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, blocks_for_truncate(inode));
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);
192 if (inode->i_nlink) {
193 truncate_inode_pages(&inode->i_data, 0);
197 if (!is_bad_inode(inode))
198 dquot_initialize(inode);
200 if (ext4_should_order_data(inode))
201 ext4_begin_ordered_truncate(inode, 0);
202 truncate_inode_pages(&inode->i_data, 0);
204 if (is_bad_inode(inode))
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL, inode);
220 ext4_handle_sync(handle);
222 err = ext4_mark_inode_dirty(handle, inode);
224 ext4_warning(inode->i_sb,
225 "couldn't mark inode dirty (err %d)", err);
229 ext4_truncate(inode);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (!ext4_handle_has_enough_credits(handle, 3)) {
238 err = ext4_journal_extend(handle, 3);
240 err = ext4_journal_restart(handle, 3);
242 ext4_warning(inode->i_sb,
243 "couldn't extend journal (err %d)", err);
245 ext4_journal_stop(handle);
246 ext4_orphan_del(NULL, inode);
252 * Kill off the orphan record which ext4_truncate created.
253 * AKPM: I think this can be inside the above `if'.
254 * Note that ext4_orphan_del() has to be able to cope with the
255 * deletion of a non-existent orphan - this is because we don't
256 * know if ext4_truncate() actually created an orphan record.
257 * (Well, we could do this if we need to, but heck - it works)
259 ext4_orphan_del(handle, inode);
260 EXT4_I(inode)->i_dtime = get_seconds();
263 * One subtle ordering requirement: if anything has gone wrong
264 * (transaction abort, IO errors, whatever), then we can still
265 * do these next steps (the fs will already have been marked as
266 * having errors), but we can't free the inode if the mark_dirty
269 if (ext4_mark_inode_dirty(handle, inode))
270 /* If that failed, just do the required in-core inode clear. */
271 ext4_clear_inode(inode);
273 ext4_free_inode(handle, inode);
274 ext4_journal_stop(handle);
277 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
283 struct buffer_head *bh;
286 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
288 p->key = *(p->p = v);
293 * ext4_block_to_path - parse the block number into array of offsets
294 * @inode: inode in question (we are only interested in its superblock)
295 * @i_block: block number to be parsed
296 * @offsets: array to store the offsets in
297 * @boundary: set this non-zero if the referred-to block is likely to be
298 * followed (on disk) by an indirect block.
300 * To store the locations of file's data ext4 uses a data structure common
301 * for UNIX filesystems - tree of pointers anchored in the inode, with
302 * data blocks at leaves and indirect blocks in intermediate nodes.
303 * This function translates the block number into path in that tree -
304 * return value is the path length and @offsets[n] is the offset of
305 * pointer to (n+1)th node in the nth one. If @block is out of range
306 * (negative or too large) warning is printed and zero returned.
308 * Note: function doesn't find node addresses, so no IO is needed. All
309 * we need to know is the capacity of indirect blocks (taken from the
314 * Portability note: the last comparison (check that we fit into triple
315 * indirect block) is spelled differently, because otherwise on an
316 * architecture with 32-bit longs and 8Kb pages we might get into trouble
317 * if our filesystem had 8Kb blocks. We might use long long, but that would
318 * kill us on x86. Oh, well, at least the sign propagation does not matter -
319 * i_block would have to be negative in the very beginning, so we would not
323 static int ext4_block_to_path(struct inode *inode,
325 ext4_lblk_t offsets[4], int *boundary)
327 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
328 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
329 const long direct_blocks = EXT4_NDIR_BLOCKS,
330 indirect_blocks = ptrs,
331 double_blocks = (1 << (ptrs_bits * 2));
335 if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ((i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT4_IND_BLOCK;
340 offsets[n++] = i_block;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT4_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT4_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
354 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks, inode->i_ino);
359 *boundary = final - 1 - (i_block & (ptrs - 1));
363 static int __ext4_check_blockref(const char *function, unsigned int line,
365 __le32 *p, unsigned int max)
367 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
371 while (bref < p+max) {
372 blk = le32_to_cpu(*bref++);
374 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
376 es->s_last_error_block = cpu_to_le64(blk);
377 ext4_error_inode(inode, function, line, blk,
386 #define ext4_check_indirect_blockref(inode, bh) \
387 __ext4_check_blockref(__func__, __LINE__, inode, \
388 (__le32 *)(bh)->b_data, \
389 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
391 #define ext4_check_inode_blockref(inode) \
392 __ext4_check_blockref(__func__, __LINE__, inode, \
393 EXT4_I(inode)->i_data, \
397 * ext4_get_branch - read the chain of indirect blocks leading to data
398 * @inode: inode in question
399 * @depth: depth of the chain (1 - direct pointer, etc.)
400 * @offsets: offsets of pointers in inode/indirect blocks
401 * @chain: place to store the result
402 * @err: here we store the error value
404 * Function fills the array of triples <key, p, bh> and returns %NULL
405 * if everything went OK or the pointer to the last filled triple
406 * (incomplete one) otherwise. Upon the return chain[i].key contains
407 * the number of (i+1)-th block in the chain (as it is stored in memory,
408 * i.e. little-endian 32-bit), chain[i].p contains the address of that
409 * number (it points into struct inode for i==0 and into the bh->b_data
410 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
411 * block for i>0 and NULL for i==0. In other words, it holds the block
412 * numbers of the chain, addresses they were taken from (and where we can
413 * verify that chain did not change) and buffer_heads hosting these
416 * Function stops when it stumbles upon zero pointer (absent block)
417 * (pointer to last triple returned, *@err == 0)
418 * or when it gets an IO error reading an indirect block
419 * (ditto, *@err == -EIO)
420 * or when it reads all @depth-1 indirect blocks successfully and finds
421 * the whole chain, all way to the data (returns %NULL, *err == 0).
423 * Need to be called with
424 * down_read(&EXT4_I(inode)->i_data_sem)
426 static Indirect *ext4_get_branch(struct inode *inode, int depth,
427 ext4_lblk_t *offsets,
428 Indirect chain[4], int *err)
430 struct super_block *sb = inode->i_sb;
432 struct buffer_head *bh;
435 /* i_data is not going away, no lock needed */
436 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
440 bh = sb_getblk(sb, le32_to_cpu(p->key));
444 if (!bh_uptodate_or_lock(bh)) {
445 if (bh_submit_read(bh) < 0) {
449 /* validate block references */
450 if (ext4_check_indirect_blockref(inode, bh)) {
456 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
470 * ext4_find_near - find a place for allocation with sufficient locality
472 * @ind: descriptor of indirect block.
474 * This function returns the preferred place for block allocation.
475 * It is used when heuristic for sequential allocation fails.
477 * + if there is a block to the left of our position - allocate near it.
478 * + if pointer will live in indirect block - allocate near that block.
479 * + if pointer will live in inode - allocate in the same
482 * In the latter case we colour the starting block by the callers PID to
483 * prevent it from clashing with concurrent allocations for a different inode
484 * in the same block group. The PID is used here so that functionally related
485 * files will be close-by on-disk.
487 * Caller must make sure that @ind is valid and will stay that way.
489 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
491 struct ext4_inode_info *ei = EXT4_I(inode);
492 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
494 ext4_fsblk_t bg_start;
495 ext4_fsblk_t last_block;
496 ext4_grpblk_t colour;
497 ext4_group_t block_group;
498 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
500 /* Try to find previous block */
501 for (p = ind->p - 1; p >= start; p--) {
503 return le32_to_cpu(*p);
506 /* No such thing, so let's try location of indirect block */
508 return ind->bh->b_blocknr;
511 * It is going to be referred to from the inode itself? OK, just put it
512 * into the same cylinder group then.
514 block_group = ei->i_block_group;
515 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
516 block_group &= ~(flex_size-1);
517 if (S_ISREG(inode->i_mode))
520 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
521 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
524 * If we are doing delayed allocation, we don't need take
525 * colour into account.
527 if (test_opt(inode->i_sb, DELALLOC))
530 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
531 colour = (current->pid % 16) *
532 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
534 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
535 return bg_start + colour;
539 * ext4_find_goal - find a preferred place for allocation.
541 * @block: block we want
542 * @partial: pointer to the last triple within a chain
544 * Normally this function find the preferred place for block allocation,
546 * Because this is only used for non-extent files, we limit the block nr
549 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
555 * XXX need to get goal block from mballoc's data structures
558 goal = ext4_find_near(inode, partial);
559 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
564 * ext4_blks_to_allocate - Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
575 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
578 unsigned int count = 0;
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
589 count += blocks_to_boundary + 1;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @handle: handle for this transaction
604 * @inode: inode which needs allocated blocks
605 * @iblock: the logical block to start allocated at
606 * @goal: preferred physical block of allocation
607 * @indirect_blks: the number of blocks need to allocate for indirect
609 * @blks: number of desired blocks
610 * @new_blocks: on return it will store the new block numbers for
611 * the indirect blocks(if needed) and the first direct block,
612 * @err: on return it will store the error code
614 * This function will return the number of blocks allocated as
615 * requested by the passed-in parameters.
617 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
618 ext4_lblk_t iblock, ext4_fsblk_t goal,
619 int indirect_blks, int blks,
620 ext4_fsblk_t new_blocks[4], int *err)
622 struct ext4_allocation_request ar;
624 unsigned long count = 0, blk_allocated = 0;
626 ext4_fsblk_t current_block = 0;
630 * Here we try to allocate the requested multiple blocks at once,
631 * on a best-effort basis.
632 * To build a branch, we should allocate blocks for
633 * the indirect blocks(if not allocated yet), and at least
634 * the first direct block of this branch. That's the
635 * minimum number of blocks need to allocate(required)
637 /* first we try to allocate the indirect blocks */
638 target = indirect_blks;
641 /* allocating blocks for indirect blocks and direct blocks */
642 current_block = ext4_new_meta_blocks(handle, inode,
647 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
648 EXT4_ERROR_INODE(inode,
649 "current_block %llu + count %lu > %d!",
650 current_block, count,
651 EXT4_MAX_BLOCK_FILE_PHYS);
657 /* allocate blocks for indirect blocks */
658 while (index < indirect_blks && count) {
659 new_blocks[index++] = current_block++;
664 * save the new block number
665 * for the first direct block
667 new_blocks[index] = current_block;
668 printk(KERN_INFO "%s returned more blocks than "
669 "requested\n", __func__);
675 target = blks - count ;
676 blk_allocated = count;
679 /* Now allocate data blocks */
680 memset(&ar, 0, sizeof(ar));
685 if (S_ISREG(inode->i_mode))
686 /* enable in-core preallocation only for regular files */
687 ar.flags = EXT4_MB_HINT_DATA;
689 current_block = ext4_mb_new_blocks(handle, &ar, err);
690 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
691 EXT4_ERROR_INODE(inode,
692 "current_block %llu + ar.len %d > %d!",
693 current_block, ar.len,
694 EXT4_MAX_BLOCK_FILE_PHYS);
699 if (*err && (target == blks)) {
701 * if the allocation failed and we didn't allocate
707 if (target == blks) {
709 * save the new block number
710 * for the first direct block
712 new_blocks[index] = current_block;
714 blk_allocated += ar.len;
717 /* total number of blocks allocated for direct blocks */
722 for (i = 0; i < index; i++)
723 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
728 * ext4_alloc_branch - allocate and set up a chain of blocks.
729 * @handle: handle for this transaction
731 * @indirect_blks: number of allocated indirect blocks
732 * @blks: number of allocated direct blocks
733 * @goal: preferred place for allocation
734 * @offsets: offsets (in the blocks) to store the pointers to next.
735 * @branch: place to store the chain in.
737 * This function allocates blocks, zeroes out all but the last one,
738 * links them into chain and (if we are synchronous) writes them to disk.
739 * In other words, it prepares a branch that can be spliced onto the
740 * inode. It stores the information about that chain in the branch[], in
741 * the same format as ext4_get_branch() would do. We are calling it after
742 * we had read the existing part of chain and partial points to the last
743 * triple of that (one with zero ->key). Upon the exit we have the same
744 * picture as after the successful ext4_get_block(), except that in one
745 * place chain is disconnected - *branch->p is still zero (we did not
746 * set the last link), but branch->key contains the number that should
747 * be placed into *branch->p to fill that gap.
749 * If allocation fails we free all blocks we've allocated (and forget
750 * their buffer_heads) and return the error value the from failed
751 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
752 * as described above and return 0.
754 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
755 ext4_lblk_t iblock, int indirect_blks,
756 int *blks, ext4_fsblk_t goal,
757 ext4_lblk_t *offsets, Indirect *branch)
759 int blocksize = inode->i_sb->s_blocksize;
762 struct buffer_head *bh;
764 ext4_fsblk_t new_blocks[4];
765 ext4_fsblk_t current_block;
767 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
768 *blks, new_blocks, &err);
772 branch[0].key = cpu_to_le32(new_blocks[0]);
774 * metadata blocks and data blocks are allocated.
776 for (n = 1; n <= indirect_blks; n++) {
778 * Get buffer_head for parent block, zero it out
779 * and set the pointer to new one, then send
782 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
790 BUFFER_TRACE(bh, "call get_create_access");
791 err = ext4_journal_get_create_access(handle, bh);
793 /* Don't brelse(bh) here; it's done in
794 * ext4_journal_forget() below */
799 memset(bh->b_data, 0, blocksize);
800 branch[n].p = (__le32 *) bh->b_data + offsets[n];
801 branch[n].key = cpu_to_le32(new_blocks[n]);
802 *branch[n].p = branch[n].key;
803 if (n == indirect_blks) {
804 current_block = new_blocks[n];
806 * End of chain, update the last new metablock of
807 * the chain to point to the new allocated
808 * data blocks numbers
810 for (i = 1; i < num; i++)
811 *(branch[n].p + i) = cpu_to_le32(++current_block);
813 BUFFER_TRACE(bh, "marking uptodate");
814 set_buffer_uptodate(bh);
817 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
818 err = ext4_handle_dirty_metadata(handle, inode, bh);
825 /* Allocation failed, free what we already allocated */
826 ext4_free_blocks(handle, inode, NULL, new_blocks[0], 1, 0);
827 for (i = 1; i <= n ; i++) {
829 * branch[i].bh is newly allocated, so there is no
830 * need to revoke the block, which is why we don't
831 * need to set EXT4_FREE_BLOCKS_METADATA.
833 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1,
834 EXT4_FREE_BLOCKS_FORGET);
836 for (i = n+1; i < indirect_blks; i++)
837 ext4_free_blocks(handle, inode, NULL, new_blocks[i], 1, 0);
839 ext4_free_blocks(handle, inode, NULL, new_blocks[i], num, 0);
845 * ext4_splice_branch - splice the allocated branch onto inode.
846 * @handle: handle for this transaction
848 * @block: (logical) number of block we are adding
849 * @chain: chain of indirect blocks (with a missing link - see
851 * @where: location of missing link
852 * @num: number of indirect blocks we are adding
853 * @blks: number of direct blocks we are adding
855 * This function fills the missing link and does all housekeeping needed in
856 * inode (->i_blocks, etc.). In case of success we end up with the full
857 * chain to new block and return 0.
859 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
860 ext4_lblk_t block, Indirect *where, int num,
865 ext4_fsblk_t current_block;
868 * If we're splicing into a [td]indirect block (as opposed to the
869 * inode) then we need to get write access to the [td]indirect block
873 BUFFER_TRACE(where->bh, "get_write_access");
874 err = ext4_journal_get_write_access(handle, where->bh);
880 *where->p = where->key;
883 * Update the host buffer_head or inode to point to more just allocated
884 * direct blocks blocks
886 if (num == 0 && blks > 1) {
887 current_block = le32_to_cpu(where->key) + 1;
888 for (i = 1; i < blks; i++)
889 *(where->p + i) = cpu_to_le32(current_block++);
892 /* We are done with atomic stuff, now do the rest of housekeeping */
893 /* had we spliced it onto indirect block? */
896 * If we spliced it onto an indirect block, we haven't
897 * altered the inode. Note however that if it is being spliced
898 * onto an indirect block at the very end of the file (the
899 * file is growing) then we *will* alter the inode to reflect
900 * the new i_size. But that is not done here - it is done in
901 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
903 jbd_debug(5, "splicing indirect only\n");
904 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
905 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
910 * OK, we spliced it into the inode itself on a direct block.
912 ext4_mark_inode_dirty(handle, inode);
913 jbd_debug(5, "splicing direct\n");
918 for (i = 1; i <= num; i++) {
920 * branch[i].bh is newly allocated, so there is no
921 * need to revoke the block, which is why we don't
922 * need to set EXT4_FREE_BLOCKS_METADATA.
924 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
925 EXT4_FREE_BLOCKS_FORGET);
927 ext4_free_blocks(handle, inode, NULL, le32_to_cpu(where[num].key),
934 * The ext4_ind_map_blocks() function handles non-extents inodes
935 * (i.e., using the traditional indirect/double-indirect i_blocks
936 * scheme) for ext4_map_blocks().
938 * Allocation strategy is simple: if we have to allocate something, we will
939 * have to go the whole way to leaf. So let's do it before attaching anything
940 * to tree, set linkage between the newborn blocks, write them if sync is
941 * required, recheck the path, free and repeat if check fails, otherwise
942 * set the last missing link (that will protect us from any truncate-generated
943 * removals - all blocks on the path are immune now) and possibly force the
944 * write on the parent block.
945 * That has a nice additional property: no special recovery from the failed
946 * allocations is needed - we simply release blocks and do not touch anything
947 * reachable from inode.
949 * `handle' can be NULL if create == 0.
951 * return > 0, # of blocks mapped or allocated.
952 * return = 0, if plain lookup failed.
953 * return < 0, error case.
955 * The ext4_ind_get_blocks() function should be called with
956 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
957 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
958 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
961 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
962 struct ext4_map_blocks *map,
966 ext4_lblk_t offsets[4];
971 int blocks_to_boundary = 0;
974 ext4_fsblk_t first_block = 0;
976 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
977 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
978 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
979 &blocks_to_boundary);
984 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
986 /* Simplest case - block found, no allocation needed */
988 first_block = le32_to_cpu(chain[depth - 1].key);
991 while (count < map->m_len && count <= blocks_to_boundary) {
994 blk = le32_to_cpu(*(chain[depth-1].p + count));
996 if (blk == first_block + count)
1004 /* Next simple case - plain lookup or failed read of indirect block */
1005 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
1009 * Okay, we need to do block allocation.
1011 goal = ext4_find_goal(inode, map->m_lblk, partial);
1013 /* the number of blocks need to allocate for [d,t]indirect blocks */
1014 indirect_blks = (chain + depth) - partial - 1;
1017 * Next look up the indirect map to count the totoal number of
1018 * direct blocks to allocate for this branch.
1020 count = ext4_blks_to_allocate(partial, indirect_blks,
1021 map->m_len, blocks_to_boundary);
1023 * Block out ext4_truncate while we alter the tree
1025 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1027 offsets + (partial - chain), partial);
1030 * The ext4_splice_branch call will free and forget any buffers
1031 * on the new chain if there is a failure, but that risks using
1032 * up transaction credits, especially for bitmaps where the
1033 * credits cannot be returned. Can we handle this somehow? We
1034 * may need to return -EAGAIN upwards in the worst case. --sct
1037 err = ext4_splice_branch(handle, inode, map->m_lblk,
1038 partial, indirect_blks, count);
1042 map->m_flags |= EXT4_MAP_NEW;
1044 ext4_update_inode_fsync_trans(handle, inode, 1);
1046 map->m_flags |= EXT4_MAP_MAPPED;
1047 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1049 if (count > blocks_to_boundary)
1050 map->m_flags |= EXT4_MAP_BOUNDARY;
1052 /* Clean up and exit */
1053 partial = chain + depth - 1; /* the whole chain */
1055 while (partial > chain) {
1056 BUFFER_TRACE(partial->bh, "call brelse");
1057 brelse(partial->bh);
1065 qsize_t *ext4_get_reserved_space(struct inode *inode)
1067 return &EXT4_I(inode)->i_reserved_quota;
1072 * Calculate the number of metadata blocks need to reserve
1073 * to allocate a new block at @lblocks for non extent file based file
1075 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1078 struct ext4_inode_info *ei = EXT4_I(inode);
1079 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1082 if (lblock < EXT4_NDIR_BLOCKS)
1085 lblock -= EXT4_NDIR_BLOCKS;
1087 if (ei->i_da_metadata_calc_len &&
1088 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1089 ei->i_da_metadata_calc_len++;
1092 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1093 ei->i_da_metadata_calc_len = 1;
1094 blk_bits = order_base_2(lblock);
1095 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1099 * Calculate the number of metadata blocks need to reserve
1100 * to allocate a block located at @lblock
1102 static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
1104 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1105 return ext4_ext_calc_metadata_amount(inode, lblock);
1107 return ext4_indirect_calc_metadata_amount(inode, lblock);
1111 * Called with i_data_sem down, which is important since we can call
1112 * ext4_discard_preallocations() from here.
1114 void ext4_da_update_reserve_space(struct inode *inode,
1115 int used, int quota_claim)
1117 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1118 struct ext4_inode_info *ei = EXT4_I(inode);
1120 spin_lock(&ei->i_block_reservation_lock);
1121 trace_ext4_da_update_reserve_space(inode, used);
1122 if (unlikely(used > ei->i_reserved_data_blocks)) {
1123 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1124 "with only %d reserved data blocks\n",
1125 __func__, inode->i_ino, used,
1126 ei->i_reserved_data_blocks);
1128 used = ei->i_reserved_data_blocks;
1131 /* Update per-inode reservations */
1132 ei->i_reserved_data_blocks -= used;
1133 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1134 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1135 used + ei->i_allocated_meta_blocks);
1136 ei->i_allocated_meta_blocks = 0;
1138 if (ei->i_reserved_data_blocks == 0) {
1140 * We can release all of the reserved metadata blocks
1141 * only when we have written all of the delayed
1142 * allocation blocks.
1144 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1145 ei->i_reserved_meta_blocks);
1146 ei->i_reserved_meta_blocks = 0;
1147 ei->i_da_metadata_calc_len = 0;
1149 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1151 /* Update quota subsystem for data blocks */
1153 dquot_claim_block(inode, used);
1156 * We did fallocate with an offset that is already delayed
1157 * allocated. So on delayed allocated writeback we should
1158 * not re-claim the quota for fallocated blocks.
1160 dquot_release_reservation_block(inode, used);
1164 * If we have done all the pending block allocations and if
1165 * there aren't any writers on the inode, we can discard the
1166 * inode's preallocations.
1168 if ((ei->i_reserved_data_blocks == 0) &&
1169 (atomic_read(&inode->i_writecount) == 0))
1170 ext4_discard_preallocations(inode);
1173 static int __check_block_validity(struct inode *inode, const char *func,
1175 struct ext4_map_blocks *map)
1177 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1179 ext4_error_inode(inode, func, line, map->m_pblk,
1180 "lblock %lu mapped to illegal pblock "
1181 "(length %d)", (unsigned long) map->m_lblk,
1188 #define check_block_validity(inode, map) \
1189 __check_block_validity((inode), __func__, __LINE__, (map))
1192 * Return the number of contiguous dirty pages in a given inode
1193 * starting at page frame idx.
1195 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1196 unsigned int max_pages)
1198 struct address_space *mapping = inode->i_mapping;
1200 struct pagevec pvec;
1202 int i, nr_pages, done = 0;
1206 pagevec_init(&pvec, 0);
1209 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1210 PAGECACHE_TAG_DIRTY,
1211 (pgoff_t)PAGEVEC_SIZE);
1214 for (i = 0; i < nr_pages; i++) {
1215 struct page *page = pvec.pages[i];
1216 struct buffer_head *bh, *head;
1219 if (unlikely(page->mapping != mapping) ||
1221 PageWriteback(page) ||
1222 page->index != idx) {
1227 if (page_has_buffers(page)) {
1228 bh = head = page_buffers(page);
1230 if (!buffer_delay(bh) &&
1231 !buffer_unwritten(bh))
1233 bh = bh->b_this_page;
1234 } while (!done && (bh != head));
1241 if (num >= max_pages) {
1246 pagevec_release(&pvec);
1252 * The ext4_map_blocks() function tries to look up the requested blocks,
1253 * and returns if the blocks are already mapped.
1255 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1256 * and store the allocated blocks in the result buffer head and mark it
1259 * If file type is extents based, it will call ext4_ext_map_blocks(),
1260 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1263 * On success, it returns the number of blocks being mapped or allocate.
1264 * if create==0 and the blocks are pre-allocated and uninitialized block,
1265 * the result buffer head is unmapped. If the create ==1, it will make sure
1266 * the buffer head is mapped.
1268 * It returns 0 if plain look up failed (blocks have not been allocated), in
1269 * that casem, buffer head is unmapped
1271 * It returns the error in case of allocation failure.
1273 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1274 struct ext4_map_blocks *map, int flags)
1279 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1280 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1281 (unsigned long) map->m_lblk);
1283 * Try to see if we can get the block without requesting a new
1284 * file system block.
1286 down_read((&EXT4_I(inode)->i_data_sem));
1287 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1288 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1290 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1292 up_read((&EXT4_I(inode)->i_data_sem));
1294 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1295 int ret = check_block_validity(inode, map);
1300 /* If it is only a block(s) look up */
1301 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1305 * Returns if the blocks have already allocated
1307 * Note that if blocks have been preallocated
1308 * ext4_ext_get_block() returns th create = 0
1309 * with buffer head unmapped.
1311 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1315 * When we call get_blocks without the create flag, the
1316 * BH_Unwritten flag could have gotten set if the blocks
1317 * requested were part of a uninitialized extent. We need to
1318 * clear this flag now that we are committed to convert all or
1319 * part of the uninitialized extent to be an initialized
1320 * extent. This is because we need to avoid the combination
1321 * of BH_Unwritten and BH_Mapped flags being simultaneously
1322 * set on the buffer_head.
1324 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1327 * New blocks allocate and/or writing to uninitialized extent
1328 * will possibly result in updating i_data, so we take
1329 * the write lock of i_data_sem, and call get_blocks()
1330 * with create == 1 flag.
1332 down_write((&EXT4_I(inode)->i_data_sem));
1335 * if the caller is from delayed allocation writeout path
1336 * we have already reserved fs blocks for allocation
1337 * let the underlying get_block() function know to
1338 * avoid double accounting
1340 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1341 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1343 * We need to check for EXT4 here because migrate
1344 * could have changed the inode type in between
1346 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1347 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1349 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1351 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1353 * We allocated new blocks which will result in
1354 * i_data's format changing. Force the migrate
1355 * to fail by clearing migrate flags
1357 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1361 * Update reserved blocks/metadata blocks after successful
1362 * block allocation which had been deferred till now. We don't
1363 * support fallocate for non extent files. So we can update
1364 * reserve space here.
1367 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1368 ext4_da_update_reserve_space(inode, retval, 1);
1370 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1371 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
1373 up_write((&EXT4_I(inode)->i_data_sem));
1374 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1375 int ret = check_block_validity(inode, map);
1382 /* Maximum number of blocks we map for direct IO at once. */
1383 #define DIO_MAX_BLOCKS 4096
1385 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1386 struct buffer_head *bh, int flags)
1388 handle_t *handle = ext4_journal_current_handle();
1389 struct ext4_map_blocks map;
1390 int ret = 0, started = 0;
1393 map.m_lblk = iblock;
1394 map.m_len = bh->b_size >> inode->i_blkbits;
1396 if (flags && !handle) {
1397 /* Direct IO write... */
1398 if (map.m_len > DIO_MAX_BLOCKS)
1399 map.m_len = DIO_MAX_BLOCKS;
1400 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1401 handle = ext4_journal_start(inode, dio_credits);
1402 if (IS_ERR(handle)) {
1403 ret = PTR_ERR(handle);
1409 ret = ext4_map_blocks(handle, inode, &map, flags);
1411 map_bh(bh, inode->i_sb, map.m_pblk);
1412 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1413 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1417 ext4_journal_stop(handle);
1421 int ext4_get_block(struct inode *inode, sector_t iblock,
1422 struct buffer_head *bh, int create)
1424 return _ext4_get_block(inode, iblock, bh,
1425 create ? EXT4_GET_BLOCKS_CREATE : 0);
1429 * `handle' can be NULL if create is zero
1431 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1432 ext4_lblk_t block, int create, int *errp)
1434 struct ext4_map_blocks map;
1435 struct buffer_head *bh;
1438 J_ASSERT(handle != NULL || create == 0);
1442 err = ext4_map_blocks(handle, inode, &map,
1443 create ? EXT4_GET_BLOCKS_CREATE : 0);
1451 bh = sb_getblk(inode->i_sb, map.m_pblk);
1456 if (map.m_flags & EXT4_MAP_NEW) {
1457 J_ASSERT(create != 0);
1458 J_ASSERT(handle != NULL);
1461 * Now that we do not always journal data, we should
1462 * keep in mind whether this should always journal the
1463 * new buffer as metadata. For now, regular file
1464 * writes use ext4_get_block instead, so it's not a
1468 BUFFER_TRACE(bh, "call get_create_access");
1469 fatal = ext4_journal_get_create_access(handle, bh);
1470 if (!fatal && !buffer_uptodate(bh)) {
1471 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1472 set_buffer_uptodate(bh);
1475 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1476 err = ext4_handle_dirty_metadata(handle, inode, bh);
1480 BUFFER_TRACE(bh, "not a new buffer");
1490 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1491 ext4_lblk_t block, int create, int *err)
1493 struct buffer_head *bh;
1495 bh = ext4_getblk(handle, inode, block, create, err);
1498 if (buffer_uptodate(bh))
1500 ll_rw_block(READ_META, 1, &bh);
1502 if (buffer_uptodate(bh))
1509 static int walk_page_buffers(handle_t *handle,
1510 struct buffer_head *head,
1514 int (*fn)(handle_t *handle,
1515 struct buffer_head *bh))
1517 struct buffer_head *bh;
1518 unsigned block_start, block_end;
1519 unsigned blocksize = head->b_size;
1521 struct buffer_head *next;
1523 for (bh = head, block_start = 0;
1524 ret == 0 && (bh != head || !block_start);
1525 block_start = block_end, bh = next) {
1526 next = bh->b_this_page;
1527 block_end = block_start + blocksize;
1528 if (block_end <= from || block_start >= to) {
1529 if (partial && !buffer_uptodate(bh))
1533 err = (*fn)(handle, bh);
1541 * To preserve ordering, it is essential that the hole instantiation and
1542 * the data write be encapsulated in a single transaction. We cannot
1543 * close off a transaction and start a new one between the ext4_get_block()
1544 * and the commit_write(). So doing the jbd2_journal_start at the start of
1545 * prepare_write() is the right place.
1547 * Also, this function can nest inside ext4_writepage() ->
1548 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1549 * has generated enough buffer credits to do the whole page. So we won't
1550 * block on the journal in that case, which is good, because the caller may
1553 * By accident, ext4 can be reentered when a transaction is open via
1554 * quota file writes. If we were to commit the transaction while thus
1555 * reentered, there can be a deadlock - we would be holding a quota
1556 * lock, and the commit would never complete if another thread had a
1557 * transaction open and was blocking on the quota lock - a ranking
1560 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1561 * will _not_ run commit under these circumstances because handle->h_ref
1562 * is elevated. We'll still have enough credits for the tiny quotafile
1565 static int do_journal_get_write_access(handle_t *handle,
1566 struct buffer_head *bh)
1568 int dirty = buffer_dirty(bh);
1571 if (!buffer_mapped(bh) || buffer_freed(bh))
1574 * __block_write_begin() could have dirtied some buffers. Clean
1575 * the dirty bit as jbd2_journal_get_write_access() could complain
1576 * otherwise about fs integrity issues. Setting of the dirty bit
1577 * by __block_write_begin() isn't a real problem here as we clear
1578 * the bit before releasing a page lock and thus writeback cannot
1579 * ever write the buffer.
1582 clear_buffer_dirty(bh);
1583 ret = ext4_journal_get_write_access(handle, bh);
1585 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1590 * Truncate blocks that were not used by write. We have to truncate the
1591 * pagecache as well so that corresponding buffers get properly unmapped.
1593 static void ext4_truncate_failed_write(struct inode *inode)
1595 truncate_inode_pages(inode->i_mapping, inode->i_size);
1596 ext4_truncate(inode);
1599 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1600 struct buffer_head *bh_result, int create);
1601 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1602 loff_t pos, unsigned len, unsigned flags,
1603 struct page **pagep, void **fsdata)
1605 struct inode *inode = mapping->host;
1606 int ret, needed_blocks;
1613 trace_ext4_write_begin(inode, pos, len, flags);
1615 * Reserve one block more for addition to orphan list in case
1616 * we allocate blocks but write fails for some reason
1618 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1619 index = pos >> PAGE_CACHE_SHIFT;
1620 from = pos & (PAGE_CACHE_SIZE - 1);
1624 handle = ext4_journal_start(inode, needed_blocks);
1625 if (IS_ERR(handle)) {
1626 ret = PTR_ERR(handle);
1630 /* We cannot recurse into the filesystem as the transaction is already
1632 flags |= AOP_FLAG_NOFS;
1634 page = grab_cache_page_write_begin(mapping, index, flags);
1636 ext4_journal_stop(handle);
1642 if (ext4_should_dioread_nolock(inode))
1643 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1645 ret = __block_write_begin(page, pos, len, ext4_get_block);
1647 if (!ret && ext4_should_journal_data(inode)) {
1648 ret = walk_page_buffers(handle, page_buffers(page),
1649 from, to, NULL, do_journal_get_write_access);
1654 page_cache_release(page);
1656 * __block_write_begin may have instantiated a few blocks
1657 * outside i_size. Trim these off again. Don't need
1658 * i_size_read because we hold i_mutex.
1660 * Add inode to orphan list in case we crash before
1663 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1664 ext4_orphan_add(handle, inode);
1666 ext4_journal_stop(handle);
1667 if (pos + len > inode->i_size) {
1668 ext4_truncate_failed_write(inode);
1670 * If truncate failed early the inode might
1671 * still be on the orphan list; we need to
1672 * make sure the inode is removed from the
1673 * orphan list in that case.
1676 ext4_orphan_del(NULL, inode);
1680 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1686 /* For write_end() in data=journal mode */
1687 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1689 if (!buffer_mapped(bh) || buffer_freed(bh))
1691 set_buffer_uptodate(bh);
1692 return ext4_handle_dirty_metadata(handle, NULL, bh);
1695 static int ext4_generic_write_end(struct file *file,
1696 struct address_space *mapping,
1697 loff_t pos, unsigned len, unsigned copied,
1698 struct page *page, void *fsdata)
1700 int i_size_changed = 0;
1701 struct inode *inode = mapping->host;
1702 handle_t *handle = ext4_journal_current_handle();
1704 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1707 * No need to use i_size_read() here, the i_size
1708 * cannot change under us because we hold i_mutex.
1710 * But it's important to update i_size while still holding page lock:
1711 * page writeout could otherwise come in and zero beyond i_size.
1713 if (pos + copied > inode->i_size) {
1714 i_size_write(inode, pos + copied);
1718 if (pos + copied > EXT4_I(inode)->i_disksize) {
1719 /* We need to mark inode dirty even if
1720 * new_i_size is less that inode->i_size
1721 * bu greater than i_disksize.(hint delalloc)
1723 ext4_update_i_disksize(inode, (pos + copied));
1727 page_cache_release(page);
1730 * Don't mark the inode dirty under page lock. First, it unnecessarily
1731 * makes the holding time of page lock longer. Second, it forces lock
1732 * ordering of page lock and transaction start for journaling
1736 ext4_mark_inode_dirty(handle, inode);
1742 * We need to pick up the new inode size which generic_commit_write gave us
1743 * `file' can be NULL - eg, when called from page_symlink().
1745 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1746 * buffers are managed internally.
1748 static int ext4_ordered_write_end(struct file *file,
1749 struct address_space *mapping,
1750 loff_t pos, unsigned len, unsigned copied,
1751 struct page *page, void *fsdata)
1753 handle_t *handle = ext4_journal_current_handle();
1754 struct inode *inode = mapping->host;
1757 trace_ext4_ordered_write_end(inode, pos, len, copied);
1758 ret = ext4_jbd2_file_inode(handle, inode);
1761 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1764 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1765 /* if we have allocated more blocks and copied
1766 * less. We will have blocks allocated outside
1767 * inode->i_size. So truncate them
1769 ext4_orphan_add(handle, inode);
1773 ret2 = ext4_journal_stop(handle);
1777 if (pos + len > inode->i_size) {
1778 ext4_truncate_failed_write(inode);
1780 * If truncate failed early the inode might still be
1781 * on the orphan list; we need to make sure the inode
1782 * is removed from the orphan list in that case.
1785 ext4_orphan_del(NULL, inode);
1789 return ret ? ret : copied;
1792 static int ext4_writeback_write_end(struct file *file,
1793 struct address_space *mapping,
1794 loff_t pos, unsigned len, unsigned copied,
1795 struct page *page, void *fsdata)
1797 handle_t *handle = ext4_journal_current_handle();
1798 struct inode *inode = mapping->host;
1801 trace_ext4_writeback_write_end(inode, pos, len, copied);
1802 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1805 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1806 /* if we have allocated more blocks and copied
1807 * less. We will have blocks allocated outside
1808 * inode->i_size. So truncate them
1810 ext4_orphan_add(handle, inode);
1815 ret2 = ext4_journal_stop(handle);
1819 if (pos + len > inode->i_size) {
1820 ext4_truncate_failed_write(inode);
1822 * If truncate failed early the inode might still be
1823 * on the orphan list; we need to make sure the inode
1824 * is removed from the orphan list in that case.
1827 ext4_orphan_del(NULL, inode);
1830 return ret ? ret : copied;
1833 static int ext4_journalled_write_end(struct file *file,
1834 struct address_space *mapping,
1835 loff_t pos, unsigned len, unsigned copied,
1836 struct page *page, void *fsdata)
1838 handle_t *handle = ext4_journal_current_handle();
1839 struct inode *inode = mapping->host;
1845 trace_ext4_journalled_write_end(inode, pos, len, copied);
1846 from = pos & (PAGE_CACHE_SIZE - 1);
1850 if (!PageUptodate(page))
1852 page_zero_new_buffers(page, from+copied, to);
1855 ret = walk_page_buffers(handle, page_buffers(page), from,
1856 to, &partial, write_end_fn);
1858 SetPageUptodate(page);
1859 new_i_size = pos + copied;
1860 if (new_i_size > inode->i_size)
1861 i_size_write(inode, pos+copied);
1862 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1863 if (new_i_size > EXT4_I(inode)->i_disksize) {
1864 ext4_update_i_disksize(inode, new_i_size);
1865 ret2 = ext4_mark_inode_dirty(handle, inode);
1871 page_cache_release(page);
1872 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1873 /* if we have allocated more blocks and copied
1874 * less. We will have blocks allocated outside
1875 * inode->i_size. So truncate them
1877 ext4_orphan_add(handle, inode);
1879 ret2 = ext4_journal_stop(handle);
1882 if (pos + len > inode->i_size) {
1883 ext4_truncate_failed_write(inode);
1885 * If truncate failed early the inode might still be
1886 * on the orphan list; we need to make sure the inode
1887 * is removed from the orphan list in that case.
1890 ext4_orphan_del(NULL, inode);
1893 return ret ? ret : copied;
1897 * Reserve a single block located at lblock
1899 static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
1902 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1903 struct ext4_inode_info *ei = EXT4_I(inode);
1904 unsigned long md_needed;
1908 * recalculate the amount of metadata blocks to reserve
1909 * in order to allocate nrblocks
1910 * worse case is one extent per block
1913 spin_lock(&ei->i_block_reservation_lock);
1914 md_needed = ext4_calc_metadata_amount(inode, lblock);
1915 trace_ext4_da_reserve_space(inode, md_needed);
1916 spin_unlock(&ei->i_block_reservation_lock);
1919 * We will charge metadata quota at writeout time; this saves
1920 * us from metadata over-estimation, though we may go over by
1921 * a small amount in the end. Here we just reserve for data.
1923 ret = dquot_reserve_block(inode, 1);
1927 * We do still charge estimated metadata to the sb though;
1928 * we cannot afford to run out of free blocks.
1930 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1931 dquot_release_reservation_block(inode, 1);
1932 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1938 spin_lock(&ei->i_block_reservation_lock);
1939 ei->i_reserved_data_blocks++;
1940 ei->i_reserved_meta_blocks += md_needed;
1941 spin_unlock(&ei->i_block_reservation_lock);
1943 return 0; /* success */
1946 static void ext4_da_release_space(struct inode *inode, int to_free)
1948 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1949 struct ext4_inode_info *ei = EXT4_I(inode);
1952 return; /* Nothing to release, exit */
1954 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1956 trace_ext4_da_release_space(inode, to_free);
1957 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1959 * if there aren't enough reserved blocks, then the
1960 * counter is messed up somewhere. Since this
1961 * function is called from invalidate page, it's
1962 * harmless to return without any action.
1964 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1965 "ino %lu, to_free %d with only %d reserved "
1966 "data blocks\n", inode->i_ino, to_free,
1967 ei->i_reserved_data_blocks);
1969 to_free = ei->i_reserved_data_blocks;
1971 ei->i_reserved_data_blocks -= to_free;
1973 if (ei->i_reserved_data_blocks == 0) {
1975 * We can release all of the reserved metadata blocks
1976 * only when we have written all of the delayed
1977 * allocation blocks.
1979 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1980 ei->i_reserved_meta_blocks);
1981 ei->i_reserved_meta_blocks = 0;
1982 ei->i_da_metadata_calc_len = 0;
1985 /* update fs dirty data blocks counter */
1986 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1988 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1990 dquot_release_reservation_block(inode, to_free);
1993 static void ext4_da_page_release_reservation(struct page *page,
1994 unsigned long offset)
1997 struct buffer_head *head, *bh;
1998 unsigned int curr_off = 0;
2000 head = page_buffers(page);
2003 unsigned int next_off = curr_off + bh->b_size;
2005 if ((offset <= curr_off) && (buffer_delay(bh))) {
2007 clear_buffer_delay(bh);
2009 curr_off = next_off;
2010 } while ((bh = bh->b_this_page) != head);
2011 ext4_da_release_space(page->mapping->host, to_release);
2015 * Delayed allocation stuff
2019 * mpage_da_submit_io - walks through extent of pages and try to write
2020 * them with writepage() call back
2022 * @mpd->inode: inode
2023 * @mpd->first_page: first page of the extent
2024 * @mpd->next_page: page after the last page of the extent
2026 * By the time mpage_da_submit_io() is called we expect all blocks
2027 * to be allocated. this may be wrong if allocation failed.
2029 * As pages are already locked by write_cache_pages(), we can't use it
2031 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2032 struct ext4_map_blocks *map)
2034 struct pagevec pvec;
2035 unsigned long index, end;
2036 int ret = 0, err, nr_pages, i;
2037 struct inode *inode = mpd->inode;
2038 struct address_space *mapping = inode->i_mapping;
2039 loff_t size = i_size_read(inode);
2040 unsigned int len, block_start;
2041 struct buffer_head *bh, *page_bufs = NULL;
2042 int journal_data = ext4_should_journal_data(inode);
2043 sector_t pblock = 0, cur_logical = 0;
2044 struct ext4_io_submit io_submit;
2046 BUG_ON(mpd->next_page <= mpd->first_page);
2047 memset(&io_submit, 0, sizeof(io_submit));
2049 * We need to start from the first_page to the next_page - 1
2050 * to make sure we also write the mapped dirty buffer_heads.
2051 * If we look at mpd->b_blocknr we would only be looking
2052 * at the currently mapped buffer_heads.
2054 index = mpd->first_page;
2055 end = mpd->next_page - 1;
2057 pagevec_init(&pvec, 0);
2058 while (index <= end) {
2059 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2062 for (i = 0; i < nr_pages; i++) {
2063 int commit_write = 0, skip_page = 0;
2064 struct page *page = pvec.pages[i];
2066 index = page->index;
2070 if (index == size >> PAGE_CACHE_SHIFT)
2071 len = size & ~PAGE_CACHE_MASK;
2073 len = PAGE_CACHE_SIZE;
2075 cur_logical = index << (PAGE_CACHE_SHIFT -
2077 pblock = map->m_pblk + (cur_logical -
2082 BUG_ON(!PageLocked(page));
2083 BUG_ON(PageWriteback(page));
2086 * If the page does not have buffers (for
2087 * whatever reason), try to create them using
2088 * __block_write_begin. If this fails,
2089 * skip the page and move on.
2091 if (!page_has_buffers(page)) {
2092 if (__block_write_begin(page, 0, len,
2093 noalloc_get_block_write)) {
2101 bh = page_bufs = page_buffers(page);
2106 if (map && (cur_logical >= map->m_lblk) &&
2107 (cur_logical <= (map->m_lblk +
2108 (map->m_len - 1)))) {
2109 if (buffer_delay(bh)) {
2110 clear_buffer_delay(bh);
2111 bh->b_blocknr = pblock;
2113 if (buffer_unwritten(bh) ||
2115 BUG_ON(bh->b_blocknr != pblock);
2116 if (map->m_flags & EXT4_MAP_UNINIT)
2117 set_buffer_uninit(bh);
2118 clear_buffer_unwritten(bh);
2121 /* skip page if block allocation undone */
2122 if (buffer_delay(bh) || buffer_unwritten(bh))
2124 bh = bh->b_this_page;
2125 block_start += bh->b_size;
2128 } while (bh != page_bufs);
2134 /* mark the buffer_heads as dirty & uptodate */
2135 block_commit_write(page, 0, len);
2137 clear_page_dirty_for_io(page);
2139 * Delalloc doesn't support data journalling,
2140 * but eventually maybe we'll lift this
2143 if (unlikely(journal_data && PageChecked(page)))
2144 err = __ext4_journalled_writepage(page, len);
2145 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
2146 err = ext4_bio_write_page(&io_submit, page,
2149 err = block_write_full_page(page,
2150 noalloc_get_block_write, mpd->wbc);
2153 mpd->pages_written++;
2155 * In error case, we have to continue because
2156 * remaining pages are still locked
2161 pagevec_release(&pvec);
2163 ext4_io_submit(&io_submit);
2167 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
2171 struct pagevec pvec;
2172 struct inode *inode = mpd->inode;
2173 struct address_space *mapping = inode->i_mapping;
2175 index = mpd->first_page;
2176 end = mpd->next_page - 1;
2177 while (index <= end) {
2178 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2181 for (i = 0; i < nr_pages; i++) {
2182 struct page *page = pvec.pages[i];
2183 if (page->index > end)
2185 BUG_ON(!PageLocked(page));
2186 BUG_ON(PageWriteback(page));
2187 block_invalidatepage(page, 0);
2188 ClearPageUptodate(page);
2191 index = pvec.pages[nr_pages - 1]->index + 1;
2192 pagevec_release(&pvec);
2197 static void ext4_print_free_blocks(struct inode *inode)
2199 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2200 printk(KERN_CRIT "Total free blocks count %lld\n",
2201 ext4_count_free_blocks(inode->i_sb));
2202 printk(KERN_CRIT "Free/Dirty block details\n");
2203 printk(KERN_CRIT "free_blocks=%lld\n",
2204 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2205 printk(KERN_CRIT "dirty_blocks=%lld\n",
2206 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2207 printk(KERN_CRIT "Block reservation details\n");
2208 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2209 EXT4_I(inode)->i_reserved_data_blocks);
2210 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2211 EXT4_I(inode)->i_reserved_meta_blocks);
2216 * mpage_da_map_and_submit - go through given space, map them
2217 * if necessary, and then submit them for I/O
2219 * @mpd - bh describing space
2221 * The function skips space we know is already mapped to disk blocks.
2224 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2226 int err, blks, get_blocks_flags;
2227 struct ext4_map_blocks map, *mapp = NULL;
2228 sector_t next = mpd->b_blocknr;
2229 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2230 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2231 handle_t *handle = NULL;
2234 * If the blocks are mapped already, or we couldn't accumulate
2235 * any blocks, then proceed immediately to the submission stage.
2237 if ((mpd->b_size == 0) ||
2238 ((mpd->b_state & (1 << BH_Mapped)) &&
2239 !(mpd->b_state & (1 << BH_Delay)) &&
2240 !(mpd->b_state & (1 << BH_Unwritten))))
2243 handle = ext4_journal_current_handle();
2247 * Call ext4_map_blocks() to allocate any delayed allocation
2248 * blocks, or to convert an uninitialized extent to be
2249 * initialized (in the case where we have written into
2250 * one or more preallocated blocks).
2252 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2253 * indicate that we are on the delayed allocation path. This
2254 * affects functions in many different parts of the allocation
2255 * call path. This flag exists primarily because we don't
2256 * want to change *many* call functions, so ext4_map_blocks()
2257 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
2258 * inode's allocation semaphore is taken.
2260 * If the blocks in questions were delalloc blocks, set
2261 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2262 * variables are updated after the blocks have been allocated.
2265 map.m_len = max_blocks;
2266 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2267 if (ext4_should_dioread_nolock(mpd->inode))
2268 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2269 if (mpd->b_state & (1 << BH_Delay))
2270 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2272 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2274 struct super_block *sb = mpd->inode->i_sb;
2278 * If get block returns EAGAIN or ENOSPC and there
2279 * appears to be free blocks we will just let
2280 * mpage_da_submit_io() unlock all of the pages.
2285 if (err == -ENOSPC &&
2286 ext4_count_free_blocks(sb)) {
2292 * get block failure will cause us to loop in
2293 * writepages, because a_ops->writepage won't be able
2294 * to make progress. The page will be redirtied by
2295 * writepage and writepages will again try to write
2298 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2299 ext4_msg(sb, KERN_CRIT,
2300 "delayed block allocation failed for inode %lu "
2301 "at logical offset %llu with max blocks %zd "
2302 "with error %d", mpd->inode->i_ino,
2303 (unsigned long long) next,
2304 mpd->b_size >> mpd->inode->i_blkbits, err);
2305 ext4_msg(sb, KERN_CRIT,
2306 "This should not happen!! Data will be lost\n");
2308 ext4_print_free_blocks(mpd->inode);
2310 /* invalidate all the pages */
2311 ext4_da_block_invalidatepages(mpd);
2313 /* Mark this page range as having been completed */
2320 if (map.m_flags & EXT4_MAP_NEW) {
2321 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2324 for (i = 0; i < map.m_len; i++)
2325 unmap_underlying_metadata(bdev, map.m_pblk + i);
2328 if (ext4_should_order_data(mpd->inode)) {
2329 err = ext4_jbd2_file_inode(handle, mpd->inode);
2331 /* This only happens if the journal is aborted */
2336 * Update on-disk size along with block allocation.
2338 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2339 if (disksize > i_size_read(mpd->inode))
2340 disksize = i_size_read(mpd->inode);
2341 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2342 ext4_update_i_disksize(mpd->inode, disksize);
2343 err = ext4_mark_inode_dirty(handle, mpd->inode);
2345 ext4_error(mpd->inode->i_sb,
2346 "Failed to mark inode %lu dirty",
2351 mpage_da_submit_io(mpd, mapp);
2355 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2356 (1 << BH_Delay) | (1 << BH_Unwritten))
2359 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2361 * @mpd->lbh - extent of blocks
2362 * @logical - logical number of the block in the file
2363 * @bh - bh of the block (used to access block's state)
2365 * the function is used to collect contig. blocks in same state
2367 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2368 sector_t logical, size_t b_size,
2369 unsigned long b_state)
2372 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2375 * XXX Don't go larger than mballoc is willing to allocate
2376 * This is a stopgap solution. We eventually need to fold
2377 * mpage_da_submit_io() into this function and then call
2378 * ext4_map_blocks() multiple times in a loop
2380 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2383 /* check if thereserved journal credits might overflow */
2384 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2385 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2387 * With non-extent format we are limited by the journal
2388 * credit available. Total credit needed to insert
2389 * nrblocks contiguous blocks is dependent on the
2390 * nrblocks. So limit nrblocks.
2393 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2394 EXT4_MAX_TRANS_DATA) {
2396 * Adding the new buffer_head would make it cross the
2397 * allowed limit for which we have journal credit
2398 * reserved. So limit the new bh->b_size
2400 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2401 mpd->inode->i_blkbits;
2402 /* we will do mpage_da_submit_io in the next loop */
2406 * First block in the extent
2408 if (mpd->b_size == 0) {
2409 mpd->b_blocknr = logical;
2410 mpd->b_size = b_size;
2411 mpd->b_state = b_state & BH_FLAGS;
2415 next = mpd->b_blocknr + nrblocks;
2417 * Can we merge the block to our big extent?
2419 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2420 mpd->b_size += b_size;
2426 * We couldn't merge the block to our extent, so we
2427 * need to flush current extent and start new one
2429 mpage_da_map_and_submit(mpd);
2433 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2435 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2439 * This is a special get_blocks_t callback which is used by
2440 * ext4_da_write_begin(). It will either return mapped block or
2441 * reserve space for a single block.
2443 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2444 * We also have b_blocknr = -1 and b_bdev initialized properly
2446 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2447 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2448 * initialized properly.
2450 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2451 struct buffer_head *bh, int create)
2453 struct ext4_map_blocks map;
2455 sector_t invalid_block = ~((sector_t) 0xffff);
2457 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2460 BUG_ON(create == 0);
2461 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2463 map.m_lblk = iblock;
2467 * first, we need to know whether the block is allocated already
2468 * preallocated blocks are unmapped but should treated
2469 * the same as allocated blocks.
2471 ret = ext4_map_blocks(NULL, inode, &map, 0);
2475 if (buffer_delay(bh))
2476 return 0; /* Not sure this could or should happen */
2478 * XXX: __block_write_begin() unmaps passed block, is it OK?
2480 ret = ext4_da_reserve_space(inode, iblock);
2482 /* not enough space to reserve */
2485 map_bh(bh, inode->i_sb, invalid_block);
2487 set_buffer_delay(bh);
2491 map_bh(bh, inode->i_sb, map.m_pblk);
2492 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2494 if (buffer_unwritten(bh)) {
2495 /* A delayed write to unwritten bh should be marked
2496 * new and mapped. Mapped ensures that we don't do
2497 * get_block multiple times when we write to the same
2498 * offset and new ensures that we do proper zero out
2499 * for partial write.
2502 set_buffer_mapped(bh);
2508 * This function is used as a standard get_block_t calback function
2509 * when there is no desire to allocate any blocks. It is used as a
2510 * callback function for block_write_begin() and block_write_full_page().
2511 * These functions should only try to map a single block at a time.
2513 * Since this function doesn't do block allocations even if the caller
2514 * requests it by passing in create=1, it is critically important that
2515 * any caller checks to make sure that any buffer heads are returned
2516 * by this function are either all already mapped or marked for
2517 * delayed allocation before calling block_write_full_page(). Otherwise,
2518 * b_blocknr could be left unitialized, and the page write functions will
2519 * be taken by surprise.
2521 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2522 struct buffer_head *bh_result, int create)
2524 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2525 return _ext4_get_block(inode, iblock, bh_result, 0);
2528 static int bget_one(handle_t *handle, struct buffer_head *bh)
2534 static int bput_one(handle_t *handle, struct buffer_head *bh)
2540 static int __ext4_journalled_writepage(struct page *page,
2543 struct address_space *mapping = page->mapping;
2544 struct inode *inode = mapping->host;
2545 struct buffer_head *page_bufs;
2546 handle_t *handle = NULL;
2550 ClearPageChecked(page);
2551 page_bufs = page_buffers(page);
2553 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2554 /* As soon as we unlock the page, it can go away, but we have
2555 * references to buffers so we are safe */
2558 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2559 if (IS_ERR(handle)) {
2560 ret = PTR_ERR(handle);
2564 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2565 do_journal_get_write_access);
2567 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2571 err = ext4_journal_stop(handle);
2575 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2576 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2581 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2582 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2585 * Note that we don't need to start a transaction unless we're journaling data
2586 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2587 * need to file the inode to the transaction's list in ordered mode because if
2588 * we are writing back data added by write(), the inode is already there and if
2589 * we are writing back data modified via mmap(), noone guarantees in which
2590 * transaction the data will hit the disk. In case we are journaling data, we
2591 * cannot start transaction directly because transaction start ranks above page
2592 * lock so we have to do some magic.
2594 * This function can get called via...
2595 * - ext4_da_writepages after taking page lock (have journal handle)
2596 * - journal_submit_inode_data_buffers (no journal handle)
2597 * - shrink_page_list via pdflush (no journal handle)
2598 * - grab_page_cache when doing write_begin (have journal handle)
2600 * We don't do any block allocation in this function. If we have page with
2601 * multiple blocks we need to write those buffer_heads that are mapped. This
2602 * is important for mmaped based write. So if we do with blocksize 1K
2603 * truncate(f, 1024);
2604 * a = mmap(f, 0, 4096);
2606 * truncate(f, 4096);
2607 * we have in the page first buffer_head mapped via page_mkwrite call back
2608 * but other bufer_heads would be unmapped but dirty(dirty done via the
2609 * do_wp_page). So writepage should write the first block. If we modify
2610 * the mmap area beyond 1024 we will again get a page_fault and the
2611 * page_mkwrite callback will do the block allocation and mark the
2612 * buffer_heads mapped.
2614 * We redirty the page if we have any buffer_heads that is either delay or
2615 * unwritten in the page.
2617 * We can get recursively called as show below.
2619 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2622 * But since we don't do any block allocation we should not deadlock.
2623 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2625 static int ext4_writepage(struct page *page,
2626 struct writeback_control *wbc)
2628 int ret = 0, commit_write = 0;
2631 struct buffer_head *page_bufs = NULL;
2632 struct inode *inode = page->mapping->host;
2634 trace_ext4_writepage(inode, page);
2635 size = i_size_read(inode);
2636 if (page->index == size >> PAGE_CACHE_SHIFT)
2637 len = size & ~PAGE_CACHE_MASK;
2639 len = PAGE_CACHE_SIZE;
2642 * If the page does not have buffers (for whatever reason),
2643 * try to create them using __block_write_begin. If this
2644 * fails, redirty the page and move on.
2646 if (!page_has_buffers(page)) {
2647 if (__block_write_begin(page, 0, len,
2648 noalloc_get_block_write)) {
2650 redirty_page_for_writepage(wbc, page);
2656 page_bufs = page_buffers(page);
2657 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2658 ext4_bh_delay_or_unwritten)) {
2660 * We don't want to do block allocation, so redirty
2661 * the page and return. We may reach here when we do
2662 * a journal commit via journal_submit_inode_data_buffers.
2663 * We can also reach here via shrink_page_list
2668 /* now mark the buffer_heads as dirty and uptodate */
2669 block_commit_write(page, 0, len);
2671 if (PageChecked(page) && ext4_should_journal_data(inode))
2673 * It's mmapped pagecache. Add buffers and journal it. There
2674 * doesn't seem much point in redirtying the page here.
2676 return __ext4_journalled_writepage(page, len);
2678 if (buffer_uninit(page_bufs)) {
2679 ext4_set_bh_endio(page_bufs, inode);
2680 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2681 wbc, ext4_end_io_buffer_write);
2683 ret = block_write_full_page(page, noalloc_get_block_write,
2690 * This is called via ext4_da_writepages() to
2691 * calulate the total number of credits to reserve to fit
2692 * a single extent allocation into a single transaction,
2693 * ext4_da_writpeages() will loop calling this before
2694 * the block allocation.
2697 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2699 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2702 * With non-extent format the journal credit needed to
2703 * insert nrblocks contiguous block is dependent on
2704 * number of contiguous block. So we will limit
2705 * number of contiguous block to a sane value
2707 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2708 (max_blocks > EXT4_MAX_TRANS_DATA))
2709 max_blocks = EXT4_MAX_TRANS_DATA;
2711 return ext4_chunk_trans_blocks(inode, max_blocks);
2715 * write_cache_pages_da - walk the list of dirty pages of the given
2716 * address space and accumulate pages that need writing, and call
2717 * mpage_da_map_and_submit to map a single contiguous memory region
2718 * and then write them.
2720 static int write_cache_pages_da(struct address_space *mapping,
2721 struct writeback_control *wbc,
2722 struct mpage_da_data *mpd,
2723 pgoff_t *done_index)
2725 struct buffer_head *bh, *head;
2726 struct inode *inode = mapping->host;
2727 struct pagevec pvec;
2728 unsigned int nr_pages;
2731 long nr_to_write = wbc->nr_to_write;
2732 int i, tag, ret = 0;
2734 memset(mpd, 0, sizeof(struct mpage_da_data));
2737 pagevec_init(&pvec, 0);
2738 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2739 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2741 if (wbc->sync_mode == WB_SYNC_ALL)
2742 tag = PAGECACHE_TAG_TOWRITE;
2744 tag = PAGECACHE_TAG_DIRTY;
2746 *done_index = index;
2747 while (index <= end) {
2748 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2749 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2753 for (i = 0; i < nr_pages; i++) {
2754 struct page *page = pvec.pages[i];
2757 * At this point, the page may be truncated or
2758 * invalidated (changing page->mapping to NULL), or
2759 * even swizzled back from swapper_space to tmpfs file
2760 * mapping. However, page->index will not change
2761 * because we have a reference on the page.
2763 if (page->index > end)
2766 *done_index = page->index + 1;
2769 * If we can't merge this page, and we have
2770 * accumulated an contiguous region, write it
2772 if ((mpd->next_page != page->index) &&
2773 (mpd->next_page != mpd->first_page)) {
2774 mpage_da_map_and_submit(mpd);
2775 goto ret_extent_tail;
2781 * If the page is no longer dirty, or its
2782 * mapping no longer corresponds to inode we
2783 * are writing (which means it has been
2784 * truncated or invalidated), or the page is
2785 * already under writeback and we are not
2786 * doing a data integrity writeback, skip the page
2788 if (!PageDirty(page) ||
2789 (PageWriteback(page) &&
2790 (wbc->sync_mode == WB_SYNC_NONE)) ||
2791 unlikely(page->mapping != mapping)) {
2796 if (PageWriteback(page))
2797 wait_on_page_writeback(page);
2799 BUG_ON(PageWriteback(page));
2801 if (mpd->next_page != page->index)
2802 mpd->first_page = page->index;
2803 mpd->next_page = page->index + 1;
2804 logical = (sector_t) page->index <<
2805 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2807 if (!page_has_buffers(page)) {
2808 mpage_add_bh_to_extent(mpd, logical,
2810 (1 << BH_Dirty) | (1 << BH_Uptodate));
2812 goto ret_extent_tail;
2815 * Page with regular buffer heads,
2816 * just add all dirty ones
2818 head = page_buffers(page);
2821 BUG_ON(buffer_locked(bh));
2823 * We need to try to allocate
2824 * unmapped blocks in the same page.
2825 * Otherwise we won't make progress
2826 * with the page in ext4_writepage
2828 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2829 mpage_add_bh_to_extent(mpd, logical,
2833 goto ret_extent_tail;
2834 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2836 * mapped dirty buffer. We need
2837 * to update the b_state
2838 * because we look at b_state
2839 * in mpage_da_map_blocks. We
2840 * don't update b_size because
2841 * if we find an unmapped
2842 * buffer_head later we need to
2843 * use the b_state flag of that
2846 if (mpd->b_size == 0)
2847 mpd->b_state = bh->b_state & BH_FLAGS;
2850 } while ((bh = bh->b_this_page) != head);
2853 if (nr_to_write > 0) {
2855 if (nr_to_write == 0 &&
2856 wbc->sync_mode == WB_SYNC_NONE)
2858 * We stop writing back only if we are
2859 * not doing integrity sync. In case of
2860 * integrity sync we have to keep going
2861 * because someone may be concurrently
2862 * dirtying pages, and we might have
2863 * synced a lot of newly appeared dirty
2864 * pages, but have not synced all of the
2870 pagevec_release(&pvec);
2875 ret = MPAGE_DA_EXTENT_TAIL;
2877 pagevec_release(&pvec);
2883 static int ext4_da_writepages(struct address_space *mapping,
2884 struct writeback_control *wbc)
2887 int range_whole = 0;
2888 handle_t *handle = NULL;
2889 struct mpage_da_data mpd;
2890 struct inode *inode = mapping->host;
2891 int pages_written = 0;
2892 unsigned int max_pages;
2893 int range_cyclic, cycled = 1, io_done = 0;
2894 int needed_blocks, ret = 0;
2895 long desired_nr_to_write, nr_to_writebump = 0;
2896 loff_t range_start = wbc->range_start;
2897 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2898 pgoff_t done_index = 0;
2901 trace_ext4_da_writepages(inode, wbc);
2904 * No pages to write? This is mainly a kludge to avoid starting
2905 * a transaction for special inodes like journal inode on last iput()
2906 * because that could violate lock ordering on umount
2908 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2912 * If the filesystem has aborted, it is read-only, so return
2913 * right away instead of dumping stack traces later on that
2914 * will obscure the real source of the problem. We test
2915 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2916 * the latter could be true if the filesystem is mounted
2917 * read-only, and in that case, ext4_da_writepages should
2918 * *never* be called, so if that ever happens, we would want
2921 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2924 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2927 range_cyclic = wbc->range_cyclic;
2928 if (wbc->range_cyclic) {
2929 index = mapping->writeback_index;
2932 wbc->range_start = index << PAGE_CACHE_SHIFT;
2933 wbc->range_end = LLONG_MAX;
2934 wbc->range_cyclic = 0;
2937 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2938 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2942 * This works around two forms of stupidity. The first is in
2943 * the writeback code, which caps the maximum number of pages
2944 * written to be 1024 pages. This is wrong on multiple
2945 * levels; different architectues have a different page size,
2946 * which changes the maximum amount of data which gets
2947 * written. Secondly, 4 megabytes is way too small. XFS
2948 * forces this value to be 16 megabytes by multiplying
2949 * nr_to_write parameter by four, and then relies on its
2950 * allocator to allocate larger extents to make them
2951 * contiguous. Unfortunately this brings us to the second
2952 * stupidity, which is that ext4's mballoc code only allocates
2953 * at most 2048 blocks. So we force contiguous writes up to
2954 * the number of dirty blocks in the inode, or
2955 * sbi->max_writeback_mb_bump whichever is smaller.
2957 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2958 if (!range_cyclic && range_whole) {
2959 if (wbc->nr_to_write == LONG_MAX)
2960 desired_nr_to_write = wbc->nr_to_write;
2962 desired_nr_to_write = wbc->nr_to_write * 8;
2964 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2966 if (desired_nr_to_write > max_pages)
2967 desired_nr_to_write = max_pages;
2969 if (wbc->nr_to_write < desired_nr_to_write) {
2970 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2971 wbc->nr_to_write = desired_nr_to_write;
2975 if (wbc->sync_mode == WB_SYNC_ALL)
2976 tag_pages_for_writeback(mapping, index, end);
2978 while (!ret && wbc->nr_to_write > 0) {
2981 * we insert one extent at a time. So we need
2982 * credit needed for single extent allocation.
2983 * journalled mode is currently not supported
2986 BUG_ON(ext4_should_journal_data(inode));
2987 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2989 /* start a new transaction*/
2990 handle = ext4_journal_start(inode, needed_blocks);
2991 if (IS_ERR(handle)) {
2992 ret = PTR_ERR(handle);
2993 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2994 "%ld pages, ino %lu; err %d", __func__,
2995 wbc->nr_to_write, inode->i_ino, ret);
2996 goto out_writepages;
3000 * Now call write_cache_pages_da() to find the next
3001 * contiguous region of logical blocks that need
3002 * blocks to be allocated by ext4 and submit them.
3004 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3006 * If we have a contiguous extent of pages and we
3007 * haven't done the I/O yet, map the blocks and submit
3010 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3011 mpage_da_map_and_submit(&mpd);
3012 ret = MPAGE_DA_EXTENT_TAIL;
3014 trace_ext4_da_write_pages(inode, &mpd);
3015 wbc->nr_to_write -= mpd.pages_written;
3017 ext4_journal_stop(handle);
3019 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3020 /* commit the transaction which would
3021 * free blocks released in the transaction
3024 jbd2_journal_force_commit_nested(sbi->s_journal);
3026 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3028 * got one extent now try with
3031 pages_written += mpd.pages_written;
3034 } else if (wbc->nr_to_write)
3036 * There is no more writeout needed
3037 * or we requested for a noblocking writeout
3038 * and we found the device congested
3042 if (!io_done && !cycled) {
3045 wbc->range_start = index << PAGE_CACHE_SHIFT;
3046 wbc->range_end = mapping->writeback_index - 1;
3051 wbc->range_cyclic = range_cyclic;
3052 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3054 * set the writeback_index so that range_cyclic
3055 * mode will write it back later
3057 mapping->writeback_index = done_index;
3060 wbc->nr_to_write -= nr_to_writebump;
3061 wbc->range_start = range_start;
3062 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3066 #define FALL_BACK_TO_NONDELALLOC 1
3067 static int ext4_nonda_switch(struct super_block *sb)
3069 s64 free_blocks, dirty_blocks;
3070 struct ext4_sb_info *sbi = EXT4_SB(sb);
3073 * switch to non delalloc mode if we are running low
3074 * on free block. The free block accounting via percpu
3075 * counters can get slightly wrong with percpu_counter_batch getting
3076 * accumulated on each CPU without updating global counters
3077 * Delalloc need an accurate free block accounting. So switch
3078 * to non delalloc when we are near to error range.
3080 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3081 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3082 if (2 * free_blocks < 3 * dirty_blocks ||
3083 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3085 * free block count is less than 150% of dirty blocks
3086 * or free blocks is less than watermark
3091 * Even if we don't switch but are nearing capacity,
3092 * start pushing delalloc when 1/2 of free blocks are dirty.
3094 if (free_blocks < 2 * dirty_blocks)
3095 writeback_inodes_sb_if_idle(sb);
3100 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3101 loff_t pos, unsigned len, unsigned flags,
3102 struct page **pagep, void **fsdata)
3104 int ret, retries = 0;
3107 struct inode *inode = mapping->host;
3110 index = pos >> PAGE_CACHE_SHIFT;
3112 if (ext4_nonda_switch(inode->i_sb)) {
3113 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3114 return ext4_write_begin(file, mapping, pos,
3115 len, flags, pagep, fsdata);
3117 *fsdata = (void *)0;
3118 trace_ext4_da_write_begin(inode, pos, len, flags);
3121 * With delayed allocation, we don't log the i_disksize update
3122 * if there is delayed block allocation. But we still need
3123 * to journalling the i_disksize update if writes to the end
3124 * of file which has an already mapped buffer.
3126 handle = ext4_journal_start(inode, 1);
3127 if (IS_ERR(handle)) {
3128 ret = PTR_ERR(handle);
3131 /* We cannot recurse into the filesystem as the transaction is already
3133 flags |= AOP_FLAG_NOFS;
3135 page = grab_cache_page_write_begin(mapping, index, flags);
3137 ext4_journal_stop(handle);
3143 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3146 ext4_journal_stop(handle);
3147 page_cache_release(page);
3149 * block_write_begin may have instantiated a few blocks
3150 * outside i_size. Trim these off again. Don't need
3151 * i_size_read because we hold i_mutex.
3153 if (pos + len > inode->i_size)
3154 ext4_truncate_failed_write(inode);
3157 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3164 * Check if we should update i_disksize
3165 * when write to the end of file but not require block allocation
3167 static int ext4_da_should_update_i_disksize(struct page *page,
3168 unsigned long offset)
3170 struct buffer_head *bh;
3171 struct inode *inode = page->mapping->host;
3175 bh = page_buffers(page);
3176 idx = offset >> inode->i_blkbits;
3178 for (i = 0; i < idx; i++)
3179 bh = bh->b_this_page;
3181 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3186 static int ext4_da_write_end(struct file *file,
3187 struct address_space *mapping,
3188 loff_t pos, unsigned len, unsigned copied,
3189 struct page *page, void *fsdata)
3191 struct inode *inode = mapping->host;
3193 handle_t *handle = ext4_journal_current_handle();
3195 unsigned long start, end;
3196 int write_mode = (int)(unsigned long)fsdata;
3198 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3199 if (ext4_should_order_data(inode)) {
3200 return ext4_ordered_write_end(file, mapping, pos,
3201 len, copied, page, fsdata);
3202 } else if (ext4_should_writeback_data(inode)) {
3203 return ext4_writeback_write_end(file, mapping, pos,
3204 len, copied, page, fsdata);
3210 trace_ext4_da_write_end(inode, pos, len, copied);
3211 start = pos & (PAGE_CACHE_SIZE - 1);
3212 end = start + copied - 1;
3215 * generic_write_end() will run mark_inode_dirty() if i_size
3216 * changes. So let's piggyback the i_disksize mark_inode_dirty
3220 new_i_size = pos + copied;
3221 if (new_i_size > EXT4_I(inode)->i_disksize) {
3222 if (ext4_da_should_update_i_disksize(page, end)) {
3223 down_write(&EXT4_I(inode)->i_data_sem);
3224 if (new_i_size > EXT4_I(inode)->i_disksize) {
3226 * Updating i_disksize when extending file
3227 * without needing block allocation
3229 if (ext4_should_order_data(inode))
3230 ret = ext4_jbd2_file_inode(handle,
3233 EXT4_I(inode)->i_disksize = new_i_size;
3235 up_write(&EXT4_I(inode)->i_data_sem);
3236 /* We need to mark inode dirty even if
3237 * new_i_size is less that inode->i_size
3238 * bu greater than i_disksize.(hint delalloc)
3240 ext4_mark_inode_dirty(handle, inode);
3243 ret2 = generic_write_end(file, mapping, pos, len, copied,
3248 ret2 = ext4_journal_stop(handle);
3252 return ret ? ret : copied;
3255 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3258 * Drop reserved blocks
3260 BUG_ON(!PageLocked(page));
3261 if (!page_has_buffers(page))
3264 ext4_da_page_release_reservation(page, offset);
3267 ext4_invalidatepage(page, offset);
3273 * Force all delayed allocation blocks to be allocated for a given inode.
3275 int ext4_alloc_da_blocks(struct inode *inode)
3277 trace_ext4_alloc_da_blocks(inode);
3279 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3280 !EXT4_I(inode)->i_reserved_meta_blocks)
3284 * We do something simple for now. The filemap_flush() will
3285 * also start triggering a write of the data blocks, which is
3286 * not strictly speaking necessary (and for users of
3287 * laptop_mode, not even desirable). However, to do otherwise
3288 * would require replicating code paths in:
3290 * ext4_da_writepages() ->
3291 * write_cache_pages() ---> (via passed in callback function)
3292 * __mpage_da_writepage() -->
3293 * mpage_add_bh_to_extent()
3294 * mpage_da_map_blocks()
3296 * The problem is that write_cache_pages(), located in
3297 * mm/page-writeback.c, marks pages clean in preparation for
3298 * doing I/O, which is not desirable if we're not planning on
3301 * We could call write_cache_pages(), and then redirty all of
3302 * the pages by calling redirty_page_for_writepage() but that
3303 * would be ugly in the extreme. So instead we would need to
3304 * replicate parts of the code in the above functions,
3305 * simplifying them becuase we wouldn't actually intend to
3306 * write out the pages, but rather only collect contiguous
3307 * logical block extents, call the multi-block allocator, and
3308 * then update the buffer heads with the block allocations.
3310 * For now, though, we'll cheat by calling filemap_flush(),
3311 * which will map the blocks, and start the I/O, but not
3312 * actually wait for the I/O to complete.
3314 return filemap_flush(inode->i_mapping);
3318 * bmap() is special. It gets used by applications such as lilo and by
3319 * the swapper to find the on-disk block of a specific piece of data.
3321 * Naturally, this is dangerous if the block concerned is still in the
3322 * journal. If somebody makes a swapfile on an ext4 data-journaling
3323 * filesystem and enables swap, then they may get a nasty shock when the
3324 * data getting swapped to that swapfile suddenly gets overwritten by
3325 * the original zero's written out previously to the journal and
3326 * awaiting writeback in the kernel's buffer cache.
3328 * So, if we see any bmap calls here on a modified, data-journaled file,
3329 * take extra steps to flush any blocks which might be in the cache.
3331 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3333 struct inode *inode = mapping->host;
3337 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3338 test_opt(inode->i_sb, DELALLOC)) {
3340 * With delalloc we want to sync the file
3341 * so that we can make sure we allocate
3344 filemap_write_and_wait(mapping);
3347 if (EXT4_JOURNAL(inode) &&
3348 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3350 * This is a REALLY heavyweight approach, but the use of
3351 * bmap on dirty files is expected to be extremely rare:
3352 * only if we run lilo or swapon on a freshly made file
3353 * do we expect this to happen.
3355 * (bmap requires CAP_SYS_RAWIO so this does not
3356 * represent an unprivileged user DOS attack --- we'd be
3357 * in trouble if mortal users could trigger this path at
3360 * NB. EXT4_STATE_JDATA is not set on files other than
3361 * regular files. If somebody wants to bmap a directory
3362 * or symlink and gets confused because the buffer
3363 * hasn't yet been flushed to disk, they deserve
3364 * everything they get.
3367 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3368 journal = EXT4_JOURNAL(inode);
3369 jbd2_journal_lock_updates(journal);
3370 err = jbd2_journal_flush(journal);
3371 jbd2_journal_unlock_updates(journal);
3377 return generic_block_bmap(mapping, block, ext4_get_block);
3380 static int ext4_readpage(struct file *file, struct page *page)
3382 return mpage_readpage(page, ext4_get_block);
3386 ext4_readpages(struct file *file, struct address_space *mapping,
3387 struct list_head *pages, unsigned nr_pages)
3389 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3392 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3394 struct buffer_head *head, *bh;
3395 unsigned int curr_off = 0;
3397 if (!page_has_buffers(page))
3399 head = bh = page_buffers(page);
3401 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3403 ext4_free_io_end(bh->b_private);
3404 bh->b_private = NULL;
3405 bh->b_end_io = NULL;
3407 curr_off = curr_off + bh->b_size;
3408 bh = bh->b_this_page;
3409 } while (bh != head);
3412 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3414 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3417 * free any io_end structure allocated for buffers to be discarded
3419 if (ext4_should_dioread_nolock(page->mapping->host))
3420 ext4_invalidatepage_free_endio(page, offset);
3422 * If it's a full truncate we just forget about the pending dirtying
3425 ClearPageChecked(page);
3428 jbd2_journal_invalidatepage(journal, page, offset);
3430 block_invalidatepage(page, offset);
3433 static int ext4_releasepage(struct page *page, gfp_t wait)
3435 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3437 WARN_ON(PageChecked(page));
3438 if (!page_has_buffers(page))
3441 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3443 return try_to_free_buffers(page);
3447 * O_DIRECT for ext3 (or indirect map) based files
3449 * If the O_DIRECT write will extend the file then add this inode to the
3450 * orphan list. So recovery will truncate it back to the original size
3451 * if the machine crashes during the write.
3453 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3454 * crashes then stale disk data _may_ be exposed inside the file. But current
3455 * VFS code falls back into buffered path in that case so we are safe.
3457 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3458 const struct iovec *iov, loff_t offset,
3459 unsigned long nr_segs)
3461 struct file *file = iocb->ki_filp;
3462 struct inode *inode = file->f_mapping->host;
3463 struct ext4_inode_info *ei = EXT4_I(inode);
3467 size_t count = iov_length(iov, nr_segs);
3471 loff_t final_size = offset + count;
3473 if (final_size > inode->i_size) {
3474 /* Credits for sb + inode write */
3475 handle = ext4_journal_start(inode, 2);
3476 if (IS_ERR(handle)) {
3477 ret = PTR_ERR(handle);
3480 ret = ext4_orphan_add(handle, inode);
3482 ext4_journal_stop(handle);
3486 ei->i_disksize = inode->i_size;
3487 ext4_journal_stop(handle);
3492 if (rw == READ && ext4_should_dioread_nolock(inode))
3493 ret = __blockdev_direct_IO(rw, iocb, inode,
3494 inode->i_sb->s_bdev, iov,
3496 ext4_get_block, NULL, NULL, 0);
3498 ret = blockdev_direct_IO(rw, iocb, inode,
3499 inode->i_sb->s_bdev, iov,
3501 ext4_get_block, NULL);
3503 if (unlikely((rw & WRITE) && ret < 0)) {
3504 loff_t isize = i_size_read(inode);
3505 loff_t end = offset + iov_length(iov, nr_segs);
3508 vmtruncate(inode, isize);
3511 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3517 /* Credits for sb + inode write */
3518 handle = ext4_journal_start(inode, 2);
3519 if (IS_ERR(handle)) {
3520 /* This is really bad luck. We've written the data
3521 * but cannot extend i_size. Bail out and pretend
3522 * the write failed... */
3523 ret = PTR_ERR(handle);
3525 ext4_orphan_del(NULL, inode);
3530 ext4_orphan_del(handle, inode);
3532 loff_t end = offset + ret;
3533 if (end > inode->i_size) {
3534 ei->i_disksize = end;
3535 i_size_write(inode, end);
3537 * We're going to return a positive `ret'
3538 * here due to non-zero-length I/O, so there's
3539 * no way of reporting error returns from
3540 * ext4_mark_inode_dirty() to userspace. So
3543 ext4_mark_inode_dirty(handle, inode);
3546 err = ext4_journal_stop(handle);
3555 * ext4_get_block used when preparing for a DIO write or buffer write.
3556 * We allocate an uinitialized extent if blocks haven't been allocated.
3557 * The extent will be converted to initialized after the IO is complete.
3559 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3560 struct buffer_head *bh_result, int create)
3562 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3563 inode->i_ino, create);
3564 return _ext4_get_block(inode, iblock, bh_result,
3565 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3568 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3569 ssize_t size, void *private, int ret,
3572 ext4_io_end_t *io_end = iocb->private;
3573 struct workqueue_struct *wq;
3574 unsigned long flags;
3575 struct ext4_inode_info *ei;
3577 /* if not async direct IO or dio with 0 bytes write, just return */
3578 if (!io_end || !size)
3581 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3582 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3583 iocb->private, io_end->inode->i_ino, iocb, offset,
3586 /* if not aio dio with unwritten extents, just free io and return */
3587 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3588 ext4_free_io_end(io_end);
3589 iocb->private = NULL;
3592 aio_complete(iocb, ret, 0);
3596 io_end->offset = offset;
3597 io_end->size = size;
3599 io_end->iocb = iocb;
3600 io_end->result = ret;
3602 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3604 /* Add the io_end to per-inode completed aio dio list*/
3605 ei = EXT4_I(io_end->inode);
3606 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3607 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3608 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3610 /* queue the work to convert unwritten extents to written */
3611 queue_work(wq, &io_end->work);
3612 iocb->private = NULL;
3615 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3617 ext4_io_end_t *io_end = bh->b_private;
3618 struct workqueue_struct *wq;
3619 struct inode *inode;
3620 unsigned long flags;
3622 if (!test_clear_buffer_uninit(bh) || !io_end)
3625 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3626 printk("sb umounted, discard end_io request for inode %lu\n",
3627 io_end->inode->i_ino);
3628 ext4_free_io_end(io_end);
3632 io_end->flag = EXT4_IO_END_UNWRITTEN;
3633 inode = io_end->inode;
3635 /* Add the io_end to per-inode completed io list*/
3636 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3637 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3638 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3640 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3641 /* queue the work to convert unwritten extents to written */
3642 queue_work(wq, &io_end->work);
3644 bh->b_private = NULL;
3645 bh->b_end_io = NULL;
3646 clear_buffer_uninit(bh);
3647 end_buffer_async_write(bh, uptodate);
3650 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3652 ext4_io_end_t *io_end;
3653 struct page *page = bh->b_page;
3654 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3655 size_t size = bh->b_size;
3658 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3660 pr_warn_ratelimited("%s: allocation fail\n", __func__);
3664 io_end->offset = offset;
3665 io_end->size = size;
3667 * We need to hold a reference to the page to make sure it
3668 * doesn't get evicted before ext4_end_io_work() has a chance
3669 * to convert the extent from written to unwritten.
3671 io_end->page = page;
3672 get_page(io_end->page);
3674 bh->b_private = io_end;
3675 bh->b_end_io = ext4_end_io_buffer_write;
3680 * For ext4 extent files, ext4 will do direct-io write to holes,
3681 * preallocated extents, and those write extend the file, no need to
3682 * fall back to buffered IO.
3684 * For holes, we fallocate those blocks, mark them as uninitialized
3685 * If those blocks were preallocated, we mark sure they are splited, but
3686 * still keep the range to write as uninitialized.
3688 * The unwrritten extents will be converted to written when DIO is completed.
3689 * For async direct IO, since the IO may still pending when return, we
3690 * set up an end_io call back function, which will do the convertion
3691 * when async direct IO completed.
3693 * If the O_DIRECT write will extend the file then add this inode to the
3694 * orphan list. So recovery will truncate it back to the original size
3695 * if the machine crashes during the write.
3698 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3699 const struct iovec *iov, loff_t offset,
3700 unsigned long nr_segs)
3702 struct file *file = iocb->ki_filp;
3703 struct inode *inode = file->f_mapping->host;
3705 size_t count = iov_length(iov, nr_segs);
3707 loff_t final_size = offset + count;
3708 if (rw == WRITE && final_size <= inode->i_size) {
3710 * We could direct write to holes and fallocate.
3712 * Allocated blocks to fill the hole are marked as uninitialized
3713 * to prevent paralel buffered read to expose the stale data
3714 * before DIO complete the data IO.
3716 * As to previously fallocated extents, ext4 get_block
3717 * will just simply mark the buffer mapped but still
3718 * keep the extents uninitialized.
3720 * for non AIO case, we will convert those unwritten extents
3721 * to written after return back from blockdev_direct_IO.
3723 * for async DIO, the conversion needs to be defered when
3724 * the IO is completed. The ext4 end_io callback function
3725 * will be called to take care of the conversion work.
3726 * Here for async case, we allocate an io_end structure to
3729 iocb->private = NULL;
3730 EXT4_I(inode)->cur_aio_dio = NULL;
3731 if (!is_sync_kiocb(iocb)) {
3732 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3736 * we save the io structure for current async
3737 * direct IO, so that later ext4_map_blocks()
3738 * could flag the io structure whether there
3739 * is a unwritten extents needs to be converted
3740 * when IO is completed.
3742 EXT4_I(inode)->cur_aio_dio = iocb->private;
3745 ret = blockdev_direct_IO(rw, iocb, inode,
3746 inode->i_sb->s_bdev, iov,
3748 ext4_get_block_write,
3751 EXT4_I(inode)->cur_aio_dio = NULL;
3753 * The io_end structure takes a reference to the inode,
3754 * that structure needs to be destroyed and the
3755 * reference to the inode need to be dropped, when IO is
3756 * complete, even with 0 byte write, or failed.
3758 * In the successful AIO DIO case, the io_end structure will be
3759 * desctroyed and the reference to the inode will be dropped
3760 * after the end_io call back function is called.
3762 * In the case there is 0 byte write, or error case, since
3763 * VFS direct IO won't invoke the end_io call back function,
3764 * we need to free the end_io structure here.
3766 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3767 ext4_free_io_end(iocb->private);
3768 iocb->private = NULL;
3769 } else if (ret > 0 && ext4_test_inode_state(inode,
3770 EXT4_STATE_DIO_UNWRITTEN)) {
3773 * for non AIO case, since the IO is already
3774 * completed, we could do the convertion right here
3776 err = ext4_convert_unwritten_extents(inode,
3780 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3785 /* for write the the end of file case, we fall back to old way */
3786 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3789 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3790 const struct iovec *iov, loff_t offset,
3791 unsigned long nr_segs)
3793 struct file *file = iocb->ki_filp;
3794 struct inode *inode = file->f_mapping->host;
3796 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3797 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3799 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3803 * Pages can be marked dirty completely asynchronously from ext4's journalling
3804 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3805 * much here because ->set_page_dirty is called under VFS locks. The page is
3806 * not necessarily locked.
3808 * We cannot just dirty the page and leave attached buffers clean, because the
3809 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3810 * or jbddirty because all the journalling code will explode.
3812 * So what we do is to mark the page "pending dirty" and next time writepage
3813 * is called, propagate that into the buffers appropriately.
3815 static int ext4_journalled_set_page_dirty(struct page *page)
3817 SetPageChecked(page);
3818 return __set_page_dirty_nobuffers(page);
3821 static const struct address_space_operations ext4_ordered_aops = {
3822 .readpage = ext4_readpage,
3823 .readpages = ext4_readpages,
3824 .writepage = ext4_writepage,
3825 .sync_page = block_sync_page,
3826 .write_begin = ext4_write_begin,
3827 .write_end = ext4_ordered_write_end,
3829 .invalidatepage = ext4_invalidatepage,
3830 .releasepage = ext4_releasepage,
3831 .direct_IO = ext4_direct_IO,
3832 .migratepage = buffer_migrate_page,
3833 .is_partially_uptodate = block_is_partially_uptodate,
3834 .error_remove_page = generic_error_remove_page,
3837 static const struct address_space_operations ext4_writeback_aops = {
3838 .readpage = ext4_readpage,
3839 .readpages = ext4_readpages,
3840 .writepage = ext4_writepage,
3841 .sync_page = block_sync_page,
3842 .write_begin = ext4_write_begin,
3843 .write_end = ext4_writeback_write_end,
3845 .invalidatepage = ext4_invalidatepage,
3846 .releasepage = ext4_releasepage,
3847 .direct_IO = ext4_direct_IO,
3848 .migratepage = buffer_migrate_page,
3849 .is_partially_uptodate = block_is_partially_uptodate,
3850 .error_remove_page = generic_error_remove_page,
3853 static const struct address_space_operations ext4_journalled_aops = {
3854 .readpage = ext4_readpage,
3855 .readpages = ext4_readpages,
3856 .writepage = ext4_writepage,
3857 .sync_page = block_sync_page,
3858 .write_begin = ext4_write_begin,
3859 .write_end = ext4_journalled_write_end,
3860 .set_page_dirty = ext4_journalled_set_page_dirty,
3862 .invalidatepage = ext4_invalidatepage,
3863 .releasepage = ext4_releasepage,
3864 .is_partially_uptodate = block_is_partially_uptodate,
3865 .error_remove_page = generic_error_remove_page,
3868 static const struct address_space_operations ext4_da_aops = {
3869 .readpage = ext4_readpage,
3870 .readpages = ext4_readpages,
3871 .writepage = ext4_writepage,
3872 .writepages = ext4_da_writepages,
3873 .sync_page = block_sync_page,
3874 .write_begin = ext4_da_write_begin,
3875 .write_end = ext4_da_write_end,
3877 .invalidatepage = ext4_da_invalidatepage,
3878 .releasepage = ext4_releasepage,
3879 .direct_IO = ext4_direct_IO,
3880 .migratepage = buffer_migrate_page,
3881 .is_partially_uptodate = block_is_partially_uptodate,
3882 .error_remove_page = generic_error_remove_page,
3885 void ext4_set_aops(struct inode *inode)
3887 if (ext4_should_order_data(inode) &&
3888 test_opt(inode->i_sb, DELALLOC))
3889 inode->i_mapping->a_ops = &ext4_da_aops;
3890 else if (ext4_should_order_data(inode))
3891 inode->i_mapping->a_ops = &ext4_ordered_aops;
3892 else if (ext4_should_writeback_data(inode) &&
3893 test_opt(inode->i_sb, DELALLOC))
3894 inode->i_mapping->a_ops = &ext4_da_aops;
3895 else if (ext4_should_writeback_data(inode))
3896 inode->i_mapping->a_ops = &ext4_writeback_aops;
3898 inode->i_mapping->a_ops = &ext4_journalled_aops;
3902 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3903 * up to the end of the block which corresponds to `from'.
3904 * This required during truncate. We need to physically zero the tail end
3905 * of that block so it doesn't yield old data if the file is later grown.
3907 int ext4_block_truncate_page(handle_t *handle,
3908 struct address_space *mapping, loff_t from)
3910 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3911 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3912 unsigned blocksize, length, pos;
3914 struct inode *inode = mapping->host;
3915 struct buffer_head *bh;
3919 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3920 mapping_gfp_mask(mapping) & ~__GFP_FS);
3924 blocksize = inode->i_sb->s_blocksize;
3925 length = blocksize - (offset & (blocksize - 1));
3926 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3928 if (!page_has_buffers(page))
3929 create_empty_buffers(page, blocksize, 0);
3931 /* Find the buffer that contains "offset" */
3932 bh = page_buffers(page);
3934 while (offset >= pos) {
3935 bh = bh->b_this_page;
3941 if (buffer_freed(bh)) {
3942 BUFFER_TRACE(bh, "freed: skip");
3946 if (!buffer_mapped(bh)) {
3947 BUFFER_TRACE(bh, "unmapped");
3948 ext4_get_block(inode, iblock, bh, 0);
3949 /* unmapped? It's a hole - nothing to do */
3950 if (!buffer_mapped(bh)) {
3951 BUFFER_TRACE(bh, "still unmapped");
3956 /* Ok, it's mapped. Make sure it's up-to-date */
3957 if (PageUptodate(page))
3958 set_buffer_uptodate(bh);
3960 if (!buffer_uptodate(bh)) {
3962 ll_rw_block(READ, 1, &bh);
3964 /* Uhhuh. Read error. Complain and punt. */
3965 if (!buffer_uptodate(bh))
3969 if (ext4_should_journal_data(inode)) {
3970 BUFFER_TRACE(bh, "get write access");
3971 err = ext4_journal_get_write_access(handle, bh);
3976 zero_user(page, offset, length);
3978 BUFFER_TRACE(bh, "zeroed end of block");
3981 if (ext4_should_journal_data(inode)) {
3982 err = ext4_handle_dirty_metadata(handle, inode, bh);
3984 if (ext4_should_order_data(inode) && EXT4_I(inode)->jinode)
3985 err = ext4_jbd2_file_inode(handle, inode);
3986 mark_buffer_dirty(bh);
3991 page_cache_release(page);
3996 * Probably it should be a library function... search for first non-zero word
3997 * or memcmp with zero_page, whatever is better for particular architecture.
4000 static inline int all_zeroes(__le32 *p, __le32 *q)
4009 * ext4_find_shared - find the indirect blocks for partial truncation.
4010 * @inode: inode in question
4011 * @depth: depth of the affected branch
4012 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4013 * @chain: place to store the pointers to partial indirect blocks
4014 * @top: place to the (detached) top of branch
4016 * This is a helper function used by ext4_truncate().
4018 * When we do truncate() we may have to clean the ends of several
4019 * indirect blocks but leave the blocks themselves alive. Block is
4020 * partially truncated if some data below the new i_size is refered
4021 * from it (and it is on the path to the first completely truncated
4022 * data block, indeed). We have to free the top of that path along
4023 * with everything to the right of the path. Since no allocation
4024 * past the truncation point is possible until ext4_truncate()
4025 * finishes, we may safely do the latter, but top of branch may
4026 * require special attention - pageout below the truncation point
4027 * might try to populate it.
4029 * We atomically detach the top of branch from the tree, store the
4030 * block number of its root in *@top, pointers to buffer_heads of
4031 * partially truncated blocks - in @chain[].bh and pointers to
4032 * their last elements that should not be removed - in
4033 * @chain[].p. Return value is the pointer to last filled element
4036 * The work left to caller to do the actual freeing of subtrees:
4037 * a) free the subtree starting from *@top
4038 * b) free the subtrees whose roots are stored in
4039 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4040 * c) free the subtrees growing from the inode past the @chain[0].
4041 * (no partially truncated stuff there). */
4043 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4044 ext4_lblk_t offsets[4], Indirect chain[4],
4047 Indirect *partial, *p;
4051 /* Make k index the deepest non-null offset + 1 */
4052 for (k = depth; k > 1 && !offsets[k-1]; k--)
4054 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4055 /* Writer: pointers */
4057 partial = chain + k-1;
4059 * If the branch acquired continuation since we've looked at it -
4060 * fine, it should all survive and (new) top doesn't belong to us.
4062 if (!partial->key && *partial->p)
4065 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4068 * OK, we've found the last block that must survive. The rest of our
4069 * branch should be detached before unlocking. However, if that rest
4070 * of branch is all ours and does not grow immediately from the inode
4071 * it's easier to cheat and just decrement partial->p.
4073 if (p == chain + k - 1 && p > chain) {
4077 /* Nope, don't do this in ext4. Must leave the tree intact */
4084 while (partial > p) {
4085 brelse(partial->bh);
4093 * Zero a number of block pointers in either an inode or an indirect block.
4094 * If we restart the transaction we must again get write access to the
4095 * indirect block for further modification.
4097 * We release `count' blocks on disk, but (last - first) may be greater
4098 * than `count' because there can be holes in there.
4100 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4101 struct buffer_head *bh,
4102 ext4_fsblk_t block_to_free,
4103 unsigned long count, __le32 *first,
4107 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4110 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4111 flags |= EXT4_FREE_BLOCKS_METADATA;
4113 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4115 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4116 "blocks %llu len %lu",
4117 (unsigned long long) block_to_free, count);
4121 if (try_to_extend_transaction(handle, inode)) {
4123 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4124 err = ext4_handle_dirty_metadata(handle, inode, bh);
4125 if (unlikely(err)) {
4126 ext4_std_error(inode->i_sb, err);
4130 err = ext4_mark_inode_dirty(handle, inode);
4131 if (unlikely(err)) {
4132 ext4_std_error(inode->i_sb, err);
4135 err = ext4_truncate_restart_trans(handle, inode,
4136 blocks_for_truncate(inode));
4137 if (unlikely(err)) {
4138 ext4_std_error(inode->i_sb, err);
4142 BUFFER_TRACE(bh, "retaking write access");
4143 ext4_journal_get_write_access(handle, bh);
4147 for (p = first; p < last; p++)
4150 ext4_free_blocks(handle, inode, NULL, block_to_free, count, flags);
4155 * ext4_free_data - free a list of data blocks
4156 * @handle: handle for this transaction
4157 * @inode: inode we are dealing with
4158 * @this_bh: indirect buffer_head which contains *@first and *@last
4159 * @first: array of block numbers
4160 * @last: points immediately past the end of array
4162 * We are freeing all blocks refered from that array (numbers are stored as
4163 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4165 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4166 * blocks are contiguous then releasing them at one time will only affect one
4167 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4168 * actually use a lot of journal space.
4170 * @this_bh will be %NULL if @first and @last point into the inode's direct
4173 static void ext4_free_data(handle_t *handle, struct inode *inode,
4174 struct buffer_head *this_bh,
4175 __le32 *first, __le32 *last)
4177 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4178 unsigned long count = 0; /* Number of blocks in the run */
4179 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4182 ext4_fsblk_t nr; /* Current block # */
4183 __le32 *p; /* Pointer into inode/ind
4184 for current block */
4187 if (this_bh) { /* For indirect block */
4188 BUFFER_TRACE(this_bh, "get_write_access");
4189 err = ext4_journal_get_write_access(handle, this_bh);
4190 /* Important: if we can't update the indirect pointers
4191 * to the blocks, we can't free them. */
4196 for (p = first; p < last; p++) {
4197 nr = le32_to_cpu(*p);
4199 /* accumulate blocks to free if they're contiguous */
4202 block_to_free_p = p;
4204 } else if (nr == block_to_free + count) {
4207 if (ext4_clear_blocks(handle, inode, this_bh,
4208 block_to_free, count,
4209 block_to_free_p, p))
4212 block_to_free_p = p;
4219 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4220 count, block_to_free_p, p);
4223 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4226 * The buffer head should have an attached journal head at this
4227 * point. However, if the data is corrupted and an indirect
4228 * block pointed to itself, it would have been detached when
4229 * the block was cleared. Check for this instead of OOPSing.
4231 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4232 ext4_handle_dirty_metadata(handle, inode, this_bh);
4234 EXT4_ERROR_INODE(inode,
4235 "circular indirect block detected at "
4237 (unsigned long long) this_bh->b_blocknr);
4242 * ext4_free_branches - free an array of branches
4243 * @handle: JBD handle for this transaction
4244 * @inode: inode we are dealing with
4245 * @parent_bh: the buffer_head which contains *@first and *@last
4246 * @first: array of block numbers
4247 * @last: pointer immediately past the end of array
4248 * @depth: depth of the branches to free
4250 * We are freeing all blocks refered from these branches (numbers are
4251 * stored as little-endian 32-bit) and updating @inode->i_blocks
4254 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4255 struct buffer_head *parent_bh,
4256 __le32 *first, __le32 *last, int depth)
4261 if (ext4_handle_is_aborted(handle))
4265 struct buffer_head *bh;
4266 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4268 while (--p >= first) {
4269 nr = le32_to_cpu(*p);
4271 continue; /* A hole */
4273 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4275 EXT4_ERROR_INODE(inode,
4276 "invalid indirect mapped "
4277 "block %lu (level %d)",
4278 (unsigned long) nr, depth);
4282 /* Go read the buffer for the next level down */
4283 bh = sb_bread(inode->i_sb, nr);
4286 * A read failure? Report error and clear slot
4290 EXT4_ERROR_INODE_BLOCK(inode, nr,
4295 /* This zaps the entire block. Bottom up. */
4296 BUFFER_TRACE(bh, "free child branches");
4297 ext4_free_branches(handle, inode, bh,
4298 (__le32 *) bh->b_data,
4299 (__le32 *) bh->b_data + addr_per_block,
4304 * Everything below this this pointer has been
4305 * released. Now let this top-of-subtree go.
4307 * We want the freeing of this indirect block to be
4308 * atomic in the journal with the updating of the
4309 * bitmap block which owns it. So make some room in
4312 * We zero the parent pointer *after* freeing its
4313 * pointee in the bitmaps, so if extend_transaction()
4314 * for some reason fails to put the bitmap changes and
4315 * the release into the same transaction, recovery
4316 * will merely complain about releasing a free block,
4317 * rather than leaking blocks.
4319 if (ext4_handle_is_aborted(handle))
4321 if (try_to_extend_transaction(handle, inode)) {
4322 ext4_mark_inode_dirty(handle, inode);
4323 ext4_truncate_restart_trans(handle, inode,
4324 blocks_for_truncate(inode));
4328 * The forget flag here is critical because if
4329 * we are journaling (and not doing data
4330 * journaling), we have to make sure a revoke
4331 * record is written to prevent the journal
4332 * replay from overwriting the (former)
4333 * indirect block if it gets reallocated as a
4334 * data block. This must happen in the same
4335 * transaction where the data blocks are
4338 ext4_free_blocks(handle, inode, NULL, nr, 1,
4339 EXT4_FREE_BLOCKS_METADATA|
4340 EXT4_FREE_BLOCKS_FORGET);
4344 * The block which we have just freed is
4345 * pointed to by an indirect block: journal it
4347 BUFFER_TRACE(parent_bh, "get_write_access");
4348 if (!ext4_journal_get_write_access(handle,
4351 BUFFER_TRACE(parent_bh,
4352 "call ext4_handle_dirty_metadata");
4353 ext4_handle_dirty_metadata(handle,
4360 /* We have reached the bottom of the tree. */
4361 BUFFER_TRACE(parent_bh, "free data blocks");
4362 ext4_free_data(handle, inode, parent_bh, first, last);
4366 int ext4_can_truncate(struct inode *inode)
4368 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4370 if (S_ISREG(inode->i_mode))
4372 if (S_ISDIR(inode->i_mode))
4374 if (S_ISLNK(inode->i_mode))
4375 return !ext4_inode_is_fast_symlink(inode);
4382 * We block out ext4_get_block() block instantiations across the entire
4383 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4384 * simultaneously on behalf of the same inode.
4386 * As we work through the truncate and commmit bits of it to the journal there
4387 * is one core, guiding principle: the file's tree must always be consistent on
4388 * disk. We must be able to restart the truncate after a crash.
4390 * The file's tree may be transiently inconsistent in memory (although it
4391 * probably isn't), but whenever we close off and commit a journal transaction,
4392 * the contents of (the filesystem + the journal) must be consistent and
4393 * restartable. It's pretty simple, really: bottom up, right to left (although
4394 * left-to-right works OK too).
4396 * Note that at recovery time, journal replay occurs *before* the restart of
4397 * truncate against the orphan inode list.
4399 * The committed inode has the new, desired i_size (which is the same as
4400 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4401 * that this inode's truncate did not complete and it will again call
4402 * ext4_truncate() to have another go. So there will be instantiated blocks
4403 * to the right of the truncation point in a crashed ext4 filesystem. But
4404 * that's fine - as long as they are linked from the inode, the post-crash
4405 * ext4_truncate() run will find them and release them.
4407 void ext4_truncate(struct inode *inode)
4410 struct ext4_inode_info *ei = EXT4_I(inode);
4411 __le32 *i_data = ei->i_data;
4412 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4413 struct address_space *mapping = inode->i_mapping;
4414 ext4_lblk_t offsets[4];
4419 ext4_lblk_t last_block;
4420 unsigned blocksize = inode->i_sb->s_blocksize;
4422 if (!ext4_can_truncate(inode))
4425 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4427 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4428 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4430 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4431 ext4_ext_truncate(inode);
4435 handle = start_transaction(inode);
4437 return; /* AKPM: return what? */
4439 last_block = (inode->i_size + blocksize-1)
4440 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4442 if (inode->i_size & (blocksize - 1))
4443 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4446 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4448 goto out_stop; /* error */
4451 * OK. This truncate is going to happen. We add the inode to the
4452 * orphan list, so that if this truncate spans multiple transactions,
4453 * and we crash, we will resume the truncate when the filesystem
4454 * recovers. It also marks the inode dirty, to catch the new size.
4456 * Implication: the file must always be in a sane, consistent
4457 * truncatable state while each transaction commits.
4459 if (ext4_orphan_add(handle, inode))
4463 * From here we block out all ext4_get_block() callers who want to
4464 * modify the block allocation tree.
4466 down_write(&ei->i_data_sem);
4468 ext4_discard_preallocations(inode);
4471 * The orphan list entry will now protect us from any crash which
4472 * occurs before the truncate completes, so it is now safe to propagate
4473 * the new, shorter inode size (held for now in i_size) into the
4474 * on-disk inode. We do this via i_disksize, which is the value which
4475 * ext4 *really* writes onto the disk inode.
4477 ei->i_disksize = inode->i_size;
4479 if (n == 1) { /* direct blocks */
4480 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4481 i_data + EXT4_NDIR_BLOCKS);
4485 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4486 /* Kill the top of shared branch (not detached) */
4488 if (partial == chain) {
4489 /* Shared branch grows from the inode */
4490 ext4_free_branches(handle, inode, NULL,
4491 &nr, &nr+1, (chain+n-1) - partial);
4494 * We mark the inode dirty prior to restart,
4495 * and prior to stop. No need for it here.
4498 /* Shared branch grows from an indirect block */
4499 BUFFER_TRACE(partial->bh, "get_write_access");
4500 ext4_free_branches(handle, inode, partial->bh,
4502 partial->p+1, (chain+n-1) - partial);
4505 /* Clear the ends of indirect blocks on the shared branch */
4506 while (partial > chain) {
4507 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4508 (__le32*)partial->bh->b_data+addr_per_block,
4509 (chain+n-1) - partial);
4510 BUFFER_TRACE(partial->bh, "call brelse");
4511 brelse(partial->bh);
4515 /* Kill the remaining (whole) subtrees */
4516 switch (offsets[0]) {
4518 nr = i_data[EXT4_IND_BLOCK];
4520 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4521 i_data[EXT4_IND_BLOCK] = 0;
4523 case EXT4_IND_BLOCK:
4524 nr = i_data[EXT4_DIND_BLOCK];
4526 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4527 i_data[EXT4_DIND_BLOCK] = 0;
4529 case EXT4_DIND_BLOCK:
4530 nr = i_data[EXT4_TIND_BLOCK];
4532 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4533 i_data[EXT4_TIND_BLOCK] = 0;
4535 case EXT4_TIND_BLOCK:
4539 up_write(&ei->i_data_sem);
4540 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4541 ext4_mark_inode_dirty(handle, inode);
4544 * In a multi-transaction truncate, we only make the final transaction
4548 ext4_handle_sync(handle);
4551 * If this was a simple ftruncate(), and the file will remain alive
4552 * then we need to clear up the orphan record which we created above.
4553 * However, if this was a real unlink then we were called by
4554 * ext4_delete_inode(), and we allow that function to clean up the
4555 * orphan info for us.
4558 ext4_orphan_del(handle, inode);
4560 ext4_journal_stop(handle);
4564 * ext4_get_inode_loc returns with an extra refcount against the inode's
4565 * underlying buffer_head on success. If 'in_mem' is true, we have all
4566 * data in memory that is needed to recreate the on-disk version of this
4569 static int __ext4_get_inode_loc(struct inode *inode,
4570 struct ext4_iloc *iloc, int in_mem)
4572 struct ext4_group_desc *gdp;
4573 struct buffer_head *bh;
4574 struct super_block *sb = inode->i_sb;
4576 int inodes_per_block, inode_offset;
4579 if (!ext4_valid_inum(sb, inode->i_ino))
4582 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4583 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4588 * Figure out the offset within the block group inode table
4590 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4591 inode_offset = ((inode->i_ino - 1) %
4592 EXT4_INODES_PER_GROUP(sb));
4593 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4594 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4596 bh = sb_getblk(sb, block);
4598 EXT4_ERROR_INODE_BLOCK(inode, block,
4599 "unable to read itable block");
4602 if (!buffer_uptodate(bh)) {
4606 * If the buffer has the write error flag, we have failed
4607 * to write out another inode in the same block. In this
4608 * case, we don't have to read the block because we may
4609 * read the old inode data successfully.
4611 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4612 set_buffer_uptodate(bh);
4614 if (buffer_uptodate(bh)) {
4615 /* someone brought it uptodate while we waited */
4621 * If we have all information of the inode in memory and this
4622 * is the only valid inode in the block, we need not read the
4626 struct buffer_head *bitmap_bh;
4629 start = inode_offset & ~(inodes_per_block - 1);
4631 /* Is the inode bitmap in cache? */
4632 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4637 * If the inode bitmap isn't in cache then the
4638 * optimisation may end up performing two reads instead
4639 * of one, so skip it.
4641 if (!buffer_uptodate(bitmap_bh)) {
4645 for (i = start; i < start + inodes_per_block; i++) {
4646 if (i == inode_offset)
4648 if (ext4_test_bit(i, bitmap_bh->b_data))
4652 if (i == start + inodes_per_block) {
4653 /* all other inodes are free, so skip I/O */
4654 memset(bh->b_data, 0, bh->b_size);
4655 set_buffer_uptodate(bh);
4663 * If we need to do any I/O, try to pre-readahead extra
4664 * blocks from the inode table.
4666 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4667 ext4_fsblk_t b, end, table;
4670 table = ext4_inode_table(sb, gdp);
4671 /* s_inode_readahead_blks is always a power of 2 */
4672 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4675 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4676 num = EXT4_INODES_PER_GROUP(sb);
4677 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4678 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4679 num -= ext4_itable_unused_count(sb, gdp);
4680 table += num / inodes_per_block;
4684 sb_breadahead(sb, b++);
4688 * There are other valid inodes in the buffer, this inode
4689 * has in-inode xattrs, or we don't have this inode in memory.
4690 * Read the block from disk.
4693 bh->b_end_io = end_buffer_read_sync;
4694 submit_bh(READ_META, bh);
4696 if (!buffer_uptodate(bh)) {
4697 EXT4_ERROR_INODE_BLOCK(inode, block,
4698 "unable to read itable block");
4708 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4710 /* We have all inode data except xattrs in memory here. */
4711 return __ext4_get_inode_loc(inode, iloc,
4712 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4715 void ext4_set_inode_flags(struct inode *inode)
4717 unsigned int flags = EXT4_I(inode)->i_flags;
4719 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4720 if (flags & EXT4_SYNC_FL)
4721 inode->i_flags |= S_SYNC;
4722 if (flags & EXT4_APPEND_FL)
4723 inode->i_flags |= S_APPEND;
4724 if (flags & EXT4_IMMUTABLE_FL)
4725 inode->i_flags |= S_IMMUTABLE;
4726 if (flags & EXT4_NOATIME_FL)
4727 inode->i_flags |= S_NOATIME;
4728 if (flags & EXT4_DIRSYNC_FL)
4729 inode->i_flags |= S_DIRSYNC;
4732 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4733 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4735 unsigned int vfs_fl;
4736 unsigned long old_fl, new_fl;
4739 vfs_fl = ei->vfs_inode.i_flags;
4740 old_fl = ei->i_flags;
4741 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4742 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4744 if (vfs_fl & S_SYNC)
4745 new_fl |= EXT4_SYNC_FL;
4746 if (vfs_fl & S_APPEND)
4747 new_fl |= EXT4_APPEND_FL;
4748 if (vfs_fl & S_IMMUTABLE)
4749 new_fl |= EXT4_IMMUTABLE_FL;
4750 if (vfs_fl & S_NOATIME)
4751 new_fl |= EXT4_NOATIME_FL;
4752 if (vfs_fl & S_DIRSYNC)
4753 new_fl |= EXT4_DIRSYNC_FL;
4754 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4757 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4758 struct ext4_inode_info *ei)
4761 struct inode *inode = &(ei->vfs_inode);
4762 struct super_block *sb = inode->i_sb;
4764 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4765 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4766 /* we are using combined 48 bit field */
4767 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4768 le32_to_cpu(raw_inode->i_blocks_lo);
4769 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4770 /* i_blocks represent file system block size */
4771 return i_blocks << (inode->i_blkbits - 9);
4776 return le32_to_cpu(raw_inode->i_blocks_lo);
4780 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4782 struct ext4_iloc iloc;
4783 struct ext4_inode *raw_inode;
4784 struct ext4_inode_info *ei;
4785 struct inode *inode;
4786 journal_t *journal = EXT4_SB(sb)->s_journal;
4790 inode = iget_locked(sb, ino);
4792 return ERR_PTR(-ENOMEM);
4793 if (!(inode->i_state & I_NEW))
4799 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4802 raw_inode = ext4_raw_inode(&iloc);
4803 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4804 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4805 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4806 if (!(test_opt(inode->i_sb, NO_UID32))) {
4807 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4808 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4810 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4812 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4813 ei->i_dir_start_lookup = 0;
4814 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4815 /* We now have enough fields to check if the inode was active or not.
4816 * This is needed because nfsd might try to access dead inodes
4817 * the test is that same one that e2fsck uses
4818 * NeilBrown 1999oct15
4820 if (inode->i_nlink == 0) {
4821 if (inode->i_mode == 0 ||
4822 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4823 /* this inode is deleted */
4827 /* The only unlinked inodes we let through here have
4828 * valid i_mode and are being read by the orphan
4829 * recovery code: that's fine, we're about to complete
4830 * the process of deleting those. */
4832 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4833 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4834 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4835 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4837 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4838 inode->i_size = ext4_isize(raw_inode);
4839 ei->i_disksize = inode->i_size;
4841 ei->i_reserved_quota = 0;
4843 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4844 ei->i_block_group = iloc.block_group;
4845 ei->i_last_alloc_group = ~0;
4847 * NOTE! The in-memory inode i_data array is in little-endian order
4848 * even on big-endian machines: we do NOT byteswap the block numbers!
4850 for (block = 0; block < EXT4_N_BLOCKS; block++)
4851 ei->i_data[block] = raw_inode->i_block[block];
4852 INIT_LIST_HEAD(&ei->i_orphan);
4855 * Set transaction id's of transactions that have to be committed
4856 * to finish f[data]sync. We set them to currently running transaction
4857 * as we cannot be sure that the inode or some of its metadata isn't
4858 * part of the transaction - the inode could have been reclaimed and
4859 * now it is reread from disk.
4862 transaction_t *transaction;
4865 read_lock(&journal->j_state_lock);
4866 if (journal->j_running_transaction)
4867 transaction = journal->j_running_transaction;
4869 transaction = journal->j_committing_transaction;
4871 tid = transaction->t_tid;
4873 tid = journal->j_commit_sequence;
4874 read_unlock(&journal->j_state_lock);
4875 ei->i_sync_tid = tid;
4876 ei->i_datasync_tid = tid;
4879 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4880 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4881 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4882 EXT4_INODE_SIZE(inode->i_sb)) {
4886 if (ei->i_extra_isize == 0) {
4887 /* The extra space is currently unused. Use it. */
4888 ei->i_extra_isize = sizeof(struct ext4_inode) -
4889 EXT4_GOOD_OLD_INODE_SIZE;
4891 __le32 *magic = (void *)raw_inode +
4892 EXT4_GOOD_OLD_INODE_SIZE +
4894 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4895 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4898 ei->i_extra_isize = 0;
4900 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4901 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4902 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4903 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4905 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4906 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4907 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4909 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4913 if (ei->i_file_acl &&
4914 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4915 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4919 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4920 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4921 (S_ISLNK(inode->i_mode) &&
4922 !ext4_inode_is_fast_symlink(inode)))
4923 /* Validate extent which is part of inode */
4924 ret = ext4_ext_check_inode(inode);
4925 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4926 (S_ISLNK(inode->i_mode) &&
4927 !ext4_inode_is_fast_symlink(inode))) {
4928 /* Validate block references which are part of inode */
4929 ret = ext4_check_inode_blockref(inode);
4934 if (S_ISREG(inode->i_mode)) {
4935 inode->i_op = &ext4_file_inode_operations;
4936 inode->i_fop = &ext4_file_operations;
4937 ext4_set_aops(inode);
4938 } else if (S_ISDIR(inode->i_mode)) {
4939 inode->i_op = &ext4_dir_inode_operations;
4940 inode->i_fop = &ext4_dir_operations;
4941 } else if (S_ISLNK(inode->i_mode)) {
4942 if (ext4_inode_is_fast_symlink(inode)) {
4943 inode->i_op = &ext4_fast_symlink_inode_operations;
4944 nd_terminate_link(ei->i_data, inode->i_size,
4945 sizeof(ei->i_data) - 1);
4947 inode->i_op = &ext4_symlink_inode_operations;
4948 ext4_set_aops(inode);
4950 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4951 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4952 inode->i_op = &ext4_special_inode_operations;
4953 if (raw_inode->i_block[0])
4954 init_special_inode(inode, inode->i_mode,
4955 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4957 init_special_inode(inode, inode->i_mode,
4958 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4961 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
4965 ext4_set_inode_flags(inode);
4966 unlock_new_inode(inode);
4972 return ERR_PTR(ret);
4975 static int ext4_inode_blocks_set(handle_t *handle,
4976 struct ext4_inode *raw_inode,
4977 struct ext4_inode_info *ei)
4979 struct inode *inode = &(ei->vfs_inode);
4980 u64 i_blocks = inode->i_blocks;
4981 struct super_block *sb = inode->i_sb;
4983 if (i_blocks <= ~0U) {
4985 * i_blocks can be represnted in a 32 bit variable
4986 * as multiple of 512 bytes
4988 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4989 raw_inode->i_blocks_high = 0;
4990 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4993 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4996 if (i_blocks <= 0xffffffffffffULL) {
4998 * i_blocks can be represented in a 48 bit variable
4999 * as multiple of 512 bytes
5001 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5002 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5003 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5005 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5006 /* i_block is stored in file system block size */
5007 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5008 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5009 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5015 * Post the struct inode info into an on-disk inode location in the
5016 * buffer-cache. This gobbles the caller's reference to the
5017 * buffer_head in the inode location struct.
5019 * The caller must have write access to iloc->bh.
5021 static int ext4_do_update_inode(handle_t *handle,
5022 struct inode *inode,
5023 struct ext4_iloc *iloc)
5025 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5026 struct ext4_inode_info *ei = EXT4_I(inode);
5027 struct buffer_head *bh = iloc->bh;
5028 int err = 0, rc, block;
5030 /* For fields not not tracking in the in-memory inode,
5031 * initialise them to zero for new inodes. */
5032 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5033 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5035 ext4_get_inode_flags(ei);
5036 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5037 if (!(test_opt(inode->i_sb, NO_UID32))) {
5038 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5039 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5041 * Fix up interoperability with old kernels. Otherwise, old inodes get
5042 * re-used with the upper 16 bits of the uid/gid intact
5045 raw_inode->i_uid_high =
5046 cpu_to_le16(high_16_bits(inode->i_uid));
5047 raw_inode->i_gid_high =
5048 cpu_to_le16(high_16_bits(inode->i_gid));
5050 raw_inode->i_uid_high = 0;
5051 raw_inode->i_gid_high = 0;
5054 raw_inode->i_uid_low =
5055 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5056 raw_inode->i_gid_low =
5057 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5058 raw_inode->i_uid_high = 0;
5059 raw_inode->i_gid_high = 0;
5061 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5063 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5064 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5065 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5066 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5068 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5070 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5071 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5072 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5073 cpu_to_le32(EXT4_OS_HURD))
5074 raw_inode->i_file_acl_high =
5075 cpu_to_le16(ei->i_file_acl >> 32);
5076 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5077 ext4_isize_set(raw_inode, ei->i_disksize);
5078 if (ei->i_disksize > 0x7fffffffULL) {
5079 struct super_block *sb = inode->i_sb;
5080 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5081 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5082 EXT4_SB(sb)->s_es->s_rev_level ==
5083 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5084 /* If this is the first large file
5085 * created, add a flag to the superblock.
5087 err = ext4_journal_get_write_access(handle,
5088 EXT4_SB(sb)->s_sbh);
5091 ext4_update_dynamic_rev(sb);
5092 EXT4_SET_RO_COMPAT_FEATURE(sb,
5093 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5095 ext4_handle_sync(handle);
5096 err = ext4_handle_dirty_metadata(handle, NULL,
5097 EXT4_SB(sb)->s_sbh);
5100 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5101 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5102 if (old_valid_dev(inode->i_rdev)) {
5103 raw_inode->i_block[0] =
5104 cpu_to_le32(old_encode_dev(inode->i_rdev));
5105 raw_inode->i_block[1] = 0;
5107 raw_inode->i_block[0] = 0;
5108 raw_inode->i_block[1] =
5109 cpu_to_le32(new_encode_dev(inode->i_rdev));
5110 raw_inode->i_block[2] = 0;
5113 for (block = 0; block < EXT4_N_BLOCKS; block++)
5114 raw_inode->i_block[block] = ei->i_data[block];
5116 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5117 if (ei->i_extra_isize) {
5118 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5119 raw_inode->i_version_hi =
5120 cpu_to_le32(inode->i_version >> 32);
5121 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5124 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5125 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5128 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5130 ext4_update_inode_fsync_trans(handle, inode, 0);
5133 ext4_std_error(inode->i_sb, err);
5138 * ext4_write_inode()
5140 * We are called from a few places:
5142 * - Within generic_file_write() for O_SYNC files.
5143 * Here, there will be no transaction running. We wait for any running
5144 * trasnaction to commit.
5146 * - Within sys_sync(), kupdate and such.
5147 * We wait on commit, if tol to.
5149 * - Within prune_icache() (PF_MEMALLOC == true)
5150 * Here we simply return. We can't afford to block kswapd on the
5153 * In all cases it is actually safe for us to return without doing anything,
5154 * because the inode has been copied into a raw inode buffer in
5155 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5158 * Note that we are absolutely dependent upon all inode dirtiers doing the
5159 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5160 * which we are interested.
5162 * It would be a bug for them to not do this. The code:
5164 * mark_inode_dirty(inode)
5166 * inode->i_size = expr;
5168 * is in error because a kswapd-driven write_inode() could occur while
5169 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5170 * will no longer be on the superblock's dirty inode list.
5172 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5176 if (current->flags & PF_MEMALLOC)
5179 if (EXT4_SB(inode->i_sb)->s_journal) {
5180 if (ext4_journal_current_handle()) {
5181 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5186 if (wbc->sync_mode != WB_SYNC_ALL)
5189 err = ext4_force_commit(inode->i_sb);
5191 struct ext4_iloc iloc;
5193 err = __ext4_get_inode_loc(inode, &iloc, 0);
5196 if (wbc->sync_mode == WB_SYNC_ALL)
5197 sync_dirty_buffer(iloc.bh);
5198 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5199 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5200 "IO error syncing inode");
5211 * Called from notify_change.
5213 * We want to trap VFS attempts to truncate the file as soon as
5214 * possible. In particular, we want to make sure that when the VFS
5215 * shrinks i_size, we put the inode on the orphan list and modify
5216 * i_disksize immediately, so that during the subsequent flushing of
5217 * dirty pages and freeing of disk blocks, we can guarantee that any
5218 * commit will leave the blocks being flushed in an unused state on
5219 * disk. (On recovery, the inode will get truncated and the blocks will
5220 * be freed, so we have a strong guarantee that no future commit will
5221 * leave these blocks visible to the user.)
5223 * Another thing we have to assure is that if we are in ordered mode
5224 * and inode is still attached to the committing transaction, we must
5225 * we start writeout of all the dirty pages which are being truncated.
5226 * This way we are sure that all the data written in the previous
5227 * transaction are already on disk (truncate waits for pages under
5230 * Called with inode->i_mutex down.
5232 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5234 struct inode *inode = dentry->d_inode;
5237 const unsigned int ia_valid = attr->ia_valid;
5239 error = inode_change_ok(inode, attr);
5243 if (is_quota_modification(inode, attr))
5244 dquot_initialize(inode);
5245 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5246 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5249 /* (user+group)*(old+new) structure, inode write (sb,
5250 * inode block, ? - but truncate inode update has it) */
5251 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5252 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5253 if (IS_ERR(handle)) {
5254 error = PTR_ERR(handle);
5257 error = dquot_transfer(inode, attr);
5259 ext4_journal_stop(handle);
5262 /* Update corresponding info in inode so that everything is in
5263 * one transaction */
5264 if (attr->ia_valid & ATTR_UID)
5265 inode->i_uid = attr->ia_uid;
5266 if (attr->ia_valid & ATTR_GID)
5267 inode->i_gid = attr->ia_gid;
5268 error = ext4_mark_inode_dirty(handle, inode);
5269 ext4_journal_stop(handle);
5272 if (attr->ia_valid & ATTR_SIZE) {
5273 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5274 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5276 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5281 if (S_ISREG(inode->i_mode) &&
5282 attr->ia_valid & ATTR_SIZE &&
5283 (attr->ia_size < inode->i_size ||
5284 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5287 handle = ext4_journal_start(inode, 3);
5288 if (IS_ERR(handle)) {
5289 error = PTR_ERR(handle);
5292 if (ext4_handle_valid(handle)) {
5293 error = ext4_orphan_add(handle, inode);
5296 EXT4_I(inode)->i_disksize = attr->ia_size;
5297 rc = ext4_mark_inode_dirty(handle, inode);
5300 ext4_journal_stop(handle);
5302 if (ext4_should_order_data(inode)) {
5303 error = ext4_begin_ordered_truncate(inode,
5306 /* Do as much error cleanup as possible */
5307 handle = ext4_journal_start(inode, 3);
5308 if (IS_ERR(handle)) {
5309 ext4_orphan_del(NULL, inode);
5312 ext4_orphan_del(handle, inode);
5314 ext4_journal_stop(handle);
5318 /* ext4_truncate will clear the flag */
5319 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5320 ext4_truncate(inode);
5323 if ((attr->ia_valid & ATTR_SIZE) &&
5324 attr->ia_size != i_size_read(inode))
5325 rc = vmtruncate(inode, attr->ia_size);
5328 setattr_copy(inode, attr);
5329 mark_inode_dirty(inode);
5333 * If the call to ext4_truncate failed to get a transaction handle at
5334 * all, we need to clean up the in-core orphan list manually.
5336 if (orphan && inode->i_nlink)
5337 ext4_orphan_del(NULL, inode);
5339 if (!rc && (ia_valid & ATTR_MODE))
5340 rc = ext4_acl_chmod(inode);
5343 ext4_std_error(inode->i_sb, error);
5349 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5352 struct inode *inode;
5353 unsigned long delalloc_blocks;
5355 inode = dentry->d_inode;
5356 generic_fillattr(inode, stat);
5359 * We can't update i_blocks if the block allocation is delayed
5360 * otherwise in the case of system crash before the real block
5361 * allocation is done, we will have i_blocks inconsistent with
5362 * on-disk file blocks.
5363 * We always keep i_blocks updated together with real
5364 * allocation. But to not confuse with user, stat
5365 * will return the blocks that include the delayed allocation
5366 * blocks for this file.
5368 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5370 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5374 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5379 /* if nrblocks are contiguous */
5382 * With N contiguous data blocks, it need at most
5383 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5384 * 2 dindirect blocks
5387 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5388 return indirects + 3;
5391 * if nrblocks are not contiguous, worse case, each block touch
5392 * a indirect block, and each indirect block touch a double indirect
5393 * block, plus a triple indirect block
5395 indirects = nrblocks * 2 + 1;
5399 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5401 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5402 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5403 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5407 * Account for index blocks, block groups bitmaps and block group
5408 * descriptor blocks if modify datablocks and index blocks
5409 * worse case, the indexs blocks spread over different block groups
5411 * If datablocks are discontiguous, they are possible to spread over
5412 * different block groups too. If they are contiuguous, with flexbg,
5413 * they could still across block group boundary.
5415 * Also account for superblock, inode, quota and xattr blocks
5417 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5419 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5425 * How many index blocks need to touch to modify nrblocks?
5426 * The "Chunk" flag indicating whether the nrblocks is
5427 * physically contiguous on disk
5429 * For Direct IO and fallocate, they calls get_block to allocate
5430 * one single extent at a time, so they could set the "Chunk" flag
5432 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5437 * Now let's see how many group bitmaps and group descriptors need
5447 if (groups > ngroups)
5449 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5450 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5452 /* bitmaps and block group descriptor blocks */
5453 ret += groups + gdpblocks;
5455 /* Blocks for super block, inode, quota and xattr blocks */
5456 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5462 * Calulate the total number of credits to reserve to fit
5463 * the modification of a single pages into a single transaction,
5464 * which may include multiple chunks of block allocations.
5466 * This could be called via ext4_write_begin()
5468 * We need to consider the worse case, when
5469 * one new block per extent.
5471 int ext4_writepage_trans_blocks(struct inode *inode)
5473 int bpp = ext4_journal_blocks_per_page(inode);
5476 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5478 /* Account for data blocks for journalled mode */
5479 if (ext4_should_journal_data(inode))
5485 * Calculate the journal credits for a chunk of data modification.
5487 * This is called from DIO, fallocate or whoever calling
5488 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5490 * journal buffers for data blocks are not included here, as DIO
5491 * and fallocate do no need to journal data buffers.
5493 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5495 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5499 * The caller must have previously called ext4_reserve_inode_write().
5500 * Give this, we know that the caller already has write access to iloc->bh.
5502 int ext4_mark_iloc_dirty(handle_t *handle,
5503 struct inode *inode, struct ext4_iloc *iloc)
5507 if (test_opt(inode->i_sb, I_VERSION))
5508 inode_inc_iversion(inode);
5510 /* the do_update_inode consumes one bh->b_count */
5513 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5514 err = ext4_do_update_inode(handle, inode, iloc);
5520 * On success, We end up with an outstanding reference count against
5521 * iloc->bh. This _must_ be cleaned up later.
5525 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5526 struct ext4_iloc *iloc)
5530 err = ext4_get_inode_loc(inode, iloc);
5532 BUFFER_TRACE(iloc->bh, "get_write_access");
5533 err = ext4_journal_get_write_access(handle, iloc->bh);
5539 ext4_std_error(inode->i_sb, err);
5544 * Expand an inode by new_extra_isize bytes.
5545 * Returns 0 on success or negative error number on failure.
5547 static int ext4_expand_extra_isize(struct inode *inode,
5548 unsigned int new_extra_isize,
5549 struct ext4_iloc iloc,
5552 struct ext4_inode *raw_inode;
5553 struct ext4_xattr_ibody_header *header;
5555 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5558 raw_inode = ext4_raw_inode(&iloc);
5560 header = IHDR(inode, raw_inode);
5562 /* No extended attributes present */
5563 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5564 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5565 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5567 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5571 /* try to expand with EAs present */
5572 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5577 * What we do here is to mark the in-core inode as clean with respect to inode
5578 * dirtiness (it may still be data-dirty).
5579 * This means that the in-core inode may be reaped by prune_icache
5580 * without having to perform any I/O. This is a very good thing,
5581 * because *any* task may call prune_icache - even ones which
5582 * have a transaction open against a different journal.
5584 * Is this cheating? Not really. Sure, we haven't written the
5585 * inode out, but prune_icache isn't a user-visible syncing function.
5586 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5587 * we start and wait on commits.
5589 * Is this efficient/effective? Well, we're being nice to the system
5590 * by cleaning up our inodes proactively so they can be reaped
5591 * without I/O. But we are potentially leaving up to five seconds'
5592 * worth of inodes floating about which prune_icache wants us to
5593 * write out. One way to fix that would be to get prune_icache()
5594 * to do a write_super() to free up some memory. It has the desired
5597 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5599 struct ext4_iloc iloc;
5600 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5601 static unsigned int mnt_count;
5605 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5606 err = ext4_reserve_inode_write(handle, inode, &iloc);
5607 if (ext4_handle_valid(handle) &&
5608 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5609 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5611 * We need extra buffer credits since we may write into EA block
5612 * with this same handle. If journal_extend fails, then it will
5613 * only result in a minor loss of functionality for that inode.
5614 * If this is felt to be critical, then e2fsck should be run to
5615 * force a large enough s_min_extra_isize.
5617 if ((jbd2_journal_extend(handle,
5618 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5619 ret = ext4_expand_extra_isize(inode,
5620 sbi->s_want_extra_isize,
5623 ext4_set_inode_state(inode,
5624 EXT4_STATE_NO_EXPAND);
5626 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5627 ext4_warning(inode->i_sb,
5628 "Unable to expand inode %lu. Delete"
5629 " some EAs or run e2fsck.",
5632 le16_to_cpu(sbi->s_es->s_mnt_count);
5638 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5643 * ext4_dirty_inode() is called from __mark_inode_dirty()
5645 * We're really interested in the case where a file is being extended.
5646 * i_size has been changed by generic_commit_write() and we thus need
5647 * to include the updated inode in the current transaction.
5649 * Also, dquot_alloc_block() will always dirty the inode when blocks
5650 * are allocated to the file.
5652 * If the inode is marked synchronous, we don't honour that here - doing
5653 * so would cause a commit on atime updates, which we don't bother doing.
5654 * We handle synchronous inodes at the highest possible level.
5656 void ext4_dirty_inode(struct inode *inode)
5660 handle = ext4_journal_start(inode, 2);
5664 ext4_mark_inode_dirty(handle, inode);
5666 ext4_journal_stop(handle);
5673 * Bind an inode's backing buffer_head into this transaction, to prevent
5674 * it from being flushed to disk early. Unlike
5675 * ext4_reserve_inode_write, this leaves behind no bh reference and
5676 * returns no iloc structure, so the caller needs to repeat the iloc
5677 * lookup to mark the inode dirty later.
5679 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5681 struct ext4_iloc iloc;
5685 err = ext4_get_inode_loc(inode, &iloc);
5687 BUFFER_TRACE(iloc.bh, "get_write_access");
5688 err = jbd2_journal_get_write_access(handle, iloc.bh);
5690 err = ext4_handle_dirty_metadata(handle,
5696 ext4_std_error(inode->i_sb, err);
5701 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5708 * We have to be very careful here: changing a data block's
5709 * journaling status dynamically is dangerous. If we write a
5710 * data block to the journal, change the status and then delete
5711 * that block, we risk forgetting to revoke the old log record
5712 * from the journal and so a subsequent replay can corrupt data.
5713 * So, first we make sure that the journal is empty and that
5714 * nobody is changing anything.
5717 journal = EXT4_JOURNAL(inode);
5720 if (is_journal_aborted(journal))
5723 jbd2_journal_lock_updates(journal);
5724 jbd2_journal_flush(journal);
5727 * OK, there are no updates running now, and all cached data is
5728 * synced to disk. We are now in a completely consistent state
5729 * which doesn't have anything in the journal, and we know that
5730 * no filesystem updates are running, so it is safe to modify
5731 * the inode's in-core data-journaling state flag now.
5735 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5737 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5738 ext4_set_aops(inode);
5740 jbd2_journal_unlock_updates(journal);
5742 /* Finally we can mark the inode as dirty. */
5744 handle = ext4_journal_start(inode, 1);
5746 return PTR_ERR(handle);
5748 err = ext4_mark_inode_dirty(handle, inode);
5749 ext4_handle_sync(handle);
5750 ext4_journal_stop(handle);
5751 ext4_std_error(inode->i_sb, err);
5756 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5758 return !buffer_mapped(bh);
5761 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5763 struct page *page = vmf->page;
5768 struct file *file = vma->vm_file;
5769 struct inode *inode = file->f_path.dentry->d_inode;
5770 struct address_space *mapping = inode->i_mapping;
5773 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5774 * get i_mutex because we are already holding mmap_sem.
5776 down_read(&inode->i_alloc_sem);
5777 size = i_size_read(inode);
5778 if (page->mapping != mapping || size <= page_offset(page)
5779 || !PageUptodate(page)) {
5780 /* page got truncated from under us? */
5784 if (PageMappedToDisk(page))
5787 if (page->index == size >> PAGE_CACHE_SHIFT)
5788 len = size & ~PAGE_CACHE_MASK;
5790 len = PAGE_CACHE_SIZE;
5794 * return if we have all the buffers mapped. This avoid
5795 * the need to call write_begin/write_end which does a
5796 * journal_start/journal_stop which can block and take
5799 if (page_has_buffers(page)) {
5800 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5801 ext4_bh_unmapped)) {
5808 * OK, we need to fill the hole... Do write_begin write_end
5809 * to do block allocation/reservation.We are not holding
5810 * inode.i__mutex here. That allow * parallel write_begin,
5811 * write_end call. lock_page prevent this from happening
5812 * on the same page though
5814 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5815 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5818 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5819 len, len, page, fsdata);
5825 ret = VM_FAULT_SIGBUS;
5826 up_read(&inode->i_alloc_sem);