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/printk.h>
43 #include <linux/slab.h>
44 #include <linux/ratelimit.h>
46 #include "ext4_jbd2.h"
49 #include "ext4_extents.h"
51 #include <trace/events/ext4.h>
53 #define MPAGE_DA_EXTENT_TAIL 0x01
55 static inline int ext4_begin_ordered_truncate(struct inode *inode,
58 trace_ext4_begin_ordered_truncate(inode, new_size);
60 * If jinode is zero, then we never opened the file for
61 * writing, so there's no need to call
62 * jbd2_journal_begin_ordered_truncate() since there's no
63 * outstanding writes we need to flush.
65 if (!EXT4_I(inode)->jinode)
67 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
68 EXT4_I(inode)->jinode,
72 static void ext4_invalidatepage(struct page *page, unsigned long offset);
73 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
74 struct buffer_head *bh_result, int create);
75 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
76 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
77 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
78 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
81 * Test whether an inode is a fast symlink.
83 static int ext4_inode_is_fast_symlink(struct inode *inode)
85 int ea_blocks = EXT4_I(inode)->i_file_acl ?
86 (inode->i_sb->s_blocksize >> 9) : 0;
88 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
92 * Work out how many blocks we need to proceed with the next chunk of a
93 * truncate transaction.
95 static unsigned long blocks_for_truncate(struct inode *inode)
99 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
101 /* Give ourselves just enough room to cope with inodes in which
102 * i_blocks is corrupt: we've seen disk corruptions in the past
103 * which resulted in random data in an inode which looked enough
104 * like a regular file for ext4 to try to delete it. Things
105 * will go a bit crazy if that happens, but at least we should
106 * try not to panic the whole kernel. */
110 /* But we need to bound the transaction so we don't overflow the
112 if (needed > EXT4_MAX_TRANS_DATA)
113 needed = EXT4_MAX_TRANS_DATA;
115 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
119 * Truncate transactions can be complex and absolutely huge. So we need to
120 * be able to restart the transaction at a conventient checkpoint to make
121 * sure we don't overflow the journal.
123 * start_transaction gets us a new handle for a truncate transaction,
124 * and extend_transaction tries to extend the existing one a bit. If
125 * extend fails, we need to propagate the failure up and restart the
126 * transaction in the top-level truncate loop. --sct
128 static handle_t *start_transaction(struct inode *inode)
132 result = ext4_journal_start(inode, blocks_for_truncate(inode));
136 ext4_std_error(inode->i_sb, PTR_ERR(result));
141 * Try to extend this transaction for the purposes of truncation.
143 * Returns 0 if we managed to create more room. If we can't create more
144 * room, and the transaction must be restarted we return 1.
146 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
148 if (!ext4_handle_valid(handle))
150 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
152 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
162 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
174 jbd_debug(2, "restarting handle %p\n", handle);
175 up_write(&EXT4_I(inode)->i_data_sem);
176 ret = ext4_journal_restart(handle, nblocks);
177 down_write(&EXT4_I(inode)->i_data_sem);
178 ext4_discard_preallocations(inode);
184 * Called at the last iput() if i_nlink is zero.
186 void ext4_evict_inode(struct inode *inode)
191 trace_ext4_evict_inode(inode);
193 mutex_lock(&inode->i_mutex);
194 ext4_flush_completed_IO(inode);
195 mutex_unlock(&inode->i_mutex);
196 ext4_ioend_wait(inode);
198 if (inode->i_nlink) {
199 truncate_inode_pages(&inode->i_data, 0);
203 if (!is_bad_inode(inode))
204 dquot_initialize(inode);
206 if (ext4_should_order_data(inode))
207 ext4_begin_ordered_truncate(inode, 0);
208 truncate_inode_pages(&inode->i_data, 0);
210 if (is_bad_inode(inode))
213 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
214 if (IS_ERR(handle)) {
215 ext4_std_error(inode->i_sb, PTR_ERR(handle));
217 * If we're going to skip the normal cleanup, we still need to
218 * make sure that the in-core orphan linked list is properly
221 ext4_orphan_del(NULL, inode);
226 ext4_handle_sync(handle);
228 err = ext4_mark_inode_dirty(handle, inode);
230 ext4_warning(inode->i_sb,
231 "couldn't mark inode dirty (err %d)", err);
235 ext4_truncate(inode);
238 * ext4_ext_truncate() doesn't reserve any slop when it
239 * restarts journal transactions; therefore there may not be
240 * enough credits left in the handle to remove the inode from
241 * the orphan list and set the dtime field.
243 if (!ext4_handle_has_enough_credits(handle, 3)) {
244 err = ext4_journal_extend(handle, 3);
246 err = ext4_journal_restart(handle, 3);
248 ext4_warning(inode->i_sb,
249 "couldn't extend journal (err %d)", err);
251 ext4_journal_stop(handle);
252 ext4_orphan_del(NULL, inode);
258 * Kill off the orphan record which ext4_truncate created.
259 * AKPM: I think this can be inside the above `if'.
260 * Note that ext4_orphan_del() has to be able to cope with the
261 * deletion of a non-existent orphan - this is because we don't
262 * know if ext4_truncate() actually created an orphan record.
263 * (Well, we could do this if we need to, but heck - it works)
265 ext4_orphan_del(handle, inode);
266 EXT4_I(inode)->i_dtime = get_seconds();
269 * One subtle ordering requirement: if anything has gone wrong
270 * (transaction abort, IO errors, whatever), then we can still
271 * do these next steps (the fs will already have been marked as
272 * having errors), but we can't free the inode if the mark_dirty
275 if (ext4_mark_inode_dirty(handle, inode))
276 /* If that failed, just do the required in-core inode clear. */
277 ext4_clear_inode(inode);
279 ext4_free_inode(handle, inode);
280 ext4_journal_stop(handle);
283 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
289 struct buffer_head *bh;
292 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
294 p->key = *(p->p = v);
299 * ext4_block_to_path - parse the block number into array of offsets
300 * @inode: inode in question (we are only interested in its superblock)
301 * @i_block: block number to be parsed
302 * @offsets: array to store the offsets in
303 * @boundary: set this non-zero if the referred-to block is likely to be
304 * followed (on disk) by an indirect block.
306 * To store the locations of file's data ext4 uses a data structure common
307 * for UNIX filesystems - tree of pointers anchored in the inode, with
308 * data blocks at leaves and indirect blocks in intermediate nodes.
309 * This function translates the block number into path in that tree -
310 * return value is the path length and @offsets[n] is the offset of
311 * pointer to (n+1)th node in the nth one. If @block is out of range
312 * (negative or too large) warning is printed and zero returned.
314 * Note: function doesn't find node addresses, so no IO is needed. All
315 * we need to know is the capacity of indirect blocks (taken from the
320 * Portability note: the last comparison (check that we fit into triple
321 * indirect block) is spelled differently, because otherwise on an
322 * architecture with 32-bit longs and 8Kb pages we might get into trouble
323 * if our filesystem had 8Kb blocks. We might use long long, but that would
324 * kill us on x86. Oh, well, at least the sign propagation does not matter -
325 * i_block would have to be negative in the very beginning, so we would not
329 static int ext4_block_to_path(struct inode *inode,
331 ext4_lblk_t offsets[4], int *boundary)
333 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
334 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
335 const long direct_blocks = EXT4_NDIR_BLOCKS,
336 indirect_blocks = ptrs,
337 double_blocks = (1 << (ptrs_bits * 2));
341 if (i_block < direct_blocks) {
342 offsets[n++] = i_block;
343 final = direct_blocks;
344 } else if ((i_block -= direct_blocks) < indirect_blocks) {
345 offsets[n++] = EXT4_IND_BLOCK;
346 offsets[n++] = i_block;
348 } else if ((i_block -= indirect_blocks) < double_blocks) {
349 offsets[n++] = EXT4_DIND_BLOCK;
350 offsets[n++] = i_block >> ptrs_bits;
351 offsets[n++] = i_block & (ptrs - 1);
353 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
354 offsets[n++] = EXT4_TIND_BLOCK;
355 offsets[n++] = i_block >> (ptrs_bits * 2);
356 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
357 offsets[n++] = i_block & (ptrs - 1);
360 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
361 i_block + direct_blocks +
362 indirect_blocks + double_blocks, inode->i_ino);
365 *boundary = final - 1 - (i_block & (ptrs - 1));
369 static int __ext4_check_blockref(const char *function, unsigned int line,
371 __le32 *p, unsigned int max)
373 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
377 while (bref < p+max) {
378 blk = le32_to_cpu(*bref++);
380 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
382 es->s_last_error_block = cpu_to_le64(blk);
383 ext4_error_inode(inode, function, line, blk,
392 #define ext4_check_indirect_blockref(inode, bh) \
393 __ext4_check_blockref(__func__, __LINE__, inode, \
394 (__le32 *)(bh)->b_data, \
395 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
397 #define ext4_check_inode_blockref(inode) \
398 __ext4_check_blockref(__func__, __LINE__, inode, \
399 EXT4_I(inode)->i_data, \
403 * ext4_get_branch - read the chain of indirect blocks leading to data
404 * @inode: inode in question
405 * @depth: depth of the chain (1 - direct pointer, etc.)
406 * @offsets: offsets of pointers in inode/indirect blocks
407 * @chain: place to store the result
408 * @err: here we store the error value
410 * Function fills the array of triples <key, p, bh> and returns %NULL
411 * if everything went OK or the pointer to the last filled triple
412 * (incomplete one) otherwise. Upon the return chain[i].key contains
413 * the number of (i+1)-th block in the chain (as it is stored in memory,
414 * i.e. little-endian 32-bit), chain[i].p contains the address of that
415 * number (it points into struct inode for i==0 and into the bh->b_data
416 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
417 * block for i>0 and NULL for i==0. In other words, it holds the block
418 * numbers of the chain, addresses they were taken from (and where we can
419 * verify that chain did not change) and buffer_heads hosting these
422 * Function stops when it stumbles upon zero pointer (absent block)
423 * (pointer to last triple returned, *@err == 0)
424 * or when it gets an IO error reading an indirect block
425 * (ditto, *@err == -EIO)
426 * or when it reads all @depth-1 indirect blocks successfully and finds
427 * the whole chain, all way to the data (returns %NULL, *err == 0).
429 * Need to be called with
430 * down_read(&EXT4_I(inode)->i_data_sem)
432 static Indirect *ext4_get_branch(struct inode *inode, int depth,
433 ext4_lblk_t *offsets,
434 Indirect chain[4], int *err)
436 struct super_block *sb = inode->i_sb;
438 struct buffer_head *bh;
441 /* i_data is not going away, no lock needed */
442 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
446 bh = sb_getblk(sb, le32_to_cpu(p->key));
450 if (!bh_uptodate_or_lock(bh)) {
451 if (bh_submit_read(bh) < 0) {
455 /* validate block references */
456 if (ext4_check_indirect_blockref(inode, bh)) {
462 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
476 * ext4_find_near - find a place for allocation with sufficient locality
478 * @ind: descriptor of indirect block.
480 * This function returns the preferred place for block allocation.
481 * It is used when heuristic for sequential allocation fails.
483 * + if there is a block to the left of our position - allocate near it.
484 * + if pointer will live in indirect block - allocate near that block.
485 * + if pointer will live in inode - allocate in the same
488 * In the latter case we colour the starting block by the callers PID to
489 * prevent it from clashing with concurrent allocations for a different inode
490 * in the same block group. The PID is used here so that functionally related
491 * files will be close-by on-disk.
493 * Caller must make sure that @ind is valid and will stay that way.
495 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
497 struct ext4_inode_info *ei = EXT4_I(inode);
498 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
500 ext4_fsblk_t bg_start;
501 ext4_fsblk_t last_block;
502 ext4_grpblk_t colour;
503 ext4_group_t block_group;
504 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
506 /* Try to find previous block */
507 for (p = ind->p - 1; p >= start; p--) {
509 return le32_to_cpu(*p);
512 /* No such thing, so let's try location of indirect block */
514 return ind->bh->b_blocknr;
517 * It is going to be referred to from the inode itself? OK, just put it
518 * into the same cylinder group then.
520 block_group = ei->i_block_group;
521 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
522 block_group &= ~(flex_size-1);
523 if (S_ISREG(inode->i_mode))
526 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
527 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
530 * If we are doing delayed allocation, we don't need take
531 * colour into account.
533 if (test_opt(inode->i_sb, DELALLOC))
536 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
537 colour = (current->pid % 16) *
538 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
540 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
541 return bg_start + colour;
545 * ext4_find_goal - find a preferred place for allocation.
547 * @block: block we want
548 * @partial: pointer to the last triple within a chain
550 * Normally this function find the preferred place for block allocation,
552 * Because this is only used for non-extent files, we limit the block nr
555 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
561 * XXX need to get goal block from mballoc's data structures
564 goal = ext4_find_near(inode, partial);
565 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
570 * ext4_blks_to_allocate - Look up the block map and count the number
571 * of direct blocks need to be allocated for the given branch.
573 * @branch: chain of indirect blocks
574 * @k: number of blocks need for indirect blocks
575 * @blks: number of data blocks to be mapped.
576 * @blocks_to_boundary: the offset in the indirect block
578 * return the total number of blocks to be allocate, including the
579 * direct and indirect blocks.
581 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
582 int blocks_to_boundary)
584 unsigned int count = 0;
587 * Simple case, [t,d]Indirect block(s) has not allocated yet
588 * then it's clear blocks on that path have not allocated
591 /* right now we don't handle cross boundary allocation */
592 if (blks < blocks_to_boundary + 1)
595 count += blocks_to_boundary + 1;
600 while (count < blks && count <= blocks_to_boundary &&
601 le32_to_cpu(*(branch[0].p + count)) == 0) {
608 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
609 * @handle: handle for this transaction
610 * @inode: inode which needs allocated blocks
611 * @iblock: the logical block to start allocated at
612 * @goal: preferred physical block of allocation
613 * @indirect_blks: the number of blocks need to allocate for indirect
615 * @blks: number of desired blocks
616 * @new_blocks: on return it will store the new block numbers for
617 * the indirect blocks(if needed) and the first direct block,
618 * @err: on return it will store the error code
620 * This function will return the number of blocks allocated as
621 * requested by the passed-in parameters.
623 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
624 ext4_lblk_t iblock, ext4_fsblk_t goal,
625 int indirect_blks, int blks,
626 ext4_fsblk_t new_blocks[4], int *err)
628 struct ext4_allocation_request ar;
630 unsigned long count = 0, blk_allocated = 0;
632 ext4_fsblk_t current_block = 0;
636 * Here we try to allocate the requested multiple blocks at once,
637 * on a best-effort basis.
638 * To build a branch, we should allocate blocks for
639 * the indirect blocks(if not allocated yet), and at least
640 * the first direct block of this branch. That's the
641 * minimum number of blocks need to allocate(required)
643 /* first we try to allocate the indirect blocks */
644 target = indirect_blks;
647 /* allocating blocks for indirect blocks and direct blocks */
648 current_block = ext4_new_meta_blocks(handle, inode, goal,
653 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
654 EXT4_ERROR_INODE(inode,
655 "current_block %llu + count %lu > %d!",
656 current_block, count,
657 EXT4_MAX_BLOCK_FILE_PHYS);
663 /* allocate blocks for indirect blocks */
664 while (index < indirect_blks && count) {
665 new_blocks[index++] = current_block++;
670 * save the new block number
671 * for the first direct block
673 new_blocks[index] = current_block;
674 printk(KERN_INFO "%s returned more blocks than "
675 "requested\n", __func__);
681 target = blks - count ;
682 blk_allocated = count;
685 /* Now allocate data blocks */
686 memset(&ar, 0, sizeof(ar));
691 if (S_ISREG(inode->i_mode))
692 /* enable in-core preallocation only for regular files */
693 ar.flags = EXT4_MB_HINT_DATA;
695 current_block = ext4_mb_new_blocks(handle, &ar, err);
696 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
697 EXT4_ERROR_INODE(inode,
698 "current_block %llu + ar.len %d > %d!",
699 current_block, ar.len,
700 EXT4_MAX_BLOCK_FILE_PHYS);
705 if (*err && (target == blks)) {
707 * if the allocation failed and we didn't allocate
713 if (target == blks) {
715 * save the new block number
716 * for the first direct block
718 new_blocks[index] = current_block;
720 blk_allocated += ar.len;
723 /* total number of blocks allocated for direct blocks */
728 for (i = 0; i < index; i++)
729 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
734 * ext4_alloc_branch - allocate and set up a chain of blocks.
735 * @handle: handle for this transaction
737 * @indirect_blks: number of allocated indirect blocks
738 * @blks: number of allocated direct blocks
739 * @goal: preferred place for allocation
740 * @offsets: offsets (in the blocks) to store the pointers to next.
741 * @branch: place to store the chain in.
743 * This function allocates blocks, zeroes out all but the last one,
744 * links them into chain and (if we are synchronous) writes them to disk.
745 * In other words, it prepares a branch that can be spliced onto the
746 * inode. It stores the information about that chain in the branch[], in
747 * the same format as ext4_get_branch() would do. We are calling it after
748 * we had read the existing part of chain and partial points to the last
749 * triple of that (one with zero ->key). Upon the exit we have the same
750 * picture as after the successful ext4_get_block(), except that in one
751 * place chain is disconnected - *branch->p is still zero (we did not
752 * set the last link), but branch->key contains the number that should
753 * be placed into *branch->p to fill that gap.
755 * If allocation fails we free all blocks we've allocated (and forget
756 * their buffer_heads) and return the error value the from failed
757 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
758 * as described above and return 0.
760 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
761 ext4_lblk_t iblock, int indirect_blks,
762 int *blks, ext4_fsblk_t goal,
763 ext4_lblk_t *offsets, Indirect *branch)
765 int blocksize = inode->i_sb->s_blocksize;
768 struct buffer_head *bh;
770 ext4_fsblk_t new_blocks[4];
771 ext4_fsblk_t current_block;
773 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
774 *blks, new_blocks, &err);
778 branch[0].key = cpu_to_le32(new_blocks[0]);
780 * metadata blocks and data blocks are allocated.
782 for (n = 1; n <= indirect_blks; n++) {
784 * Get buffer_head for parent block, zero it out
785 * and set the pointer to new one, then send
788 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
796 BUFFER_TRACE(bh, "call get_create_access");
797 err = ext4_journal_get_create_access(handle, bh);
799 /* Don't brelse(bh) here; it's done in
800 * ext4_journal_forget() below */
805 memset(bh->b_data, 0, blocksize);
806 branch[n].p = (__le32 *) bh->b_data + offsets[n];
807 branch[n].key = cpu_to_le32(new_blocks[n]);
808 *branch[n].p = branch[n].key;
809 if (n == indirect_blks) {
810 current_block = new_blocks[n];
812 * End of chain, update the last new metablock of
813 * the chain to point to the new allocated
814 * data blocks numbers
816 for (i = 1; i < num; i++)
817 *(branch[n].p + i) = cpu_to_le32(++current_block);
819 BUFFER_TRACE(bh, "marking uptodate");
820 set_buffer_uptodate(bh);
823 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
824 err = ext4_handle_dirty_metadata(handle, inode, bh);
831 /* Allocation failed, free what we already allocated */
832 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
833 for (i = 1; i <= n ; i++) {
835 * branch[i].bh is newly allocated, so there is no
836 * need to revoke the block, which is why we don't
837 * need to set EXT4_FREE_BLOCKS_METADATA.
839 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
840 EXT4_FREE_BLOCKS_FORGET);
842 for (i = n+1; i < indirect_blks; i++)
843 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
845 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
851 * ext4_splice_branch - splice the allocated branch onto inode.
852 * @handle: handle for this transaction
854 * @block: (logical) number of block we are adding
855 * @chain: chain of indirect blocks (with a missing link - see
857 * @where: location of missing link
858 * @num: number of indirect blocks we are adding
859 * @blks: number of direct blocks we are adding
861 * This function fills the missing link and does all housekeeping needed in
862 * inode (->i_blocks, etc.). In case of success we end up with the full
863 * chain to new block and return 0.
865 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
866 ext4_lblk_t block, Indirect *where, int num,
871 ext4_fsblk_t current_block;
874 * If we're splicing into a [td]indirect block (as opposed to the
875 * inode) then we need to get write access to the [td]indirect block
879 BUFFER_TRACE(where->bh, "get_write_access");
880 err = ext4_journal_get_write_access(handle, where->bh);
886 *where->p = where->key;
889 * Update the host buffer_head or inode to point to more just allocated
890 * direct blocks blocks
892 if (num == 0 && blks > 1) {
893 current_block = le32_to_cpu(where->key) + 1;
894 for (i = 1; i < blks; i++)
895 *(where->p + i) = cpu_to_le32(current_block++);
898 /* We are done with atomic stuff, now do the rest of housekeeping */
899 /* had we spliced it onto indirect block? */
902 * If we spliced it onto an indirect block, we haven't
903 * altered the inode. Note however that if it is being spliced
904 * onto an indirect block at the very end of the file (the
905 * file is growing) then we *will* alter the inode to reflect
906 * the new i_size. But that is not done here - it is done in
907 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
909 jbd_debug(5, "splicing indirect only\n");
910 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
911 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
916 * OK, we spliced it into the inode itself on a direct block.
918 ext4_mark_inode_dirty(handle, inode);
919 jbd_debug(5, "splicing direct\n");
924 for (i = 1; i <= num; i++) {
926 * branch[i].bh is newly allocated, so there is no
927 * need to revoke the block, which is why we don't
928 * need to set EXT4_FREE_BLOCKS_METADATA.
930 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
931 EXT4_FREE_BLOCKS_FORGET);
933 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
940 * The ext4_ind_map_blocks() function handles non-extents inodes
941 * (i.e., using the traditional indirect/double-indirect i_blocks
942 * scheme) for ext4_map_blocks().
944 * Allocation strategy is simple: if we have to allocate something, we will
945 * have to go the whole way to leaf. So let's do it before attaching anything
946 * to tree, set linkage between the newborn blocks, write them if sync is
947 * required, recheck the path, free and repeat if check fails, otherwise
948 * set the last missing link (that will protect us from any truncate-generated
949 * removals - all blocks on the path are immune now) and possibly force the
950 * write on the parent block.
951 * That has a nice additional property: no special recovery from the failed
952 * allocations is needed - we simply release blocks and do not touch anything
953 * reachable from inode.
955 * `handle' can be NULL if create == 0.
957 * return > 0, # of blocks mapped or allocated.
958 * return = 0, if plain lookup failed.
959 * return < 0, error case.
961 * The ext4_ind_get_blocks() function should be called with
962 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
963 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
964 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
967 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
968 struct ext4_map_blocks *map,
972 ext4_lblk_t offsets[4];
977 int blocks_to_boundary = 0;
980 ext4_fsblk_t first_block = 0;
982 trace_ext4_ind_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
983 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
984 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
985 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
986 &blocks_to_boundary);
991 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
993 /* Simplest case - block found, no allocation needed */
995 first_block = le32_to_cpu(chain[depth - 1].key);
998 while (count < map->m_len && count <= blocks_to_boundary) {
1001 blk = le32_to_cpu(*(chain[depth-1].p + count));
1003 if (blk == first_block + count)
1011 /* Next simple case - plain lookup or failed read of indirect block */
1012 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
1016 * Okay, we need to do block allocation.
1018 goal = ext4_find_goal(inode, map->m_lblk, partial);
1020 /* the number of blocks need to allocate for [d,t]indirect blocks */
1021 indirect_blks = (chain + depth) - partial - 1;
1024 * Next look up the indirect map to count the totoal number of
1025 * direct blocks to allocate for this branch.
1027 count = ext4_blks_to_allocate(partial, indirect_blks,
1028 map->m_len, blocks_to_boundary);
1030 * Block out ext4_truncate while we alter the tree
1032 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1034 offsets + (partial - chain), partial);
1037 * The ext4_splice_branch call will free and forget any buffers
1038 * on the new chain if there is a failure, but that risks using
1039 * up transaction credits, especially for bitmaps where the
1040 * credits cannot be returned. Can we handle this somehow? We
1041 * may need to return -EAGAIN upwards in the worst case. --sct
1044 err = ext4_splice_branch(handle, inode, map->m_lblk,
1045 partial, indirect_blks, count);
1049 map->m_flags |= EXT4_MAP_NEW;
1051 ext4_update_inode_fsync_trans(handle, inode, 1);
1053 map->m_flags |= EXT4_MAP_MAPPED;
1054 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1056 if (count > blocks_to_boundary)
1057 map->m_flags |= EXT4_MAP_BOUNDARY;
1059 /* Clean up and exit */
1060 partial = chain + depth - 1; /* the whole chain */
1062 while (partial > chain) {
1063 BUFFER_TRACE(partial->bh, "call brelse");
1064 brelse(partial->bh);
1068 trace_ext4_ind_map_blocks_exit(inode, map->m_lblk,
1069 map->m_pblk, map->m_len, err);
1074 qsize_t *ext4_get_reserved_space(struct inode *inode)
1076 return &EXT4_I(inode)->i_reserved_quota;
1081 * Calculate the number of metadata blocks need to reserve
1082 * to allocate a new block at @lblocks for non extent file based file
1084 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1087 struct ext4_inode_info *ei = EXT4_I(inode);
1088 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1091 if (lblock < EXT4_NDIR_BLOCKS)
1094 lblock -= EXT4_NDIR_BLOCKS;
1096 if (ei->i_da_metadata_calc_len &&
1097 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1098 ei->i_da_metadata_calc_len++;
1101 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1102 ei->i_da_metadata_calc_len = 1;
1103 blk_bits = order_base_2(lblock);
1104 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1108 * Calculate the number of metadata blocks need to reserve
1109 * to allocate a block located at @lblock
1111 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
1113 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1114 return ext4_ext_calc_metadata_amount(inode, lblock);
1116 return ext4_indirect_calc_metadata_amount(inode, lblock);
1120 * Called with i_data_sem down, which is important since we can call
1121 * ext4_discard_preallocations() from here.
1123 void ext4_da_update_reserve_space(struct inode *inode,
1124 int used, int quota_claim)
1126 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1127 struct ext4_inode_info *ei = EXT4_I(inode);
1129 spin_lock(&ei->i_block_reservation_lock);
1130 trace_ext4_da_update_reserve_space(inode, used);
1131 if (unlikely(used > ei->i_reserved_data_blocks)) {
1132 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1133 "with only %d reserved data blocks\n",
1134 __func__, inode->i_ino, used,
1135 ei->i_reserved_data_blocks);
1137 used = ei->i_reserved_data_blocks;
1140 /* Update per-inode reservations */
1141 ei->i_reserved_data_blocks -= used;
1142 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1143 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1144 used + ei->i_allocated_meta_blocks);
1145 ei->i_allocated_meta_blocks = 0;
1147 if (ei->i_reserved_data_blocks == 0) {
1149 * We can release all of the reserved metadata blocks
1150 * only when we have written all of the delayed
1151 * allocation blocks.
1153 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1154 ei->i_reserved_meta_blocks);
1155 ei->i_reserved_meta_blocks = 0;
1156 ei->i_da_metadata_calc_len = 0;
1158 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1160 /* Update quota subsystem for data blocks */
1162 dquot_claim_block(inode, used);
1165 * We did fallocate with an offset that is already delayed
1166 * allocated. So on delayed allocated writeback we should
1167 * not re-claim the quota for fallocated blocks.
1169 dquot_release_reservation_block(inode, used);
1173 * If we have done all the pending block allocations and if
1174 * there aren't any writers on the inode, we can discard the
1175 * inode's preallocations.
1177 if ((ei->i_reserved_data_blocks == 0) &&
1178 (atomic_read(&inode->i_writecount) == 0))
1179 ext4_discard_preallocations(inode);
1182 static int __check_block_validity(struct inode *inode, const char *func,
1184 struct ext4_map_blocks *map)
1186 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1188 ext4_error_inode(inode, func, line, map->m_pblk,
1189 "lblock %lu mapped to illegal pblock "
1190 "(length %d)", (unsigned long) map->m_lblk,
1197 #define check_block_validity(inode, map) \
1198 __check_block_validity((inode), __func__, __LINE__, (map))
1201 * Return the number of contiguous dirty pages in a given inode
1202 * starting at page frame idx.
1204 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1205 unsigned int max_pages)
1207 struct address_space *mapping = inode->i_mapping;
1209 struct pagevec pvec;
1211 int i, nr_pages, done = 0;
1215 pagevec_init(&pvec, 0);
1218 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1219 PAGECACHE_TAG_DIRTY,
1220 (pgoff_t)PAGEVEC_SIZE);
1223 for (i = 0; i < nr_pages; i++) {
1224 struct page *page = pvec.pages[i];
1225 struct buffer_head *bh, *head;
1228 if (unlikely(page->mapping != mapping) ||
1230 PageWriteback(page) ||
1231 page->index != idx) {
1236 if (page_has_buffers(page)) {
1237 bh = head = page_buffers(page);
1239 if (!buffer_delay(bh) &&
1240 !buffer_unwritten(bh))
1242 bh = bh->b_this_page;
1243 } while (!done && (bh != head));
1250 if (num >= max_pages) {
1255 pagevec_release(&pvec);
1261 * The ext4_map_blocks() function tries to look up the requested blocks,
1262 * and returns if the blocks are already mapped.
1264 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1265 * and store the allocated blocks in the result buffer head and mark it
1268 * If file type is extents based, it will call ext4_ext_map_blocks(),
1269 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1272 * On success, it returns the number of blocks being mapped or allocate.
1273 * if create==0 and the blocks are pre-allocated and uninitialized block,
1274 * the result buffer head is unmapped. If the create ==1, it will make sure
1275 * the buffer head is mapped.
1277 * It returns 0 if plain look up failed (blocks have not been allocated), in
1278 * that casem, buffer head is unmapped
1280 * It returns the error in case of allocation failure.
1282 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1283 struct ext4_map_blocks *map, int flags)
1288 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1289 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1290 (unsigned long) map->m_lblk);
1292 * Try to see if we can get the block without requesting a new
1293 * file system block.
1295 down_read((&EXT4_I(inode)->i_data_sem));
1296 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1297 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1299 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1301 up_read((&EXT4_I(inode)->i_data_sem));
1303 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1304 int ret = check_block_validity(inode, map);
1309 /* If it is only a block(s) look up */
1310 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1314 * Returns if the blocks have already allocated
1316 * Note that if blocks have been preallocated
1317 * ext4_ext_get_block() returns th create = 0
1318 * with buffer head unmapped.
1320 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1324 * When we call get_blocks without the create flag, the
1325 * BH_Unwritten flag could have gotten set if the blocks
1326 * requested were part of a uninitialized extent. We need to
1327 * clear this flag now that we are committed to convert all or
1328 * part of the uninitialized extent to be an initialized
1329 * extent. This is because we need to avoid the combination
1330 * of BH_Unwritten and BH_Mapped flags being simultaneously
1331 * set on the buffer_head.
1333 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1336 * New blocks allocate and/or writing to uninitialized extent
1337 * will possibly result in updating i_data, so we take
1338 * the write lock of i_data_sem, and call get_blocks()
1339 * with create == 1 flag.
1341 down_write((&EXT4_I(inode)->i_data_sem));
1344 * if the caller is from delayed allocation writeout path
1345 * we have already reserved fs blocks for allocation
1346 * let the underlying get_block() function know to
1347 * avoid double accounting
1349 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1350 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1352 * We need to check for EXT4 here because migrate
1353 * could have changed the inode type in between
1355 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1356 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1358 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1360 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1362 * We allocated new blocks which will result in
1363 * i_data's format changing. Force the migrate
1364 * to fail by clearing migrate flags
1366 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1370 * Update reserved blocks/metadata blocks after successful
1371 * block allocation which had been deferred till now. We don't
1372 * support fallocate for non extent files. So we can update
1373 * reserve space here.
1376 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1377 ext4_da_update_reserve_space(inode, retval, 1);
1379 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1380 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1382 up_write((&EXT4_I(inode)->i_data_sem));
1383 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1384 int ret = check_block_validity(inode, map);
1391 /* Maximum number of blocks we map for direct IO at once. */
1392 #define DIO_MAX_BLOCKS 4096
1394 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1395 struct buffer_head *bh, int flags)
1397 handle_t *handle = ext4_journal_current_handle();
1398 struct ext4_map_blocks map;
1399 int ret = 0, started = 0;
1402 map.m_lblk = iblock;
1403 map.m_len = bh->b_size >> inode->i_blkbits;
1405 if (flags && !handle) {
1406 /* Direct IO write... */
1407 if (map.m_len > DIO_MAX_BLOCKS)
1408 map.m_len = DIO_MAX_BLOCKS;
1409 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1410 handle = ext4_journal_start(inode, dio_credits);
1411 if (IS_ERR(handle)) {
1412 ret = PTR_ERR(handle);
1418 ret = ext4_map_blocks(handle, inode, &map, flags);
1420 map_bh(bh, inode->i_sb, map.m_pblk);
1421 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1422 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1426 ext4_journal_stop(handle);
1430 int ext4_get_block(struct inode *inode, sector_t iblock,
1431 struct buffer_head *bh, int create)
1433 return _ext4_get_block(inode, iblock, bh,
1434 create ? EXT4_GET_BLOCKS_CREATE : 0);
1438 * `handle' can be NULL if create is zero
1440 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1441 ext4_lblk_t block, int create, int *errp)
1443 struct ext4_map_blocks map;
1444 struct buffer_head *bh;
1447 J_ASSERT(handle != NULL || create == 0);
1451 err = ext4_map_blocks(handle, inode, &map,
1452 create ? EXT4_GET_BLOCKS_CREATE : 0);
1460 bh = sb_getblk(inode->i_sb, map.m_pblk);
1465 if (map.m_flags & EXT4_MAP_NEW) {
1466 J_ASSERT(create != 0);
1467 J_ASSERT(handle != NULL);
1470 * Now that we do not always journal data, we should
1471 * keep in mind whether this should always journal the
1472 * new buffer as metadata. For now, regular file
1473 * writes use ext4_get_block instead, so it's not a
1477 BUFFER_TRACE(bh, "call get_create_access");
1478 fatal = ext4_journal_get_create_access(handle, bh);
1479 if (!fatal && !buffer_uptodate(bh)) {
1480 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1481 set_buffer_uptodate(bh);
1484 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1485 err = ext4_handle_dirty_metadata(handle, inode, bh);
1489 BUFFER_TRACE(bh, "not a new buffer");
1499 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1500 ext4_lblk_t block, int create, int *err)
1502 struct buffer_head *bh;
1504 bh = ext4_getblk(handle, inode, block, create, err);
1507 if (buffer_uptodate(bh))
1509 ll_rw_block(READ_META, 1, &bh);
1511 if (buffer_uptodate(bh))
1518 static int walk_page_buffers(handle_t *handle,
1519 struct buffer_head *head,
1523 int (*fn)(handle_t *handle,
1524 struct buffer_head *bh))
1526 struct buffer_head *bh;
1527 unsigned block_start, block_end;
1528 unsigned blocksize = head->b_size;
1530 struct buffer_head *next;
1532 for (bh = head, block_start = 0;
1533 ret == 0 && (bh != head || !block_start);
1534 block_start = block_end, bh = next) {
1535 next = bh->b_this_page;
1536 block_end = block_start + blocksize;
1537 if (block_end <= from || block_start >= to) {
1538 if (partial && !buffer_uptodate(bh))
1542 err = (*fn)(handle, bh);
1550 * To preserve ordering, it is essential that the hole instantiation and
1551 * the data write be encapsulated in a single transaction. We cannot
1552 * close off a transaction and start a new one between the ext4_get_block()
1553 * and the commit_write(). So doing the jbd2_journal_start at the start of
1554 * prepare_write() is the right place.
1556 * Also, this function can nest inside ext4_writepage() ->
1557 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1558 * has generated enough buffer credits to do the whole page. So we won't
1559 * block on the journal in that case, which is good, because the caller may
1562 * By accident, ext4 can be reentered when a transaction is open via
1563 * quota file writes. If we were to commit the transaction while thus
1564 * reentered, there can be a deadlock - we would be holding a quota
1565 * lock, and the commit would never complete if another thread had a
1566 * transaction open and was blocking on the quota lock - a ranking
1569 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1570 * will _not_ run commit under these circumstances because handle->h_ref
1571 * is elevated. We'll still have enough credits for the tiny quotafile
1574 static int do_journal_get_write_access(handle_t *handle,
1575 struct buffer_head *bh)
1577 int dirty = buffer_dirty(bh);
1580 if (!buffer_mapped(bh) || buffer_freed(bh))
1583 * __block_write_begin() could have dirtied some buffers. Clean
1584 * the dirty bit as jbd2_journal_get_write_access() could complain
1585 * otherwise about fs integrity issues. Setting of the dirty bit
1586 * by __block_write_begin() isn't a real problem here as we clear
1587 * the bit before releasing a page lock and thus writeback cannot
1588 * ever write the buffer.
1591 clear_buffer_dirty(bh);
1592 ret = ext4_journal_get_write_access(handle, bh);
1594 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1599 * Truncate blocks that were not used by write. We have to truncate the
1600 * pagecache as well so that corresponding buffers get properly unmapped.
1602 static void ext4_truncate_failed_write(struct inode *inode)
1604 truncate_inode_pages(inode->i_mapping, inode->i_size);
1605 ext4_truncate(inode);
1608 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1609 struct buffer_head *bh_result, int create);
1610 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1611 loff_t pos, unsigned len, unsigned flags,
1612 struct page **pagep, void **fsdata)
1614 struct inode *inode = mapping->host;
1615 int ret, needed_blocks;
1622 trace_ext4_write_begin(inode, pos, len, flags);
1624 * Reserve one block more for addition to orphan list in case
1625 * we allocate blocks but write fails for some reason
1627 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1628 index = pos >> PAGE_CACHE_SHIFT;
1629 from = pos & (PAGE_CACHE_SIZE - 1);
1633 handle = ext4_journal_start(inode, needed_blocks);
1634 if (IS_ERR(handle)) {
1635 ret = PTR_ERR(handle);
1639 /* We cannot recurse into the filesystem as the transaction is already
1641 flags |= AOP_FLAG_NOFS;
1643 page = grab_cache_page_write_begin(mapping, index, flags);
1645 ext4_journal_stop(handle);
1651 if (ext4_should_dioread_nolock(inode))
1652 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1654 ret = __block_write_begin(page, pos, len, ext4_get_block);
1656 if (!ret && ext4_should_journal_data(inode)) {
1657 ret = walk_page_buffers(handle, page_buffers(page),
1658 from, to, NULL, do_journal_get_write_access);
1663 page_cache_release(page);
1665 * __block_write_begin may have instantiated a few blocks
1666 * outside i_size. Trim these off again. Don't need
1667 * i_size_read because we hold i_mutex.
1669 * Add inode to orphan list in case we crash before
1672 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1673 ext4_orphan_add(handle, inode);
1675 ext4_journal_stop(handle);
1676 if (pos + len > inode->i_size) {
1677 ext4_truncate_failed_write(inode);
1679 * If truncate failed early the inode might
1680 * still be on the orphan list; we need to
1681 * make sure the inode is removed from the
1682 * orphan list in that case.
1685 ext4_orphan_del(NULL, inode);
1689 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1695 /* For write_end() in data=journal mode */
1696 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1698 if (!buffer_mapped(bh) || buffer_freed(bh))
1700 set_buffer_uptodate(bh);
1701 return ext4_handle_dirty_metadata(handle, NULL, bh);
1704 static int ext4_generic_write_end(struct file *file,
1705 struct address_space *mapping,
1706 loff_t pos, unsigned len, unsigned copied,
1707 struct page *page, void *fsdata)
1709 int i_size_changed = 0;
1710 struct inode *inode = mapping->host;
1711 handle_t *handle = ext4_journal_current_handle();
1713 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1716 * No need to use i_size_read() here, the i_size
1717 * cannot change under us because we hold i_mutex.
1719 * But it's important to update i_size while still holding page lock:
1720 * page writeout could otherwise come in and zero beyond i_size.
1722 if (pos + copied > inode->i_size) {
1723 i_size_write(inode, pos + copied);
1727 if (pos + copied > EXT4_I(inode)->i_disksize) {
1728 /* We need to mark inode dirty even if
1729 * new_i_size is less that inode->i_size
1730 * bu greater than i_disksize.(hint delalloc)
1732 ext4_update_i_disksize(inode, (pos + copied));
1736 page_cache_release(page);
1739 * Don't mark the inode dirty under page lock. First, it unnecessarily
1740 * makes the holding time of page lock longer. Second, it forces lock
1741 * ordering of page lock and transaction start for journaling
1745 ext4_mark_inode_dirty(handle, inode);
1751 * We need to pick up the new inode size which generic_commit_write gave us
1752 * `file' can be NULL - eg, when called from page_symlink().
1754 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1755 * buffers are managed internally.
1757 static int ext4_ordered_write_end(struct file *file,
1758 struct address_space *mapping,
1759 loff_t pos, unsigned len, unsigned copied,
1760 struct page *page, void *fsdata)
1762 handle_t *handle = ext4_journal_current_handle();
1763 struct inode *inode = mapping->host;
1766 trace_ext4_ordered_write_end(inode, pos, len, copied);
1767 ret = ext4_jbd2_file_inode(handle, inode);
1770 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1773 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1774 /* if we have allocated more blocks and copied
1775 * less. We will have blocks allocated outside
1776 * inode->i_size. So truncate them
1778 ext4_orphan_add(handle, inode);
1782 ret2 = ext4_journal_stop(handle);
1786 if (pos + len > inode->i_size) {
1787 ext4_truncate_failed_write(inode);
1789 * If truncate failed early the inode might still be
1790 * on the orphan list; we need to make sure the inode
1791 * is removed from the orphan list in that case.
1794 ext4_orphan_del(NULL, inode);
1798 return ret ? ret : copied;
1801 static int ext4_writeback_write_end(struct file *file,
1802 struct address_space *mapping,
1803 loff_t pos, unsigned len, unsigned copied,
1804 struct page *page, void *fsdata)
1806 handle_t *handle = ext4_journal_current_handle();
1807 struct inode *inode = mapping->host;
1810 trace_ext4_writeback_write_end(inode, pos, len, copied);
1811 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1814 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1815 /* if we have allocated more blocks and copied
1816 * less. We will have blocks allocated outside
1817 * inode->i_size. So truncate them
1819 ext4_orphan_add(handle, inode);
1824 ret2 = ext4_journal_stop(handle);
1828 if (pos + len > inode->i_size) {
1829 ext4_truncate_failed_write(inode);
1831 * If truncate failed early the inode might still be
1832 * on the orphan list; we need to make sure the inode
1833 * is removed from the orphan list in that case.
1836 ext4_orphan_del(NULL, inode);
1839 return ret ? ret : copied;
1842 static int ext4_journalled_write_end(struct file *file,
1843 struct address_space *mapping,
1844 loff_t pos, unsigned len, unsigned copied,
1845 struct page *page, void *fsdata)
1847 handle_t *handle = ext4_journal_current_handle();
1848 struct inode *inode = mapping->host;
1854 trace_ext4_journalled_write_end(inode, pos, len, copied);
1855 from = pos & (PAGE_CACHE_SIZE - 1);
1858 BUG_ON(!ext4_handle_valid(handle));
1861 if (!PageUptodate(page))
1863 page_zero_new_buffers(page, from+copied, to);
1866 ret = walk_page_buffers(handle, page_buffers(page), from,
1867 to, &partial, write_end_fn);
1869 SetPageUptodate(page);
1870 new_i_size = pos + copied;
1871 if (new_i_size > inode->i_size)
1872 i_size_write(inode, pos+copied);
1873 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1874 if (new_i_size > EXT4_I(inode)->i_disksize) {
1875 ext4_update_i_disksize(inode, new_i_size);
1876 ret2 = ext4_mark_inode_dirty(handle, inode);
1882 page_cache_release(page);
1883 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1884 /* if we have allocated more blocks and copied
1885 * less. We will have blocks allocated outside
1886 * inode->i_size. So truncate them
1888 ext4_orphan_add(handle, inode);
1890 ret2 = ext4_journal_stop(handle);
1893 if (pos + len > inode->i_size) {
1894 ext4_truncate_failed_write(inode);
1896 * If truncate failed early the inode might still be
1897 * on the orphan list; we need to make sure the inode
1898 * is removed from the orphan list in that case.
1901 ext4_orphan_del(NULL, inode);
1904 return ret ? ret : copied;
1908 * Reserve a single block located at lblock
1910 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1913 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1914 struct ext4_inode_info *ei = EXT4_I(inode);
1915 unsigned long md_needed;
1919 * recalculate the amount of metadata blocks to reserve
1920 * in order to allocate nrblocks
1921 * worse case is one extent per block
1924 spin_lock(&ei->i_block_reservation_lock);
1925 md_needed = ext4_calc_metadata_amount(inode, lblock);
1926 trace_ext4_da_reserve_space(inode, md_needed);
1927 spin_unlock(&ei->i_block_reservation_lock);
1930 * We will charge metadata quota at writeout time; this saves
1931 * us from metadata over-estimation, though we may go over by
1932 * a small amount in the end. Here we just reserve for data.
1934 ret = dquot_reserve_block(inode, 1);
1938 * We do still charge estimated metadata to the sb though;
1939 * we cannot afford to run out of free blocks.
1941 if (ext4_claim_free_blocks(sbi, md_needed + 1, 0)) {
1942 dquot_release_reservation_block(inode, 1);
1943 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1949 spin_lock(&ei->i_block_reservation_lock);
1950 ei->i_reserved_data_blocks++;
1951 ei->i_reserved_meta_blocks += md_needed;
1952 spin_unlock(&ei->i_block_reservation_lock);
1954 return 0; /* success */
1957 static void ext4_da_release_space(struct inode *inode, int to_free)
1959 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1960 struct ext4_inode_info *ei = EXT4_I(inode);
1963 return; /* Nothing to release, exit */
1965 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1967 trace_ext4_da_release_space(inode, to_free);
1968 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1970 * if there aren't enough reserved blocks, then the
1971 * counter is messed up somewhere. Since this
1972 * function is called from invalidate page, it's
1973 * harmless to return without any action.
1975 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1976 "ino %lu, to_free %d with only %d reserved "
1977 "data blocks\n", inode->i_ino, to_free,
1978 ei->i_reserved_data_blocks);
1980 to_free = ei->i_reserved_data_blocks;
1982 ei->i_reserved_data_blocks -= to_free;
1984 if (ei->i_reserved_data_blocks == 0) {
1986 * We can release all of the reserved metadata blocks
1987 * only when we have written all of the delayed
1988 * allocation blocks.
1990 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1991 ei->i_reserved_meta_blocks);
1992 ei->i_reserved_meta_blocks = 0;
1993 ei->i_da_metadata_calc_len = 0;
1996 /* update fs dirty data blocks counter */
1997 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1999 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
2001 dquot_release_reservation_block(inode, to_free);
2004 static void ext4_da_page_release_reservation(struct page *page,
2005 unsigned long offset)
2008 struct buffer_head *head, *bh;
2009 unsigned int curr_off = 0;
2011 head = page_buffers(page);
2014 unsigned int next_off = curr_off + bh->b_size;
2016 if ((offset <= curr_off) && (buffer_delay(bh))) {
2018 clear_buffer_delay(bh);
2020 curr_off = next_off;
2021 } while ((bh = bh->b_this_page) != head);
2022 ext4_da_release_space(page->mapping->host, to_release);
2026 * Delayed allocation stuff
2030 * mpage_da_submit_io - walks through extent of pages and try to write
2031 * them with writepage() call back
2033 * @mpd->inode: inode
2034 * @mpd->first_page: first page of the extent
2035 * @mpd->next_page: page after the last page of the extent
2037 * By the time mpage_da_submit_io() is called we expect all blocks
2038 * to be allocated. this may be wrong if allocation failed.
2040 * As pages are already locked by write_cache_pages(), we can't use it
2042 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2043 struct ext4_map_blocks *map)
2045 struct pagevec pvec;
2046 unsigned long index, end;
2047 int ret = 0, err, nr_pages, i;
2048 struct inode *inode = mpd->inode;
2049 struct address_space *mapping = inode->i_mapping;
2050 loff_t size = i_size_read(inode);
2051 unsigned int len, block_start;
2052 struct buffer_head *bh, *page_bufs = NULL;
2053 int journal_data = ext4_should_journal_data(inode);
2054 sector_t pblock = 0, cur_logical = 0;
2055 struct ext4_io_submit io_submit;
2057 BUG_ON(mpd->next_page <= mpd->first_page);
2058 memset(&io_submit, 0, sizeof(io_submit));
2060 * We need to start from the first_page to the next_page - 1
2061 * to make sure we also write the mapped dirty buffer_heads.
2062 * If we look at mpd->b_blocknr we would only be looking
2063 * at the currently mapped buffer_heads.
2065 index = mpd->first_page;
2066 end = mpd->next_page - 1;
2068 pagevec_init(&pvec, 0);
2069 while (index <= end) {
2070 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2073 for (i = 0; i < nr_pages; i++) {
2074 int commit_write = 0, skip_page = 0;
2075 struct page *page = pvec.pages[i];
2077 index = page->index;
2081 if (index == size >> PAGE_CACHE_SHIFT)
2082 len = size & ~PAGE_CACHE_MASK;
2084 len = PAGE_CACHE_SIZE;
2086 cur_logical = index << (PAGE_CACHE_SHIFT -
2088 pblock = map->m_pblk + (cur_logical -
2093 BUG_ON(!PageLocked(page));
2094 BUG_ON(PageWriteback(page));
2097 * If the page does not have buffers (for
2098 * whatever reason), try to create them using
2099 * __block_write_begin. If this fails,
2100 * skip the page and move on.
2102 if (!page_has_buffers(page)) {
2103 if (__block_write_begin(page, 0, len,
2104 noalloc_get_block_write)) {
2112 bh = page_bufs = page_buffers(page);
2117 if (map && (cur_logical >= map->m_lblk) &&
2118 (cur_logical <= (map->m_lblk +
2119 (map->m_len - 1)))) {
2120 if (buffer_delay(bh)) {
2121 clear_buffer_delay(bh);
2122 bh->b_blocknr = pblock;
2124 if (buffer_unwritten(bh) ||
2126 BUG_ON(bh->b_blocknr != pblock);
2127 if (map->m_flags & EXT4_MAP_UNINIT)
2128 set_buffer_uninit(bh);
2129 clear_buffer_unwritten(bh);
2132 /* skip page if block allocation undone */
2133 if (buffer_delay(bh) || buffer_unwritten(bh))
2135 bh = bh->b_this_page;
2136 block_start += bh->b_size;
2139 } while (bh != page_bufs);
2145 /* mark the buffer_heads as dirty & uptodate */
2146 block_commit_write(page, 0, len);
2148 clear_page_dirty_for_io(page);
2150 * Delalloc doesn't support data journalling,
2151 * but eventually maybe we'll lift this
2154 if (unlikely(journal_data && PageChecked(page)))
2155 err = __ext4_journalled_writepage(page, len);
2156 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
2157 err = ext4_bio_write_page(&io_submit, page,
2159 else if (buffer_uninit(page_bufs)) {
2160 ext4_set_bh_endio(page_bufs, inode);
2161 err = block_write_full_page_endio(page,
2162 noalloc_get_block_write,
2163 mpd->wbc, ext4_end_io_buffer_write);
2165 err = block_write_full_page(page,
2166 noalloc_get_block_write, mpd->wbc);
2169 mpd->pages_written++;
2171 * In error case, we have to continue because
2172 * remaining pages are still locked
2177 pagevec_release(&pvec);
2179 ext4_io_submit(&io_submit);
2183 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
2187 struct pagevec pvec;
2188 struct inode *inode = mpd->inode;
2189 struct address_space *mapping = inode->i_mapping;
2191 index = mpd->first_page;
2192 end = mpd->next_page - 1;
2193 while (index <= end) {
2194 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2197 for (i = 0; i < nr_pages; i++) {
2198 struct page *page = pvec.pages[i];
2199 if (page->index > end)
2201 BUG_ON(!PageLocked(page));
2202 BUG_ON(PageWriteback(page));
2203 block_invalidatepage(page, 0);
2204 ClearPageUptodate(page);
2207 index = pvec.pages[nr_pages - 1]->index + 1;
2208 pagevec_release(&pvec);
2213 static void ext4_print_free_blocks(struct inode *inode)
2215 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2216 printk(KERN_CRIT "Total free blocks count %lld\n",
2217 ext4_count_free_blocks(inode->i_sb));
2218 printk(KERN_CRIT "Free/Dirty block details\n");
2219 printk(KERN_CRIT "free_blocks=%lld\n",
2220 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2221 printk(KERN_CRIT "dirty_blocks=%lld\n",
2222 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2223 printk(KERN_CRIT "Block reservation details\n");
2224 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2225 EXT4_I(inode)->i_reserved_data_blocks);
2226 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2227 EXT4_I(inode)->i_reserved_meta_blocks);
2232 * mpage_da_map_and_submit - go through given space, map them
2233 * if necessary, and then submit them for I/O
2235 * @mpd - bh describing space
2237 * The function skips space we know is already mapped to disk blocks.
2240 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2242 int err, blks, get_blocks_flags;
2243 struct ext4_map_blocks map, *mapp = NULL;
2244 sector_t next = mpd->b_blocknr;
2245 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2246 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2247 handle_t *handle = NULL;
2250 * If the blocks are mapped already, or we couldn't accumulate
2251 * any blocks, then proceed immediately to the submission stage.
2253 if ((mpd->b_size == 0) ||
2254 ((mpd->b_state & (1 << BH_Mapped)) &&
2255 !(mpd->b_state & (1 << BH_Delay)) &&
2256 !(mpd->b_state & (1 << BH_Unwritten))))
2259 handle = ext4_journal_current_handle();
2263 * Call ext4_map_blocks() to allocate any delayed allocation
2264 * blocks, or to convert an uninitialized extent to be
2265 * initialized (in the case where we have written into
2266 * one or more preallocated blocks).
2268 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2269 * indicate that we are on the delayed allocation path. This
2270 * affects functions in many different parts of the allocation
2271 * call path. This flag exists primarily because we don't
2272 * want to change *many* call functions, so ext4_map_blocks()
2273 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2274 * inode's allocation semaphore is taken.
2276 * If the blocks in questions were delalloc blocks, set
2277 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2278 * variables are updated after the blocks have been allocated.
2281 map.m_len = max_blocks;
2282 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2283 if (ext4_should_dioread_nolock(mpd->inode))
2284 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2285 if (mpd->b_state & (1 << BH_Delay))
2286 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2288 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2290 struct super_block *sb = mpd->inode->i_sb;
2294 * If get block returns EAGAIN or ENOSPC and there
2295 * appears to be free blocks we will just let
2296 * mpage_da_submit_io() unlock all of the pages.
2301 if (err == -ENOSPC &&
2302 ext4_count_free_blocks(sb)) {
2308 * get block failure will cause us to loop in
2309 * writepages, because a_ops->writepage won't be able
2310 * to make progress. The page will be redirtied by
2311 * writepage and writepages will again try to write
2314 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2315 ext4_msg(sb, KERN_CRIT,
2316 "delayed block allocation failed for inode %lu "
2317 "at logical offset %llu with max blocks %zd "
2318 "with error %d", mpd->inode->i_ino,
2319 (unsigned long long) next,
2320 mpd->b_size >> mpd->inode->i_blkbits, err);
2321 ext4_msg(sb, KERN_CRIT,
2322 "This should not happen!! Data will be lost\n");
2324 ext4_print_free_blocks(mpd->inode);
2326 /* invalidate all the pages */
2327 ext4_da_block_invalidatepages(mpd);
2329 /* Mark this page range as having been completed */
2336 if (map.m_flags & EXT4_MAP_NEW) {
2337 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2340 for (i = 0; i < map.m_len; i++)
2341 unmap_underlying_metadata(bdev, map.m_pblk + i);
2344 if (ext4_should_order_data(mpd->inode)) {
2345 err = ext4_jbd2_file_inode(handle, mpd->inode);
2347 /* This only happens if the journal is aborted */
2352 * Update on-disk size along with block allocation.
2354 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2355 if (disksize > i_size_read(mpd->inode))
2356 disksize = i_size_read(mpd->inode);
2357 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2358 ext4_update_i_disksize(mpd->inode, disksize);
2359 err = ext4_mark_inode_dirty(handle, mpd->inode);
2361 ext4_error(mpd->inode->i_sb,
2362 "Failed to mark inode %lu dirty",
2367 mpage_da_submit_io(mpd, mapp);
2371 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2372 (1 << BH_Delay) | (1 << BH_Unwritten))
2375 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2377 * @mpd->lbh - extent of blocks
2378 * @logical - logical number of the block in the file
2379 * @bh - bh of the block (used to access block's state)
2381 * the function is used to collect contig. blocks in same state
2383 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2384 sector_t logical, size_t b_size,
2385 unsigned long b_state)
2388 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2391 * XXX Don't go larger than mballoc is willing to allocate
2392 * This is a stopgap solution. We eventually need to fold
2393 * mpage_da_submit_io() into this function and then call
2394 * ext4_map_blocks() multiple times in a loop
2396 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2399 /* check if thereserved journal credits might overflow */
2400 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2401 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2403 * With non-extent format we are limited by the journal
2404 * credit available. Total credit needed to insert
2405 * nrblocks contiguous blocks is dependent on the
2406 * nrblocks. So limit nrblocks.
2409 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2410 EXT4_MAX_TRANS_DATA) {
2412 * Adding the new buffer_head would make it cross the
2413 * allowed limit for which we have journal credit
2414 * reserved. So limit the new bh->b_size
2416 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2417 mpd->inode->i_blkbits;
2418 /* we will do mpage_da_submit_io in the next loop */
2422 * First block in the extent
2424 if (mpd->b_size == 0) {
2425 mpd->b_blocknr = logical;
2426 mpd->b_size = b_size;
2427 mpd->b_state = b_state & BH_FLAGS;
2431 next = mpd->b_blocknr + nrblocks;
2433 * Can we merge the block to our big extent?
2435 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2436 mpd->b_size += b_size;
2442 * We couldn't merge the block to our extent, so we
2443 * need to flush current extent and start new one
2445 mpage_da_map_and_submit(mpd);
2449 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2451 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2455 * This is a special get_blocks_t callback which is used by
2456 * ext4_da_write_begin(). It will either return mapped block or
2457 * reserve space for a single block.
2459 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2460 * We also have b_blocknr = -1 and b_bdev initialized properly
2462 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2463 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2464 * initialized properly.
2466 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2467 struct buffer_head *bh, int create)
2469 struct ext4_map_blocks map;
2471 sector_t invalid_block = ~((sector_t) 0xffff);
2473 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2476 BUG_ON(create == 0);
2477 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2479 map.m_lblk = iblock;
2483 * first, we need to know whether the block is allocated already
2484 * preallocated blocks are unmapped but should treated
2485 * the same as allocated blocks.
2487 ret = ext4_map_blocks(NULL, inode, &map, 0);
2491 if (buffer_delay(bh))
2492 return 0; /* Not sure this could or should happen */
2494 * XXX: __block_write_begin() unmaps passed block, is it OK?
2496 ret = ext4_da_reserve_space(inode, iblock);
2498 /* not enough space to reserve */
2501 map_bh(bh, inode->i_sb, invalid_block);
2503 set_buffer_delay(bh);
2507 map_bh(bh, inode->i_sb, map.m_pblk);
2508 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2510 if (buffer_unwritten(bh)) {
2511 /* A delayed write to unwritten bh should be marked
2512 * new and mapped. Mapped ensures that we don't do
2513 * get_block multiple times when we write to the same
2514 * offset and new ensures that we do proper zero out
2515 * for partial write.
2518 set_buffer_mapped(bh);
2524 * This function is used as a standard get_block_t calback function
2525 * when there is no desire to allocate any blocks. It is used as a
2526 * callback function for block_write_begin() and block_write_full_page().
2527 * These functions should only try to map a single block at a time.
2529 * Since this function doesn't do block allocations even if the caller
2530 * requests it by passing in create=1, it is critically important that
2531 * any caller checks to make sure that any buffer heads are returned
2532 * by this function are either all already mapped or marked for
2533 * delayed allocation before calling block_write_full_page(). Otherwise,
2534 * b_blocknr could be left unitialized, and the page write functions will
2535 * be taken by surprise.
2537 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2538 struct buffer_head *bh_result, int create)
2540 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2541 return _ext4_get_block(inode, iblock, bh_result, 0);
2544 static int bget_one(handle_t *handle, struct buffer_head *bh)
2550 static int bput_one(handle_t *handle, struct buffer_head *bh)
2556 static int __ext4_journalled_writepage(struct page *page,
2559 struct address_space *mapping = page->mapping;
2560 struct inode *inode = mapping->host;
2561 struct buffer_head *page_bufs;
2562 handle_t *handle = NULL;
2566 ClearPageChecked(page);
2567 page_bufs = page_buffers(page);
2569 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2570 /* As soon as we unlock the page, it can go away, but we have
2571 * references to buffers so we are safe */
2574 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2575 if (IS_ERR(handle)) {
2576 ret = PTR_ERR(handle);
2580 BUG_ON(!ext4_handle_valid(handle));
2582 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2583 do_journal_get_write_access);
2585 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2589 err = ext4_journal_stop(handle);
2593 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2594 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2599 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2600 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2603 * Note that we don't need to start a transaction unless we're journaling data
2604 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2605 * need to file the inode to the transaction's list in ordered mode because if
2606 * we are writing back data added by write(), the inode is already there and if
2607 * we are writing back data modified via mmap(), no one guarantees in which
2608 * transaction the data will hit the disk. In case we are journaling data, we
2609 * cannot start transaction directly because transaction start ranks above page
2610 * lock so we have to do some magic.
2612 * This function can get called via...
2613 * - ext4_da_writepages after taking page lock (have journal handle)
2614 * - journal_submit_inode_data_buffers (no journal handle)
2615 * - shrink_page_list via pdflush (no journal handle)
2616 * - grab_page_cache when doing write_begin (have journal handle)
2618 * We don't do any block allocation in this function. If we have page with
2619 * multiple blocks we need to write those buffer_heads that are mapped. This
2620 * is important for mmaped based write. So if we do with blocksize 1K
2621 * truncate(f, 1024);
2622 * a = mmap(f, 0, 4096);
2624 * truncate(f, 4096);
2625 * we have in the page first buffer_head mapped via page_mkwrite call back
2626 * but other bufer_heads would be unmapped but dirty(dirty done via the
2627 * do_wp_page). So writepage should write the first block. If we modify
2628 * the mmap area beyond 1024 we will again get a page_fault and the
2629 * page_mkwrite callback will do the block allocation and mark the
2630 * buffer_heads mapped.
2632 * We redirty the page if we have any buffer_heads that is either delay or
2633 * unwritten in the page.
2635 * We can get recursively called as show below.
2637 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2640 * But since we don't do any block allocation we should not deadlock.
2641 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2643 static int ext4_writepage(struct page *page,
2644 struct writeback_control *wbc)
2646 int ret = 0, commit_write = 0;
2649 struct buffer_head *page_bufs = NULL;
2650 struct inode *inode = page->mapping->host;
2652 trace_ext4_writepage(page);
2653 size = i_size_read(inode);
2654 if (page->index == size >> PAGE_CACHE_SHIFT)
2655 len = size & ~PAGE_CACHE_MASK;
2657 len = PAGE_CACHE_SIZE;
2660 * If the page does not have buffers (for whatever reason),
2661 * try to create them using __block_write_begin. If this
2662 * fails, redirty the page and move on.
2664 if (!page_has_buffers(page)) {
2665 if (__block_write_begin(page, 0, len,
2666 noalloc_get_block_write)) {
2668 redirty_page_for_writepage(wbc, page);
2674 page_bufs = page_buffers(page);
2675 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2676 ext4_bh_delay_or_unwritten)) {
2678 * We don't want to do block allocation, so redirty
2679 * the page and return. We may reach here when we do
2680 * a journal commit via journal_submit_inode_data_buffers.
2681 * We can also reach here via shrink_page_list
2686 /* now mark the buffer_heads as dirty and uptodate */
2687 block_commit_write(page, 0, len);
2689 if (PageChecked(page) && ext4_should_journal_data(inode))
2691 * It's mmapped pagecache. Add buffers and journal it. There
2692 * doesn't seem much point in redirtying the page here.
2694 return __ext4_journalled_writepage(page, len);
2696 if (buffer_uninit(page_bufs)) {
2697 ext4_set_bh_endio(page_bufs, inode);
2698 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2699 wbc, ext4_end_io_buffer_write);
2701 ret = block_write_full_page(page, noalloc_get_block_write,
2708 * This is called via ext4_da_writepages() to
2709 * calculate the total number of credits to reserve to fit
2710 * a single extent allocation into a single transaction,
2711 * ext4_da_writpeages() will loop calling this before
2712 * the block allocation.
2715 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2717 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2720 * With non-extent format the journal credit needed to
2721 * insert nrblocks contiguous block is dependent on
2722 * number of contiguous block. So we will limit
2723 * number of contiguous block to a sane value
2725 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2726 (max_blocks > EXT4_MAX_TRANS_DATA))
2727 max_blocks = EXT4_MAX_TRANS_DATA;
2729 return ext4_chunk_trans_blocks(inode, max_blocks);
2733 * write_cache_pages_da - walk the list of dirty pages of the given
2734 * address space and accumulate pages that need writing, and call
2735 * mpage_da_map_and_submit to map a single contiguous memory region
2736 * and then write them.
2738 static int write_cache_pages_da(struct address_space *mapping,
2739 struct writeback_control *wbc,
2740 struct mpage_da_data *mpd,
2741 pgoff_t *done_index)
2743 struct buffer_head *bh, *head;
2744 struct inode *inode = mapping->host;
2745 struct pagevec pvec;
2746 unsigned int nr_pages;
2749 long nr_to_write = wbc->nr_to_write;
2750 int i, tag, ret = 0;
2752 memset(mpd, 0, sizeof(struct mpage_da_data));
2755 pagevec_init(&pvec, 0);
2756 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2757 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2759 if (wbc->sync_mode == WB_SYNC_ALL)
2760 tag = PAGECACHE_TAG_TOWRITE;
2762 tag = PAGECACHE_TAG_DIRTY;
2764 *done_index = index;
2765 while (index <= end) {
2766 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2767 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2771 for (i = 0; i < nr_pages; i++) {
2772 struct page *page = pvec.pages[i];
2775 * At this point, the page may be truncated or
2776 * invalidated (changing page->mapping to NULL), or
2777 * even swizzled back from swapper_space to tmpfs file
2778 * mapping. However, page->index will not change
2779 * because we have a reference on the page.
2781 if (page->index > end)
2784 *done_index = page->index + 1;
2787 * If we can't merge this page, and we have
2788 * accumulated an contiguous region, write it
2790 if ((mpd->next_page != page->index) &&
2791 (mpd->next_page != mpd->first_page)) {
2792 mpage_da_map_and_submit(mpd);
2793 goto ret_extent_tail;
2799 * If the page is no longer dirty, or its
2800 * mapping no longer corresponds to inode we
2801 * are writing (which means it has been
2802 * truncated or invalidated), or the page is
2803 * already under writeback and we are not
2804 * doing a data integrity writeback, skip the page
2806 if (!PageDirty(page) ||
2807 (PageWriteback(page) &&
2808 (wbc->sync_mode == WB_SYNC_NONE)) ||
2809 unlikely(page->mapping != mapping)) {
2814 wait_on_page_writeback(page);
2815 BUG_ON(PageWriteback(page));
2817 if (mpd->next_page != page->index)
2818 mpd->first_page = page->index;
2819 mpd->next_page = page->index + 1;
2820 logical = (sector_t) page->index <<
2821 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2823 if (!page_has_buffers(page)) {
2824 mpage_add_bh_to_extent(mpd, logical,
2826 (1 << BH_Dirty) | (1 << BH_Uptodate));
2828 goto ret_extent_tail;
2831 * Page with regular buffer heads,
2832 * just add all dirty ones
2834 head = page_buffers(page);
2837 BUG_ON(buffer_locked(bh));
2839 * We need to try to allocate
2840 * unmapped blocks in the same page.
2841 * Otherwise we won't make progress
2842 * with the page in ext4_writepage
2844 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2845 mpage_add_bh_to_extent(mpd, logical,
2849 goto ret_extent_tail;
2850 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2852 * mapped dirty buffer. We need
2853 * to update the b_state
2854 * because we look at b_state
2855 * in mpage_da_map_blocks. We
2856 * don't update b_size because
2857 * if we find an unmapped
2858 * buffer_head later we need to
2859 * use the b_state flag of that
2862 if (mpd->b_size == 0)
2863 mpd->b_state = bh->b_state & BH_FLAGS;
2866 } while ((bh = bh->b_this_page) != head);
2869 if (nr_to_write > 0) {
2871 if (nr_to_write == 0 &&
2872 wbc->sync_mode == WB_SYNC_NONE)
2874 * We stop writing back only if we are
2875 * not doing integrity sync. In case of
2876 * integrity sync we have to keep going
2877 * because someone may be concurrently
2878 * dirtying pages, and we might have
2879 * synced a lot of newly appeared dirty
2880 * pages, but have not synced all of the
2886 pagevec_release(&pvec);
2891 ret = MPAGE_DA_EXTENT_TAIL;
2893 pagevec_release(&pvec);
2899 static int ext4_da_writepages(struct address_space *mapping,
2900 struct writeback_control *wbc)
2903 int range_whole = 0;
2904 handle_t *handle = NULL;
2905 struct mpage_da_data mpd;
2906 struct inode *inode = mapping->host;
2907 int pages_written = 0;
2908 unsigned int max_pages;
2909 int range_cyclic, cycled = 1, io_done = 0;
2910 int needed_blocks, ret = 0;
2911 long desired_nr_to_write, nr_to_writebump = 0;
2912 loff_t range_start = wbc->range_start;
2913 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2914 pgoff_t done_index = 0;
2917 trace_ext4_da_writepages(inode, wbc);
2920 * No pages to write? This is mainly a kludge to avoid starting
2921 * a transaction for special inodes like journal inode on last iput()
2922 * because that could violate lock ordering on umount
2924 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2928 * If the filesystem has aborted, it is read-only, so return
2929 * right away instead of dumping stack traces later on that
2930 * will obscure the real source of the problem. We test
2931 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2932 * the latter could be true if the filesystem is mounted
2933 * read-only, and in that case, ext4_da_writepages should
2934 * *never* be called, so if that ever happens, we would want
2937 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2940 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2943 range_cyclic = wbc->range_cyclic;
2944 if (wbc->range_cyclic) {
2945 index = mapping->writeback_index;
2948 wbc->range_start = index << PAGE_CACHE_SHIFT;
2949 wbc->range_end = LLONG_MAX;
2950 wbc->range_cyclic = 0;
2953 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2954 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2958 * This works around two forms of stupidity. The first is in
2959 * the writeback code, which caps the maximum number of pages
2960 * written to be 1024 pages. This is wrong on multiple
2961 * levels; different architectues have a different page size,
2962 * which changes the maximum amount of data which gets
2963 * written. Secondly, 4 megabytes is way too small. XFS
2964 * forces this value to be 16 megabytes by multiplying
2965 * nr_to_write parameter by four, and then relies on its
2966 * allocator to allocate larger extents to make them
2967 * contiguous. Unfortunately this brings us to the second
2968 * stupidity, which is that ext4's mballoc code only allocates
2969 * at most 2048 blocks. So we force contiguous writes up to
2970 * the number of dirty blocks in the inode, or
2971 * sbi->max_writeback_mb_bump whichever is smaller.
2973 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2974 if (!range_cyclic && range_whole) {
2975 if (wbc->nr_to_write == LONG_MAX)
2976 desired_nr_to_write = wbc->nr_to_write;
2978 desired_nr_to_write = wbc->nr_to_write * 8;
2980 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2982 if (desired_nr_to_write > max_pages)
2983 desired_nr_to_write = max_pages;
2985 if (wbc->nr_to_write < desired_nr_to_write) {
2986 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2987 wbc->nr_to_write = desired_nr_to_write;
2991 if (wbc->sync_mode == WB_SYNC_ALL)
2992 tag_pages_for_writeback(mapping, index, end);
2994 while (!ret && wbc->nr_to_write > 0) {
2997 * we insert one extent at a time. So we need
2998 * credit needed for single extent allocation.
2999 * journalled mode is currently not supported
3002 BUG_ON(ext4_should_journal_data(inode));
3003 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3005 /* start a new transaction*/
3006 handle = ext4_journal_start(inode, needed_blocks);
3007 if (IS_ERR(handle)) {
3008 ret = PTR_ERR(handle);
3009 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3010 "%ld pages, ino %lu; err %d", __func__,
3011 wbc->nr_to_write, inode->i_ino, ret);
3012 goto out_writepages;
3016 * Now call write_cache_pages_da() to find the next
3017 * contiguous region of logical blocks that need
3018 * blocks to be allocated by ext4 and submit them.
3020 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3022 * If we have a contiguous extent of pages and we
3023 * haven't done the I/O yet, map the blocks and submit
3026 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3027 mpage_da_map_and_submit(&mpd);
3028 ret = MPAGE_DA_EXTENT_TAIL;
3030 trace_ext4_da_write_pages(inode, &mpd);
3031 wbc->nr_to_write -= mpd.pages_written;
3033 ext4_journal_stop(handle);
3035 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3036 /* commit the transaction which would
3037 * free blocks released in the transaction
3040 jbd2_journal_force_commit_nested(sbi->s_journal);
3042 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3044 * got one extent now try with
3047 pages_written += mpd.pages_written;
3050 } else if (wbc->nr_to_write)
3052 * There is no more writeout needed
3053 * or we requested for a noblocking writeout
3054 * and we found the device congested
3058 if (!io_done && !cycled) {
3061 wbc->range_start = index << PAGE_CACHE_SHIFT;
3062 wbc->range_end = mapping->writeback_index - 1;
3067 wbc->range_cyclic = range_cyclic;
3068 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3070 * set the writeback_index so that range_cyclic
3071 * mode will write it back later
3073 mapping->writeback_index = done_index;
3076 wbc->nr_to_write -= nr_to_writebump;
3077 wbc->range_start = range_start;
3078 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3082 #define FALL_BACK_TO_NONDELALLOC 1
3083 static int ext4_nonda_switch(struct super_block *sb)
3085 s64 free_blocks, dirty_blocks;
3086 struct ext4_sb_info *sbi = EXT4_SB(sb);
3089 * switch to non delalloc mode if we are running low
3090 * on free block. The free block accounting via percpu
3091 * counters can get slightly wrong with percpu_counter_batch getting
3092 * accumulated on each CPU without updating global counters
3093 * Delalloc need an accurate free block accounting. So switch
3094 * to non delalloc when we are near to error range.
3096 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3097 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3098 if (2 * free_blocks < 3 * dirty_blocks ||
3099 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3101 * free block count is less than 150% of dirty blocks
3102 * or free blocks is less than watermark
3107 * Even if we don't switch but are nearing capacity,
3108 * start pushing delalloc when 1/2 of free blocks are dirty.
3110 if (free_blocks < 2 * dirty_blocks)
3111 writeback_inodes_sb_if_idle(sb);
3116 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3117 loff_t pos, unsigned len, unsigned flags,
3118 struct page **pagep, void **fsdata)
3120 int ret, retries = 0;
3123 struct inode *inode = mapping->host;
3126 index = pos >> PAGE_CACHE_SHIFT;
3128 if (ext4_nonda_switch(inode->i_sb)) {
3129 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3130 return ext4_write_begin(file, mapping, pos,
3131 len, flags, pagep, fsdata);
3133 *fsdata = (void *)0;
3134 trace_ext4_da_write_begin(inode, pos, len, flags);
3137 * With delayed allocation, we don't log the i_disksize update
3138 * if there is delayed block allocation. But we still need
3139 * to journalling the i_disksize update if writes to the end
3140 * of file which has an already mapped buffer.
3142 handle = ext4_journal_start(inode, 1);
3143 if (IS_ERR(handle)) {
3144 ret = PTR_ERR(handle);
3147 /* We cannot recurse into the filesystem as the transaction is already
3149 flags |= AOP_FLAG_NOFS;
3151 page = grab_cache_page_write_begin(mapping, index, flags);
3153 ext4_journal_stop(handle);
3159 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3162 ext4_journal_stop(handle);
3163 page_cache_release(page);
3165 * block_write_begin may have instantiated a few blocks
3166 * outside i_size. Trim these off again. Don't need
3167 * i_size_read because we hold i_mutex.
3169 if (pos + len > inode->i_size)
3170 ext4_truncate_failed_write(inode);
3173 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3180 * Check if we should update i_disksize
3181 * when write to the end of file but not require block allocation
3183 static int ext4_da_should_update_i_disksize(struct page *page,
3184 unsigned long offset)
3186 struct buffer_head *bh;
3187 struct inode *inode = page->mapping->host;
3191 bh = page_buffers(page);
3192 idx = offset >> inode->i_blkbits;
3194 for (i = 0; i < idx; i++)
3195 bh = bh->b_this_page;
3197 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3202 static int ext4_da_write_end(struct file *file,
3203 struct address_space *mapping,
3204 loff_t pos, unsigned len, unsigned copied,
3205 struct page *page, void *fsdata)
3207 struct inode *inode = mapping->host;
3209 handle_t *handle = ext4_journal_current_handle();
3211 unsigned long start, end;
3212 int write_mode = (int)(unsigned long)fsdata;
3214 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3215 if (ext4_should_order_data(inode)) {
3216 return ext4_ordered_write_end(file, mapping, pos,
3217 len, copied, page, fsdata);
3218 } else if (ext4_should_writeback_data(inode)) {
3219 return ext4_writeback_write_end(file, mapping, pos,
3220 len, copied, page, fsdata);
3226 trace_ext4_da_write_end(inode, pos, len, copied);
3227 start = pos & (PAGE_CACHE_SIZE - 1);
3228 end = start + copied - 1;
3231 * generic_write_end() will run mark_inode_dirty() if i_size
3232 * changes. So let's piggyback the i_disksize mark_inode_dirty
3236 new_i_size = pos + copied;
3237 if (new_i_size > EXT4_I(inode)->i_disksize) {
3238 if (ext4_da_should_update_i_disksize(page, end)) {
3239 down_write(&EXT4_I(inode)->i_data_sem);
3240 if (new_i_size > EXT4_I(inode)->i_disksize) {
3242 * Updating i_disksize when extending file
3243 * without needing block allocation
3245 if (ext4_should_order_data(inode))
3246 ret = ext4_jbd2_file_inode(handle,
3249 EXT4_I(inode)->i_disksize = new_i_size;
3251 up_write(&EXT4_I(inode)->i_data_sem);
3252 /* We need to mark inode dirty even if
3253 * new_i_size is less that inode->i_size
3254 * bu greater than i_disksize.(hint delalloc)
3256 ext4_mark_inode_dirty(handle, inode);
3259 ret2 = generic_write_end(file, mapping, pos, len, copied,
3264 ret2 = ext4_journal_stop(handle);
3268 return ret ? ret : copied;
3271 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3274 * Drop reserved blocks
3276 BUG_ON(!PageLocked(page));
3277 if (!page_has_buffers(page))
3280 ext4_da_page_release_reservation(page, offset);
3283 ext4_invalidatepage(page, offset);
3289 * Force all delayed allocation blocks to be allocated for a given inode.
3291 int ext4_alloc_da_blocks(struct inode *inode)
3293 trace_ext4_alloc_da_blocks(inode);
3295 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3296 !EXT4_I(inode)->i_reserved_meta_blocks)
3300 * We do something simple for now. The filemap_flush() will
3301 * also start triggering a write of the data blocks, which is
3302 * not strictly speaking necessary (and for users of
3303 * laptop_mode, not even desirable). However, to do otherwise
3304 * would require replicating code paths in:
3306 * ext4_da_writepages() ->
3307 * write_cache_pages() ---> (via passed in callback function)
3308 * __mpage_da_writepage() -->
3309 * mpage_add_bh_to_extent()
3310 * mpage_da_map_blocks()
3312 * The problem is that write_cache_pages(), located in
3313 * mm/page-writeback.c, marks pages clean in preparation for
3314 * doing I/O, which is not desirable if we're not planning on
3317 * We could call write_cache_pages(), and then redirty all of
3318 * the pages by calling redirty_page_for_writepage() but that
3319 * would be ugly in the extreme. So instead we would need to
3320 * replicate parts of the code in the above functions,
3321 * simplifying them because we wouldn't actually intend to
3322 * write out the pages, but rather only collect contiguous
3323 * logical block extents, call the multi-block allocator, and
3324 * then update the buffer heads with the block allocations.
3326 * For now, though, we'll cheat by calling filemap_flush(),
3327 * which will map the blocks, and start the I/O, but not
3328 * actually wait for the I/O to complete.
3330 return filemap_flush(inode->i_mapping);
3334 * bmap() is special. It gets used by applications such as lilo and by
3335 * the swapper to find the on-disk block of a specific piece of data.
3337 * Naturally, this is dangerous if the block concerned is still in the
3338 * journal. If somebody makes a swapfile on an ext4 data-journaling
3339 * filesystem and enables swap, then they may get a nasty shock when the
3340 * data getting swapped to that swapfile suddenly gets overwritten by
3341 * the original zero's written out previously to the journal and
3342 * awaiting writeback in the kernel's buffer cache.
3344 * So, if we see any bmap calls here on a modified, data-journaled file,
3345 * take extra steps to flush any blocks which might be in the cache.
3347 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3349 struct inode *inode = mapping->host;
3353 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3354 test_opt(inode->i_sb, DELALLOC)) {
3356 * With delalloc we want to sync the file
3357 * so that we can make sure we allocate
3360 filemap_write_and_wait(mapping);
3363 if (EXT4_JOURNAL(inode) &&
3364 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3366 * This is a REALLY heavyweight approach, but the use of
3367 * bmap on dirty files is expected to be extremely rare:
3368 * only if we run lilo or swapon on a freshly made file
3369 * do we expect this to happen.
3371 * (bmap requires CAP_SYS_RAWIO so this does not
3372 * represent an unprivileged user DOS attack --- we'd be
3373 * in trouble if mortal users could trigger this path at
3376 * NB. EXT4_STATE_JDATA is not set on files other than
3377 * regular files. If somebody wants to bmap a directory
3378 * or symlink and gets confused because the buffer
3379 * hasn't yet been flushed to disk, they deserve
3380 * everything they get.
3383 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3384 journal = EXT4_JOURNAL(inode);
3385 jbd2_journal_lock_updates(journal);
3386 err = jbd2_journal_flush(journal);
3387 jbd2_journal_unlock_updates(journal);
3393 return generic_block_bmap(mapping, block, ext4_get_block);
3396 static int ext4_readpage(struct file *file, struct page *page)
3398 trace_ext4_readpage(page);
3399 return mpage_readpage(page, ext4_get_block);
3403 ext4_readpages(struct file *file, struct address_space *mapping,
3404 struct list_head *pages, unsigned nr_pages)
3406 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3409 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3411 struct buffer_head *head, *bh;
3412 unsigned int curr_off = 0;
3414 if (!page_has_buffers(page))
3416 head = bh = page_buffers(page);
3418 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3420 ext4_free_io_end(bh->b_private);
3421 bh->b_private = NULL;
3422 bh->b_end_io = NULL;
3424 curr_off = curr_off + bh->b_size;
3425 bh = bh->b_this_page;
3426 } while (bh != head);
3429 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3431 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3433 trace_ext4_invalidatepage(page, offset);
3436 * free any io_end structure allocated for buffers to be discarded
3438 if (ext4_should_dioread_nolock(page->mapping->host))
3439 ext4_invalidatepage_free_endio(page, offset);
3441 * If it's a full truncate we just forget about the pending dirtying
3444 ClearPageChecked(page);
3447 jbd2_journal_invalidatepage(journal, page, offset);
3449 block_invalidatepage(page, offset);
3452 static int ext4_releasepage(struct page *page, gfp_t wait)
3454 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3456 trace_ext4_releasepage(page);
3458 WARN_ON(PageChecked(page));
3459 if (!page_has_buffers(page))
3462 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3464 return try_to_free_buffers(page);
3468 * O_DIRECT for ext3 (or indirect map) based files
3470 * If the O_DIRECT write will extend the file then add this inode to the
3471 * orphan list. So recovery will truncate it back to the original size
3472 * if the machine crashes during the write.
3474 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3475 * crashes then stale disk data _may_ be exposed inside the file. But current
3476 * VFS code falls back into buffered path in that case so we are safe.
3478 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3479 const struct iovec *iov, loff_t offset,
3480 unsigned long nr_segs)
3482 struct file *file = iocb->ki_filp;
3483 struct inode *inode = file->f_mapping->host;
3484 struct ext4_inode_info *ei = EXT4_I(inode);
3488 size_t count = iov_length(iov, nr_segs);
3492 loff_t final_size = offset + count;
3494 if (final_size > inode->i_size) {
3495 /* Credits for sb + inode write */
3496 handle = ext4_journal_start(inode, 2);
3497 if (IS_ERR(handle)) {
3498 ret = PTR_ERR(handle);
3501 ret = ext4_orphan_add(handle, inode);
3503 ext4_journal_stop(handle);
3507 ei->i_disksize = inode->i_size;
3508 ext4_journal_stop(handle);
3513 if (rw == READ && ext4_should_dioread_nolock(inode))
3514 ret = __blockdev_direct_IO(rw, iocb, inode,
3515 inode->i_sb->s_bdev, iov,
3517 ext4_get_block, NULL, NULL, 0);
3519 ret = blockdev_direct_IO(rw, iocb, inode,
3520 inode->i_sb->s_bdev, iov,
3522 ext4_get_block, NULL);
3524 if (unlikely((rw & WRITE) && ret < 0)) {
3525 loff_t isize = i_size_read(inode);
3526 loff_t end = offset + iov_length(iov, nr_segs);
3529 ext4_truncate_failed_write(inode);
3532 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3538 /* Credits for sb + inode write */
3539 handle = ext4_journal_start(inode, 2);
3540 if (IS_ERR(handle)) {
3541 /* This is really bad luck. We've written the data
3542 * but cannot extend i_size. Bail out and pretend
3543 * the write failed... */
3544 ret = PTR_ERR(handle);
3546 ext4_orphan_del(NULL, inode);
3551 ext4_orphan_del(handle, inode);
3553 loff_t end = offset + ret;
3554 if (end > inode->i_size) {
3555 ei->i_disksize = end;
3556 i_size_write(inode, end);
3558 * We're going to return a positive `ret'
3559 * here due to non-zero-length I/O, so there's
3560 * no way of reporting error returns from
3561 * ext4_mark_inode_dirty() to userspace. So
3564 ext4_mark_inode_dirty(handle, inode);
3567 err = ext4_journal_stop(handle);
3576 * ext4_get_block used when preparing for a DIO write or buffer write.
3577 * We allocate an uinitialized extent if blocks haven't been allocated.
3578 * The extent will be converted to initialized after the IO is complete.
3580 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3581 struct buffer_head *bh_result, int create)
3583 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3584 inode->i_ino, create);
3585 return _ext4_get_block(inode, iblock, bh_result,
3586 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3589 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3590 ssize_t size, void *private, int ret,
3593 ext4_io_end_t *io_end = iocb->private;
3594 struct workqueue_struct *wq;
3595 unsigned long flags;
3596 struct ext4_inode_info *ei;
3598 /* if not async direct IO or dio with 0 bytes write, just return */
3599 if (!io_end || !size)
3602 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3603 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3604 iocb->private, io_end->inode->i_ino, iocb, offset,
3607 /* if not aio dio with unwritten extents, just free io and return */
3608 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3609 ext4_free_io_end(io_end);
3610 iocb->private = NULL;
3613 aio_complete(iocb, ret, 0);
3617 io_end->offset = offset;
3618 io_end->size = size;
3620 io_end->iocb = iocb;
3621 io_end->result = ret;
3623 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3625 /* Add the io_end to per-inode completed aio dio list*/
3626 ei = EXT4_I(io_end->inode);
3627 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3628 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3629 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3631 /* queue the work to convert unwritten extents to written */
3632 queue_work(wq, &io_end->work);
3633 iocb->private = NULL;
3636 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3638 ext4_io_end_t *io_end = bh->b_private;
3639 struct workqueue_struct *wq;
3640 struct inode *inode;
3641 unsigned long flags;
3643 if (!test_clear_buffer_uninit(bh) || !io_end)
3646 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3647 printk("sb umounted, discard end_io request for inode %lu\n",
3648 io_end->inode->i_ino);
3649 ext4_free_io_end(io_end);
3654 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
3655 * but being more careful is always safe for the future change.
3657 inode = io_end->inode;
3658 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3659 io_end->flag |= EXT4_IO_END_UNWRITTEN;
3660 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
3663 /* Add the io_end to per-inode completed io list*/
3664 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3665 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3666 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3668 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3669 /* queue the work to convert unwritten extents to written */
3670 queue_work(wq, &io_end->work);
3672 bh->b_private = NULL;
3673 bh->b_end_io = NULL;
3674 clear_buffer_uninit(bh);
3675 end_buffer_async_write(bh, uptodate);
3678 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3680 ext4_io_end_t *io_end;
3681 struct page *page = bh->b_page;
3682 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3683 size_t size = bh->b_size;
3686 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3688 pr_warn_ratelimited("%s: allocation fail\n", __func__);
3692 io_end->offset = offset;
3693 io_end->size = size;
3695 * We need to hold a reference to the page to make sure it
3696 * doesn't get evicted before ext4_end_io_work() has a chance
3697 * to convert the extent from written to unwritten.
3699 io_end->page = page;
3700 get_page(io_end->page);
3702 bh->b_private = io_end;
3703 bh->b_end_io = ext4_end_io_buffer_write;
3708 * For ext4 extent files, ext4 will do direct-io write to holes,
3709 * preallocated extents, and those write extend the file, no need to
3710 * fall back to buffered IO.
3712 * For holes, we fallocate those blocks, mark them as uninitialized
3713 * If those blocks were preallocated, we mark sure they are splited, but
3714 * still keep the range to write as uninitialized.
3716 * The unwrritten extents will be converted to written when DIO is completed.
3717 * For async direct IO, since the IO may still pending when return, we
3718 * set up an end_io call back function, which will do the conversion
3719 * when async direct IO completed.
3721 * If the O_DIRECT write will extend the file then add this inode to the
3722 * orphan list. So recovery will truncate it back to the original size
3723 * if the machine crashes during the write.
3726 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3727 const struct iovec *iov, loff_t offset,
3728 unsigned long nr_segs)
3730 struct file *file = iocb->ki_filp;
3731 struct inode *inode = file->f_mapping->host;
3733 size_t count = iov_length(iov, nr_segs);
3735 loff_t final_size = offset + count;
3736 if (rw == WRITE && final_size <= inode->i_size) {
3738 * We could direct write to holes and fallocate.
3740 * Allocated blocks to fill the hole are marked as uninitialized
3741 * to prevent parallel buffered read to expose the stale data
3742 * before DIO complete the data IO.
3744 * As to previously fallocated extents, ext4 get_block
3745 * will just simply mark the buffer mapped but still
3746 * keep the extents uninitialized.
3748 * for non AIO case, we will convert those unwritten extents
3749 * to written after return back from blockdev_direct_IO.
3751 * for async DIO, the conversion needs to be defered when
3752 * the IO is completed. The ext4 end_io callback function
3753 * will be called to take care of the conversion work.
3754 * Here for async case, we allocate an io_end structure to
3757 iocb->private = NULL;
3758 EXT4_I(inode)->cur_aio_dio = NULL;
3759 if (!is_sync_kiocb(iocb)) {
3760 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3764 * we save the io structure for current async
3765 * direct IO, so that later ext4_map_blocks()
3766 * could flag the io structure whether there
3767 * is a unwritten extents needs to be converted
3768 * when IO is completed.
3770 EXT4_I(inode)->cur_aio_dio = iocb->private;
3773 ret = blockdev_direct_IO(rw, iocb, inode,
3774 inode->i_sb->s_bdev, iov,
3776 ext4_get_block_write,
3779 EXT4_I(inode)->cur_aio_dio = NULL;
3781 * The io_end structure takes a reference to the inode,
3782 * that structure needs to be destroyed and the
3783 * reference to the inode need to be dropped, when IO is
3784 * complete, even with 0 byte write, or failed.
3786 * In the successful AIO DIO case, the io_end structure will be
3787 * desctroyed and the reference to the inode will be dropped
3788 * after the end_io call back function is called.
3790 * In the case there is 0 byte write, or error case, since
3791 * VFS direct IO won't invoke the end_io call back function,
3792 * we need to free the end_io structure here.
3794 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3795 ext4_free_io_end(iocb->private);
3796 iocb->private = NULL;
3797 } else if (ret > 0 && ext4_test_inode_state(inode,
3798 EXT4_STATE_DIO_UNWRITTEN)) {
3801 * for non AIO case, since the IO is already
3802 * completed, we could do the conversion right here
3804 err = ext4_convert_unwritten_extents(inode,
3808 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3813 /* for write the the end of file case, we fall back to old way */
3814 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3817 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3818 const struct iovec *iov, loff_t offset,
3819 unsigned long nr_segs)
3821 struct file *file = iocb->ki_filp;
3822 struct inode *inode = file->f_mapping->host;
3825 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
3826 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3827 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3829 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3830 trace_ext4_direct_IO_exit(inode, offset,
3831 iov_length(iov, nr_segs), rw, ret);
3836 * Pages can be marked dirty completely asynchronously from ext4's journalling
3837 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3838 * much here because ->set_page_dirty is called under VFS locks. The page is
3839 * not necessarily locked.
3841 * We cannot just dirty the page and leave attached buffers clean, because the
3842 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3843 * or jbddirty because all the journalling code will explode.
3845 * So what we do is to mark the page "pending dirty" and next time writepage
3846 * is called, propagate that into the buffers appropriately.
3848 static int ext4_journalled_set_page_dirty(struct page *page)
3850 SetPageChecked(page);
3851 return __set_page_dirty_nobuffers(page);
3854 static const struct address_space_operations ext4_ordered_aops = {
3855 .readpage = ext4_readpage,
3856 .readpages = ext4_readpages,
3857 .writepage = ext4_writepage,
3858 .write_begin = ext4_write_begin,
3859 .write_end = ext4_ordered_write_end,
3861 .invalidatepage = ext4_invalidatepage,
3862 .releasepage = ext4_releasepage,
3863 .direct_IO = ext4_direct_IO,
3864 .migratepage = buffer_migrate_page,
3865 .is_partially_uptodate = block_is_partially_uptodate,
3866 .error_remove_page = generic_error_remove_page,
3869 static const struct address_space_operations ext4_writeback_aops = {
3870 .readpage = ext4_readpage,
3871 .readpages = ext4_readpages,
3872 .writepage = ext4_writepage,
3873 .write_begin = ext4_write_begin,
3874 .write_end = ext4_writeback_write_end,
3876 .invalidatepage = ext4_invalidatepage,
3877 .releasepage = ext4_releasepage,
3878 .direct_IO = ext4_direct_IO,
3879 .migratepage = buffer_migrate_page,
3880 .is_partially_uptodate = block_is_partially_uptodate,
3881 .error_remove_page = generic_error_remove_page,
3884 static const struct address_space_operations ext4_journalled_aops = {
3885 .readpage = ext4_readpage,
3886 .readpages = ext4_readpages,
3887 .writepage = ext4_writepage,
3888 .write_begin = ext4_write_begin,
3889 .write_end = ext4_journalled_write_end,
3890 .set_page_dirty = ext4_journalled_set_page_dirty,
3892 .invalidatepage = ext4_invalidatepage,
3893 .releasepage = ext4_releasepage,
3894 .is_partially_uptodate = block_is_partially_uptodate,
3895 .error_remove_page = generic_error_remove_page,
3898 static const struct address_space_operations ext4_da_aops = {
3899 .readpage = ext4_readpage,
3900 .readpages = ext4_readpages,
3901 .writepage = ext4_writepage,
3902 .writepages = ext4_da_writepages,
3903 .write_begin = ext4_da_write_begin,
3904 .write_end = ext4_da_write_end,
3906 .invalidatepage = ext4_da_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 void ext4_set_aops(struct inode *inode)
3916 if (ext4_should_order_data(inode) &&
3917 test_opt(inode->i_sb, DELALLOC))
3918 inode->i_mapping->a_ops = &ext4_da_aops;
3919 else if (ext4_should_order_data(inode))
3920 inode->i_mapping->a_ops = &ext4_ordered_aops;
3921 else if (ext4_should_writeback_data(inode) &&
3922 test_opt(inode->i_sb, DELALLOC))
3923 inode->i_mapping->a_ops = &ext4_da_aops;
3924 else if (ext4_should_writeback_data(inode))
3925 inode->i_mapping->a_ops = &ext4_writeback_aops;
3927 inode->i_mapping->a_ops = &ext4_journalled_aops;
3931 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3932 * up to the end of the block which corresponds to `from'.
3933 * This required during truncate. We need to physically zero the tail end
3934 * of that block so it doesn't yield old data if the file is later grown.
3936 int ext4_block_truncate_page(handle_t *handle,
3937 struct address_space *mapping, loff_t from)
3939 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3942 struct inode *inode = mapping->host;
3944 blocksize = inode->i_sb->s_blocksize;
3945 length = blocksize - (offset & (blocksize - 1));
3947 return ext4_block_zero_page_range(handle, mapping, from, length);
3951 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3952 * starting from file offset 'from'. The range to be zero'd must
3953 * be contained with in one block. If the specified range exceeds
3954 * the end of the block it will be shortened to end of the block
3955 * that cooresponds to 'from'
3957 int ext4_block_zero_page_range(handle_t *handle,
3958 struct address_space *mapping, loff_t from, loff_t length)
3960 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3961 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3962 unsigned blocksize, max, pos;
3964 struct inode *inode = mapping->host;
3965 struct buffer_head *bh;
3969 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3970 mapping_gfp_mask(mapping) & ~__GFP_FS);
3974 blocksize = inode->i_sb->s_blocksize;
3975 max = blocksize - (offset & (blocksize - 1));
3978 * correct length if it does not fall between
3979 * 'from' and the end of the block
3981 if (length > max || length < 0)
3984 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3986 if (!page_has_buffers(page))
3987 create_empty_buffers(page, blocksize, 0);
3989 /* Find the buffer that contains "offset" */
3990 bh = page_buffers(page);
3992 while (offset >= pos) {
3993 bh = bh->b_this_page;
3999 if (buffer_freed(bh)) {
4000 BUFFER_TRACE(bh, "freed: skip");
4004 if (!buffer_mapped(bh)) {
4005 BUFFER_TRACE(bh, "unmapped");
4006 ext4_get_block(inode, iblock, bh, 0);
4007 /* unmapped? It's a hole - nothing to do */
4008 if (!buffer_mapped(bh)) {
4009 BUFFER_TRACE(bh, "still unmapped");
4014 /* Ok, it's mapped. Make sure it's up-to-date */
4015 if (PageUptodate(page))
4016 set_buffer_uptodate(bh);
4018 if (!buffer_uptodate(bh)) {
4020 ll_rw_block(READ, 1, &bh);
4022 /* Uhhuh. Read error. Complain and punt. */
4023 if (!buffer_uptodate(bh))
4027 if (ext4_should_journal_data(inode)) {
4028 BUFFER_TRACE(bh, "get write access");
4029 err = ext4_journal_get_write_access(handle, bh);
4034 zero_user(page, offset, length);
4036 BUFFER_TRACE(bh, "zeroed end of block");
4039 if (ext4_should_journal_data(inode)) {
4040 err = ext4_handle_dirty_metadata(handle, inode, bh);
4042 if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
4043 err = ext4_jbd2_file_inode(handle, inode);
4044 mark_buffer_dirty(bh);
4049 page_cache_release(page);
4054 * Probably it should be a library function... search for first non-zero word
4055 * or memcmp with zero_page, whatever is better for particular architecture.
4058 static inline int all_zeroes(__le32 *p, __le32 *q)
4067 * ext4_find_shared - find the indirect blocks for partial truncation.
4068 * @inode: inode in question
4069 * @depth: depth of the affected branch
4070 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4071 * @chain: place to store the pointers to partial indirect blocks
4072 * @top: place to the (detached) top of branch
4074 * This is a helper function used by ext4_truncate().
4076 * When we do truncate() we may have to clean the ends of several
4077 * indirect blocks but leave the blocks themselves alive. Block is
4078 * partially truncated if some data below the new i_size is referred
4079 * from it (and it is on the path to the first completely truncated
4080 * data block, indeed). We have to free the top of that path along
4081 * with everything to the right of the path. Since no allocation
4082 * past the truncation point is possible until ext4_truncate()
4083 * finishes, we may safely do the latter, but top of branch may
4084 * require special attention - pageout below the truncation point
4085 * might try to populate it.
4087 * We atomically detach the top of branch from the tree, store the
4088 * block number of its root in *@top, pointers to buffer_heads of
4089 * partially truncated blocks - in @chain[].bh and pointers to
4090 * their last elements that should not be removed - in
4091 * @chain[].p. Return value is the pointer to last filled element
4094 * The work left to caller to do the actual freeing of subtrees:
4095 * a) free the subtree starting from *@top
4096 * b) free the subtrees whose roots are stored in
4097 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4098 * c) free the subtrees growing from the inode past the @chain[0].
4099 * (no partially truncated stuff there). */
4101 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4102 ext4_lblk_t offsets[4], Indirect chain[4],
4105 Indirect *partial, *p;
4109 /* Make k index the deepest non-null offset + 1 */
4110 for (k = depth; k > 1 && !offsets[k-1]; k--)
4112 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4113 /* Writer: pointers */
4115 partial = chain + k-1;
4117 * If the branch acquired continuation since we've looked at it -
4118 * fine, it should all survive and (new) top doesn't belong to us.
4120 if (!partial->key && *partial->p)
4123 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4126 * OK, we've found the last block that must survive. The rest of our
4127 * branch should be detached before unlocking. However, if that rest
4128 * of branch is all ours and does not grow immediately from the inode
4129 * it's easier to cheat and just decrement partial->p.
4131 if (p == chain + k - 1 && p > chain) {
4135 /* Nope, don't do this in ext4. Must leave the tree intact */
4142 while (partial > p) {
4143 brelse(partial->bh);
4151 * Zero a number of block pointers in either an inode or an indirect block.
4152 * If we restart the transaction we must again get write access to the
4153 * indirect block for further modification.
4155 * We release `count' blocks on disk, but (last - first) may be greater
4156 * than `count' because there can be holes in there.
4158 * Return 0 on success, 1 on invalid block range
4159 * and < 0 on fatal error.
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;
4171 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4172 flags |= EXT4_FREE_BLOCKS_METADATA;
4174 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4176 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4177 "blocks %llu len %lu",
4178 (unsigned long long) block_to_free, count);
4182 if (try_to_extend_transaction(handle, inode)) {
4184 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4185 err = ext4_handle_dirty_metadata(handle, inode, bh);
4189 err = ext4_mark_inode_dirty(handle, inode);
4192 err = ext4_truncate_restart_trans(handle, inode,
4193 blocks_for_truncate(inode));
4197 BUFFER_TRACE(bh, "retaking write access");
4198 err = ext4_journal_get_write_access(handle, bh);
4204 for (p = first; p < last; p++)
4207 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4210 ext4_std_error(inode->i_sb, err);
4215 * ext4_free_data - free a list of data blocks
4216 * @handle: handle for this transaction
4217 * @inode: inode we are dealing with
4218 * @this_bh: indirect buffer_head which contains *@first and *@last
4219 * @first: array of block numbers
4220 * @last: points immediately past the end of array
4222 * We are freeing all blocks referred from that array (numbers are stored as
4223 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4225 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4226 * blocks are contiguous then releasing them at one time will only affect one
4227 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4228 * actually use a lot of journal space.
4230 * @this_bh will be %NULL if @first and @last point into the inode's direct
4233 static void ext4_free_data(handle_t *handle, struct inode *inode,
4234 struct buffer_head *this_bh,
4235 __le32 *first, __le32 *last)
4237 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4238 unsigned long count = 0; /* Number of blocks in the run */
4239 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4242 ext4_fsblk_t nr; /* Current block # */
4243 __le32 *p; /* Pointer into inode/ind
4244 for current block */
4247 if (this_bh) { /* For indirect block */
4248 BUFFER_TRACE(this_bh, "get_write_access");
4249 err = ext4_journal_get_write_access(handle, this_bh);
4250 /* Important: if we can't update the indirect pointers
4251 * to the blocks, we can't free them. */
4256 for (p = first; p < last; p++) {
4257 nr = le32_to_cpu(*p);
4259 /* accumulate blocks to free if they're contiguous */
4262 block_to_free_p = p;
4264 } else if (nr == block_to_free + count) {
4267 err = ext4_clear_blocks(handle, inode, this_bh,
4268 block_to_free, count,
4269 block_to_free_p, p);
4273 block_to_free_p = p;
4279 if (!err && count > 0)
4280 err = ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4281 count, block_to_free_p, p);
4287 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4290 * The buffer head should have an attached journal head at this
4291 * point. However, if the data is corrupted and an indirect
4292 * block pointed to itself, it would have been detached when
4293 * the block was cleared. Check for this instead of OOPSing.
4295 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4296 ext4_handle_dirty_metadata(handle, inode, this_bh);
4298 EXT4_ERROR_INODE(inode,
4299 "circular indirect block detected at "
4301 (unsigned long long) this_bh->b_blocknr);
4306 * ext4_free_branches - free an array of branches
4307 * @handle: JBD handle for this transaction
4308 * @inode: inode we are dealing with
4309 * @parent_bh: the buffer_head which contains *@first and *@last
4310 * @first: array of block numbers
4311 * @last: pointer immediately past the end of array
4312 * @depth: depth of the branches to free
4314 * We are freeing all blocks referred from these branches (numbers are
4315 * stored as little-endian 32-bit) and updating @inode->i_blocks
4318 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4319 struct buffer_head *parent_bh,
4320 __le32 *first, __le32 *last, int depth)
4325 if (ext4_handle_is_aborted(handle))
4329 struct buffer_head *bh;
4330 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4332 while (--p >= first) {
4333 nr = le32_to_cpu(*p);
4335 continue; /* A hole */
4337 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4339 EXT4_ERROR_INODE(inode,
4340 "invalid indirect mapped "
4341 "block %lu (level %d)",
4342 (unsigned long) nr, depth);
4346 /* Go read the buffer for the next level down */
4347 bh = sb_bread(inode->i_sb, nr);
4350 * A read failure? Report error and clear slot
4354 EXT4_ERROR_INODE_BLOCK(inode, nr,
4359 /* This zaps the entire block. Bottom up. */
4360 BUFFER_TRACE(bh, "free child branches");
4361 ext4_free_branches(handle, inode, bh,
4362 (__le32 *) bh->b_data,
4363 (__le32 *) bh->b_data + addr_per_block,
4368 * Everything below this this pointer has been
4369 * released. Now let this top-of-subtree go.
4371 * We want the freeing of this indirect block to be
4372 * atomic in the journal with the updating of the
4373 * bitmap block which owns it. So make some room in
4376 * We zero the parent pointer *after* freeing its
4377 * pointee in the bitmaps, so if extend_transaction()
4378 * for some reason fails to put the bitmap changes and
4379 * the release into the same transaction, recovery
4380 * will merely complain about releasing a free block,
4381 * rather than leaking blocks.
4383 if (ext4_handle_is_aborted(handle))
4385 if (try_to_extend_transaction(handle, inode)) {
4386 ext4_mark_inode_dirty(handle, inode);
4387 ext4_truncate_restart_trans(handle, inode,
4388 blocks_for_truncate(inode));
4392 * The forget flag here is critical because if
4393 * we are journaling (and not doing data
4394 * journaling), we have to make sure a revoke
4395 * record is written to prevent the journal
4396 * replay from overwriting the (former)
4397 * indirect block if it gets reallocated as a
4398 * data block. This must happen in the same
4399 * transaction where the data blocks are
4402 ext4_free_blocks(handle, inode, NULL, nr, 1,
4403 EXT4_FREE_BLOCKS_METADATA|
4404 EXT4_FREE_BLOCKS_FORGET);
4408 * The block which we have just freed is
4409 * pointed to by an indirect block: journal it
4411 BUFFER_TRACE(parent_bh, "get_write_access");
4412 if (!ext4_journal_get_write_access(handle,
4415 BUFFER_TRACE(parent_bh,
4416 "call ext4_handle_dirty_metadata");
4417 ext4_handle_dirty_metadata(handle,
4424 /* We have reached the bottom of the tree. */
4425 BUFFER_TRACE(parent_bh, "free data blocks");
4426 ext4_free_data(handle, inode, parent_bh, first, last);
4430 int ext4_can_truncate(struct inode *inode)
4432 if (S_ISREG(inode->i_mode))
4434 if (S_ISDIR(inode->i_mode))
4436 if (S_ISLNK(inode->i_mode))
4437 return !ext4_inode_is_fast_symlink(inode);
4442 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
4443 * associated with the given offset and length
4445 * @inode: File inode
4446 * @offset: The offset where the hole will begin
4447 * @len: The length of the hole
4449 * Returns: 0 on sucess or negative on failure
4452 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
4454 struct inode *inode = file->f_path.dentry->d_inode;
4455 if (!S_ISREG(inode->i_mode))
4458 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4459 /* TODO: Add support for non extent hole punching */
4463 return ext4_ext_punch_hole(file, offset, length);
4469 * We block out ext4_get_block() block instantiations across the entire
4470 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4471 * simultaneously on behalf of the same inode.
4473 * As we work through the truncate and commmit bits of it to the journal there
4474 * is one core, guiding principle: the file's tree must always be consistent on
4475 * disk. We must be able to restart the truncate after a crash.
4477 * The file's tree may be transiently inconsistent in memory (although it
4478 * probably isn't), but whenever we close off and commit a journal transaction,
4479 * the contents of (the filesystem + the journal) must be consistent and
4480 * restartable. It's pretty simple, really: bottom up, right to left (although
4481 * left-to-right works OK too).
4483 * Note that at recovery time, journal replay occurs *before* the restart of
4484 * truncate against the orphan inode list.
4486 * The committed inode has the new, desired i_size (which is the same as
4487 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4488 * that this inode's truncate did not complete and it will again call
4489 * ext4_truncate() to have another go. So there will be instantiated blocks
4490 * to the right of the truncation point in a crashed ext4 filesystem. But
4491 * that's fine - as long as they are linked from the inode, the post-crash
4492 * ext4_truncate() run will find them and release them.
4494 void ext4_truncate(struct inode *inode)
4497 struct ext4_inode_info *ei = EXT4_I(inode);
4498 __le32 *i_data = ei->i_data;
4499 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4500 struct address_space *mapping = inode->i_mapping;
4501 ext4_lblk_t offsets[4];
4506 ext4_lblk_t last_block, max_block;
4507 unsigned blocksize = inode->i_sb->s_blocksize;
4509 trace_ext4_truncate_enter(inode);
4511 if (!ext4_can_truncate(inode))
4514 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4516 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4517 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4519 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4520 ext4_ext_truncate(inode);
4521 trace_ext4_truncate_exit(inode);
4525 handle = start_transaction(inode);
4527 return; /* AKPM: return what? */
4529 last_block = (inode->i_size + blocksize-1)
4530 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4531 max_block = (EXT4_SB(inode->i_sb)->s_bitmap_maxbytes + blocksize-1)
4532 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4534 if (inode->i_size & (blocksize - 1))
4535 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4538 if (last_block != max_block) {
4539 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4541 goto out_stop; /* error */
4545 * OK. This truncate is going to happen. We add the inode to the
4546 * orphan list, so that if this truncate spans multiple transactions,
4547 * and we crash, we will resume the truncate when the filesystem
4548 * recovers. It also marks the inode dirty, to catch the new size.
4550 * Implication: the file must always be in a sane, consistent
4551 * truncatable state while each transaction commits.
4553 if (ext4_orphan_add(handle, inode))
4557 * From here we block out all ext4_get_block() callers who want to
4558 * modify the block allocation tree.
4560 down_write(&ei->i_data_sem);
4562 ext4_discard_preallocations(inode);
4565 * The orphan list entry will now protect us from any crash which
4566 * occurs before the truncate completes, so it is now safe to propagate
4567 * the new, shorter inode size (held for now in i_size) into the
4568 * on-disk inode. We do this via i_disksize, which is the value which
4569 * ext4 *really* writes onto the disk inode.
4571 ei->i_disksize = inode->i_size;
4573 if (last_block == max_block) {
4575 * It is unnecessary to free any data blocks if last_block is
4576 * equal to the indirect block limit.
4579 } else if (n == 1) { /* direct blocks */
4580 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4581 i_data + EXT4_NDIR_BLOCKS);
4585 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4586 /* Kill the top of shared branch (not detached) */
4588 if (partial == chain) {
4589 /* Shared branch grows from the inode */
4590 ext4_free_branches(handle, inode, NULL,
4591 &nr, &nr+1, (chain+n-1) - partial);
4594 * We mark the inode dirty prior to restart,
4595 * and prior to stop. No need for it here.
4598 /* Shared branch grows from an indirect block */
4599 BUFFER_TRACE(partial->bh, "get_write_access");
4600 ext4_free_branches(handle, inode, partial->bh,
4602 partial->p+1, (chain+n-1) - partial);
4605 /* Clear the ends of indirect blocks on the shared branch */
4606 while (partial > chain) {
4607 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4608 (__le32*)partial->bh->b_data+addr_per_block,
4609 (chain+n-1) - partial);
4610 BUFFER_TRACE(partial->bh, "call brelse");
4611 brelse(partial->bh);
4615 /* Kill the remaining (whole) subtrees */
4616 switch (offsets[0]) {
4618 nr = i_data[EXT4_IND_BLOCK];
4620 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4621 i_data[EXT4_IND_BLOCK] = 0;
4623 case EXT4_IND_BLOCK:
4624 nr = i_data[EXT4_DIND_BLOCK];
4626 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4627 i_data[EXT4_DIND_BLOCK] = 0;
4629 case EXT4_DIND_BLOCK:
4630 nr = i_data[EXT4_TIND_BLOCK];
4632 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4633 i_data[EXT4_TIND_BLOCK] = 0;
4635 case EXT4_TIND_BLOCK:
4640 up_write(&ei->i_data_sem);
4641 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4642 ext4_mark_inode_dirty(handle, inode);
4645 * In a multi-transaction truncate, we only make the final transaction
4649 ext4_handle_sync(handle);
4652 * If this was a simple ftruncate(), and the file will remain alive
4653 * then we need to clear up the orphan record which we created above.
4654 * However, if this was a real unlink then we were called by
4655 * ext4_delete_inode(), and we allow that function to clean up the
4656 * orphan info for us.
4659 ext4_orphan_del(handle, inode);
4661 ext4_journal_stop(handle);
4662 trace_ext4_truncate_exit(inode);
4666 * ext4_get_inode_loc returns with an extra refcount against the inode's
4667 * underlying buffer_head on success. If 'in_mem' is true, we have all
4668 * data in memory that is needed to recreate the on-disk version of this
4671 static int __ext4_get_inode_loc(struct inode *inode,
4672 struct ext4_iloc *iloc, int in_mem)
4674 struct ext4_group_desc *gdp;
4675 struct buffer_head *bh;
4676 struct super_block *sb = inode->i_sb;
4678 int inodes_per_block, inode_offset;
4681 if (!ext4_valid_inum(sb, inode->i_ino))
4684 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4685 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4690 * Figure out the offset within the block group inode table
4692 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4693 inode_offset = ((inode->i_ino - 1) %
4694 EXT4_INODES_PER_GROUP(sb));
4695 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4696 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4698 bh = sb_getblk(sb, block);
4700 EXT4_ERROR_INODE_BLOCK(inode, block,
4701 "unable to read itable block");
4704 if (!buffer_uptodate(bh)) {
4708 * If the buffer has the write error flag, we have failed
4709 * to write out another inode in the same block. In this
4710 * case, we don't have to read the block because we may
4711 * read the old inode data successfully.
4713 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4714 set_buffer_uptodate(bh);
4716 if (buffer_uptodate(bh)) {
4717 /* someone brought it uptodate while we waited */
4723 * If we have all information of the inode in memory and this
4724 * is the only valid inode in the block, we need not read the
4728 struct buffer_head *bitmap_bh;
4731 start = inode_offset & ~(inodes_per_block - 1);
4733 /* Is the inode bitmap in cache? */
4734 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4739 * If the inode bitmap isn't in cache then the
4740 * optimisation may end up performing two reads instead
4741 * of one, so skip it.
4743 if (!buffer_uptodate(bitmap_bh)) {
4747 for (i = start; i < start + inodes_per_block; i++) {
4748 if (i == inode_offset)
4750 if (ext4_test_bit(i, bitmap_bh->b_data))
4754 if (i == start + inodes_per_block) {
4755 /* all other inodes are free, so skip I/O */
4756 memset(bh->b_data, 0, bh->b_size);
4757 set_buffer_uptodate(bh);
4765 * If we need to do any I/O, try to pre-readahead extra
4766 * blocks from the inode table.
4768 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4769 ext4_fsblk_t b, end, table;
4772 table = ext4_inode_table(sb, gdp);
4773 /* s_inode_readahead_blks is always a power of 2 */
4774 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4777 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4778 num = EXT4_INODES_PER_GROUP(sb);
4779 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4780 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4781 num -= ext4_itable_unused_count(sb, gdp);
4782 table += num / inodes_per_block;
4786 sb_breadahead(sb, b++);
4790 * There are other valid inodes in the buffer, this inode
4791 * has in-inode xattrs, or we don't have this inode in memory.
4792 * Read the block from disk.
4794 trace_ext4_load_inode(inode);
4796 bh->b_end_io = end_buffer_read_sync;
4797 submit_bh(READ_META, bh);
4799 if (!buffer_uptodate(bh)) {
4800 EXT4_ERROR_INODE_BLOCK(inode, block,
4801 "unable to read itable block");
4811 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4813 /* We have all inode data except xattrs in memory here. */
4814 return __ext4_get_inode_loc(inode, iloc,
4815 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4818 void ext4_set_inode_flags(struct inode *inode)
4820 unsigned int flags = EXT4_I(inode)->i_flags;
4822 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4823 if (flags & EXT4_SYNC_FL)
4824 inode->i_flags |= S_SYNC;
4825 if (flags & EXT4_APPEND_FL)
4826 inode->i_flags |= S_APPEND;
4827 if (flags & EXT4_IMMUTABLE_FL)
4828 inode->i_flags |= S_IMMUTABLE;
4829 if (flags & EXT4_NOATIME_FL)
4830 inode->i_flags |= S_NOATIME;
4831 if (flags & EXT4_DIRSYNC_FL)
4832 inode->i_flags |= S_DIRSYNC;
4835 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4836 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4838 unsigned int vfs_fl;
4839 unsigned long old_fl, new_fl;
4842 vfs_fl = ei->vfs_inode.i_flags;
4843 old_fl = ei->i_flags;
4844 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4845 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4847 if (vfs_fl & S_SYNC)
4848 new_fl |= EXT4_SYNC_FL;
4849 if (vfs_fl & S_APPEND)
4850 new_fl |= EXT4_APPEND_FL;
4851 if (vfs_fl & S_IMMUTABLE)
4852 new_fl |= EXT4_IMMUTABLE_FL;
4853 if (vfs_fl & S_NOATIME)
4854 new_fl |= EXT4_NOATIME_FL;
4855 if (vfs_fl & S_DIRSYNC)
4856 new_fl |= EXT4_DIRSYNC_FL;
4857 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4860 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4861 struct ext4_inode_info *ei)
4864 struct inode *inode = &(ei->vfs_inode);
4865 struct super_block *sb = inode->i_sb;
4867 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4868 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4869 /* we are using combined 48 bit field */
4870 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4871 le32_to_cpu(raw_inode->i_blocks_lo);
4872 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4873 /* i_blocks represent file system block size */
4874 return i_blocks << (inode->i_blkbits - 9);
4879 return le32_to_cpu(raw_inode->i_blocks_lo);
4883 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4885 struct ext4_iloc iloc;
4886 struct ext4_inode *raw_inode;
4887 struct ext4_inode_info *ei;
4888 struct inode *inode;
4889 journal_t *journal = EXT4_SB(sb)->s_journal;
4893 inode = iget_locked(sb, ino);
4895 return ERR_PTR(-ENOMEM);
4896 if (!(inode->i_state & I_NEW))
4902 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4905 raw_inode = ext4_raw_inode(&iloc);
4906 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4907 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4908 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4909 if (!(test_opt(inode->i_sb, NO_UID32))) {
4910 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4911 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4913 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4915 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4916 ei->i_dir_start_lookup = 0;
4917 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4918 /* We now have enough fields to check if the inode was active or not.
4919 * This is needed because nfsd might try to access dead inodes
4920 * the test is that same one that e2fsck uses
4921 * NeilBrown 1999oct15
4923 if (inode->i_nlink == 0) {
4924 if (inode->i_mode == 0 ||
4925 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4926 /* this inode is deleted */
4930 /* The only unlinked inodes we let through here have
4931 * valid i_mode and are being read by the orphan
4932 * recovery code: that's fine, we're about to complete
4933 * the process of deleting those. */
4935 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4936 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4937 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4938 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4940 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4941 inode->i_size = ext4_isize(raw_inode);
4942 ei->i_disksize = inode->i_size;
4944 ei->i_reserved_quota = 0;
4946 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4947 ei->i_block_group = iloc.block_group;
4948 ei->i_last_alloc_group = ~0;
4950 * NOTE! The in-memory inode i_data array is in little-endian order
4951 * even on big-endian machines: we do NOT byteswap the block numbers!
4953 for (block = 0; block < EXT4_N_BLOCKS; block++)
4954 ei->i_data[block] = raw_inode->i_block[block];
4955 INIT_LIST_HEAD(&ei->i_orphan);
4958 * Set transaction id's of transactions that have to be committed
4959 * to finish f[data]sync. We set them to currently running transaction
4960 * as we cannot be sure that the inode or some of its metadata isn't
4961 * part of the transaction - the inode could have been reclaimed and
4962 * now it is reread from disk.
4965 transaction_t *transaction;
4968 read_lock(&journal->j_state_lock);
4969 if (journal->j_running_transaction)
4970 transaction = journal->j_running_transaction;
4972 transaction = journal->j_committing_transaction;
4974 tid = transaction->t_tid;
4976 tid = journal->j_commit_sequence;
4977 read_unlock(&journal->j_state_lock);
4978 ei->i_sync_tid = tid;
4979 ei->i_datasync_tid = tid;
4982 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4983 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4984 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4985 EXT4_INODE_SIZE(inode->i_sb)) {
4989 if (ei->i_extra_isize == 0) {
4990 /* The extra space is currently unused. Use it. */
4991 ei->i_extra_isize = sizeof(struct ext4_inode) -
4992 EXT4_GOOD_OLD_INODE_SIZE;
4994 __le32 *magic = (void *)raw_inode +
4995 EXT4_GOOD_OLD_INODE_SIZE +
4997 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4998 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
5001 ei->i_extra_isize = 0;
5003 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5004 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5005 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5006 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5008 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
5009 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5010 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5012 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5016 if (ei->i_file_acl &&
5017 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5018 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
5022 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5023 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5024 (S_ISLNK(inode->i_mode) &&
5025 !ext4_inode_is_fast_symlink(inode)))
5026 /* Validate extent which is part of inode */
5027 ret = ext4_ext_check_inode(inode);
5028 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5029 (S_ISLNK(inode->i_mode) &&
5030 !ext4_inode_is_fast_symlink(inode))) {
5031 /* Validate block references which are part of inode */
5032 ret = ext4_check_inode_blockref(inode);
5037 if (S_ISREG(inode->i_mode)) {
5038 inode->i_op = &ext4_file_inode_operations;
5039 inode->i_fop = &ext4_file_operations;
5040 ext4_set_aops(inode);
5041 } else if (S_ISDIR(inode->i_mode)) {
5042 inode->i_op = &ext4_dir_inode_operations;
5043 inode->i_fop = &ext4_dir_operations;
5044 } else if (S_ISLNK(inode->i_mode)) {
5045 if (ext4_inode_is_fast_symlink(inode)) {
5046 inode->i_op = &ext4_fast_symlink_inode_operations;
5047 nd_terminate_link(ei->i_data, inode->i_size,
5048 sizeof(ei->i_data) - 1);
5050 inode->i_op = &ext4_symlink_inode_operations;
5051 ext4_set_aops(inode);
5053 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5054 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5055 inode->i_op = &ext4_special_inode_operations;
5056 if (raw_inode->i_block[0])
5057 init_special_inode(inode, inode->i_mode,
5058 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5060 init_special_inode(inode, inode->i_mode,
5061 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5064 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5068 ext4_set_inode_flags(inode);
5069 unlock_new_inode(inode);
5075 return ERR_PTR(ret);
5078 static int ext4_inode_blocks_set(handle_t *handle,
5079 struct ext4_inode *raw_inode,
5080 struct ext4_inode_info *ei)
5082 struct inode *inode = &(ei->vfs_inode);
5083 u64 i_blocks = inode->i_blocks;
5084 struct super_block *sb = inode->i_sb;
5086 if (i_blocks <= ~0U) {
5088 * i_blocks can be represnted in a 32 bit variable
5089 * as multiple of 512 bytes
5091 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5092 raw_inode->i_blocks_high = 0;
5093 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5096 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5099 if (i_blocks <= 0xffffffffffffULL) {
5101 * i_blocks can be represented in a 48 bit variable
5102 * as multiple of 512 bytes
5104 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5105 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5106 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5108 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5109 /* i_block is stored in file system block size */
5110 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5111 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5112 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5118 * Post the struct inode info into an on-disk inode location in the
5119 * buffer-cache. This gobbles the caller's reference to the
5120 * buffer_head in the inode location struct.
5122 * The caller must have write access to iloc->bh.
5124 static int ext4_do_update_inode(handle_t *handle,
5125 struct inode *inode,
5126 struct ext4_iloc *iloc)
5128 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5129 struct ext4_inode_info *ei = EXT4_I(inode);
5130 struct buffer_head *bh = iloc->bh;
5131 int err = 0, rc, block;
5133 /* For fields not not tracking in the in-memory inode,
5134 * initialise them to zero for new inodes. */
5135 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5136 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5138 ext4_get_inode_flags(ei);
5139 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5140 if (!(test_opt(inode->i_sb, NO_UID32))) {
5141 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5142 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5144 * Fix up interoperability with old kernels. Otherwise, old inodes get
5145 * re-used with the upper 16 bits of the uid/gid intact
5148 raw_inode->i_uid_high =
5149 cpu_to_le16(high_16_bits(inode->i_uid));
5150 raw_inode->i_gid_high =
5151 cpu_to_le16(high_16_bits(inode->i_gid));
5153 raw_inode->i_uid_high = 0;
5154 raw_inode->i_gid_high = 0;
5157 raw_inode->i_uid_low =
5158 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5159 raw_inode->i_gid_low =
5160 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5161 raw_inode->i_uid_high = 0;
5162 raw_inode->i_gid_high = 0;
5164 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5166 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5167 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5168 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5169 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5171 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5173 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5174 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5175 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5176 cpu_to_le32(EXT4_OS_HURD))
5177 raw_inode->i_file_acl_high =
5178 cpu_to_le16(ei->i_file_acl >> 32);
5179 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5180 ext4_isize_set(raw_inode, ei->i_disksize);
5181 if (ei->i_disksize > 0x7fffffffULL) {
5182 struct super_block *sb = inode->i_sb;
5183 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5184 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5185 EXT4_SB(sb)->s_es->s_rev_level ==
5186 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5187 /* If this is the first large file
5188 * created, add a flag to the superblock.
5190 err = ext4_journal_get_write_access(handle,
5191 EXT4_SB(sb)->s_sbh);
5194 ext4_update_dynamic_rev(sb);
5195 EXT4_SET_RO_COMPAT_FEATURE(sb,
5196 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5198 ext4_handle_sync(handle);
5199 err = ext4_handle_dirty_metadata(handle, NULL,
5200 EXT4_SB(sb)->s_sbh);
5203 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5204 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5205 if (old_valid_dev(inode->i_rdev)) {
5206 raw_inode->i_block[0] =
5207 cpu_to_le32(old_encode_dev(inode->i_rdev));
5208 raw_inode->i_block[1] = 0;
5210 raw_inode->i_block[0] = 0;
5211 raw_inode->i_block[1] =
5212 cpu_to_le32(new_encode_dev(inode->i_rdev));
5213 raw_inode->i_block[2] = 0;
5216 for (block = 0; block < EXT4_N_BLOCKS; block++)
5217 raw_inode->i_block[block] = ei->i_data[block];
5219 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5220 if (ei->i_extra_isize) {
5221 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5222 raw_inode->i_version_hi =
5223 cpu_to_le32(inode->i_version >> 32);
5224 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5227 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5228 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5231 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5233 ext4_update_inode_fsync_trans(handle, inode, 0);
5236 ext4_std_error(inode->i_sb, err);
5241 * ext4_write_inode()
5243 * We are called from a few places:
5245 * - Within generic_file_write() for O_SYNC files.
5246 * Here, there will be no transaction running. We wait for any running
5247 * trasnaction to commit.
5249 * - Within sys_sync(), kupdate and such.
5250 * We wait on commit, if tol to.
5252 * - Within prune_icache() (PF_MEMALLOC == true)
5253 * Here we simply return. We can't afford to block kswapd on the
5256 * In all cases it is actually safe for us to return without doing anything,
5257 * because the inode has been copied into a raw inode buffer in
5258 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5261 * Note that we are absolutely dependent upon all inode dirtiers doing the
5262 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5263 * which we are interested.
5265 * It would be a bug for them to not do this. The code:
5267 * mark_inode_dirty(inode)
5269 * inode->i_size = expr;
5271 * is in error because a kswapd-driven write_inode() could occur while
5272 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5273 * will no longer be on the superblock's dirty inode list.
5275 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5279 if (current->flags & PF_MEMALLOC)
5282 if (EXT4_SB(inode->i_sb)->s_journal) {
5283 if (ext4_journal_current_handle()) {
5284 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5289 if (wbc->sync_mode != WB_SYNC_ALL)
5292 err = ext4_force_commit(inode->i_sb);
5294 struct ext4_iloc iloc;
5296 err = __ext4_get_inode_loc(inode, &iloc, 0);
5299 if (wbc->sync_mode == WB_SYNC_ALL)
5300 sync_dirty_buffer(iloc.bh);
5301 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5302 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5303 "IO error syncing inode");
5314 * Called from notify_change.
5316 * We want to trap VFS attempts to truncate the file as soon as
5317 * possible. In particular, we want to make sure that when the VFS
5318 * shrinks i_size, we put the inode on the orphan list and modify
5319 * i_disksize immediately, so that during the subsequent flushing of
5320 * dirty pages and freeing of disk blocks, we can guarantee that any
5321 * commit will leave the blocks being flushed in an unused state on
5322 * disk. (On recovery, the inode will get truncated and the blocks will
5323 * be freed, so we have a strong guarantee that no future commit will
5324 * leave these blocks visible to the user.)
5326 * Another thing we have to assure is that if we are in ordered mode
5327 * and inode is still attached to the committing transaction, we must
5328 * we start writeout of all the dirty pages which are being truncated.
5329 * This way we are sure that all the data written in the previous
5330 * transaction are already on disk (truncate waits for pages under
5333 * Called with inode->i_mutex down.
5335 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5337 struct inode *inode = dentry->d_inode;
5340 const unsigned int ia_valid = attr->ia_valid;
5342 error = inode_change_ok(inode, attr);
5346 if (is_quota_modification(inode, attr))
5347 dquot_initialize(inode);
5348 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5349 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5352 /* (user+group)*(old+new) structure, inode write (sb,
5353 * inode block, ? - but truncate inode update has it) */
5354 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5355 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5356 if (IS_ERR(handle)) {
5357 error = PTR_ERR(handle);
5360 error = dquot_transfer(inode, attr);
5362 ext4_journal_stop(handle);
5365 /* Update corresponding info in inode so that everything is in
5366 * one transaction */
5367 if (attr->ia_valid & ATTR_UID)
5368 inode->i_uid = attr->ia_uid;
5369 if (attr->ia_valid & ATTR_GID)
5370 inode->i_gid = attr->ia_gid;
5371 error = ext4_mark_inode_dirty(handle, inode);
5372 ext4_journal_stop(handle);
5375 if (attr->ia_valid & ATTR_SIZE) {
5376 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5377 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5379 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5384 if (S_ISREG(inode->i_mode) &&
5385 attr->ia_valid & ATTR_SIZE &&
5386 (attr->ia_size < inode->i_size)) {
5389 handle = ext4_journal_start(inode, 3);
5390 if (IS_ERR(handle)) {
5391 error = PTR_ERR(handle);
5394 if (ext4_handle_valid(handle)) {
5395 error = ext4_orphan_add(handle, inode);
5398 EXT4_I(inode)->i_disksize = attr->ia_size;
5399 rc = ext4_mark_inode_dirty(handle, inode);
5402 ext4_journal_stop(handle);
5404 if (ext4_should_order_data(inode)) {
5405 error = ext4_begin_ordered_truncate(inode,
5408 /* Do as much error cleanup as possible */
5409 handle = ext4_journal_start(inode, 3);
5410 if (IS_ERR(handle)) {
5411 ext4_orphan_del(NULL, inode);
5414 ext4_orphan_del(handle, inode);
5416 ext4_journal_stop(handle);
5422 if (attr->ia_valid & ATTR_SIZE) {
5423 if (attr->ia_size != i_size_read(inode)) {
5424 truncate_setsize(inode, attr->ia_size);
5425 ext4_truncate(inode);
5426 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
5427 ext4_truncate(inode);
5431 setattr_copy(inode, attr);
5432 mark_inode_dirty(inode);
5436 * If the call to ext4_truncate failed to get a transaction handle at
5437 * all, we need to clean up the in-core orphan list manually.
5439 if (orphan && inode->i_nlink)
5440 ext4_orphan_del(NULL, inode);
5442 if (!rc && (ia_valid & ATTR_MODE))
5443 rc = ext4_acl_chmod(inode);
5446 ext4_std_error(inode->i_sb, error);
5452 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5455 struct inode *inode;
5456 unsigned long delalloc_blocks;
5458 inode = dentry->d_inode;
5459 generic_fillattr(inode, stat);
5462 * We can't update i_blocks if the block allocation is delayed
5463 * otherwise in the case of system crash before the real block
5464 * allocation is done, we will have i_blocks inconsistent with
5465 * on-disk file blocks.
5466 * We always keep i_blocks updated together with real
5467 * allocation. But to not confuse with user, stat
5468 * will return the blocks that include the delayed allocation
5469 * blocks for this file.
5471 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5473 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5477 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5482 /* if nrblocks are contiguous */
5485 * With N contiguous data blocks, we need at most
5486 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) + 1 indirect blocks,
5487 * 2 dindirect blocks, and 1 tindirect block
5489 return DIV_ROUND_UP(nrblocks,
5490 EXT4_ADDR_PER_BLOCK(inode->i_sb)) + 4;
5493 * if nrblocks are not contiguous, worse case, each block touch
5494 * a indirect block, and each indirect block touch a double indirect
5495 * block, plus a triple indirect block
5497 indirects = nrblocks * 2 + 1;
5501 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5503 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5504 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5505 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5509 * Account for index blocks, block groups bitmaps and block group
5510 * descriptor blocks if modify datablocks and index blocks
5511 * worse case, the indexs blocks spread over different block groups
5513 * If datablocks are discontiguous, they are possible to spread over
5514 * different block groups too. If they are contiuguous, with flexbg,
5515 * they could still across block group boundary.
5517 * Also account for superblock, inode, quota and xattr blocks
5519 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5521 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5527 * How many index blocks need to touch to modify nrblocks?
5528 * The "Chunk" flag indicating whether the nrblocks is
5529 * physically contiguous on disk
5531 * For Direct IO and fallocate, they calls get_block to allocate
5532 * one single extent at a time, so they could set the "Chunk" flag
5534 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5539 * Now let's see how many group bitmaps and group descriptors need
5549 if (groups > ngroups)
5551 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5552 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5554 /* bitmaps and block group descriptor blocks */
5555 ret += groups + gdpblocks;
5557 /* Blocks for super block, inode, quota and xattr blocks */
5558 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5564 * Calculate the total number of credits to reserve to fit
5565 * the modification of a single pages into a single transaction,
5566 * which may include multiple chunks of block allocations.
5568 * This could be called via ext4_write_begin()
5570 * We need to consider the worse case, when
5571 * one new block per extent.
5573 int ext4_writepage_trans_blocks(struct inode *inode)
5575 int bpp = ext4_journal_blocks_per_page(inode);
5578 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5580 /* Account for data blocks for journalled mode */
5581 if (ext4_should_journal_data(inode))
5587 * Calculate the journal credits for a chunk of data modification.
5589 * This is called from DIO, fallocate or whoever calling
5590 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5592 * journal buffers for data blocks are not included here, as DIO
5593 * and fallocate do no need to journal data buffers.
5595 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5597 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5601 * The caller must have previously called ext4_reserve_inode_write().
5602 * Give this, we know that the caller already has write access to iloc->bh.
5604 int ext4_mark_iloc_dirty(handle_t *handle,
5605 struct inode *inode, struct ext4_iloc *iloc)
5609 if (test_opt(inode->i_sb, I_VERSION))
5610 inode_inc_iversion(inode);
5612 /* the do_update_inode consumes one bh->b_count */
5615 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5616 err = ext4_do_update_inode(handle, inode, iloc);
5622 * On success, We end up with an outstanding reference count against
5623 * iloc->bh. This _must_ be cleaned up later.
5627 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5628 struct ext4_iloc *iloc)
5632 err = ext4_get_inode_loc(inode, iloc);
5634 BUFFER_TRACE(iloc->bh, "get_write_access");
5635 err = ext4_journal_get_write_access(handle, iloc->bh);
5641 ext4_std_error(inode->i_sb, err);
5646 * Expand an inode by new_extra_isize bytes.
5647 * Returns 0 on success or negative error number on failure.
5649 static int ext4_expand_extra_isize(struct inode *inode,
5650 unsigned int new_extra_isize,
5651 struct ext4_iloc iloc,
5654 struct ext4_inode *raw_inode;
5655 struct ext4_xattr_ibody_header *header;
5657 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5660 raw_inode = ext4_raw_inode(&iloc);
5662 header = IHDR(inode, raw_inode);
5664 /* No extended attributes present */
5665 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5666 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5667 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5669 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5673 /* try to expand with EAs present */
5674 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5679 * What we do here is to mark the in-core inode as clean with respect to inode
5680 * dirtiness (it may still be data-dirty).
5681 * This means that the in-core inode may be reaped by prune_icache
5682 * without having to perform any I/O. This is a very good thing,
5683 * because *any* task may call prune_icache - even ones which
5684 * have a transaction open against a different journal.
5686 * Is this cheating? Not really. Sure, we haven't written the
5687 * inode out, but prune_icache isn't a user-visible syncing function.
5688 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5689 * we start and wait on commits.
5691 * Is this efficient/effective? Well, we're being nice to the system
5692 * by cleaning up our inodes proactively so they can be reaped
5693 * without I/O. But we are potentially leaving up to five seconds'
5694 * worth of inodes floating about which prune_icache wants us to
5695 * write out. One way to fix that would be to get prune_icache()
5696 * to do a write_super() to free up some memory. It has the desired
5699 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5701 struct ext4_iloc iloc;
5702 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5703 static unsigned int mnt_count;
5707 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5708 err = ext4_reserve_inode_write(handle, inode, &iloc);
5709 if (ext4_handle_valid(handle) &&
5710 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5711 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5713 * We need extra buffer credits since we may write into EA block
5714 * with this same handle. If journal_extend fails, then it will
5715 * only result in a minor loss of functionality for that inode.
5716 * If this is felt to be critical, then e2fsck should be run to
5717 * force a large enough s_min_extra_isize.
5719 if ((jbd2_journal_extend(handle,
5720 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5721 ret = ext4_expand_extra_isize(inode,
5722 sbi->s_want_extra_isize,
5725 ext4_set_inode_state(inode,
5726 EXT4_STATE_NO_EXPAND);
5728 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5729 ext4_warning(inode->i_sb,
5730 "Unable to expand inode %lu. Delete"
5731 " some EAs or run e2fsck.",
5734 le16_to_cpu(sbi->s_es->s_mnt_count);
5740 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5745 * ext4_dirty_inode() is called from __mark_inode_dirty()
5747 * We're really interested in the case where a file is being extended.
5748 * i_size has been changed by generic_commit_write() and we thus need
5749 * to include the updated inode in the current transaction.
5751 * Also, dquot_alloc_block() will always dirty the inode when blocks
5752 * are allocated to the file.
5754 * If the inode is marked synchronous, we don't honour that here - doing
5755 * so would cause a commit on atime updates, which we don't bother doing.
5756 * We handle synchronous inodes at the highest possible level.
5758 void ext4_dirty_inode(struct inode *inode, int flags)
5762 handle = ext4_journal_start(inode, 2);
5766 ext4_mark_inode_dirty(handle, inode);
5768 ext4_journal_stop(handle);
5775 * Bind an inode's backing buffer_head into this transaction, to prevent
5776 * it from being flushed to disk early. Unlike
5777 * ext4_reserve_inode_write, this leaves behind no bh reference and
5778 * returns no iloc structure, so the caller needs to repeat the iloc
5779 * lookup to mark the inode dirty later.
5781 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5783 struct ext4_iloc iloc;
5787 err = ext4_get_inode_loc(inode, &iloc);
5789 BUFFER_TRACE(iloc.bh, "get_write_access");
5790 err = jbd2_journal_get_write_access(handle, iloc.bh);
5792 err = ext4_handle_dirty_metadata(handle,
5798 ext4_std_error(inode->i_sb, err);
5803 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5810 * We have to be very careful here: changing a data block's
5811 * journaling status dynamically is dangerous. If we write a
5812 * data block to the journal, change the status and then delete
5813 * that block, we risk forgetting to revoke the old log record
5814 * from the journal and so a subsequent replay can corrupt data.
5815 * So, first we make sure that the journal is empty and that
5816 * nobody is changing anything.
5819 journal = EXT4_JOURNAL(inode);
5822 if (is_journal_aborted(journal))
5825 jbd2_journal_lock_updates(journal);
5826 jbd2_journal_flush(journal);
5829 * OK, there are no updates running now, and all cached data is
5830 * synced to disk. We are now in a completely consistent state
5831 * which doesn't have anything in the journal, and we know that
5832 * no filesystem updates are running, so it is safe to modify
5833 * the inode's in-core data-journaling state flag now.
5837 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5839 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5840 ext4_set_aops(inode);
5842 jbd2_journal_unlock_updates(journal);
5844 /* Finally we can mark the inode as dirty. */
5846 handle = ext4_journal_start(inode, 1);
5848 return PTR_ERR(handle);
5850 err = ext4_mark_inode_dirty(handle, inode);
5851 ext4_handle_sync(handle);
5852 ext4_journal_stop(handle);
5853 ext4_std_error(inode->i_sb, err);
5858 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5860 return !buffer_mapped(bh);
5863 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5865 struct page *page = vmf->page;
5870 struct file *file = vma->vm_file;
5871 struct inode *inode = file->f_path.dentry->d_inode;
5872 struct address_space *mapping = inode->i_mapping;
5875 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5876 * get i_mutex because we are already holding mmap_sem.
5878 down_read(&inode->i_alloc_sem);
5879 size = i_size_read(inode);
5880 if (page->mapping != mapping || size <= page_offset(page)
5881 || !PageUptodate(page)) {
5882 /* page got truncated from under us? */
5888 wait_on_page_writeback(page);
5889 if (PageMappedToDisk(page)) {
5890 up_read(&inode->i_alloc_sem);
5891 return VM_FAULT_LOCKED;
5894 if (page->index == size >> PAGE_CACHE_SHIFT)
5895 len = size & ~PAGE_CACHE_MASK;
5897 len = PAGE_CACHE_SIZE;
5900 * return if we have all the buffers mapped. This avoid
5901 * the need to call write_begin/write_end which does a
5902 * journal_start/journal_stop which can block and take
5905 if (page_has_buffers(page)) {
5906 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5907 ext4_bh_unmapped)) {
5908 up_read(&inode->i_alloc_sem);
5909 return VM_FAULT_LOCKED;
5914 * OK, we need to fill the hole... Do write_begin write_end
5915 * to do block allocation/reservation.We are not holding
5916 * inode.i__mutex here. That allow * parallel write_begin,
5917 * write_end call. lock_page prevent this from happening
5918 * on the same page though
5920 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5921 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5924 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5925 len, len, page, fsdata);
5931 * write_begin/end might have created a dirty page and someone
5932 * could wander in and start the IO. Make sure that hasn't
5936 wait_on_page_writeback(page);
5937 up_read(&inode->i_alloc_sem);
5938 return VM_FAULT_LOCKED;
5941 ret = VM_FAULT_SIGBUS;
5942 up_read(&inode->i_alloc_sem);