 * sure we don't overflow the journal.
114 * start_transaction gets us a new handle for a truncate transaction,
115 * and extend_transaction tries to extend the existing one a bit. If
116 * extend fails, we need to propagate the failure up and restart the
117 * transaction in the top-level truncate loop. --sct
119 static handle_t *start_transaction(struct inode *inode)
123 result = ext4_journal_start(inode, blocks_for_truncate(inode));
127 ext4_std_error(inode->i_sb, PTR_ERR(result));
132 * Try to extend this transaction for the purposes of truncation.
134 * Returns 0 if we managed to create more room. If we can't create more
135 * room, and the transaction must be restarted we return 1.
137 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
139 if (!ext4_handle_valid(handle))
141 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
143 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
149 * Restart the transaction associated with *handle. This does a commit,
150 * so before we call here everything must be consistently dirtied against
153 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
159 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
160 * moment, get_block can be called only for blocks inside i_size since
161 * page cache has been already dropped and writes are blocked by
162 * i_mutex. So we can safely drop the i_data_sem here.
164 BUG_ON(EXT4_JOURNAL(inode) == NULL);
165 jbd_debug(2, "restarting handle %p\n", handle);
166 up_write(&EXT4_I(inode)->i_data_sem);
167 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
168 down_write(&EXT4_I(inode)->i_data_sem);
169 ext4_discard_preallocations(inode);
175 * Called at the last iput() if i_nlink is zero.
177 void ext4_evict_inode(struct inode *inode)
182 trace_ext4_evict_inode(inode);
183 if (inode->i_nlink) {
184 truncate_inode_pages(&inode->i_data, 0);
188 if (!is_bad_inode(inode))
189 dquot_initialize(inode);
191 if (ext4_should_order_data(inode))
192 ext4_begin_ordered_truncate(inode, 0);
193 truncate_inode_pages(&inode->i_data, 0);
195 if (is_bad_inode(inode))
198 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
199 if (IS_ERR(handle)) {
200 ext4_std_error(inode->i_sb, PTR_ERR(handle));
202 * If we're going to skip the normal cleanup, we still need to
203 * make sure that the in-core orphan linked list is properly
206 ext4_orphan_del(NULL, inode);
211 ext4_handle_sync(handle);
213 err = ext4_mark_inode_dirty(handle, inode);
215 ext4_warning(inode->i_sb,
216 "couldn't mark inode dirty (err %d)", err);
220 ext4_truncate(inode);
223 * ext4_ext_truncate() doesn't reserve any slop when it
224 * restarts journal transactions; therefore there may not be
225 * enough credits left in the handle to remove the inode from
226 * the orphan list and set the dtime field.
228 if (!ext4_handle_has_enough_credits(handle, 3)) {
229 err = ext4_journal_extend(handle, 3);
231 err = ext4_journal_restart(handle, 3);
233 ext4_warning(inode->i_sb,
234 "couldn't extend journal (err %d)", err);
236 ext4_journal_stop(handle);
237 ext4_orphan_del(NULL, inode);
243 * Kill off the orphan record which ext4_truncate created.
244 * AKPM: I think this can be inside the above `if'.
245 * Note that ext4_orphan_del() has to be able to cope with the
246 * deletion of a non-existent orphan - this is because we don't
247 * know if ext4_truncate() actually created an orphan record.
248 * (Well, we could do this if we need to, but heck - it works)
250 ext4_orphan_del(handle, inode);
251 EXT4_I(inode)->i_dtime = get_seconds();
254 * One subtle ordering requirement: if anything has gone wrong
255 * (transaction abort, IO errors, whatever), then we can still
256 * do these next steps (the fs will already have been marked as
257 * having errors), but we can't free the inode if the mark_dirty
260 if (ext4_mark_inode_dirty(handle, inode))
261 /* If that failed, just do the required in-core inode clear. */
262 ext4_clear_inode(inode);
264 ext4_free_inode(handle, inode);
265 ext4_journal_stop(handle);
268 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
274 struct buffer_head *bh;
277 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
279 p->key = *(p->p = v);
284 * ext4_block_to_path - parse the block number into array of offsets
285 * @inode: inode in question (we are only interested in its superblock)
286 * @i_block: block number to be parsed
287 * @offsets: array to store the offsets in
288 * @boundary: set this non-zero if the referred-to block is likely to be
289 * followed (on disk) by an indirect block.
291 * To store the locations of file's data ext4 uses a data structure common
292 * for UNIX filesystems - tree of pointers anchored in the inode, with
293 * data blocks at leaves and indirect blocks in intermediate nodes.
294 * This function translates the block number into path in that tree -
295 * return value is the path length and @offsets[n] is the offset of
296 * pointer to (n+1)th node in the nth one. If @block is out of range
297 * (negative or too large) warning is printed and zero returned.
299 * Note: function doesn't find node addresses, so no IO is needed. All
300 * we need to know is the capacity of indirect blocks (taken from the
305 * Portability note: the last comparison (check that we fit into triple
306 * indirect block) is spelled differently, because otherwise on an
307 * architecture with 32-bit longs and 8Kb pages we might get into trouble
308 * if our filesystem had 8Kb blocks. We might use long long, but that would
309 * kill us on x86. Oh, well, at least the sign propagation does not matter -
310 * i_block would have to be negative in the very beginning, so we would not
314 static int ext4_block_to_path(struct inode *inode,
316 ext4_lblk_t offsets[4], int *boundary)
318 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
319 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
320 const long direct_blocks = EXT4_NDIR_BLOCKS,
321 indirect_blocks = ptrs,
322 double_blocks = (1 << (ptrs_bits * 2));
326 if (i_block < direct_blocks) {
327 offsets[n++] = i_block;
328 final = direct_blocks;
329 } else if ((i_block -= direct_blocks) < indirect_blocks) {
330 offsets[n++] = EXT4_IND_BLOCK;
331 offsets[n++] = i_block;
333 } else if ((i_block -= indirect_blocks) < double_blocks) {
334 offsets[n++] = EXT4_DIND_BLOCK;
335 offsets[n++] = i_block >> ptrs_bits;
336 offsets[n++] = i_block & (ptrs - 1);
338 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
339 offsets[n++] = EXT4_TIND_BLOCK;
340 offsets[n++] = i_block >> (ptrs_bits * 2);
341 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
342 offsets[n++] = i_block & (ptrs - 1);
345 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
346 i_block + direct_blocks +
347 indirect_blocks + double_blocks, inode->i_ino);
350 *boundary = final - 1 - (i_block & (ptrs - 1));
354 static int __ext4_check_blockref(const char *function, unsigned int line,
356 __le32 *p, unsigned int max)
358 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
362 while (bref < p+max) {
363 blk = le32_to_cpu(*bref++);
365 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
367 es->s_last_error_block = cpu_to_le64(blk);
368 ext4_error_inode(inode, function, line, blk,
377 #define ext4_check_indirect_blockref(inode, bh) \
378 __ext4_check_blockref(__func__, __LINE__, inode, \
379 (__le32 *)(bh)->b_data, \
380 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
382 #define ext4_check_inode_blockref(inode) \
383 __ext4_check_blockref(__func__, __LINE__, inode, \
384 EXT4_I(inode)->i_data, \
388 * ext4_get_branch - read the chain of indirect blocks leading to data
389 * @inode: inode in question
390 * @depth: depth of the chain (1 - direct pointer, etc.)
391 * @offsets: offsets of pointers in inode/indirect blocks
392 * @chain: place to store the result
393 * @err: here we store the error value
395 * Function fills the array of triples <key, p, bh> and returns %NULL
396 * if everything went OK or the pointer to the last filled triple
397 * (incomplete one) otherwise. Upon the return chain[i].key contains
398 * the number of (i+1)-th block in the chain (as it is stored in memory,
399 * i.e. little-endian 32-bit), chain[i].p contains the address of that
400 * number (it points into struct inode for i==0 and into the bh->b_data
401 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
402 * block for i>0 and NULL for i==0. In other words, it holds the block
403 * numbers of the chain, addresses they were taken from (and where we can
404 * verify that chain did not change) and buffer_heads hosting these
407 * Function stops when it stumbles upon zero pointer (absent block)
408 * (pointer to last triple returned, *@err == 0)
409 * or when it gets an IO error reading an indirect block
410 * (ditto, *@err == -EIO)
411 * or when it reads all @depth-1 indirect blocks successfully and finds
412 * the whole chain, all way to the data (returns %NULL, *err == 0).
414 * Need to be called with
415 * down_read(&EXT4_I(inode)->i_data_sem)
417 static Indirect *ext4_get_branch(struct inode *inode, int depth,
418 ext4_lblk_t *offsets,
419 Indirect chain[4], int *err)
421 struct super_block *sb = inode->i_sb;
423 struct buffer_head *bh;
426 /* i_data is not going away, no lock needed */
427 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
431 bh = sb_getblk(sb, le32_to_cpu(p->key));
435 if (!bh_uptodate_or_lock(bh)) {
436 if (bh_submit_read(bh) < 0) {
440 /* validate block references */
441 if (ext4_check_indirect_blockref(inode, bh)) {
447 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
461 * ext4_find_near - find a place for allocation with sufficient locality
463 * @ind: descriptor of indirect block.
465 * This function returns the preferred place for block allocation.
466 * It is used when heuristic for sequential allocation fails.
468 * + if there is a block to the left of our position - allocate near it.
469 * + if pointer will live in indirect block - allocate near that block.
470 * + if pointer will live in inode - allocate in the same
473 * In the latter case we colour the starting block by the callers PID to
474 * prevent it from clashing with concurrent allocations for a different inode
475 * in the same block group. The PID is used here so that functionally related
476 * files will be close-by on-disk.
478 * Caller must make sure that @ind is valid and will stay that way.
480 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
482 struct ext4_inode_info *ei = EXT4_I(inode);
483 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
485 ext4_fsblk_t bg_start;
486 ext4_fsblk_t last_block;
487 ext4_grpblk_t colour;
488 ext4_group_t block_group;
489 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
491 /* Try to find previous block */
492 for (p = ind->p - 1; p >= start; p--) {
494 return le32_to_cpu(*p);
497 /* No such thing, so let's try location of indirect block */
499 return ind->bh->b_blocknr;
502 * It is going to be referred to from the inode itself? OK, just put it
503 * into the same cylinder group then.
505 block_group = ei->i_block_group;
506 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
507 block_group &= ~(flex_size-1);
508 if (S_ISREG(inode->i_mode))
511 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
512 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
515 * If we are doing delayed allocation, we don't need take
516 * colour into account.
518 if (test_opt(inode->i_sb, DELALLOC))
521 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
522 colour = (current->pid % 16) *
523 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
525 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
526 return bg_start + colour;
530 * ext4_find_goal - find a preferred place for allocation.
532 * @block: block we want
533 * @partial: pointer to the last triple within a chain
535 * Normally this function find the preferred place for block allocation,
537 * Because this is only used for non-extent files, we limit the block nr
540 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
546 * XXX need to get goal block from mballoc's data structures
549 goal = ext4_find_near(inode, partial);
550 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
555 * ext4_blks_to_allocate: Look up the block map and count the number
556 * of direct blocks need to be allocated for the given branch.
558 * @branch: chain of indirect blocks
559 * @k: number of blocks need for indirect blocks
560 * @blks: number of data blocks to be mapped.
561 * @blocks_to_boundary: the offset in the indirect block
563 * return the total number of blocks to be allocate, including the
564 * direct and indirect blocks.
566 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
567 int blocks_to_boundary)
569 unsigned int count = 0;
572 * Simple case, [t,d]Indirect block(s) has not allocated yet
573 * then it's clear blocks on that path have not allocated
576 /* right now we don't handle cross boundary allocation */
577 if (blks < blocks_to_boundary + 1)
580 count += blocks_to_boundary + 1;
585 while (count < blks && count <= blocks_to_boundary &&
586 le32_to_cpu(*(branch[0].p + count)) == 0) {
593 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
594 * @indirect_blks: the number of blocks need to allocate for indirect
597 * @new_blocks: on return it will store the new block numbers for
598 * the indirect blocks(if needed) and the first direct block,
599 * @blks: on return it will store the total number of allocated
602 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
603 ext4_lblk_t iblock, ext4_fsblk_t goal,
604 int indirect_blks, int blks,
605 ext4_fsblk_t new_blocks[4], int *err)
607 struct ext4_allocation_request ar;
609 unsigned long count = 0, blk_allocated = 0;
611 ext4_fsblk_t current_block = 0;
615 * Here we try to allocate the requested multiple blocks at once,
616 * on a best-effort basis.
617 * To build a branch, we should allocate blocks for
618 * the indirect blocks(if not allocated yet), and at least
619 * the first direct block of this branch. That's the
620 * minimum number of blocks need to allocate(required)
622 /* first we try to allocate the indirect blocks */
623 target = indirect_blks;
626 /* allocating blocks for indirect blocks and direct blocks */
627 current_block = ext4_new_meta_blocks(handle, inode,
632 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
633 EXT4_ERROR_INODE(inode,
634 "current_block %llu + count %lu > %d!",
635 current_block, count,
636 EXT4_MAX_BLOCK_FILE_PHYS);
642 /* allocate blocks for indirect blocks */
643 while (index < indirect_blks && count) {
644 new_blocks[index++] = current_block++;
649 * save the new block number
650 * for the first direct block
652 new_blocks[index] = current_block;
653 printk(KERN_INFO "%s returned more blocks than "
654 "requested\n", __func__);
660 target = blks - count ;
661 blk_allocated = count;
664 /* Now allocate data blocks */
665 memset(&ar, 0, sizeof(ar));
670 if (S_ISREG(inode->i_mode))
671 /* enable in-core preallocation only for regular files */
672 ar.flags = EXT4_MB_HINT_DATA;
674 current_block = ext4_mb_new_blocks(handle, &ar, err);
675 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
676 EXT4_ERROR_INODE(inode,
677 "current_block %llu + ar.len %d > %d!",
678 current_block, ar.len,
679 EXT4_MAX_BLOCK_FILE_PHYS);
684 if (*err && (target == blks)) {
686 * if the allocation failed and we didn't allocate
692 if (target == blks) {
694 * save the new block number
695 * for the first direct block
697 new_blocks[index] = current_block;
699 blk_allocated += ar.len;
702 /* total number of blocks allocated for direct blocks */
707 for (i = 0; i < index; i++)
708 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
713 * ext4_alloc_branch - allocate and set up a chain of blocks.
715 * @indirect_blks: number of allocated indirect blocks
716 * @blks: number of allocated direct blocks
717 * @offsets: offsets (in the blocks) to store the pointers to next.
718 * @branch: place to store the chain in.
720 * This function allocates blocks, zeroes out all but the last one,
721 * links them into chain and (if we are synchronous) writes them to disk.
722 * In other words, it prepares a branch that can be spliced onto the
723 * inode. It stores the information about that chain in the branch[], in
724 * the same format as ext4_get_branch() would do. We are calling it after
725 * we had read the existing part of chain and partial points to the last
726 * triple of that (one with zero ->key). Upon the exit we have the same
727 * picture as after the successful ext4_get_block(), except that in one
728 * place chain is disconnected - *branch->p is still zero (we did not
729 * set the last link), but branch->key contains the number that should
730 * be placed into *branch->p to fill that gap.
732 * If allocation fails we free all blocks we've allocated (and forget
733 * their buffer_heads) and return the error value the from failed
734 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
735 * as described above and return 0.
737 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
738 ext4_lblk_t iblock, int indirect_blks,
739 int *blks, ext4_fsblk_t goal,
740 ext4_lblk_t *offsets, Indirect *branch)
742 int blocksize = inode->i_sb->s_blocksize;
745 struct buffer_head *bh;
747 ext4_fsblk_t new_blocks[4];
748 ext4_fsblk_t current_block;
750 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
751 *blks, new_blocks, &err);
755 branch[0].key = cpu_to_le32(new_blocks[0]);
757 * metadata blocks and data blocks are allocated.
759 for (n = 1; n <= indirect_blks; n++) {
761 * Get buffer_head for parent block, zero it out
762 * and set the pointer to new one, then send
765 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
773 BUFFER_TRACE(bh, "call get_create_access");
774 err = ext4_journal_get_create_access(handle, bh);
776 /* Don't brelse(bh) here; it's done in
777 * ext4_journal_forget() below */
782 memset(bh->b_data, 0, blocksize);
783 branch[n].p = (__le32 *) bh->b_data + offsets[n];
784 branch[n].key = cpu_to_le32(new_blocks[n]);
785 *branch[n].p = branch[n].key;
786 if (n == indirect_blks) {
787 current_block = new_blocks[n];
789 * End of chain, update the last new metablock of
790 * the chain to point to the new allocated
791 * data blocks numbers
793 for (i = 1; i < num; i++)
794 *(branch[n].p + i) = cpu_to_le32(++current_block);
796 BUFFER_TRACE(bh, "marking uptodate");
797 set_buffer_uptodate(bh);
800 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
801 err = ext4_handle_dirty_metadata(handle, inode, bh);
808 /* Allocation failed, free what we already allocated */
809 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
810 for (i = 1; i <= n ; i++) {
812 * branch[i].bh is newly allocated, so there is no
813 * need to revoke the block, which is why we don't
814 * need to set EXT4_FREE_BLOCKS_METADATA.
816 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
817 EXT4_FREE_BLOCKS_FORGET);
819 for (i = n+1; i < indirect_blks; i++)
820 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
822 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
828 * ext4_splice_branch - splice the allocated branch onto inode.
830 * @block: (logical) number of block we are adding
831 * @chain: chain of indirect blocks (with a missing link - see
833 * @where: location of missing link
834 * @num: number of indirect blocks we are adding
835 * @blks: number of direct blocks we are adding
837 * This function fills the missing link and does all housekeeping needed in
838 * inode (->i_blocks, etc.). In case of success we end up with the full
839 * chain to new block and return 0.
841 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
842 ext4_lblk_t block, Indirect *where, int num,
847 ext4_fsblk_t current_block;
850 * If we're splicing into a [td]indirect block (as opposed to the
851 * inode) then we need to get write access to the [td]indirect block
855 BUFFER_TRACE(where->bh, "get_write_access");
856 err = ext4_journal_get_write_access(handle, where->bh);
862 *where->p = where->key;
865 * Update the host buffer_head or inode to point to more just allocated
866 * direct blocks blocks
868 if (num == 0 && blks > 1) {
869 current_block = le32_to_cpu(where->key) + 1;
870 for (i = 1; i < blks; i++)
871 *(where->p + i) = cpu_to_le32(current_block++);
874 /* We are done with atomic stuff, now do the rest of housekeeping */
875 /* had we spliced it onto indirect block? */
878 * If we spliced it onto an indirect block, we haven't
879 * altered the inode. Note however that if it is being spliced
880 * onto an indirect block at the very end of the file (the
881 * file is growing) then we *will* alter the inode to reflect
882 * the new i_size. But that is not done here - it is done in
883 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
885 jbd_debug(5, "splicing indirect only\n");
886 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
887 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
892 * OK, we spliced it into the inode itself on a direct block.
894 ext4_mark_inode_dirty(handle, inode);
895 jbd_debug(5, "splicing direct\n");
900 for (i = 1; i <= num; i++) {
902 * branch[i].bh is newly allocated, so there is no
903 * need to revoke the block, which is why we don't
904 * need to set EXT4_FREE_BLOCKS_METADATA.
906 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
907 EXT4_FREE_BLOCKS_FORGET);
909 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
916 * The ext4_ind_map_blocks() function handles non-extents inodes
917 * (i.e., using the traditional indirect/double-indirect i_blocks
918 * scheme) for ext4_map_blocks().
920 * Allocation strategy is simple: if we have to allocate something, we will
921 * have to go the whole way to leaf. So let's do it before attaching anything
922 * to tree, set linkage between the newborn blocks, write them if sync is
923 * required, recheck the path, free and repeat if check fails, otherwise
924 * set the last missing link (that will protect us from any truncate-generated
925 * removals - all blocks on the path are immune now) and possibly force the
926 * write on the parent block.
927 * That has a nice additional property: no special recovery from the failed
928 * allocations is needed - we simply release blocks and do not touch anything
929 * reachable from inode.
931 * `handle' can be NULL if create == 0.
933 * return > 0, # of blocks mapped or allocated.
934 * return = 0, if plain lookup failed.
935 * return < 0, error case.
937 * The ext4_ind_get_blocks() function should be called with
938 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
939 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
940 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
943 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
944 struct ext4_map_blocks *map,
948 ext4_lblk_t offsets[4];
953 int blocks_to_boundary = 0;
956 ext4_fsblk_t first_block = 0;
958 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
959 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
960 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
961 &blocks_to_boundary);
966 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
968 /* Simplest case - block found, no allocation needed */
970 first_block = le32_to_cpu(chain[depth - 1].key);
973 while (count < map->m_len && count <= blocks_to_boundary) {
976 blk = le32_to_cpu(*(chain[depth-1].p + count));
978 if (blk == first_block + count)
986 /* Next simple case - plain lookup or failed read of indirect block */
987 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
991 * Okay, we need to do block allocation.
993 goal = ext4_find_goal(inode, map->m_lblk, partial);
995 /* the number of blocks need to allocate for [d,t]indirect blocks */
996 indirect_blks = (chain + depth) - partial - 1;
999 * Next look up the indirect map to count the totoal number of
1000 * direct blocks to allocate for this branch.
1002 count = ext4_blks_to_allocate(partial, indirect_blks,
1003 map->m_len, blocks_to_boundary);
1005 * Block out ext4_truncate while we alter the tree
1007 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1009 offsets + (partial - chain), partial);
1012 * The ext4_splice_branch call will free and forget any buffers
1013 * on the new chain if there is a failure, but that risks using
1014 * up transaction credits, especially for bitmaps where the
1015 * credits cannot be returned. Can we handle this somehow? We
1016 * may need to return -EAGAIN upwards in the worst case. --sct
1019 err = ext4_splice_branch(handle, inode, map->m_lblk,
1020 partial, indirect_blks, count);
1024 map->m_flags |= EXT4_MAP_NEW;
1026 ext4_update_inode_fsync_trans(handle, inode, 1);
1028 map->m_flags |= EXT4_MAP_MAPPED;
1029 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1031 if (count > blocks_to_boundary)
1032 map->m_flags |= EXT4_MAP_BOUNDARY;
1034 /* Clean up and exit */
1035 partial = chain + depth - 1; /* the whole chain */
1037 while (partial > chain) {
1038 BUFFER_TRACE(partial->bh, "call brelse");
1039 brelse(partial->bh);
1047 qsize_t *ext4_get_reserved_space(struct inode *inode)
1049 return &EXT4_I(inode)->i_reserved_quota;
1054 * Calculate the number of metadata blocks need to reserve
1055 * to allocate a new block at @lblocks for non extent file based file
1057 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1060 struct ext4_inode_info *ei = EXT4_I(inode);
1061 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1064 if (lblock < EXT4_NDIR_BLOCKS)
1067 lblock -= EXT4_NDIR_BLOCKS;
1069 if (ei->i_da_metadata_calc_len &&
1070 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1071 ei->i_da_metadata_calc_len++;
1074 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1075 ei->i_da_metadata_calc_len = 1;
1076 blk_bits = order_base_2(lblock);
1077 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1081 * Calculate the number of metadata blocks need to reserve
1082 * to allocate a block located at @lblock
1084 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1086 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1087 return ext4_ext_calc_metadata_amount(inode, lblock);
1089 return ext4_indirect_calc_metadata_amount(inode, lblock);
1093 * Called with i_data_sem down, which is important since we can call
1094 * ext4_discard_preallocations() from here.
1096 void ext4_da_update_reserve_space(struct inode *inode,
1097 int used, int quota_claim)
1099 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1100 struct ext4_inode_info *ei = EXT4_I(inode);
1102 spin_lock(&ei->i_block_reservation_lock);
1103 trace_ext4_da_update_reserve_space(inode, used);
1104 if (unlikely(used > ei->i_reserved_data_blocks)) {
1105 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1106 "with only %d reserved data blocks\n",
1107 __func__, inode->i_ino, used,
1108 ei->i_reserved_data_blocks);
1110 used = ei->i_reserved_data_blocks;
1113 /* Update per-inode reservations */
1114 ei->i_reserved_data_blocks -= used;
1115 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1116 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1117 used + ei->i_allocated_meta_blocks);
1118 ei->i_allocated_meta_blocks = 0;
1120 if (ei->i_reserved_data_blocks == 0) {
1122 * We can release all of the reserved metadata blocks
1123 * only when we have written all of the delayed
1124 * allocation blocks.
1126 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1127 ei->i_reserved_meta_blocks);
1128 ei->i_reserved_meta_blocks = 0;
1129 ei->i_da_metadata_calc_len = 0;
1131 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1133 /* Update quota subsystem for data blocks */
1135 dquot_claim_block(inode, used);
1138 * We did fallocate with an offset that is already delayed
1139 * allocated. So on delayed allocated writeback we should
1140 * not re-claim the quota for fallocated blocks.
1142 dquot_release_reservation_block(inode, used);
1146 * If we have done all the pending block allocations and if
1147 * there aren't any writers on the inode, we can discard the
1148 * inode's preallocations.
1150 if ((ei->i_reserved_data_blocks == 0) &&
1151 (atomic_read(&inode->i_writecount) == 0))
1152 ext4_discard_preallocations(inode);
1155 static int __check_block_validity(struct inode *inode, const char *func,
1157 struct ext4_map_blocks *map)
1159 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1161 ext4_error_inode(inode, func, line, map->m_pblk,
1162 "lblock %lu mapped to illegal pblock "
1163 "(length %d)", (unsigned long) map->m_lblk,
1170 #define check_block_validity(inode, map) \
1171 __check_block_validity((inode), __func__, __LINE__, (map))
1174 * Return the number of contiguous dirty pages in a given inode
1175 * starting at page frame idx.
1177 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1178 unsigned int max_pages)
1180 struct address_space *mapping = inode->i_mapping;
1182 struct pagevec pvec;
1184 int i, nr_pages, done = 0;
1188 pagevec_init(&pvec, 0);
1191 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1192 PAGECACHE_TAG_DIRTY,
1193 (pgoff_t)PAGEVEC_SIZE);
1196 for (i = 0; i < nr_pages; i++) {
1197 struct page *page = pvec.pages[i];
1198 struct buffer_head *bh, *head;
1201 if (unlikely(page->mapping != mapping) ||
1203 PageWriteback(page) ||
1204 page->index != idx) {
1209 if (page_has_buffers(page)) {
1210 bh = head = page_buffers(page);
1212 if (!buffer_delay(bh) &&
1213 !buffer_unwritten(bh))
1215 bh = bh->b_this_page;
1216 } while (!done && (bh != head));
1223 if (num >= max_pages) {
1228 pagevec_release(&pvec);
1234 * The ext4_map_blocks() function tries to look up the requested blocks,
1235 * and returns if the blocks are already mapped.
1237 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1238 * and store the allocated blocks in the result buffer head and mark it
1241 * If file type is extents based, it will call ext4_ext_map_blocks(),
1242 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1245 * On success, it returns the number of blocks being mapped or allocate.
1246 * if create==0 and the blocks are pre-allocated and uninitialized block,
1247 * the result buffer head is unmapped. If the create ==1, it will make sure
1248 * the buffer head is mapped.
1250 * It returns 0 if plain look up failed (blocks have not been allocated), in
1251 * that casem, buffer head is unmapped
1253 * It returns the error in case of allocation failure.
1255 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1256 struct ext4_map_blocks *map, int flags)
1261 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1262 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1263 (unsigned long) map->m_lblk);
1265 * Try to see if we can get the block without requesting a new
1266 * file system block.
1268 down_read((&EXT4_I(inode)->i_data_sem));
1269 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1270 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1272 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1274 up_read((&EXT4_I(inode)->i_data_sem));
1276 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1277 int ret = check_block_validity(inode, map);
1282 /* If it is only a block(s) look up */
1283 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1287 * Returns if the blocks have already allocated
1289 * Note that if blocks have been preallocated
1290 * ext4_ext_get_block() returns th create = 0
1291 * with buffer head unmapped.
1293 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1297 * When we call get_blocks without the create flag, the
1298 * BH_Unwritten flag could have gotten set if the blocks
1299 * requested were part of a uninitialized extent. We need to
1300 * clear this flag now that we are committed to convert all or
1301 * part of the uninitialized extent to be an initialized
1302 * extent. This is because we need to avoid the combination
1303 * of BH_Unwritten and BH_Mapped flags being simultaneously
1304 * set on the buffer_head.
1306 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1309 * New blocks allocate and/or writing to uninitialized extent
1310 * will possibly result in updating i_data, so we take
1311 * the write lock of i_data_sem, and call get_blocks()
1312 * with create == 1 flag.
1314 down_write((&EXT4_I(inode)->i_data_sem));
1317 * if the caller is from delayed allocation writeout path
1318 * we have already reserved fs blocks for allocation
1319 * let the underlying get_block() function know to
1320 * avoid double accounting
1322 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1323 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1325 * We need to check for EXT4 here because migrate
1326 * could have changed the inode type in between
1328 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1329 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1331 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1333 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1335 * We allocated new blocks which will result in
1336 * i_data's format changing. Force the migrate
1337 * to fail by clearing migrate flags
1339 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1343 * Update reserved blocks/metadata blocks after successful
1344 * block allocation which had been deferred till now. We don't
1345 * support fallocate for non extent files. So we can update
1346 * reserve space here.
1349 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1350 ext4_da_update_reserve_space(inode, retval, 1);
1352 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1353 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1355 up_write((&EXT4_I(inode)->i_data_sem));
1356 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1357 int ret = check_block_validity(inode, map);
1364 /* Maximum number of blocks we map for direct IO at once. */
1365 #define DIO_MAX_BLOCKS 4096
1367 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1368 struct buffer_head *bh, int flags)
1370 handle_t *handle = ext4_journal_current_handle();
1371 struct ext4_map_blocks map;
1372 int ret = 0, started = 0;
1375 map.m_lblk = iblock;
1376 map.m_len = bh->b_size >> inode->i_blkbits;
1378 if (flags && !handle) {
1379 /* Direct IO write... */
1380 if (map.m_len > DIO_MAX_BLOCKS)
1381 map.m_len = DIO_MAX_BLOCKS;
1382 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1383 handle = ext4_journal_start(inode, dio_credits);
1384 if (IS_ERR(handle)) {
1385 ret = PTR_ERR(handle);
1391 ret = ext4_map_blocks(handle, inode, &map, flags);
1393 map_bh(bh, inode->i_sb, map.m_pblk);
1394 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1395 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1399 ext4_journal_stop(handle);
1403 int ext4_get_block(struct inode *inode, sector_t iblock,
1404 struct buffer_head *bh, int create)
1406 return _ext4_get_block(inode, iblock, bh,
1407 create ? EXT4_GET_BLOCKS_CREATE : 0);
1411 * `handle' can be NULL if create is zero
1413 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1414 ext4_lblk_t block, int create, int *errp)
1416 struct ext4_map_blocks map;
1417 struct buffer_head *bh;
1420 J_ASSERT(handle != NULL || create == 0);
1424 err = ext4_map_blocks(handle, inode, &map,
1425 create ? EXT4_GET_BLOCKS_CREATE : 0);
1433 bh = sb_getblk(inode->i_sb, map.m_pblk);
1438 if (map.m_flags & EXT4_MAP_NEW) {
1439 J_ASSERT(create != 0);
1440 J_ASSERT(handle != NULL);
1443 * Now that we do not always journal data, we should
1444 * keep in mind whether this should always journal the
1445 * new buffer as metadata. For now, regular file
1446 * writes use ext4_get_block instead, so it's not a
1450 BUFFER_TRACE(bh, "call get_create_access");
1451 fatal = ext4_journal_get_create_access(handle, bh);
1452 if (!fatal && !buffer_uptodate(bh)) {
1453 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1454 set_buffer_uptodate(bh);
1457 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1458 err = ext4_handle_dirty_metadata(handle, inode, bh);
1462 BUFFER_TRACE(bh, "not a new buffer");
1472 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1473 ext4_lblk_t block, int create, int *err)
1475 struct buffer_head *bh;
1477 bh = ext4_getblk(handle, inode, block, create, err);
1480 if (buffer_uptodate(bh))
1482 ll_rw_block(READ_META, 1, &bh);
1484 if (buffer_uptodate(bh))
1491 static int walk_page_buffers(handle_t *handle,
1492 struct buffer_head *head,
1496 int (*fn)(handle_t *handle,
1497 struct buffer_head *bh))
1499 struct buffer_head *bh;
1500 unsigned block_start, block_end;
1501 unsigned blocksize = head->b_size;
1503 struct buffer_head *next;
1505 for (bh = head, block_start = 0;
1506 ret == 0 && (bh != head || !block_start);
1507 block_start = block_end, bh = next) {
1508 next = bh->b_this_page;
1509 block_end = block_start + blocksize;
1510 if (block_end <= from || block_start >= to) {
1511 if (partial && !buffer_uptodate(bh))
1515 err = (*fn)(handle, bh);
1523 * To preserve ordering, it is essential that the hole instantiation and
1524 * the data write be encapsulated in a single transaction. We cannot
1525 * close off a transaction and start a new one between the ext4_get_block()
1526 * and the commit_write(). So doing the jbd2_journal_start at the start of
1527 * prepare_write() is the right place.
1529 * Also, this function can nest inside ext4_writepage() ->
1530 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1531 * has generated enough buffer credits to do the whole page. So we won't
1532 * block on the journal in that case, which is good, because the caller may
1535 * By accident, ext4 can be reentered when a transaction is open via
1536 * quota file writes. If we were to commit the transaction while thus
1537 * reentered, there can be a deadlock - we would be holding a quota
1538 * lock, and the commit would never complete if another thread had a
1539 * transaction open and was blocking on the quota lock - a ranking
1542 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1543 * will _not_ run commit under these circumstances because handle->h_ref
1544 * is elevated. We'll still have enough credits for the tiny quotafile
1547 static int do_journal_get_write_access(handle_t *handle,
1548 struct buffer_head *bh)
1550 int dirty = buffer_dirty(bh);
1553 if (!buffer_mapped(bh) || buffer_freed(bh))
1556 * __block_write_begin() could have dirtied some buffers. Clean
1557 * the dirty bit as jbd2_journal_get_write_access() could complain
1558 * otherwise about fs integrity issues. Setting of the dirty bit
1559 * by __block_write_begin() isn't a real problem here as we clear
1560 * the bit before releasing a page lock and thus writeback cannot
1561 * ever write the buffer.
1564 clear_buffer_dirty(bh);
1565 ret = ext4_journal_get_write_access(handle, bh);
1567 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1572 * Truncate blocks that were not used by write. We have to truncate the
1573 * pagecache as well so that corresponding buffers get properly unmapped.
1575 static void ext4_truncate_failed_write(struct inode *inode)
1577 truncate_inode_pages(inode->i_mapping, inode->i_size);
1578 ext4_truncate(inode);
1581 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1582 struct buffer_head *bh_result, int create);
1583 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1584 loff_t pos, unsigned len, unsigned flags,
1585 struct page **pagep, void **fsdata)
1587 struct inode *inode = mapping->host;
1588 int ret, needed_blocks;
1595 trace_ext4_write_begin(inode, pos, len, flags);
1597 * Reserve one block more for addition to orphan list in case
1598 * we allocate blocks but write fails for some reason
1600 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1601 index = pos >> PAGE_CACHE_SHIFT;
1602 from = pos & (PAGE_CACHE_SIZE - 1);
1606 handle = ext4_journal_start(inode, needed_blocks);
1607 if (IS_ERR(handle)) {
1608 ret = PTR_ERR(handle);
1612 /* We cannot recurse into the filesystem as the transaction is already
1614 flags |= AOP_FLAG_NOFS;
1616 page = grab_cache_page_write_begin(mapping, index, flags);
1618 ext4_journal_stop(handle);
1624 if (ext4_should_dioread_nolock(inode))
1625 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1627 ret = __block_write_begin(page, pos, len, ext4_get_block);
1629 if (!ret && ext4_should_journal_data(inode)) {
1630 ret = walk_page_buffers(handle, page_buffers(page),
1631 from, to, NULL, do_journal_get_write_access);
1636 page_cache_release(page);
1638 * __block_write_begin may have instantiated a few blocks
1639 * outside i_size. Trim these off again. Don't need
1640 * i_size_read because we hold i_mutex.
1642 * Add inode to orphan list in case we crash before
1645 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1646 ext4_orphan_add(handle, inode);
1648 ext4_journal_stop(handle);
1649 if (pos + len > inode->i_size) {
1650 ext4_truncate_failed_write(inode);
1652 * If truncate failed early the inode might
1653 * still be on the orphan list; we need to
1654 * make sure the inode is removed from the
1655 * orphan list in that case.
1658 ext4_orphan_del(NULL, inode);
1662 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1668 /* For write_end() in data=journal mode */
1669 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1671 if (!buffer_mapped(bh) || buffer_freed(bh))
1673 set_buffer_uptodate(bh);
1674 return ext4_handle_dirty_metadata(handle, NULL, bh);
1677 static int ext4_generic_write_end(struct file *file,
1678 struct address_space *mapping,
1679 loff_t pos, unsigned len, unsigned copied,
1680 struct page *page, void *fsdata)
1682 int i_size_changed = 0;
1683 struct inode *inode = mapping->host;
1684 handle_t *handle = ext4_journal_current_handle();
1686 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1689 * No need to use i_size_read() here, the i_size
1690 * cannot change under us because we hold i_mutex.
1692 * But it's important to update i_size while still holding page lock:
1693 * page writeout could otherwise come in and zero beyond i_size.
1695 if (pos + copied > inode->i_size) {
1696 i_size_write(inode, pos + copied);
1700 if (pos + copied > EXT4_I(inode)->i_disksize) {
1701 /* We need to mark inode dirty even if
1702 * new_i_size is less that inode->i_size
1703 * bu greater than i_disksize.(hint delalloc)
1705 ext4_update_i_disksize(inode, (pos + copied));
1709 page_cache_release(page);
1712 * Don't mark the inode dirty under page lock. First, it unnecessarily
1713 * makes the holding time of page lock longer. Second, it forces lock
1714 * ordering of page lock and transaction start for journaling
1718 ext4_mark_inode_dirty(handle, inode);
1724 * We need to pick up the new inode size which generic_commit_write gave us
1725 * `file' can be NULL - eg, when called from page_symlink().
1727 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1728 * buffers are managed internally.
1730 static int ext4_ordered_write_end(struct file *file,
1731 struct address_space *mapping,
1732 loff_t pos, unsigned len, unsigned copied,
1733 struct page *page, void *fsdata)
1735 handle_t *handle = ext4_journal_current_handle();
1736 struct inode *inode = mapping->host;
1739 trace_ext4_ordered_write_end(inode, pos, len, copied);
1740 ret = ext4_jbd2_file_inode(handle, inode);
1743 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1746 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1747 /* if we have allocated more blocks and copied
1748 * less. We will have blocks allocated outside
1749 * inode->i_size. So truncate them
1751 ext4_orphan_add(handle, inode);
1755 ret2 = ext4_journal_stop(handle);
1759 if (pos + len > inode->i_size) {
1760 ext4_truncate_failed_write(inode);
1762 * If truncate failed early the inode might still be
1763 * on the orphan list; we need to make sure the inode
1764 * is removed from the orphan list in that case.
1767 ext4_orphan_del(NULL, inode);
1771 return ret ? ret : copied;
1774 static int ext4_writeback_write_end(struct file *file,
1775 struct address_space *mapping,
1776 loff_t pos, unsigned len, unsigned copied,
1777 struct page *page, void *fsdata)
1779 handle_t *handle = ext4_journal_current_handle();
1780 struct inode *inode = mapping->host;
1783 trace_ext4_writeback_write_end(inode, pos, len, copied);
1784 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1787 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1788 /* if we have allocated more blocks and copied
1789 * less. We will have blocks allocated outside
1790 * inode->i_size. So truncate them
1792 ext4_orphan_add(handle, inode);
1797 ret2 = ext4_journal_stop(handle);
1801 if (pos + len > inode->i_size) {
1802 ext4_truncate_failed_write(inode);
1804 * If truncate failed early the inode might still be
1805 * on the orphan list; we need to make sure the inode
1806 * is removed from the orphan list in that case.
1809 ext4_orphan_del(NULL, inode);
1812 return ret ? ret : copied;
1815 static int ext4_journalled_write_end(struct file *file,
1816 struct address_space *mapping,
1817 loff_t pos, unsigned len, unsigned copied,
1818 struct page *page, void *fsdata)
1820 handle_t *handle = ext4_journal_current_handle();
1821 struct inode *inode = mapping->host;
1827 trace_ext4_journalled_write_end(inode, pos, len, copied);
1828 from = pos & (PAGE_CACHE_SIZE - 1);
1832 if (!PageUptodate(page))
1834 page_zero_new_buffers(page, from+copied, to);
1837 ret = walk_page_buffers(handle, page_buffers(page), from,
1838 to, &partial, write_end_fn);
1840 SetPageUptodate(page);
1841 new_i_size = pos + copied;
1842 if (new_i_size > inode->i_size)
1843 i_size_write(inode, pos+copied);
1844 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1845 if (new_i_size > EXT4_I(inode)->i_disksize) {
1846 ext4_update_i_disksize(inode, new_i_size);
1847 ret2 = ext4_mark_inode_dirty(handle, inode);
1853 page_cache_release(page);
1854 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1855 /* if we have allocated more blocks and copied
1856 * less. We will have blocks allocated outside
1857 * inode->i_size. So truncate them
1859 ext4_orphan_add(handle, inode);
1861 ret2 = ext4_journal_stop(handle);
1864 if (pos + len > inode->i_size) {
1865 ext4_truncate_failed_write(inode);
1867 * If truncate failed early the inode might still be
1868 * on the orphan list; we need to make sure the inode
1869 * is removed from the orphan list in that case.
1872 ext4_orphan_del(NULL, inode);
1875 return ret ? ret : copied;
1879 * Reserve a single block located at lblock
1881 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1884 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1885 struct ext4_inode_info *ei = EXT4_I(inode);
1886 unsigned long md_needed;
1890 * recalculate the amount of metadata blocks to reserve
1891 * in order to allocate nrblocks
1892 * worse case is one extent per block
1895 spin_lock(&ei->i_block_reservation_lock);
1896 md_needed = ext4_calc_metadata_amount(inode, lblock);
1897 trace_ext4_da_reserve_space(inode, md_needed);
1898 spin_unlock(&ei->i_block_reservation_lock);
1901 * We will charge metadata quota at writeout time; this saves
1902 * us from metadata over-estimation, though we may go over by
1903 * a small amount in the end. Here we just reserve for data.
1905 ret = dquot_reserve_block(inode, 1);
1909 * We do still charge estimated metadata to the sb though;
1910 * we cannot afford to run out of free blocks.
1912 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1913 dquot_release_reservation_block(inode, 1);
1914 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1920 spin_lock(&ei->i_block_reservation_lock);
1921 ei->i_reserved_data_blocks++;
1922 ei->i_reserved_meta_blocks += md_needed;
1923 spin_unlock(&ei->i_block_reservation_lock);
1925 return 0; /* success */
1928 static void ext4_da_release_space(struct inode *inode, int to_free)
1930 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1931 struct ext4_inode_info *ei = EXT4_I(inode);
1934 return; /* Nothing to release, exit */
1936 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1938 trace_ext4_da_release_space(inode, to_free);
1939 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1941 * if there aren't enough reserved blocks, then the
1942 * counter is messed up somewhere. Since this
1943 * function is called from invalidate page, it's
1944 * harmless to return without any action.
1946 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1947 "ino %lu, to_free %d with only %d reserved "
1948 "data blocks\n", inode->i_ino, to_free,
1949 ei->i_reserved_data_blocks);
1951 to_free = ei->i_reserved_data_blocks;
1953 ei->i_reserved_data_blocks -= to_free;
1955 if (ei->i_reserved_data_blocks == 0) {
1957 * We can release all of the reserved metadata blocks
1958 * only when we have written all of the delayed
1959 * allocation blocks.
1961 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1962 ei->i_reserved_meta_blocks);
1963 ei->i_reserved_meta_blocks = 0;
1964 ei->i_da_metadata_calc_len = 0;
1967 /* update fs dirty data blocks counter */
1968 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1970 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1972 dquot_release_reservation_block(inode, to_free);
1975 static void ext4_da_page_release_reservation(struct page *page,
1976 unsigned long offset)
1979 struct buffer_head *head, *bh;
1980 unsigned int curr_off = 0;
1982 head = page_buffers(page);
1985 unsigned int next_off = curr_off + bh->b_size;
1987 if ((offset <= curr_off) && (buffer_delay(bh))) {
1989 clear_buffer_delay(bh);
1991 curr_off = next_off;
1992 } while ((bh = bh->b_this_page) != head);
1993 ext4_da_release_space(page->mapping->host, to_release);
1997 * Delayed allocation stuff
2001 * mpage_da_submit_io - walks through extent of pages and try to write
2002 * them with writepage() call back
2004 * @mpd->inode: inode
2005 * @mpd->first_page: first page of the extent
2006 * @mpd->next_page: page after the last page of the extent
2008 * By the time mpage_da_submit_io() is called we expect all blocks
2009 * to be allocated. this may be wrong if allocation failed.
2011 * As pages are already locked by write_cache_pages(), we can't use it
2013 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2014 struct ext4_map_blocks *map)
2016 struct pagevec pvec;
2017 unsigned long index, end;
2018 int ret = 0, err, nr_pages, i;
2019 struct inode *inode = mpd->inode;
2020 struct address_space *mapping = inode->i_mapping;
2021 loff_t size = i_size_read(inode);
2022 unsigned int len, block_start;
2023 struct buffer_head *bh, *page_bufs = NULL;
2024 int journal_data = ext4_should_journal_data(inode);
2025 sector_t pblock = 0, cur_logical = 0;
2026 struct ext4_io_submit io_submit;
2028 BUG_ON(mpd->next_page <= mpd->first_page);
2029 memset(&io_submit, 0, sizeof(io_submit));
2031 * We need to start from the first_page to the next_page - 1
2032 * to make sure we also write the mapped dirty buffer_heads.
2033 * If we look at mpd->b_blocknr we would only be looking
2034 * at the currently mapped buffer_heads.
2036 index = mpd->first_page;
2037 end = mpd->next_page - 1;
2039 pagevec_init(&pvec, 0);
2040 while (index <= end) {
2041 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2044 for (i = 0; i < nr_pages; i++) {
2045 int commit_write = 0, redirty_page = 0;
2046 struct page *page = pvec.pages[i];
2048 index = page->index;
2052 if (index == size >> PAGE_CACHE_SHIFT)
2053 len = size & ~PAGE_CACHE_MASK;
2055 len = PAGE_CACHE_SIZE;
2057 cur_logical = index << (PAGE_CACHE_SHIFT -
2059 pblock = map->m_pblk + (cur_logical -
2064 BUG_ON(!PageLocked(page));
2065 BUG_ON(PageWriteback(page));
2068 * If the page does not have buffers (for
2069 * whatever reason), try to create them using
2070 * __block_write_begin. If this fails,
2071 * redirty the page and move on.
2073 if (!page_has_buffers(page)) {
2074 if (__block_write_begin(page, 0, len,
2075 noalloc_get_block_write)) {
2077 redirty_page_for_writepage(mpd->wbc,
2085 bh = page_bufs = page_buffers(page);
2090 if (map && (cur_logical >= map->m_lblk) &&
2091 (cur_logical <= (map->m_lblk +
2092 (map->m_len - 1)))) {
2093 if (buffer_delay(bh)) {
2094 clear_buffer_delay(bh);
2095 bh->b_blocknr = pblock;
2097 if (buffer_unwritten(bh) ||
2099 BUG_ON(bh->b_blocknr != pblock);
2100 if (map->m_flags & EXT4_MAP_UNINIT)
2101 set_buffer_uninit(bh);
2102 clear_buffer_unwritten(bh);
2105 /* redirty page if block allocation undone */
2106 if (buffer_delay(bh) || buffer_unwritten(bh))
2108 bh = bh->b_this_page;
2109 block_start += bh->b_size;
2112 } while (bh != page_bufs);
2118 /* mark the buffer_heads as dirty & uptodate */
2119 block_commit_write(page, 0, len);
2122 * Delalloc doesn't support data journalling,
2123 * but eventually maybe we'll lift this
2126 if (unlikely(journal_data && PageChecked(page)))
2127 err = __ext4_journalled_writepage(page, len);
2129 err = ext4_bio_write_page(&io_submit, page,
2133 mpd->pages_written++;
2135 * In error case, we have to continue because
2136 * remaining pages are still locked
2141 pagevec_release(&pvec);
2143 ext4_io_submit(&io_submit);
2147 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2148 sector_t logical, long blk_cnt)
2152 struct pagevec pvec;
2153 struct inode *inode = mpd->inode;
2154 struct address_space *mapping = inode->i_mapping;
2156 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2157 end = (logical + blk_cnt - 1) >>
2158 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2159 while (index <= end) {
2160 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2163 for (i = 0; i < nr_pages; i++) {
2164 struct page *page = pvec.pages[i];
2165 if (page->index > end)
2167 BUG_ON(!PageLocked(page));
2168 BUG_ON(PageWriteback(page));
2169 block_invalidatepage(page, 0);
2170 ClearPageUptodate(page);
2173 index = pvec.pages[nr_pages - 1]->index + 1;
2174 pagevec_release(&pvec);
2179 static void ext4_print_free_blocks(struct inode *inode)
2181 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2182 printk(KERN_CRIT "Total free blocks count %lld\n",
2183 ext4_count_free_blocks(inode->i_sb));
2184 printk(KERN_CRIT "Free/Dirty block details\n");
2185 printk(KERN_CRIT "free_blocks=%lld\n",
2186 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2187 printk(KERN_CRIT "dirty_blocks=%lld\n",
2188 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2189 printk(KERN_CRIT "Block reservation details\n");
2190 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2191 EXT4_I(inode)->i_reserved_data_blocks);
2192 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2193 EXT4_I(inode)->i_reserved_meta_blocks);
2198 * mpage_da_map_and_submit - go through given space, map them
2199 * if necessary, and then submit them for I/O
2201 * @mpd - bh describing space
2203 * The function skips space we know is already mapped to disk blocks.
2206 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2208 int err, blks, get_blocks_flags;
2209 struct ext4_map_blocks map, *mapp = NULL;
2210 sector_t next = mpd->b_blocknr;
2211 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2212 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2213 handle_t *handle = NULL;
2216 * If the blocks are mapped already, or we couldn't accumulate
2217 * any blocks, then proceed immediately to the submission stage.
2219 if ((mpd->b_size == 0) ||
2220 ((mpd->b_state & (1 << BH_Mapped)) &&
2221 !(mpd->b_state & (1 << BH_Delay)) &&
2222 !(mpd->b_state & (1 << BH_Unwritten))))
2225 handle = ext4_journal_current_handle();
2229 * Call ext4_map_blocks() to allocate any delayed allocation
2230 * blocks, or to convert an uninitialized extent to be
2231 * initialized (in the case where we have written into
2232 * one or more preallocated blocks).
2234 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2235 * indicate that we are on the delayed allocation path. This
2236 * affects functions in many different parts of the allocation
2237 * call path. This flag exists primarily because we don't
2238 * want to change *many* call functions, so ext4_map_blocks()
2239 * will set the magic i_delalloc_reserved_flag once the
2240 * inode's allocation semaphore is taken.
2242 * If the blocks in questions were delalloc blocks, set
2243 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2244 * variables are updated after the blocks have been allocated.
2247 map.m_len = max_blocks;
2248 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2249 if (ext4_should_dioread_nolock(mpd->inode))
2250 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2251 if (mpd->b_state & (1 << BH_Delay))
2252 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2254 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2256 struct super_block *sb = mpd->inode->i_sb;
2260 * If get block returns EAGAIN or ENOSPC and there
2261 * appears to be free blocks we will call
2262 * ext4_writepage() for all of the pages which will
2263 * just redirty the pages.
2268 if (err == -ENOSPC &&
2269 ext4_count_free_blocks(sb)) {
2275 * get block failure will cause us to loop in
2276 * writepages, because a_ops->writepage won't be able
2277 * to make progress. The page will be redirtied by
2278 * writepage and writepages will again try to write
2281 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2282 ext4_msg(sb, KERN_CRIT,
2283 "delayed block allocation failed for inode %lu "
2284 "at logical offset %llu with max blocks %zd "
2285 "with error %d", mpd->inode->i_ino,
2286 (unsigned long long) next,
2287 mpd->b_size >> mpd->inode->i_blkbits, err);
2288 ext4_msg(sb, KERN_CRIT,
2289 "This should not happen!! Data will be lost\n");
2291 ext4_print_free_blocks(mpd->inode);
2293 /* invalidate all the pages */
2294 ext4_da_block_invalidatepages(mpd, next,
2295 mpd->b_size >> mpd->inode->i_blkbits);
2301 if (map.m_flags & EXT4_MAP_NEW) {
2302 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2305 for (i = 0; i < map.m_len; i++)
2306 unmap_underlying_metadata(bdev, map.m_pblk + i);
2309 if (ext4_should_order_data(mpd->inode)) {
2310 err = ext4_jbd2_file_inode(handle, mpd->inode);
2312 /* This only happens if the journal is aborted */
2317 * Update on-disk size along with block allocation.
2319 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2320 if (disksize > i_size_read(mpd->inode))
2321 disksize = i_size_read(mpd->inode);
2322 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2323 ext4_update_i_disksize(mpd->inode, disksize);
2324 err = ext4_mark_inode_dirty(handle, mpd->inode);
2326 ext4_error(mpd->inode->i_sb,
2327 "Failed to mark inode %lu dirty",
2332 mpage_da_submit_io(mpd, mapp);
2336 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2337 (1 << BH_Delay) | (1 << BH_Unwritten))
2340 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2342 * @mpd->lbh - extent of blocks
2343 * @logical - logical number of the block in the file
2344 * @bh - bh of the block (used to access block's state)
2346 * the function is used to collect contig. blocks in same state
2348 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2349 sector_t logical, size_t b_size,
2350 unsigned long b_state)
2353 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2356 * XXX Don't go larger than mballoc is willing to allocate
2357 * This is a stopgap solution. We eventually need to fold
2358 * mpage_da_submit_io() into this function and then call
2359 * ext4_map_blocks() multiple times in a loop
2361 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2364 /* check if thereserved journal credits might overflow */
2365 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2366 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2368 * With non-extent format we are limited by the journal
2369 * credit available. Total credit needed to insert
2370 * nrblocks contiguous blocks is dependent on the
2371 * nrblocks. So limit nrblocks.
2374 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2375 EXT4_MAX_TRANS_DATA) {
2377 * Adding the new buffer_head would make it cross the
2378 * allowed limit for which we have journal credit
2379 * reserved. So limit the new bh->b_size
2381 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2382 mpd->inode->i_blkbits;
2383 /* we will do mpage_da_submit_io in the next loop */
2387 * First block in the extent
2389 if (mpd->b_size == 0) {
2390 mpd->b_blocknr = logical;
2391 mpd->b_size = b_size;
2392 mpd->b_state = b_state & BH_FLAGS;
2396 next = mpd->b_blocknr + nrblocks;
2398 * Can we merge the block to our big extent?
2400 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2401 mpd->b_size += b_size;
2407 * We couldn't merge the block to our extent, so we
2408 * need to flush current extent and start new one
2410 mpage_da_map_and_submit(mpd);
2414 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2416 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2420 * __mpage_da_writepage - finds extent of pages and blocks
2422 * @page: page to consider
2423 * @wbc: not used, we just follow rules
2426 * The function finds extents of pages and scan them for all blocks.
2428 static int __mpage_da_writepage(struct page *page,
2429 struct writeback_control *wbc,
2430 struct mpage_da_data *mpd)
2432 struct inode *inode = mpd->inode;
2433 struct buffer_head *bh, *head;
2437 * Can we merge this page to current extent?
2439 if (mpd->next_page != page->index) {
2441 * Nope, we can't. So, we map non-allocated blocks
2442 * and start IO on them
2444 if (mpd->next_page != mpd->first_page) {
2445 mpage_da_map_and_submit(mpd);
2447 * skip rest of the page in the page_vec
2449 redirty_page_for_writepage(wbc, page);
2451 return MPAGE_DA_EXTENT_TAIL;
2455 * Start next extent of pages ...
2457 mpd->first_page = page->index;
2467 mpd->next_page = page->index + 1;
2468 logical = (sector_t) page->index <<
2469 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2471 if (!page_has_buffers(page)) {
2472 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2473 (1 << BH_Dirty) | (1 << BH_Uptodate));
2475 return MPAGE_DA_EXTENT_TAIL;
2478 * Page with regular buffer heads, just add all dirty ones
2480 head = page_buffers(page);
2483 BUG_ON(buffer_locked(bh));
2485 * We need to try to allocate
2486 * unmapped blocks in the same page.
2487 * Otherwise we won't make progress
2488 * with the page in ext4_writepage
2490 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2491 mpage_add_bh_to_extent(mpd, logical,
2495 return MPAGE_DA_EXTENT_TAIL;
2496 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2498 * mapped dirty buffer. We need to update
2499 * the b_state because we look at
2500 * b_state in mpage_da_map_blocks. We don't
2501 * update b_size because if we find an
2502 * unmapped buffer_head later we need to
2503 * use the b_state flag of that buffer_head.
2505 if (mpd->b_size == 0)
2506 mpd->b_state = bh->b_state & BH_FLAGS;
2509 } while ((bh = bh->b_this_page) != head);
2516 * This is a special get_blocks_t callback which is used by
2517 * ext4_da_write_begin(). It will either return mapped block or
2518 * reserve space for a single block.
2520 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2521 * We also have b_blocknr = -1 and b_bdev initialized properly
2523 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2524 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2525 * initialized properly.
2527 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2528 struct buffer_head *bh, int create)
2530 struct ext4_map_blocks map;
2532 sector_t invalid_block = ~((sector_t) 0xffff);
2534 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2537 BUG_ON(create == 0);
2538 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2540 map.m_lblk = iblock;
2544 * first, we need to know whether the block is allocated already
2545 * preallocated blocks are unmapped but should treated
2546 * the same as allocated blocks.
2548 ret = ext4_map_blocks(NULL, inode, &map, 0);
2552 if (buffer_delay(bh))
2553 return 0; /* Not sure this could or should happen */
2555 * XXX: __block_write_begin() unmaps passed block, is it OK?
2557 ret = ext4_da_reserve_space(inode, iblock);
2559 /* not enough space to reserve */
2562 map_bh(bh, inode->i_sb, invalid_block);
2564 set_buffer_delay(bh);
2568 map_bh(bh, inode->i_sb, map.m_pblk);
2569 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2571 if (buffer_unwritten(bh)) {
2572 /* A delayed write to unwritten bh should be marked
2573 * new and mapped. Mapped ensures that we don't do
2574 * get_block multiple times when we write to the same
2575 * offset and new ensures that we do proper zero out
2576 * for partial write.
2579 set_buffer_mapped(bh);
2585 * This function is used as a standard get_block_t calback function
2586 * when there is no desire to allocate any blocks. It is used as a
2587 * callback function for block_write_begin() and block_write_full_page().
2588 * These functions should only try to map a single block at a time.
2590 * Since this function doesn't do block allocations even if the caller
2591 * requests it by passing in create=1, it is critically important that
2592 * any caller checks to make sure that any buffer heads are returned
2593 * by this function are either all already mapped or marked for
2594 * delayed allocation before calling block_write_full_page(). Otherwise,
2595 * b_blocknr could be left unitialized, and the page write functions will
2596 * be taken by surprise.
2598 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2599 struct buffer_head *bh_result, int create)
2601 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2602 return _ext4_get_block(inode, iblock, bh_result, 0);
2605 static int bget_one(handle_t *handle, struct buffer_head *bh)
2611 static int bput_one(handle_t *handle, struct buffer_head *bh)
2617 static int __ext4_journalled_writepage(struct page *page,
2620 struct address_space *mapping = page->mapping;
2621 struct inode *inode = mapping->host;
2622 struct buffer_head *page_bufs;
2623 handle_t *handle = NULL;
2627 ClearPageChecked(page);
2628 page_bufs = page_buffers(page);
2630 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2631 /* As soon as we unlock the page, it can go away, but we have
2632 * references to buffers so we are safe */
2635 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2636 if (IS_ERR(handle)) {
2637 ret = PTR_ERR(handle);
2641 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2642 do_journal_get_write_access);
2644 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2648 err = ext4_journal_stop(handle);
2652 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2653 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2658 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2659 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2662 * Note that we don't need to start a transaction unless we're journaling data
2663 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2664 * need to file the inode to the transaction's list in ordered mode because if
2665 * we are writing back data added by write(), the inode is already there and if
2666 * we are writing back data modified via mmap(), noone guarantees in which
2667 * transaction the data will hit the disk. In case we are journaling data, we
2668 * cannot start transaction directly because transaction start ranks above page
2669 * lock so we have to do some magic.
2671 * This function can get called via...
2672 * - ext4_da_writepages after taking page lock (have journal handle)
2673 * - journal_submit_inode_data_buffers (no journal handle)
2674 * - shrink_page_list via pdflush (no journal handle)
2675 * - grab_page_cache when doing write_begin (have journal handle)
2677 * We don't do any block allocation in this function. If we have page with
2678 * multiple blocks we need to write those buffer_heads that are mapped. This
2679 * is important for mmaped based write. So if we do with blocksize 1K
2680 * truncate(f, 1024);
2681 * a = mmap(f, 0, 4096);
2683 * truncate(f, 4096);
2684 * we have in the page first buffer_head mapped via page_mkwrite call back
2685 * but other bufer_heads would be unmapped but dirty(dirty done via the
2686 * do_wp_page). So writepage should write the first block. If we modify
2687 * the mmap area beyond 1024 we will again get a page_fault and the
2688 * page_mkwrite callback will do the block allocation and mark the
2689 * buffer_heads mapped.
2691 * We redirty the page if we have any buffer_heads that is either delay or
2692 * unwritten in the page.
2694 * We can get recursively called as show below.
2696 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2699 * But since we don't do any block allocation we should not deadlock.
2700 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2702 static int ext4_writepage(struct page *page,
2703 struct writeback_control *wbc)
2705 int ret = 0, commit_write = 0;
2708 struct buffer_head *page_bufs = NULL;
2709 struct inode *inode = page->mapping->host;
2711 trace_ext4_writepage(inode, page);
2712 size = i_size_read(inode);
2713 if (page->index == size >> PAGE_CACHE_SHIFT)
2714 len = size & ~PAGE_CACHE_MASK;
2716 len = PAGE_CACHE_SIZE;
2719 * If the page does not have buffers (for whatever reason),
2720 * try to create them using __block_write_begin. If this
2721 * fails, redirty the page and move on.
2723 if (!page_has_buffers(page)) {
2724 if (__block_write_begin(page, 0, len,
2725 noalloc_get_block_write)) {
2727 redirty_page_for_writepage(wbc, page);
2733 page_bufs = page_buffers(page);
2734 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2735 ext4_bh_delay_or_unwritten)) {
2737 * We don't want to do block allocation, so redirty
2738 * the page and return. We may reach here when we do
2739 * a journal commit via journal_submit_inode_data_buffers.
2740 * We can also reach here via shrink_page_list
2745 /* now mark the buffer_heads as dirty and uptodate */
2746 block_commit_write(page, 0, len);
2748 if (PageChecked(page) && ext4_should_journal_data(inode))
2750 * It's mmapped pagecache. Add buffers and journal it. There
2751 * doesn't seem much point in redirtying the page here.
2753 return __ext4_journalled_writepage(page, len);
2755 if (buffer_uninit(page_bufs)) {
2756 ext4_set_bh_endio(page_bufs, inode);
2757 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2758 wbc, ext4_end_io_buffer_write);
2760 ret = block_write_full_page(page, noalloc_get_block_write,
2767 * This is called via ext4_da_writepages() to
2768 * calulate the total number of credits to reserve to fit
2769 * a single extent allocation into a single transaction,
2770 * ext4_da_writpeages() will loop calling this before
2771 * the block allocation.
2774 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2776 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2779 * With non-extent format the journal credit needed to
2780 * insert nrblocks contiguous block is dependent on
2781 * number of contiguous block. So we will limit
2782 * number of contiguous block to a sane value
2784 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2785 (max_blocks > EXT4_MAX_TRANS_DATA))
2786 max_blocks = EXT4_MAX_TRANS_DATA;
2788 return ext4_chunk_trans_blocks(inode, max_blocks);
2792 * write_cache_pages_da - walk the list of dirty pages of the given
2793 * address space and call the callback function (which usually writes
2796 * This is a forked version of write_cache_pages(). Differences:
2797 * Range cyclic is ignored.
2798 * no_nrwrite_index_update is always presumed true
2800 static int write_cache_pages_da(struct address_space *mapping,
2801 struct writeback_control *wbc,
2802 struct mpage_da_data *mpd,
2803 pgoff_t *done_index)
2807 struct pagevec pvec;
2810 pgoff_t end; /* Inclusive */
2811 long nr_to_write = wbc->nr_to_write;
2814 pagevec_init(&pvec, 0);
2815 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2816 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2818 if (wbc->sync_mode == WB_SYNC_ALL)
2819 tag = PAGECACHE_TAG_TOWRITE;
2821 tag = PAGECACHE_TAG_DIRTY;
2823 *done_index = index;
2824 while (!done && (index <= end)) {
2827 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2828 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2832 for (i = 0; i < nr_pages; i++) {
2833 struct page *page = pvec.pages[i];
2836 * At this point, the page may be truncated or
2837 * invalidated (changing page->mapping to NULL), or
2838 * even swizzled back from swapper_space to tmpfs file
2839 * mapping. However, page->index will not change
2840 * because we have a reference on the page.
2842 if (page->index > end) {
2847 *done_index = page->index + 1;
2852 * Page truncated or invalidated. We can freely skip it
2853 * then, even for data integrity operations: the page
2854 * has disappeared concurrently, so there could be no
2855 * real expectation of this data interity operation
2856 * even if there is now a new, dirty page at the same
2857 * pagecache address.
2859 if (unlikely(page->mapping != mapping)) {
2865 if (!PageDirty(page)) {
2866 /* someone wrote it for us */
2867 goto continue_unlock;
2870 if (PageWriteback(page)) {
2871 if (wbc->sync_mode != WB_SYNC_NONE)
2872 wait_on_page_writeback(page);
2874 goto continue_unlock;
2877 BUG_ON(PageWriteback(page));
2878 if (!clear_page_dirty_for_io(page))
2879 goto continue_unlock;
2881 ret = __mpage_da_writepage(page, wbc, mpd);
2882 if (unlikely(ret)) {
2883 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2892 if (nr_to_write > 0) {
2894 if (nr_to_write == 0 &&
2895 wbc->sync_mode == WB_SYNC_NONE) {
2897 * We stop writing back only if we are
2898 * not doing integrity sync. In case of
2899 * integrity sync we have to keep going
2900 * because someone may be concurrently
2901 * dirtying pages, and we might have
2902 * synced a lot of newly appeared dirty
2903 * pages, but have not synced all of the
2911 pagevec_release(&pvec);
2918 static int ext4_da_writepages(struct address_space *mapping,
2919 struct writeback_control *wbc)
2922 int range_whole = 0;
2923 handle_t *handle = NULL;
2924 struct mpage_da_data mpd;
2925 struct inode *inode = mapping->host;
2926 int pages_written = 0;
2928 unsigned int max_pages;
2929 int range_cyclic, cycled = 1, io_done = 0;
2930 int needed_blocks, ret = 0;
2931 long desired_nr_to_write, nr_to_writebump = 0;
2932 loff_t range_start = wbc->range_start;
2933 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2934 pgoff_t done_index = 0;
2937 trace_ext4_da_writepages(inode, wbc);
2940 * No pages to write? This is mainly a kludge to avoid starting
2941 * a transaction for special inodes like journal inode on last iput()
2942 * because that could violate lock ordering on umount
2944 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2948 * If the filesystem has aborted, it is read-only, so return
2949 * right away instead of dumping stack traces later on that
2950 * will obscure the real source of the problem. We test
2951 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2952 * the latter could be true if the filesystem is mounted
2953 * read-only, and in that case, ext4_da_writepages should
2954 * *never* be called, so if that ever happens, we would want
2957 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2960 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2963 range_cyclic = wbc->range_cyclic;
2964 if (wbc->range_cyclic) {
2965 index = mapping->writeback_index;
2968 wbc->range_start = index << PAGE_CACHE_SHIFT;
2969 wbc->range_end = LLONG_MAX;
2970 wbc->range_cyclic = 0;
2973 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2974 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2978 * This works around two forms of stupidity. The first is in
2979 * the writeback code, which caps the maximum number of pages
2980 * written to be 1024 pages. This is wrong on multiple
2981 * levels; different architectues have a different page size,
2982 * which changes the maximum amount of data which gets
2983 * written. Secondly, 4 megabytes is way too small. XFS
2984 * forces this value to be 16 megabytes by multiplying
2985 * nr_to_write parameter by four, and then relies on its
2986 * allocator to allocate larger extents to make them
2987 * contiguous. Unfortunately this brings us to the second
2988 * stupidity, which is that ext4's mballoc code only allocates
2989 * at most 2048 blocks. So we force contiguous writes up to
2990 * the number of dirty blocks in the inode, or
2991 * sbi->max_writeback_mb_bump whichever is smaller.
2993 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2994 if (!range_cyclic && range_whole) {
2995 if (wbc->nr_to_write == LONG_MAX)
2996 desired_nr_to_write = wbc->nr_to_write;
2998 desired_nr_to_write = wbc->nr_to_write * 8;
3000 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3002 if (desired_nr_to_write > max_pages)
3003 desired_nr_to_write = max_pages;
3005 if (wbc->nr_to_write < desired_nr_to_write) {
3006 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
3007 wbc->nr_to_write = desired_nr_to_write;
3011 mpd.inode = mapping->host;
3013 pages_skipped = wbc->pages_skipped;
3016 if (wbc->sync_mode == WB_SYNC_ALL)
3017 tag_pages_for_writeback(mapping, index, end);
3019 while (!ret && wbc->nr_to_write > 0) {
3022 * we insert one extent at a time. So we need
3023 * credit needed for single extent allocation.
3024 * journalled mode is currently not supported
3027 BUG_ON(ext4_should_journal_data(inode));
3028 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3030 /* start a new transaction*/
3031 handle = ext4_journal_start(inode, needed_blocks);
3032 if (IS_ERR(handle)) {
3033 ret = PTR_ERR(handle);
3034 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3035 "%ld pages, ino %lu; err %d", __func__,
3036 wbc->nr_to_write, inode->i_ino, ret);
3037 goto out_writepages;
3041 * Now call __mpage_da_writepage to find the next
3042 * contiguous region of logical blocks that need
3043 * blocks to be allocated by ext4. We don't actually
3044 * submit the blocks for I/O here, even though
3045 * write_cache_pages thinks it will, and will set the
3046 * pages as clean for write before calling
3047 * __mpage_da_writepage().
3055 mpd.pages_written = 0;
3057 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3059 * If we have a contiguous extent of pages and we
3060 * haven't done the I/O yet, map the blocks and submit
3063 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3064 mpage_da_map_and_submit(&mpd);
3065 ret = MPAGE_DA_EXTENT_TAIL;
3067 trace_ext4_da_write_pages(inode, &mpd);
3068 wbc->nr_to_write -= mpd.pages_written;
3070 ext4_journal_stop(handle);
3072 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3073 /* commit the transaction which would
3074 * free blocks released in the transaction
3077 jbd2_journal_force_commit_nested(sbi->s_journal);
3078 wbc->pages_skipped = pages_skipped;
3080 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3082 * got one extent now try with
3085 pages_written += mpd.pages_written;
3086 wbc->pages_skipped = pages_skipped;
3089 } else if (wbc->nr_to_write)
3091 * There is no more writeout needed
3092 * or we requested for a noblocking writeout
3093 * and we found the device congested
3097 if (!io_done && !cycled) {
3100 wbc->range_start = index << PAGE_CACHE_SHIFT;
3101 wbc->range_end = mapping->writeback_index - 1;
3104 if (pages_skipped != wbc->pages_skipped)
3105 ext4_msg(inode->i_sb, KERN_CRIT,
3106 "This should not happen leaving %s "
3107 "with nr_to_write = %ld ret = %d",
3108 __func__, wbc->nr_to_write, ret);
3111 wbc->range_cyclic = range_cyclic;
3112 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3114 * set the writeback_index so that range_cyclic
3115 * mode will write it back later
3117 mapping->writeback_index = done_index;
3120 wbc->nr_to_write -= nr_to_writebump;
3121 wbc->range_start = range_start;
3122 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3126 #define FALL_BACK_TO_NONDELALLOC 1
3127 static int ext4_nonda_switch(struct super_block *sb)
3129 s64 free_blocks, dirty_blocks;
3130 struct ext4_sb_info *sbi = EXT4_SB(sb);
3133 * switch to non delalloc mode if we are running low
3134 * on free block. The free block accounting via percpu
3135 * counters can get slightly wrong with percpu_counter_batch getting
3136 * accumulated on each CPU without updating global counters
3137 * Delalloc need an accurate free block accounting. So switch
3138 * to non delalloc when we are near to error range.
3140 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3141 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3142 if (2 * free_blocks < 3 * dirty_blocks ||
3143 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3145 * free block count is less than 150% of dirty blocks
3146 * or free blocks is less than watermark
3151 * Even if we don't switch but are nearing capacity,
3152 * start pushing delalloc when 1/2 of free blocks are dirty.
3154 if (free_blocks < 2 * dirty_blocks)
3155 writeback_inodes_sb_if_idle(sb);
3160 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3161 loff_t pos, unsigned len, unsigned flags,
3162 struct page **pagep, void **fsdata)
3164 int ret, retries = 0;
3167 struct inode *inode = mapping->host;
3170 index = pos >> PAGE_CACHE_SHIFT;
3172 if (ext4_nonda_switch(inode->i_sb)) {
3173 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3174 return ext4_write_begin(file, mapping, pos,
3175 len, flags, pagep, fsdata);
3177 *fsdata = (void *)0;
3178 trace_ext4_da_write_begin(inode, pos, len, flags);
3181 * With delayed allocation, we don't log the i_disksize update
3182 * if there is delayed block allocation. But we still need
3183 * to journalling the i_disksize update if writes to the end
3184 * of file which has an already mapped buffer.
3186 handle = ext4_journal_start(inode, 1);
3187 if (IS_ERR(handle)) {
3188 ret = PTR_ERR(handle);
3191 /* We cannot recurse into the filesystem as the transaction is already
3193 flags |= AOP_FLAG_NOFS;
3195 page = grab_cache_page_write_begin(mapping, index, flags);
3197 ext4_journal_stop(handle);
3203 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3206 ext4_journal_stop(handle);
3207 page_cache_release(page);
3209 * block_write_begin may have instantiated a few blocks
3210 * outside i_size. Trim these off again. Don't need
3211 * i_size_read because we hold i_mutex.
3213 if (pos + len > inode->i_size)
3214 ext4_truncate_failed_write(inode);
3217 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3224 * Check if we should update i_disksize
3225 * when write to the end of file but not require block allocation
3227 static int ext4_da_should_update_i_disksize(struct page *page,
3228 unsigned long offset)
3230 struct buffer_head *bh;
3231 struct inode *inode = page->mapping->host;
3235 bh = page_buffers(page);
3236 idx = offset >> inode->i_blkbits;
3238 for (i = 0; i < idx; i++)
3239 bh = bh->b_this_page;
3241 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3246 static int ext4_da_write_end(struct file *file,
3247 struct address_space *mapping,
3248 loff_t pos, unsigned len, unsigned copied,
3249 struct page *page, void *fsdata)
3251 struct inode *inode = mapping->host;
3253 handle_t *handle = ext4_journal_current_handle();
3255 unsigned long start, end;
3256 int write_mode = (int)(unsigned long)fsdata;
3258 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3259 if (ext4_should_order_data(inode)) {
3260 return ext4_ordered_write_end(file, mapping, pos,
3261 len, copied, page, fsdata);
3262 } else if (ext4_should_writeback_data(inode)) {
3263 return ext4_writeback_write_end(file, mapping, pos,
3264 len, copied, page, fsdata);
3270 trace_ext4_da_write_end(inode, pos, len, copied);
3271 start = pos & (PAGE_CACHE_SIZE - 1);
3272 end = start + copied - 1;
3275 * generic_write_end() will run mark_inode_dirty() if i_size
3276 * changes. So let's piggyback the i_disksize mark_inode_dirty
3280 new_i_size = pos + copied;
3281 if (new_i_size > EXT4_I(inode)->i_disksize) {
3282 if (ext4_da_should_update_i_disksize(page, end)) {
3283 down_write(&EXT4_I(inode)->i_data_sem);
3284 if (new_i_size > EXT4_I(inode)->i_disksize) {
3286 * Updating i_disksize when extending file
3287 * without needing block allocation
3289 if (ext4_should_order_data(inode))
3290 ret = ext4_jbd2_file_inode(handle,
3293 EXT4_I(inode)->i_disksize = new_i_size;
3295 up_write(&EXT4_I(inode)->i_data_sem);
3296 /* We need to mark inode dirty even if
3297 * new_i_size is less that inode->i_size
3298 * bu greater than i_disksize.(hint delalloc)
3300 ext4_mark_inode_dirty(handle, inode);
3303 ret2 = generic_write_end(file, mapping, pos, len, copied,
3308 ret2 = ext4_journal_stop(handle);
3312 return ret ? ret : copied;
3315 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3318 * Drop reserved blocks
3320 BUG_ON(!PageLocked(page));
3321 if (!page_has_buffers(page))
3324 ext4_da_page_release_reservation(page, offset);
3327 ext4_invalidatepage(page, offset);
3333 * Force all delayed allocation blocks to be allocated for a given inode.
3335 int ext4_alloc_da_blocks(struct inode *inode)
3337 trace_ext4_alloc_da_blocks(inode);
3339 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3340 !EXT4_I(inode)->i_reserved_meta_blocks)
3344 * We do something simple for now. The filemap_flush() will
3345 * also start triggering a write of the data blocks, which is
3346 * not strictly speaking necessary (and for users of
3347 * laptop_mode, not even desirable). However, to do otherwise
3348 * would require replicating code paths in:
3350 * ext4_da_writepages() ->
3351 * write_cache_pages() ---> (via passed in callback function)
3352 * __mpage_da_writepage() -->
3353 * mpage_add_bh_to_extent()
3354 * mpage_da_map_blocks()
3356 * The problem is that write_cache_pages(), located in
3357 * mm/page-writeback.c, marks pages clean in preparation for
3358 * doing I/O, which is not desirable if we're not planning on
3361 * We could call write_cache_pages(), and then redirty all of
3362 * the pages by calling redirty_page_for_writeback() but that
3363 * would be ugly in the extreme. So instead we would need to
3364 * replicate parts of the code in the above functions,
3365 * simplifying them becuase we wouldn't actually intend to
3366 * write out the pages, but rather only collect contiguous
3367 * logical block extents, call the multi-block allocator, and
3368 * then update the buffer heads with the block allocations.
3370 * For now, though, we'll cheat by calling filemap_flush(),
3371 * which will map the blocks, and start the I/O, but not
3372 * actually wait for the I/O to complete.
3374 return filemap_flush(inode->i_mapping);
3378 * bmap() is special. It gets used by applications such as lilo and by
3379 * the swapper to find the on-disk block of a specific piece of data.
3381 * Naturally, this is dangerous if the block concerned is still in the
3382 * journal. If somebody makes a swapfile on an ext4 data-journaling
3383 * filesystem and enables swap, then they may get a nasty shock when the
3384 * data getting swapped to that swapfile suddenly gets overwritten by
3385 * the original zero's written out previously to the journal and
3386 * awaiting writeback in the kernel's buffer cache.
3388 * So, if we see any bmap calls here on a modified, data-journaled file,
3389 * take extra steps to flush any blocks which might be in the cache.
3391 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3393 struct inode *inode = mapping->host;
3397 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3398 test_opt(inode->i_sb, DELALLOC)) {
3400 * With delalloc we want to sync the file
3401 * so that we can make sure we allocate
3404 filemap_write_and_wait(mapping);
3407 if (EXT4_JOURNAL(inode) &&
3408 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3410 * This is a REALLY heavyweight approach, but the use of
3411 * bmap on dirty files is expected to be extremely rare:
3412 * only if we run lilo or swapon on a freshly made file
3413 * do we expect this to happen.
3415 * (bmap requires CAP_SYS_RAWIO so this does not
3416 * represent an unprivileged user DOS attack --- we'd be
3417 * in trouble if mortal users could trigger this path at
3420 * NB. EXT4_STATE_JDATA is not set on files other than
3421 * regular files. If somebody wants to bmap a directory
3422 * or symlink and gets confused because the buffer
3423 * hasn't yet been flushed to disk, they deserve
3424 * everything they get.
3427 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3428 journal = EXT4_JOURNAL(inode);
3429 jbd2_journal_lock_updates(journal);
3430 err = jbd2_journal_flush(journal);
3431 jbd2_journal_unlock_updates(journal);
3437 return generic_block_bmap(mapping, block, ext4_get_block);
3440 static int ext4_readpage(struct file *file, struct page *page)
3442 return mpage_readpage(page, ext4_get_block);
3446 ext4_readpages(struct file *file, struct address_space *mapping,
3447 struct list_head *pages, unsigned nr_pages)
3449 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3452 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3454 struct buffer_head *head, *bh;
3455 unsigned int curr_off = 0;
3457 if (!page_has_buffers(page))
3459 head = bh = page_buffers(page);
3461 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3463 ext4_free_io_end(bh->b_private);
3464 bh->b_private = NULL;
3465 bh->b_end_io = NULL;
3467 curr_off = curr_off + bh->b_size;
3468 bh = bh->b_this_page;
3469 } while (bh != head);
3472 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3474 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3477 * free any io_end structure allocated for buffers to be discarded
3479 if (ext4_should_dioread_nolock(page->mapping->host))
3480 ext4_invalidatepage_free_endio(page, offset);
3482 * If it's a full truncate we just forget about the pending dirtying
3485 ClearPageChecked(page);
3488 jbd2_journal_invalidatepage(journal, page, offset);
3490 block_invalidatepage(page, offset);
3493 static int ext4_releasepage(struct page *page, gfp_t wait)
3495 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3497 WARN_ON(PageChecked(page));
3498 if (!page_has_buffers(page))
3501 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3503 return try_to_free_buffers(page);
3507 * O_DIRECT for ext3 (or indirect map) based files
3509 * If the O_DIRECT write will extend the file then add this inode to the
3510 * orphan list. So recovery will truncate it back to the original size
3511 * if the machine crashes during the write.
3513 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3514 * crashes then stale disk data _may_ be exposed inside the file. But current
3515 * VFS code falls back into buffered path in that case so we are safe.
3517 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3518 const struct iovec *iov, loff_t offset,
3519 unsigned long nr_segs)
3521 struct file *file = iocb->ki_filp;
3522 struct inode *inode = file->f_mapping->host;
3523 struct ext4_inode_info *ei = EXT4_I(inode);
3527 size_t count = iov_length(iov, nr_segs);
3531 loff_t final_size = offset + count;
3533 if (final_size > inode->i_size) {
3534 /* Credits for sb + inode write */
3535 handle = ext4_journal_start(inode, 2);
3536 if (IS_ERR(handle)) {
3537 ret = PTR_ERR(handle);
3540 ret = ext4_orphan_add(handle, inode);
3542 ext4_journal_stop(handle);
3546 ei->i_disksize = inode->i_size;
3547 ext4_journal_stop(handle);
3552 if (rw == READ && ext4_should_dioread_nolock(inode))
3553 ret = __blockdev_direct_IO(rw, iocb, inode,
3554 inode->i_sb->s_bdev, iov,
3556 ext4_get_block, NULL, NULL, 0);
3558 ret = blockdev_direct_IO(rw, iocb, inode,
3559 inode->i_sb->s_bdev, iov,
3561 ext4_get_block, NULL);
3563 if (unlikely((rw & WRITE) && ret < 0)) {
3564 loff_t isize = i_size_read(inode);
3565 loff_t end = offset + iov_length(iov, nr_segs);
3568 vmtruncate(inode, isize);
3571 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3577 /* Credits for sb + inode write */
3578 handle = ext4_journal_start(inode, 2);
3579 if (IS_ERR(handle)) {
3580 /* This is really bad luck. We've written the data
3581 * but cannot extend i_size. Bail out and pretend
3582 * the write failed... */
3583 ret = PTR_ERR(handle);
3585 ext4_orphan_del(NULL, inode);
3590 ext4_orphan_del(handle, inode);
3592 loff_t end = offset + ret;
3593 if (end > inode->i_size) {
3594 ei->i_disksize = end;
3595 i_size_write(inode, end);
3597 * We're going to return a positive `ret'
3598 * here due to non-zero-length I/O, so there's
3599 * no way of reporting error returns from
3600 * ext4_mark_inode_dirty() to userspace. So
3603 ext4_mark_inode_dirty(handle, inode);
3606 err = ext4_journal_stop(handle);
3615 * ext4_get_block used when preparing for a DIO write or buffer write.
3616 * We allocate an uinitialized extent if blocks haven't been allocated.
3617 * The extent will be converted to initialized after the IO is complete.
3619 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3620 struct buffer_head *bh_result, int create)
3622 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3623 inode->i_ino, create);
3624 return _ext4_get_block(inode, iblock, bh_result,
3625 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3628 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3629 ssize_t size, void *private, int ret,
3632 ext4_io_end_t *io_end = iocb->private;
3633 struct workqueue_struct *wq;
3634 unsigned long flags;
3635 struct ext4_inode_info *ei;
3637 /* if not async direct IO or dio with 0 bytes write, just return */
3638 if (!io_end || !size)
3641 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3642 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3643 iocb->private, io_end->inode->i_ino, iocb, offset,
3646 /* if not aio dio with unwritten extents, just free io and return */
3647 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3648 ext4_free_io_end(io_end);
3649 iocb->private = NULL;
3652 aio_complete(iocb, ret, 0);
3656 io_end->offset = offset;
3657 io_end->size = size;
3659 io_end->iocb = iocb;
3660 io_end->result = ret;
3662 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3664 /* Add the io_end to per-inode completed aio dio list*/
3665 ei = EXT4_I(io_end->inode);
3666 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3667 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3668 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3670 /* queue the work to convert unwritten extents to written */
3671 queue_work(wq, &io_end->work);
3672 iocb->private = NULL;
3675 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3677 ext4_io_end_t *io_end = bh->b_private;
3678 struct workqueue_struct *wq;
3679 struct inode *inode;
3680 unsigned long flags;
3682 if (!test_clear_buffer_uninit(bh) || !io_end)
3685 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3686 printk("sb umounted, discard end_io request for inode %lu\n",
3687 io_end->inode->i_ino);
3688 ext4_free_io_end(io_end);
3692 io_end->flag = EXT4_IO_END_UNWRITTEN;
3693 inode = io_end->inode;
3695 /* Add the io_end to per-inode completed io list*/
3696 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3697 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3698 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3700 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3701 /* queue the work to convert unwritten extents to written */
3702 queue_work(wq, &io_end->work);
3704 bh->b_private = NULL;
3705 bh->b_end_io = NULL;
3706 clear_buffer_uninit(bh);
3707 end_buffer_async_write(bh, uptodate);
3710 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3712 ext4_io_end_t *io_end;
3713 struct page *page = bh->b_page;
3714 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3715 size_t size = bh->b_size;
3718 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3720 if (printk_ratelimit())
3721 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3725 io_end->offset = offset;
3726 io_end->size = size;
3728 * We need to hold a reference to the page to make sure it
3729 * doesn't get evicted before ext4_end_io_work() has a chance
3730 * to convert the extent from written to unwritten.
3732 io_end->page = page;
3733 get_page(io_end->page);
3735 bh->b_private = io_end;
3736 bh->b_end_io = ext4_end_io_buffer_write;
3741 * For ext4 extent files, ext4 will do direct-io write to holes,
3742 * preallocated extents, and those write extend the file, no need to
3743 * fall back to buffered IO.
3745 * For holes, we fallocate those blocks, mark them as unintialized
3746 * If those blocks were preallocated, we mark sure they are splited, but
3747 * still keep the range to write as unintialized.
3749 * The unwrritten extents will be converted to written when DIO is completed.
3750 * For async direct IO, since the IO may still pending when return, we
3751 * set up an end_io call back function, which will do the convertion
3752 * when async direct IO completed.
3754 * If the O_DIRECT write will extend the file then add this inode to the
3755 * orphan list. So recovery will truncate it back to the original size
3756 * if the machine crashes during the write.
3759 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3760 const struct iovec *iov, loff_t offset,
3761 unsigned long nr_segs)
3763 struct file *file = iocb->ki_filp;
3764 struct inode *inode = file->f_mapping->host;
3766 size_t count = iov_length(iov, nr_segs);
3768 loff_t final_size = offset + count;
3769 if (rw == WRITE && final_size <= inode->i_size) {
3771 * We could direct write to holes and fallocate.
3773 * Allocated blocks to fill the hole are marked as uninitialized
3774 * to prevent paralel buffered read to expose the stale data
3775 * before DIO complete the data IO.
3777 * As to previously fallocated extents, ext4 get_block
3778 * will just simply mark the buffer mapped but still
3779 * keep the extents uninitialized.
3781 * for non AIO case, we will convert those unwritten extents
3782 * to written after return back from blockdev_direct_IO.
3784 * for async DIO, the conversion needs to be defered when
3785 * the IO is completed. The ext4 end_io callback function
3786 * will be called to take care of the conversion work.
3787 * Here for async case, we allocate an io_end structure to
3790 iocb->private = NULL;
3791 EXT4_I(inode)->cur_aio_dio = NULL;
3792 if (!is_sync_kiocb(iocb)) {
3793 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3797 * we save the io structure for current async
3798 * direct IO, so that later ext4_map_blocks()
3799 * could flag the io structure whether there
3800 * is a unwritten extents needs to be converted
3801 * when IO is completed.
3803 EXT4_I(inode)->cur_aio_dio = iocb->private;
3806 ret = blockdev_direct_IO(rw, iocb, inode,
3807 inode->i_sb->s_bdev, iov,
3809 ext4_get_block_write,
3812 EXT4_I(inode)->cur_aio_dio = NULL;
3814 * The io_end structure takes a reference to the inode,
3815 * that structure needs to be destroyed and the
3816 * reference to the inode need to be dropped, when IO is
3817 * complete, even with 0 byte write, or failed.
3819 * In the successful AIO DIO case, the io_end structure will be
3820 * desctroyed and the reference to the inode will be dropped
3821 * after the end_io call back function is called.
3823 * In the case there is 0 byte write, or error case, since
3824 * VFS direct IO won't invoke the end_io call back function,
3825 * we need to free the end_io structure here.
3827 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3828 ext4_free_io_end(iocb->private);
3829 iocb->private = NULL;
3830 } else if (ret > 0 && ext4_test_inode_state(inode,
3831 EXT4_STATE_DIO_UNWRITTEN)) {
3834 * for non AIO case, since the IO is already
3835 * completed, we could do the convertion right here
3837 err = ext4_convert_unwritten_extents(inode,
3841 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3846 /* for write the the end of file case, we fall back to old way */
3847 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3850 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3851 const struct iovec *iov, loff_t offset,
3852 unsigned long nr_segs)
3854 struct file *file = iocb->ki_filp;
3855 struct inode *inode = file->f_mapping->host;
3857 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3858 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3860 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3864 * Pages can be marked dirty completely asynchronously from ext4's journalling
3865 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3866 * much here because ->set_page_dirty is called under VFS locks. The page is
3867 * not necessarily locked.
3869 * We cannot just dirty the page and leave attached buffers clean, because the
3870 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3871 * or jbddirty because all the journalling code will explode.
3873 * So what we do is to mark the page "pending dirty" and next time writepage
3874 * is called, propagate that into the buffers appropriately.
3876 static int ext4_journalled_set_page_dirty(struct page *page)
3878 SetPageChecked(page);
3879 return __set_page_dirty_nobuffers(page);
3882 static const struct address_space_operations ext4_ordered_aops = {
3883 .readpage = ext4_readpage,
3884 .readpages = ext4_readpages,
3885 .writepage = ext4_writepage,
3886 .sync_page = block_sync_page,
3887 .write_begin = ext4_write_begin,
3888 .write_end = ext4_ordered_write_end,
3890 .invalidatepage = ext4_invalidatepage,
3891 .releasepage = ext4_releasepage,
3892 .direct_IO = ext4_direct_IO,
3893 .migratepage = buffer_migrate_page,
3894 .is_partially_uptodate = block_is_partially_uptodate,
3895 .error_remove_page = generic_error_remove_page,
3898 static const struct address_space_operations ext4_writeback_aops = {
3899 .readpage = ext4_readpage,
3900 .readpages = ext4_readpages,
3901 .writepage = ext4_writepage,
3902 .sync_page = block_sync_page,
3903 .write_begin = ext4_write_begin,
3904 .write_end = ext4_writeback_write_end,
3906 .invalidatepage = ext4_invalidatepage,
3907 .releasepage = ext4_releasepage,
3908 .direct_IO = ext4_direct_IO,
3909 .migratepage = buffer_migrate_page,
3910 .is_partially_uptodate = block_is_partially_uptodate,
3911 .error_remove_page = generic_error_remove_page,
3914 static const struct address_space_operations ext4_journalled_aops = {
3915 .readpage = ext4_readpage,
3916 .readpages = ext4_readpages,
3917 .writepage = ext4_writepage,
3918 .sync_page = block_sync_page,
3919 .write_begin = ext4_write_begin,
3920 .write_end = ext4_journalled_write_end,
3921 .set_page_dirty = ext4_journalled_set_page_dirty,
3923 .invalidatepage = ext4_invalidatepage,
3924 .releasepage = ext4_releasepage,
3925 .is_partially_uptodate = block_is_partially_uptodate,
3926 .error_remove_page = generic_error_remove_page,
3929 static const struct address_space_operations ext4_da_aops = {
3930 .readpage = ext4_readpage,
3931 .readpages = ext4_readpages,
3932 .writepage = ext4_writepage,
3933 .writepages = ext4_da_writepages,
3934 .sync_page = block_sync_page,
3935 .write_begin = ext4_da_write_begin,
3936 .write_end = ext4_da_write_end,
3938 .invalidatepage = ext4_da_invalidatepage,
3939 .releasepage = ext4_releasepage,
3940 .direct_IO = ext4_direct_IO,
3941 .migratepage = buffer_migrate_page,
3942 .is_partially_uptodate = block_is_partially_uptodate,
3943 .error_remove_page = generic_error_remove_page,
3946 void ext4_set_aops(struct inode *inode)
3948 if (ext4_should_order_data(inode) &&
3949 test_opt(inode->i_sb, DELALLOC))
3950 inode->i_mapping->a_ops = &ext4_da_aops;
3951 else if (ext4_should_order_data(inode))
3952 inode->i_mapping->a_ops = &ext4_ordered_aops;
3953 else if (ext4_should_writeback_data(inode) &&
3954 test_opt(inode->i_sb, DELALLOC))
3955 inode->i_mapping->a_ops = &ext4_da_aops;
3956 else if (ext4_should_writeback_data(inode))
3957 inode->i_mapping->a_ops = &ext4_writeback_aops;
3959 inode->i_mapping->a_ops = &ext4_journalled_aops;
3963 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3964 * up to the end of the block which corresponds to `from'.
3965 * This required during truncate. We need to physically zero the tail end
3966 * of that block so it doesn't yield old data if the file is later grown.
3968 int ext4_block_truncate_page(handle_t *handle,
3969 struct address_space *mapping, loff_t from)
3971 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3972 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3973 unsigned blocksize, length, pos;
3975 struct inode *inode = mapping->host;
3976 struct buffer_head *bh;
3980 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3981 mapping_gfp_mask(mapping) & ~__GFP_FS);
3985 blocksize = inode->i_sb->s_blocksize;
3986 length = blocksize - (offset & (blocksize - 1));
3987 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3989 if (!page_has_buffers(page))
3990 create_empty_buffers(page, blocksize, 0);
3992 /* Find the buffer that contains "offset" */
3993 bh = page_buffers(page);
3995 while (offset >= pos) {
3996 bh = bh->b_this_page;
4002 if (buffer_freed(bh)) {
4003 BUFFER_TRACE(bh, "freed: skip");
4007 if (!buffer_mapped(bh)) {
4008 BUFFER_TRACE(bh, "unmapped");
4009 ext4_get_block(inode, iblock, bh, 0);
4010 /* unmapped? It's a hole - nothing to do */
4011 if (!buffer_mapped(bh)) {
4012 BUFFER_TRACE(bh, "still unmapped");
4017 /* Ok, it's mapped. Make sure it's up-to-date */
4018 if (PageUptodate(page))
4019 set_buffer_uptodate(bh);
4021 if (!buffer_uptodate(bh)) {
4023 ll_rw_block(READ, 1, &bh);
4025 /* Uhhuh. Read error. Complain and punt. */
4026 if (!buffer_uptodate(bh))
4030 if (ext4_should_journal_data(inode)) {
4031 BUFFER_TRACE(bh, "get write access");
4032 err = ext4_journal_get_write_access(handle, bh);
4037 zero_user(page, offset, length);
4039 BUFFER_TRACE(bh, "zeroed end of block");
4042 if (ext4_should_journal_data(inode)) {
4043 err = ext4_handle_dirty_metadata(handle, inode, bh);
4045 if (ext4_should_order_data(inode))
4046 err = ext4_jbd2_file_inode(handle, inode);
4047 mark_buffer_dirty(bh);
4052 page_cache_release(page);
4057 * Probably it should be a library function... search for first non-zero word
4058 * or memcmp with zero_page, whatever is better for particular architecture.
4061 static inline int all_zeroes(__le32 *p, __le32 *q)
4070 * ext4_find_shared - find the indirect blocks for partial truncation.
4071 * @inode: inode in question
4072 * @depth: depth of the affected branch
4073 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4074 * @chain: place to store the pointers to partial indirect blocks
4075 * @top: place to the (detached) top of branch
4077 * This is a helper function used by ext4_truncate().
4079 * When we do truncate() we may have to clean the ends of several
4080 * indirect blocks but leave the blocks themselves alive. Block is
4081 * partially truncated if some data below the new i_size is refered
4082 * from it (and it is on the path to the first completely truncated
4083 * data block, indeed). We have to free the top of that path along
4084 * with everything to the right of the path. Since no allocation
4085 * past the truncation point is possible until ext4_truncate()
4086 * finishes, we may safely do the latter, but top of branch may
4087 * require special attention - pageout below the truncation point
4088 * might try to populate it.
4090 * We atomically detach the top of branch from the tree, store the
4091 * block number of its root in *@top, pointers to buffer_heads of
4092 * partially truncated blocks - in @chain[].bh and pointers to
4093 * their last elements that should not be removed - in
4094 * @chain[].p. Return value is the pointer to last filled element
4097 * The work left to caller to do the actual freeing of subtrees:
4098 * a) free the subtree starting from *@top
4099 * b) free the subtrees whose roots are stored in
4100 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4101 * c) free the subtrees growing from the inode past the @chain[0].
4102 * (no partially truncated stuff there). */
4104 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4105 ext4_lblk_t offsets[4], Indirect chain[4],
4108 Indirect *partial, *p;
4112 /* Make k index the deepest non-null offset + 1 */
4113 for (k = depth; k > 1 && !offsets[k-1]; k--)
4115 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4116 /* Writer: pointers */
4118 partial = chain + k-1;
4120 * If the branch acquired continuation since we've looked at it -
4121 * fine, it should all survive and (new) top doesn't belong to us.
4123 if (!partial->key && *partial->p)
4126 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4129 * OK, we've found the last block that must survive. The rest of our
4130 * branch should be detached before unlocking. However, if that rest
4131 * of branch is all ours and does not grow immediately from the inode
4132 * it's easier to cheat and just decrement partial->p.
4134 if (p == chain + k - 1 && p > chain) {
4138 /* Nope, don't do this in ext4. Must leave the tree intact */
4145 while (partial > p) {
4146 brelse(partial->bh);
4154 * Zero a number of block pointers in either an inode or an indirect block.
4155 * If we restart the transaction we must again get write access to the
4156 * indirect block for further modification.
4158 * We release `count' blocks on disk, but (last - first) may be greater
4159 * than `count' because there can be holes in there.
4161 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4162 struct buffer_head *bh,
4163 ext4_fsblk_t block_to_free,
4164 unsigned long count, __le32 *first,
4168 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4170 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4171 flags |= EXT4_FREE_BLOCKS_METADATA;
4173 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4175 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4176 "blocks %llu len %lu",
4177 (unsigned long long) block_to_free, count);
4181 if (try_to_extend_transaction(handle, inode)) {
4183 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4184 ext4_handle_dirty_metadata(handle, inode, bh);
4186 ext4_mark_inode_dirty(handle, inode);
4187 ext4_truncate_restart_trans(handle, inode,
4188 blocks_for_truncate(inode));
4190 BUFFER_TRACE(bh, "retaking write access");
4191 ext4_journal_get_write_access(handle, bh);
4195 for (p = first; p < last; p++)
4198 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4203 * ext4_free_data - free a list of data blocks
4204 * @handle: handle for this transaction
4205 * @inode: inode we are dealing with
4206 * @this_bh: indirect buffer_head which contains *@first and *@last
4207 * @first: array of block numbers
4208 * @last: points immediately past the end of array
4210 * We are freeing all blocks refered from that array (numbers are stored as
4211 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4213 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4214 * blocks are contiguous then releasing them at one time will only affect one
4215 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4216 * actually use a lot of journal space.
4218 * @this_bh will be %NULL if @first and @last point into the inode's direct
4221 static void ext4_free_data(handle_t *handle, struct inode *inode,
4222 struct buffer_head *this_bh,
4223 __le32 *first, __le32 *last)
4225 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4226 unsigned long count = 0; /* Number of blocks in the run */
4227 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4230 ext4_fsblk_t nr; /* Current block # */
4231 __le32 *p; /* Pointer into inode/ind
4232 for current block */
4235 if (this_bh) { /* For indirect block */
4236 BUFFER_TRACE(this_bh, "get_write_access");
4237 err = ext4_journal_get_write_access(handle, this_bh);
4238 /* Important: if we can't update the indirect pointers
4239 * to the blocks, we can't free them. */
4244 for (p = first; p < last; p++) {
4245 nr = le32_to_cpu(*p);
4247 /* accumulate blocks to free if they're contiguous */
4250 block_to_free_p = p;
4252 } else if (nr == block_to_free + count) {
4255 if (ext4_clear_blocks(handle, inode, this_bh,
4256 block_to_free, count,
4257 block_to_free_p, p))
4260 block_to_free_p = p;
4267 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4268 count, block_to_free_p, p);
4271 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4274 * The buffer head should have an attached journal head at this
4275 * point. However, if the data is corrupted and an indirect
4276 * block pointed to itself, it would have been detached when
4277 * the block was cleared. Check for this instead of OOPSing.
4279 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4280 ext4_handle_dirty_metadata(handle, inode, this_bh);
4282 EXT4_ERROR_INODE(inode,
4283 "circular indirect block detected at "
4285 (unsigned long long) this_bh->b_blocknr);
4290 * ext4_free_branches - free an array of branches
4291 * @handle: JBD handle for this transaction
4292 * @inode: inode we are dealing with
4293 * @parent_bh: the buffer_head which contains *@first and *@last
4294 * @first: array of block numbers
4295 * @last: pointer immediately past the end of array
4296 * @depth: depth of the branches to free
4298 * We are freeing all blocks refered from these branches (numbers are
4299 * stored as little-endian 32-bit) and updating @inode->i_blocks
4302 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4303 struct buffer_head *parent_bh,
4304 __le32 *first, __le32 *last, int depth)
4309 if (ext4_handle_is_aborted(handle))
4313 struct buffer_head *bh;
4314 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4316 while (--p >= first) {
4317 nr = le32_to_cpu(*p);
4319 continue; /* A hole */
4321 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4323 EXT4_ERROR_INODE(inode,
4324 "invalid indirect mapped "
4325 "block %lu (level %d)",
4326 (unsigned long) nr, depth);
4330 /* Go read the buffer for the next level down */
4331 bh = sb_bread(inode->i_sb, nr);
4334 * A read failure? Report error and clear slot
4338 EXT4_ERROR_INODE_BLOCK(inode, nr,
4343 /* This zaps the entire block. Bottom up. */
4344 BUFFER_TRACE(bh, "free child branches");
4345 ext4_free_branches(handle, inode, bh,
4346 (__le32 *) bh->b_data,
4347 (__le32 *) bh->b_data + addr_per_block,
4351 * Everything below this this pointer has been
4352 * released. Now let this top-of-subtree go.
4354 * We want the freeing of this indirect block to be
4355 * atomic in the journal with the updating of the
4356 * bitmap block which owns it. So make some room in
4359 * We zero the parent pointer *after* freeing its
4360 * pointee in the bitmaps, so if extend_transaction()
4361 * for some reason fails to put the bitmap changes and
4362 * the release into the same transaction, recovery
4363 * will merely complain about releasing a free block,
4364 * rather than leaking blocks.
4366 if (ext4_handle_is_aborted(handle))
4368 if (try_to_extend_transaction(handle, inode)) {
4369 ext4_mark_inode_dirty(handle, inode);
4370 ext4_truncate_restart_trans(handle, inode,
4371 blocks_for_truncate(inode));
4375 * The forget flag here is critical because if
4376 * we are journaling (and not doing data
4377 * journaling), we have to make sure a revoke
4378 * record is written to prevent the journal
4379 * replay from overwriting the (former)
4380 * indirect block if it gets reallocated as a
4381 * data block. This must happen in the same
4382 * transaction where the data blocks are
4385 ext4_free_blocks(handle, inode, 0, nr, 1,
4386 EXT4_FREE_BLOCKS_METADATA|
4387 EXT4_FREE_BLOCKS_FORGET);
4391 * The block which we have just freed is
4392 * pointed to by an indirect block: journal it
4394 BUFFER_TRACE(parent_bh, "get_write_access");
4395 if (!ext4_journal_get_write_access(handle,
4398 BUFFER_TRACE(parent_bh,
4399 "call ext4_handle_dirty_metadata");
4400 ext4_handle_dirty_metadata(handle,
4407 /* We have reached the bottom of the tree. */
4408 BUFFER_TRACE(parent_bh, "free data blocks");
4409 ext4_free_data(handle, inode, parent_bh, first, last);
4413 int ext4_can_truncate(struct inode *inode)
4415 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4417 if (S_ISREG(inode->i_mode))
4419 if (S_ISDIR(inode->i_mode))
4421 if (S_ISLNK(inode->i_mode))
4422 return !ext4_inode_is_fast_symlink(inode);
4429 * We block out ext4_get_block() block instantiations across the entire
4430 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4431 * simultaneously on behalf of the same inode.
4433 * As we work through the truncate and commmit bits of it to the journal there
4434 * is one core, guiding principle: the file's tree must always be consistent on
4435 * disk. We must be able to restart the truncate after a crash.
4437 * The file's tree may be transiently inconsistent in memory (although it
4438 * probably isn't), but whenever we close off and commit a journal transaction,
4439 * the contents of (the filesystem + the journal) must be consistent and
4440 * restartable. It's pretty simple, really: bottom up, right to left (although
4441 * left-to-right works OK too).
4443 * Note that at recovery time, journal replay occurs *before* the restart of
4444 * truncate against the orphan inode list.
4446 * The committed inode has the new, desired i_size (which is the same as
4447 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4448 * that this inode's truncate did not complete and it will again call
4449 * ext4_truncate() to have another go. So there will be instantiated blocks
4450 * to the right of the truncation point in a crashed ext4 filesystem. But
4451 * that's fine - as long as they are linked from the inode, the post-crash
4452 * ext4_truncate() run will find them and release them.
4454 void ext4_truncate(struct inode *inode)
4457 struct ext4_inode_info *ei = EXT4_I(inode);
4458 __le32 *i_data = ei->i_data;
4459 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4460 struct address_space *mapping = inode->i_mapping;
4461 ext4_lblk_t offsets[4];
4466 ext4_lblk_t last_block;
4467 unsigned blocksize = inode->i_sb->s_blocksize;
4469 if (!ext4_can_truncate(inode))
4472 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4474 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4475 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4477 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4478 ext4_ext_truncate(inode);
4482 handle = start_transaction(inode);
4484 return; /* AKPM: return what? */
4486 last_block = (inode->i_size + blocksize-1)
4487 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4489 if (inode->i_size & (blocksize - 1))
4490 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4493 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4495 goto out_stop; /* error */
4498 * OK. This truncate is going to happen. We add the inode to the
4499 * orphan list, so that if this truncate spans multiple transactions,
4500 * and we crash, we will resume the truncate when the filesystem
4501 * recovers. It also marks the inode dirty, to catch the new size.
4503 * Implication: the file must always be in a sane, consistent
4504 * truncatable state while each transaction commits.
4506 if (ext4_orphan_add(handle, inode))
4510 * From here we block out all ext4_get_block() callers who want to
4511 * modify the block allocation tree.
4513 down_write(&ei->i_data_sem);
4515 ext4_discard_preallocations(inode);
4518 * The orphan list entry will now protect us from any crash which
4519 * occurs before the truncate completes, so it is now safe to propagate
4520 * the new, shorter inode size (held for now in i_size) into the
4521 * on-disk inode. We do this via i_disksize, which is the value which
4522 * ext4 *really* writes onto the disk inode.
4524 ei->i_disksize = inode->i_size;
4526 if (n == 1) { /* direct blocks */
4527 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4528 i_data + EXT4_NDIR_BLOCKS);
4532 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4533 /* Kill the top of shared branch (not detached) */
4535 if (partial == chain) {
4536 /* Shared branch grows from the inode */
4537 ext4_free_branches(handle, inode, NULL,
4538 &nr, &nr+1, (chain+n-1) - partial);
4541 * We mark the inode dirty prior to restart,
4542 * and prior to stop. No need for it here.
4545 /* Shared branch grows from an indirect block */
4546 BUFFER_TRACE(partial->bh, "get_write_access");
4547 ext4_free_branches(handle, inode, partial->bh,
4549 partial->p+1, (chain+n-1) - partial);
4552 /* Clear the ends of indirect blocks on the shared branch */
4553 while (partial > chain) {
4554 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4555 (__le32*)partial->bh->b_data+addr_per_block,
4556 (chain+n-1) - partial);
4557 BUFFER_TRACE(partial->bh, "call brelse");
4558 brelse(partial->bh);
4562 /* Kill the remaining (whole) subtrees */
4563 switch (offsets[0]) {
4565 nr = i_data[EXT4_IND_BLOCK];
4567 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4568 i_data[EXT4_IND_BLOCK] = 0;
4570 case EXT4_IND_BLOCK:
4571 nr = i_data[EXT4_DIND_BLOCK];
4573 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4574 i_data[EXT4_DIND_BLOCK] = 0;
4576 case EXT4_DIND_BLOCK:
4577 nr = i_data[EXT4_TIND_BLOCK];
4579 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4580 i_data[EXT4_TIND_BLOCK] = 0;
4582 case EXT4_TIND_BLOCK:
4586 up_write(&ei->i_data_sem);
4587 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4588 ext4_mark_inode_dirty(handle, inode);
4591 * In a multi-transaction truncate, we only make the final transaction
4595 ext4_handle_sync(handle);
4598 * If this was a simple ftruncate(), and the file will remain alive
4599 * then we need to clear up the orphan record which we created above.
4600 * However, if this was a real unlink then we were called by
4601 * ext4_delete_inode(), and we allow that function to clean up the
4602 * orphan info for us.
4605 ext4_orphan_del(handle, inode);
4607 ext4_journal_stop(handle);
4611 * ext4_get_inode_loc returns with an extra refcount against the inode's
4612 * underlying buffer_head on success. If 'in_mem' is true, we have all
4613 * data in memory that is needed to recreate the on-disk version of this
4616 static int __ext4_get_inode_loc(struct inode *inode,
4617 struct ext4_iloc *iloc, int in_mem)
4619 struct ext4_group_desc *gdp;
4620 struct buffer_head *bh;
4621 struct super_block *sb = inode->i_sb;
4623 int inodes_per_block, inode_offset;
4626 if (!ext4_valid_inum(sb, inode->i_ino))
4629 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4630 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4635 * Figure out the offset within the block group inode table
4637 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4638 inode_offset = ((inode->i_ino - 1) %
4639 EXT4_INODES_PER_GROUP(sb));
4640 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4641 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4643 bh = sb_getblk(sb, block);
4645 EXT4_ERROR_INODE_BLOCK(inode, block,
4646 "unable to read itable block");
4649 if (!buffer_uptodate(bh)) {
4653 * If the buffer has the write error flag, we have failed
4654 * to write out another inode in the same block. In this
4655 * case, we don't have to read the block because we may
4656 * read the old inode data successfully.
4658 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4659 set_buffer_uptodate(bh);
4661 if (buffer_uptodate(bh)) {
4662 /* someone brought it uptodate while we waited */
4668 * If we have all information of the inode in memory and this
4669 * is the only valid inode in the block, we need not read the
4673 struct buffer_head *bitmap_bh;
4676 start = inode_offset & ~(inodes_per_block - 1);
4678 /* Is the inode bitmap in cache? */
4679 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4684 * If the inode bitmap isn't in cache then the
4685 * optimisation may end up performing two reads instead
4686 * of one, so skip it.
4688 if (!buffer_uptodate(bitmap_bh)) {
4692 for (i = start; i < start + inodes_per_block; i++) {
4693 if (i == inode_offset)
4695 if (ext4_test_bit(i, bitmap_bh->b_data))
4699 if (i == start + inodes_per_block) {
4700 /* all other inodes are free, so skip I/O */
4701 memset(bh->b_data, 0, bh->b_size);
4702 set_buffer_uptodate(bh);
4710 * If we need to do any I/O, try to pre-readahead extra
4711 * blocks from the inode table.
4713 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4714 ext4_fsblk_t b, end, table;
4717 table = ext4_inode_table(sb, gdp);
4718 /* s_inode_readahead_blks is always a power of 2 */
4719 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4722 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4723 num = EXT4_INODES_PER_GROUP(sb);
4724 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4725 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4726 num -= ext4_itable_unused_count(sb, gdp);
4727 table += num / inodes_per_block;
4731 sb_breadahead(sb, b++);
4735 * There are other valid inodes in the buffer, this inode
4736 * has in-inode xattrs, or we don't have this inode in memory.
4737 * Read the block from disk.
4740 bh->b_end_io = end_buffer_read_sync;
4741 submit_bh(READ_META, bh);
4743 if (!buffer_uptodate(bh)) {
4744 EXT4_ERROR_INODE_BLOCK(inode, block,
4745 "unable to read itable block");
4755 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4757 /* We have all inode data except xattrs in memory here. */
4758 return __ext4_get_inode_loc(inode, iloc,
4759 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4762 void ext4_set_inode_flags(struct inode *inode)
4764 unsigned int flags = EXT4_I(inode)->i_flags;
4766 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4767 if (flags & EXT4_SYNC_FL)
4768 inode->i_flags |= S_SYNC;
4769 if (flags & EXT4_APPEND_FL)
4770 inode->i_flags |= S_APPEND;
4771 if (flags & EXT4_IMMUTABLE_FL)
4772 inode->i_flags |= S_IMMUTABLE;
4773 if (flags & EXT4_NOATIME_FL)
4774 inode->i_flags |= S_NOATIME;
4775 if (flags & EXT4_DIRSYNC_FL)
4776 inode->i_flags |= S_DIRSYNC;
4779 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4780 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4782 unsigned int vfs_fl;
4783 unsigned long old_fl, new_fl;
4786 vfs_fl = ei->vfs_inode.i_flags;
4787 old_fl = ei->i_flags;
4788 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4789 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4791 if (vfs_fl & S_SYNC)
4792 new_fl |= EXT4_SYNC_FL;
4793 if (vfs_fl & S_APPEND)
4794 new_fl |= EXT4_APPEND_FL;
4795 if (vfs_fl & S_IMMUTABLE)
4796 new_fl |= EXT4_IMMUTABLE_FL;
4797 if (vfs_fl & S_NOATIME)
4798 new_fl |= EXT4_NOATIME_FL;
4799 if (vfs_fl & S_DIRSYNC)
4800 new_fl |= EXT4_DIRSYNC_FL;
4801 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4804 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4805 struct ext4_inode_info *ei)
4808 struct inode *inode = &(ei->vfs_inode);
4809 struct super_block *sb = inode->i_sb;
4811 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4812 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4813 /* we are using combined 48 bit field */
4814 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4815 le32_to_cpu(raw_inode->i_blocks_lo);
4816 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4817 /* i_blocks represent file system block size */
4818 return i_blocks << (inode->i_blkbits - 9);
4823 return le32_to_cpu(raw_inode->i_blocks_lo);
4827 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4829 struct ext4_iloc iloc;
4830 struct ext4_inode *raw_inode;
4831 struct ext4_inode_info *ei;
4832 struct inode *inode;
4833 journal_t *journal = EXT4_SB(sb)->s_journal;
4837 inode = iget_locked(sb, ino);
4839 return ERR_PTR(-ENOMEM);
4840 if (!(inode->i_state & I_NEW))
4846 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4849 raw_inode = ext4_raw_inode(&iloc);
4850 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4851 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4852 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4853 if (!(test_opt(inode->i_sb, NO_UID32))) {
4854 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4855 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4857 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4859 ei->i_state_flags = 0;
4860 ei->i_dir_start_lookup = 0;
4861 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4862 /* We now have enough fields to check if the inode was active or not.
4863 * This is needed because nfsd might try to access dead inodes
4864 * the test is that same one that e2fsck uses
4865 * NeilBrown 1999oct15
4867 if (inode->i_nlink == 0) {
4868 if (inode->i_mode == 0 ||
4869 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4870 /* this inode is deleted */
4874 /* The only unlinked inodes we let through here have
4875 * valid i_mode and are being read by the orphan
4876 * recovery code: that's fine, we're about to complete
4877 * the process of deleting those. */
4879 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4880 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4881 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4882 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4884 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4885 inode->i_size = ext4_isize(raw_inode);
4886 ei->i_disksize = inode->i_size;
4888 ei->i_reserved_quota = 0;
4890 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4891 ei->i_block_group = iloc.block_group;
4892 ei->i_last_alloc_group = ~0;
4894 * NOTE! The in-memory inode i_data array is in little-endian order
4895 * even on big-endian machines: we do NOT byteswap the block numbers!
4897 for (block = 0; block < EXT4_N_BLOCKS; block++)
4898 ei->i_data[block] = raw_inode->i_block[block];
4899 INIT_LIST_HEAD(&ei->i_orphan);
4902 * Set transaction id's of transactions that have to be committed
4903 * to finish f[data]sync. We set them to currently running transaction
4904 * as we cannot be sure that the inode or some of its metadata isn't
4905 * part of the transaction - the inode could have been reclaimed and
4906 * now it is reread from disk.
4909 transaction_t *transaction;
4912 read_lock(&journal->j_state_lock);
4913 if (journal->j_running_transaction)
4914 transaction = journal->j_running_transaction;
4916 transaction = journal->j_committing_transaction;
4918 tid = transaction->t_tid;
4920 tid = journal->j_commit_sequence;
4921 read_unlock(&journal->j_state_lock);
4922 ei->i_sync_tid = tid;
4923 ei->i_datasync_tid = tid;
4926 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4927 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4928 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4929 EXT4_INODE_SIZE(inode->i_sb)) {
4933 if (ei->i_extra_isize == 0) {
4934 /* The extra space is currently unused. Use it. */
4935 ei->i_extra_isize = sizeof(struct ext4_inode) -
4936 EXT4_GOOD_OLD_INODE_SIZE;
4938 __le32 *magic = (void *)raw_inode +
4939 EXT4_GOOD_OLD_INODE_SIZE +
4941 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4942 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4945 ei->i_extra_isize = 0;
4947 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4948 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4949 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4950 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4952 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4953 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4954 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4956 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4960 if (ei->i_file_acl &&
4961 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4962 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4966 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4967 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4968 (S_ISLNK(inode->i_mode) &&
4969 !ext4_inode_is_fast_symlink(inode)))
4970 /* Validate extent which is part of inode */
4971 ret = ext4_ext_check_inode(inode);
4972 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4973 (S_ISLNK(inode->i_mode) &&
4974 !ext4_inode_is_fast_symlink(inode))) {
4975 /* Validate block references which are part of inode */
4976 ret = ext4_check_inode_blockref(inode);
4981 if (S_ISREG(inode->i_mode)) {
4982 inode->i_op = &ext4_file_inode_operations;
4983 inode->i_fop = &ext4_file_operations;
4984 ext4_set_aops(inode);
4985 } else if (S_ISDIR(inode->i_mode)) {
4986 inode->i_op = &ext4_dir_inode_operations;
4987 inode->i_fop = &ext4_dir_operations;
4988 } else if (S_ISLNK(inode->i_mode)) {
4989 if (ext4_inode_is_fast_symlink(inode)) {
4990 inode->i_op = &ext4_fast_symlink_inode_operations;
4991 nd_terminate_link(ei->i_data, inode->i_size,
4992 sizeof(ei->i_data) - 1);
4994 inode->i_op = &ext4_symlink_inode_operations;
4995 ext4_set_aops(inode);
4997 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4998 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4999 inode->i_op = &ext4_special_inode_operations;
5000 if (raw_inode->i_block[0])
5001 init_special_inode(inode, inode->i_mode,
5002 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5004 init_special_inode(inode, inode->i_mode,
5005 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5008 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5012 ext4_set_inode_flags(inode);
5013 unlock_new_inode(inode);
5019 return ERR_PTR(ret);
5022 static int ext4_inode_blocks_set(handle_t *handle,
5023 struct ext4_inode *raw_inode,
5024 struct ext4_inode_info *ei)
5026 struct inode *inode = &(ei->vfs_inode);
5027 u64 i_blocks = inode->i_blocks;
5028 struct super_block *sb = inode->i_sb;
5030 if (i_blocks <= ~0U) {
5032 * i_blocks can be represnted in a 32 bit variable
5033 * as multiple of 512 bytes
5035 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5036 raw_inode->i_blocks_high = 0;
5037 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5040 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5043 if (i_blocks <= 0xffffffffffffULL) {
5045 * i_blocks can be represented in a 48 bit variable
5046 * as multiple of 512 bytes
5048 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5049 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5050 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5052 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5053 /* i_block is stored in file system block size */
5054 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5055 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5056 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5062 * Post the struct inode info into an on-disk inode location in the
5063 * buffer-cache. This gobbles the caller's reference to the
5064 * buffer_head in the inode location struct.
5066 * The caller must have write access to iloc->bh.
5068 static int ext4_do_update_inode(handle_t *handle,
5069 struct inode *inode,
5070 struct ext4_iloc *iloc)
5072 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5073 struct ext4_inode_info *ei = EXT4_I(inode);
5074 struct buffer_head *bh = iloc->bh;
5075 int err = 0, rc, block;
5077 /* For fields not not tracking in the in-memory inode,
5078 * initialise them to zero for new inodes. */
5079 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5080 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5082 ext4_get_inode_flags(ei);
5083 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5084 if (!(test_opt(inode->i_sb, NO_UID32))) {
5085 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5086 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5088 * Fix up interoperability with old kernels. Otherwise, old inodes get
5089 * re-used with the upper 16 bits of the uid/gid intact
5092 raw_inode->i_uid_high =
5093 cpu_to_le16(high_16_bits(inode->i_uid));
5094 raw_inode->i_gid_high =
5095 cpu_to_le16(high_16_bits(inode->i_gid));
5097 raw_inode->i_uid_high = 0;
5098 raw_inode->i_gid_high = 0;
5101 raw_inode->i_uid_low =
5102 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5103 raw_inode->i_gid_low =
5104 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5105 raw_inode->i_uid_high = 0;
5106 raw_inode->i_gid_high = 0;
5108 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5110 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5111 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5112 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5113 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5115 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5117 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5118 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5119 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5120 cpu_to_le32(EXT4_OS_HURD))
5121 raw_inode->i_file_acl_high =
5122 cpu_to_le16(ei->i_file_acl >> 32);
5123 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5124 ext4_isize_set(raw_inode, ei->i_disksize);
5125 if (ei->i_disksize > 0x7fffffffULL) {
5126 struct super_block *sb = inode->i_sb;
5127 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5128 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5129 EXT4_SB(sb)->s_es->s_rev_level ==
5130 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5131 /* If this is the first large file
5132 * created, add a flag to the superblock.
5134 err = ext4_journal_get_write_access(handle,
5135 EXT4_SB(sb)->s_sbh);
5138 ext4_update_dynamic_rev(sb);
5139 EXT4_SET_RO_COMPAT_FEATURE(sb,
5140 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5142 ext4_handle_sync(handle);
5143 err = ext4_handle_dirty_metadata(handle, NULL,
5144 EXT4_SB(sb)->s_sbh);
5147 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5148 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5149 if (old_valid_dev(inode->i_rdev)) {
5150 raw_inode->i_block[0] =
5151 cpu_to_le32(old_encode_dev(inode->i_rdev));
5152 raw_inode->i_block[1] = 0;
5154 raw_inode->i_block[0] = 0;
5155 raw_inode->i_block[1] =
5156 cpu_to_le32(new_encode_dev(inode->i_rdev));
5157 raw_inode->i_block[2] = 0;
5160 for (block = 0; block < EXT4_N_BLOCKS; block++)
5161 raw_inode->i_block[block] = ei->i_data[block];
5163 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5164 if (ei->i_extra_isize) {
5165 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5166 raw_inode->i_version_hi =
5167 cpu_to_le32(inode->i_version >> 32);
5168 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5171 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5172 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5175 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5177 ext4_update_inode_fsync_trans(handle, inode, 0);
5180 ext4_std_error(inode->i_sb, err);
5185 * ext4_write_inode()
5187 * We are called from a few places:
5189 * - Within generic_file_write() for O_SYNC files.
5190 * Here, there will be no transaction running. We wait for any running
5191 * trasnaction to commit.
5193 * - Within sys_sync(), kupdate and such.
5194 * We wait on commit, if tol to.
5196 * - Within prune_icache() (PF_MEMALLOC == true)
5197 * Here we simply return. We can't afford to block kswapd on the
5200 * In all cases it is actually safe for us to return without doing anything,
5201 * because the inode has been copied into a raw inode buffer in
5202 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5205 * Note that we are absolutely dependent upon all inode dirtiers doing the
5206 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5207 * which we are interested.
5209 * It would be a bug for them to not do this. The code:
5211 * mark_inode_dirty(inode)
5213 * inode->i_size = expr;
5215 * is in error because a kswapd-driven write_inode() could occur while
5216 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5217 * will no longer be on the superblock's dirty inode list.
5219 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5223 if (current->flags & PF_MEMALLOC)
5226 if (EXT4_SB(inode->i_sb)->s_journal) {
5227 if (ext4_journal_current_handle()) {
5228 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5233 if (wbc->sync_mode != WB_SYNC_ALL)
5236 err = ext4_force_commit(inode->i_sb);
5238 struct ext4_iloc iloc;
5240 err = __ext4_get_inode_loc(inode, &iloc, 0);
5243 if (wbc->sync_mode == WB_SYNC_ALL)
5244 sync_dirty_buffer(iloc.bh);
5245 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5246 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5247 "IO error syncing inode");
5258 * Called from notify_change.
5260 * We want to trap VFS attempts to truncate the file as soon as
5261 * possible. In particular, we want to make sure that when the VFS
5262 * shrinks i_size, we put the inode on the orphan list and modify
5263 * i_disksize immediately, so that during the subsequent flushing of
5264 * dirty pages and freeing of disk blocks, we can guarantee that any
5265 * commit will leave the blocks being flushed in an unused state on
5266 * disk. (On recovery, the inode will get truncated and the blocks will
5267 * be freed, so we have a strong guarantee that no future commit will
5268 * leave these blocks visible to the user.)
5270 * Another thing we have to assure is that if we are in ordered mode
5271 * and inode is still attached to the committing transaction, we must
5272 * we start writeout of all the dirty pages which are being truncated.
5273 * This way we are sure that all the data written in the previous
5274 * transaction are already on disk (truncate waits for pages under
5277 * Called with inode->i_mutex down.
5279 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5281 struct inode *inode = dentry->d_inode;
5284 const unsigned int ia_valid = attr->ia_valid;
5286 error = inode_change_ok(inode, attr);
5290 if (is_quota_modification(inode, attr))
5291 dquot_initialize(inode);
5292 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5293 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5296 /* (user+group)*(old+new) structure, inode write (sb,
5297 * inode block, ? - but truncate inode update has it) */
5298 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5299 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5300 if (IS_ERR(handle)) {
5301 error = PTR_ERR(handle);
5304 error = dquot_transfer(inode, attr);
5306 ext4_journal_stop(handle);
5309 /* Update corresponding info in inode so that everything is in
5310 * one transaction */
5311 if (attr->ia_valid & ATTR_UID)
5312 inode->i_uid = attr->ia_uid;
5313 if (attr->ia_valid & ATTR_GID)
5314 inode->i_gid = attr->ia_gid;
5315 error = ext4_mark_inode_dirty(handle, inode);
5316 ext4_journal_stop(handle);
5319 if (attr->ia_valid & ATTR_SIZE) {
5320 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5321 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5323 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5328 if (S_ISREG(inode->i_mode) &&
5329 attr->ia_valid & ATTR_SIZE &&
5330 (attr->ia_size < inode->i_size ||
5331 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5334 handle = ext4_journal_start(inode, 3);
5335 if (IS_ERR(handle)) {
5336 error = PTR_ERR(handle);
5339 if (ext4_handle_valid(handle)) {
5340 error = ext4_orphan_add(handle, inode);
5343 EXT4_I(inode)->i_disksize = attr->ia_size;
5344 rc = ext4_mark_inode_dirty(handle, inode);
5347 ext4_journal_stop(handle);
5349 if (ext4_should_order_data(inode)) {
5350 error = ext4_begin_ordered_truncate(inode,
5353 /* Do as much error cleanup as possible */
5354 handle = ext4_journal_start(inode, 3);
5355 if (IS_ERR(handle)) {
5356 ext4_orphan_del(NULL, inode);
5359 ext4_orphan_del(handle, inode);
5361 ext4_journal_stop(handle);
5365 /* ext4_truncate will clear the flag */
5366 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5367 ext4_truncate(inode);
5370 if ((attr->ia_valid & ATTR_SIZE) &&
5371 attr->ia_size != i_size_read(inode))
5372 rc = vmtruncate(inode, attr->ia_size);
5375 setattr_copy(inode, attr);
5376 mark_inode_dirty(inode);
5380 * If the call to ext4_truncate failed to get a transaction handle at
5381 * all, we need to clean up the in-core orphan list manually.
5383 if (orphan && inode->i_nlink)
5384 ext4_orphan_del(NULL, inode);
5386 if (!rc && (ia_valid & ATTR_MODE))
5387 rc = ext4_acl_chmod(inode);
5390 ext4_std_error(inode->i_sb, error);
5396 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5399 struct inode *inode;
5400 unsigned long delalloc_blocks;
5402 inode = dentry->d_inode;
5403 generic_fillattr(inode, stat);
5406 * We can't update i_blocks if the block allocation is delayed
5407 * otherwise in the case of system crash before the real block
5408 * allocation is done, we will have i_blocks inconsistent with
5409 * on-disk file blocks.
5410 * We always keep i_blocks updated together with real
5411 * allocation. But to not confuse with user, stat
5412 * will return the blocks that include the delayed allocation
5413 * blocks for this file.
5415 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5417 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5421 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5426 /* if nrblocks are contiguous */
5429 * With N contiguous data blocks, it need at most
5430 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5431 * 2 dindirect blocks
5434 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5435 return indirects + 3;
5438 * if nrblocks are not contiguous, worse case, each block touch
5439 * a indirect block, and each indirect block touch a double indirect
5440 * block, plus a triple indirect block
5442 indirects = nrblocks * 2 + 1;
5446 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5448 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5449 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5450 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5454 * Account for index blocks, block groups bitmaps and block group
5455 * descriptor blocks if modify datablocks and index blocks
5456 * worse case, the indexs blocks spread over different block groups
5458 * If datablocks are discontiguous, they are possible to spread over
5459 * different block groups too. If they are contiuguous, with flexbg,
5460 * they could still across block group boundary.
5462 * Also account for superblock, inode, quota and xattr blocks
5464 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5466 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5472 * How many index blocks need to touch to modify nrblocks?
5473 * The "Chunk" flag indicating whether the nrblocks is
5474 * physically contiguous on disk
5476 * For Direct IO and fallocate, they calls get_block to allocate
5477 * one single extent at a time, so they could set the "Chunk" flag
5479 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5484 * Now let's see how many group bitmaps and group descriptors need
5494 if (groups > ngroups)
5496 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5497 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5499 /* bitmaps and block group descriptor blocks */
5500 ret += groups + gdpblocks;
5502 /* Blocks for super block, inode, quota and xattr blocks */
5503 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5509 * Calulate the total number of credits to reserve to fit
5510 * the modification of a single pages into a single transaction,
5511 * which may include multiple chunks of block allocations.
5513 * This could be called via ext4_write_begin()
5515 * We need to consider the worse case, when
5516 * one new block per extent.
5518 int ext4_writepage_trans_blocks(struct inode *inode)
5520 int bpp = ext4_journal_blocks_per_page(inode);
5523 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5525 /* Account for data blocks for journalled mode */
5526 if (ext4_should_journal_data(inode))
5532 * Calculate the journal credits for a chunk of data modification.
5534 * This is called from DIO, fallocate or whoever calling
5535 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5537 * journal buffers for data blocks are not included here, as DIO
5538 * and fallocate do no need to journal data buffers.
5540 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5542 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5546 * The caller must have previously called ext4_reserve_inode_write().
5547 * Give this, we know that the caller already has write access to iloc->bh.
5549 int ext4_mark_iloc_dirty(handle_t *handle,
5550 struct inode *inode, struct ext4_iloc *iloc)
5554 if (test_opt(inode->i_sb, I_VERSION))
5555 inode_inc_iversion(inode);
5557 /* the do_update_inode consumes one bh->b_count */
5560 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5561 err = ext4_do_update_inode(handle, inode, iloc);
5567 * On success, We end up with an outstanding reference count against
5568 * iloc->bh. This _must_ be cleaned up later.
5572 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5573 struct ext4_iloc *iloc)
5577 err = ext4_get_inode_loc(inode, iloc);
5579 BUFFER_TRACE(iloc->bh, "get_write_access");
5580 err = ext4_journal_get_write_access(handle, iloc->bh);
5586 ext4_std_error(inode->i_sb, err);
5591 * Expand an inode by new_extra_isize bytes.
5592 * Returns 0 on success or negative error number on failure.
5594 static int ext4_expand_extra_isize(struct inode *inode,
5595 unsigned int new_extra_isize,
5596 struct ext4_iloc iloc,
5599 struct ext4_inode *raw_inode;
5600 struct ext4_xattr_ibody_header *header;
5602 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5605 raw_inode = ext4_raw_inode(&iloc);
5607 header = IHDR(inode, raw_inode);
5609 /* No extended attributes present */
5610 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5611 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5612 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5614 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5618 /* try to expand with EAs present */
5619 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5624 * What we do here is to mark the in-core inode as clean with respect to inode
5625 * dirtiness (it may still be data-dirty).
5626 * This means that the in-core inode may be reaped by prune_icache
5627 * without having to perform any I/O. This is a very good thing,
5628 * because *any* task may call prune_icache - even ones which
5629 * have a transaction open against a different journal.
5631 * Is this cheating? Not really. Sure, we haven't written the
5632 * inode out, but prune_icache isn't a user-visible syncing function.
5633 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5634 * we start and wait on commits.
5636 * Is this efficient/effective? Well, we're being nice to the system
5637 * by cleaning up our inodes proactively so they can be reaped
5638 * without I/O. But we are potentially leaving up to five seconds'
5639 * worth of inodes floating about which prune_icache wants us to
5640 * write out. One way to fix that would be to get prune_icache()
5641 * to do a write_super() to free up some memory. It has the desired
5644 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5646 struct ext4_iloc iloc;
5647 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5648 static unsigned int mnt_count;
5652 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5653 err = ext4_reserve_inode_write(handle, inode, &iloc);
5654 if (ext4_handle_valid(handle) &&
5655 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5656 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5658 * We need extra buffer credits since we may write into EA block
5659 * with this same handle. If journal_extend fails, then it will
5660 * only result in a minor loss of functionality for that inode.
5661 * If this is felt to be critical, then e2fsck should be run to
5662 * force a large enough s_min_extra_isize.
5664 if ((jbd2_journal_extend(handle,
5665 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5666 ret = ext4_expand_extra_isize(inode,
5667 sbi->s_want_extra_isize,
5670 ext4_set_inode_state(inode,
5671 EXT4_STATE_NO_EXPAND);
5673 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5674 ext4_warning(inode->i_sb,
5675 "Unable to expand inode %lu. Delete"
5676 " some EAs or run e2fsck.",
5679 le16_to_cpu(sbi->s_es->s_mnt_count);
5685 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5690 * ext4_dirty_inode() is called from __mark_inode_dirty()
5692 * We're really interested in the case where a file is being extended.
5693 * i_size has been changed by generic_commit_write() and we thus need
5694 * to include the updated inode in the current transaction.
5696 * Also, dquot_alloc_block() will always dirty the inode when blocks
5697 * are allocated to the file.
5699 * If the inode is marked synchronous, we don't honour that here - doing
5700 * so would cause a commit on atime updates, which we don't bother doing.
5701 * We handle synchronous inodes at the highest possible level.
5703 void ext4_dirty_inode(struct inode *inode)
5707 handle = ext4_journal_start(inode, 2);
5711 ext4_mark_inode_dirty(handle, inode);
5713 ext4_journal_stop(handle);
5720 * Bind an inode's backing buffer_head into this transaction, to prevent
5721 * it from being flushed to disk early. Unlike
5722 * ext4_reserve_inode_write, this leaves behind no bh reference and
5723 * returns no iloc structure, so the caller needs to repeat the iloc
5724 * lookup to mark the inode dirty later.
5726 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5728 struct ext4_iloc iloc;
5732 err = ext4_get_inode_loc(inode, &iloc);
5734 BUFFER_TRACE(iloc.bh, "get_write_access");
5735 err = jbd2_journal_get_write_access(handle, iloc.bh);
5737 err = ext4_handle_dirty_metadata(handle,
5743 ext4_std_error(inode->i_sb, err);
5748 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5755 * We have to be very careful here: changing a data block's
5756 * journaling status dynamically is dangerous. If we write a
5757 * data block to the journal, change the status and then delete
5758 * that block, we risk forgetting to revoke the old log record
5759 * from the journal and so a subsequent replay can corrupt data.
5760 * So, first we make sure that the journal is empty and that
5761 * nobody is changing anything.
5764 journal = EXT4_JOURNAL(inode);
5767 if (is_journal_aborted(journal))
5770 jbd2_journal_lock_updates(journal);
5771 jbd2_journal_flush(journal);
5774 * OK, there are no updates running now, and all cached data is
5775 * synced to disk. We are now in a completely consistent state
5776 * which doesn't have anything in the journal, and we know that
5777 * no filesystem updates are running, so it is safe to modify
5778 * the inode's in-core data-journaling state flag now.
5782 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5784 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5785 ext4_set_aops(inode);
5787 jbd2_journal_unlock_updates(journal);
5789 /* Finally we can mark the inode as dirty. */
5791 handle = ext4_journal_start(inode, 1);
5793 return PTR_ERR(handle);
5795 err = ext4_mark_inode_dirty(handle, inode);
5796 ext4_handle_sync(handle);
5797 ext4_journal_stop(handle);
5798 ext4_std_error(inode->i_sb, err);
5803 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5805 return !buffer_mapped(bh);
5808 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5810 struct page *page = vmf->page;
5815 struct file *file = vma->vm_file;
5816 struct inode *inode = file->f_path.dentry->d_inode;
5817 struct address_space *mapping = inode->i_mapping;
5820 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5821 * get i_mutex because we are already holding mmap_sem.
5823 down_read(&inode->i_alloc_sem);
5824 size = i_size_read(inode);
5825 if (page->mapping != mapping || size <= page_offset(page)
5826 || !PageUptodate(page)) {
5827 /* page got truncated from under us? */
5831 if (PageMappedToDisk(page))
5834 if (page->index == size >> PAGE_CACHE_SHIFT)
5835 len = size & ~PAGE_CACHE_MASK;
5837 len = PAGE_CACHE_SIZE;
5841 * return if we have all the buffers mapped. This avoid
5842 * the need to call write_begin/write_end which does a
5843 * journal_start/journal_stop which can block and take
5846 if (page_has_buffers(page)) {
5847 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5848 ext4_bh_unmapped)) {
5855 * OK, we need to fill the hole... Do write_begin write_end
5856 * to do block allocation/reservation.We are not holding
5857 * inode.i__mutex here. That allow * parallel write_begin,
5858 * write_end call. lock_page prevent this from happening
5859 * on the same page though
5861 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5862 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5865 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5866 len, len, page, fsdata);
5872 ret = VM_FAULT_SIGBUS;
5873 up_read(&inode->i_alloc_sem);