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/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode *inode,
49 return jbd2_journal_begin_ordered_truncate(
50 EXT4_SB(inode->i_sb)->s_journal,
51 &EXT4_I(inode)->jinode,
55 static void ext4_invalidatepage(struct page *page, unsigned long offset);
58 * Test whether an inode is a fast symlink.
60 static int ext4_inode_is_fast_symlink(struct inode *inode)
62 int ea_blocks = EXT4_I(inode)->i_file_acl ?
63 (inode->i_sb->s_blocksize >> 9) : 0;
65 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
69 * The ext4 forget function must perform a revoke if we are freeing data
70 * which has been journaled. Metadata (eg. indirect blocks) must be
71 * revoked in all cases.
73 * "bh" may be NULL: a metadata block may have been freed from memory
74 * but there may still be a record of it in the journal, and that record
75 * still needs to be revoked.
77 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
78 struct buffer_head *bh, ext4_fsblk_t blocknr)
84 BUFFER_TRACE(bh, "enter");
86 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
88 bh, is_metadata, inode->i_mode,
89 test_opt(inode->i_sb, DATA_FLAGS));
91 /* Never use the revoke function if we are doing full data
92 * journaling: there is no need to, and a V1 superblock won't
93 * support it. Otherwise, only skip the revoke on un-journaled
96 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
97 (!is_metadata && !ext4_should_journal_data(inode))) {
99 BUFFER_TRACE(bh, "call jbd2_journal_forget");
100 return ext4_journal_forget(handle, bh);
106 * data!=journal && (is_metadata || should_journal_data(inode))
108 BUFFER_TRACE(bh, "call ext4_journal_revoke");
109 err = ext4_journal_revoke(handle, blocknr, bh);
111 ext4_abort(inode->i_sb, __func__,
112 "error %d when attempting revoke", err);
113 BUFFER_TRACE(bh, "exit");
118 * Work out how many blocks we need to proceed with the next chunk of a
119 * truncate transaction.
121 static unsigned long blocks_for_truncate(struct inode *inode)
125 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
127 /* Give ourselves just enough room to cope with inodes in which
128 * i_blocks is corrupt: we've seen disk corruptions in the past
129 * which resulted in random data in an inode which looked enough
130 * like a regular file for ext4 to try to delete it. Things
131 * will go a bit crazy if that happens, but at least we should
132 * try not to panic the whole kernel. */
136 /* But we need to bound the transaction so we don't overflow the
138 if (needed > EXT4_MAX_TRANS_DATA)
139 needed = EXT4_MAX_TRANS_DATA;
141 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
145 * Truncate transactions can be complex and absolutely huge. So we need to
146 * be able to restart the transaction at a conventient checkpoint to make
147 * sure we don't overflow the journal.
149 * start_transaction gets us a new handle for a truncate transaction,
150 * and extend_transaction tries to extend the existing one a bit. If
151 * extend fails, we need to propagate the failure up and restart the
152 * transaction in the top-level truncate loop. --sct
154 static handle_t *start_transaction(struct inode *inode)
158 result = ext4_journal_start(inode, blocks_for_truncate(inode));
162 ext4_std_error(inode->i_sb, PTR_ERR(result));
167 * Try to extend this transaction for the purposes of truncation.
169 * Returns 0 if we managed to create more room. If we can't create more
170 * room, and the transaction must be restarted we return 1.
172 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
174 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
176 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
182 * Restart the transaction associated with *handle. This does a commit,
183 * so before we call here everything must be consistently dirtied against
186 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
188 jbd_debug(2, "restarting handle %p\n", handle);
189 return ext4_journal_restart(handle, blocks_for_truncate(inode));
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_delete_inode (struct inode * 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);
222 err = ext4_mark_inode_dirty(handle, inode);
224 ext4_warning(inode->i_sb, __func__,
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 (handle->h_buffer_credits < 3) {
238 err = ext4_journal_extend(handle, 3);
240 err = ext4_journal_restart(handle, 3);
242 ext4_warning(inode->i_sb, __func__,
243 "couldn't extend journal (err %d)", err);
245 ext4_journal_stop(handle);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle, inode);
259 EXT4_I(inode)->i_dtime = get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle, inode))
269 /* If that failed, just do the required in-core inode clear. */
272 ext4_free_inode(handle, inode);
273 ext4_journal_stop(handle);
276 clear_inode(inode); /* We must guarantee clearing of inode... */
282 struct buffer_head *bh;
285 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287 p->key = *(p->p = v);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t offsets[4], int *boundary)
326 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
327 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
328 const long direct_blocks = EXT4_NDIR_BLOCKS,
329 indirect_blocks = ptrs,
330 double_blocks = (1 << (ptrs_bits * 2));
335 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
336 } else if (i_block < direct_blocks) {
337 offsets[n++] = i_block;
338 final = direct_blocks;
339 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
340 offsets[n++] = EXT4_IND_BLOCK;
341 offsets[n++] = i_block;
343 } else if ((i_block -= indirect_blocks) < double_blocks) {
344 offsets[n++] = EXT4_DIND_BLOCK;
345 offsets[n++] = i_block >> ptrs_bits;
346 offsets[n++] = i_block & (ptrs - 1);
348 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
349 offsets[n++] = EXT4_TIND_BLOCK;
350 offsets[n++] = i_block >> (ptrs_bits * 2);
351 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
352 offsets[n++] = i_block & (ptrs - 1);
355 ext4_warning(inode->i_sb, "ext4_block_to_path",
356 "block %lu > max in inode %lu",
357 i_block + direct_blocks +
358 indirect_blocks + double_blocks, inode->i_ino);
361 *boundary = final - 1 - (i_block & (ptrs - 1));
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect *ext4_get_branch(struct inode *inode, int depth,
396 ext4_lblk_t *offsets,
397 Indirect chain[4], int *err)
399 struct super_block *sb = inode->i_sb;
401 struct buffer_head *bh;
404 /* i_data is not going away, no lock needed */
405 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
409 bh = sb_bread(sb, le32_to_cpu(p->key));
412 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
426 * ext4_find_near - find a place for allocation with sufficient locality
428 * @ind: descriptor of indirect block.
430 * This function returns the preferred place for block allocation.
431 * It is used when heuristic for sequential allocation fails.
433 * + if there is a block to the left of our position - allocate near it.
434 * + if pointer will live in indirect block - allocate near that block.
435 * + if pointer will live in inode - allocate in the same
438 * In the latter case we colour the starting block by the callers PID to
439 * prevent it from clashing with concurrent allocations for a different inode
440 * in the same block group. The PID is used here so that functionally related
441 * files will be close-by on-disk.
443 * Caller must make sure that @ind is valid and will stay that way.
445 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
447 struct ext4_inode_info *ei = EXT4_I(inode);
448 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
450 ext4_fsblk_t bg_start;
451 ext4_fsblk_t last_block;
452 ext4_grpblk_t colour;
454 /* Try to find previous block */
455 for (p = ind->p - 1; p >= start; p--) {
457 return le32_to_cpu(*p);
460 /* No such thing, so let's try location of indirect block */
462 return ind->bh->b_blocknr;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
469 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
471 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
472 colour = (current->pid % 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
475 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
476 return bg_start + colour;
480 * ext4_find_goal - find a preferred place for allocation.
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
485 * Normally this function find the preferred place for block allocation,
488 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
491 struct ext4_block_alloc_info *block_i;
493 block_i = EXT4_I(inode)->i_block_alloc_info;
496 * try the heuristic for sequential allocation,
497 * failing that at least try to get decent locality.
499 if (block_i && (block == block_i->last_alloc_logical_block + 1)
500 && (block_i->last_alloc_physical_block != 0)) {
501 return block_i->last_alloc_physical_block + 1;
504 return ext4_find_near(inode, partial);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
520 int blocks_to_boundary)
522 unsigned long count = 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks < blocks_to_boundary + 1)
533 count += blocks_to_boundary + 1;
538 while (count < blks && count <= blocks_to_boundary &&
539 le32_to_cpu(*(branch[0].p + count)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
556 ext4_lblk_t iblock, ext4_fsblk_t goal,
557 int indirect_blks, int blks,
558 ext4_fsblk_t new_blocks[4], int *err)
561 unsigned long count = 0, blk_allocated = 0;
563 ext4_fsblk_t current_block = 0;
567 * Here we try to allocate the requested multiple blocks at once,
568 * on a best-effort basis.
569 * To build a branch, we should allocate blocks for
570 * the indirect blocks(if not allocated yet), and at least
571 * the first direct block of this branch. That's the
572 * minimum number of blocks need to allocate(required)
574 /* first we try to allocate the indirect blocks */
575 target = indirect_blks;
578 /* allocating blocks for indirect blocks and direct blocks */
579 current_block = ext4_new_meta_blocks(handle, inode,
585 /* allocate blocks for indirect blocks */
586 while (index < indirect_blks && count) {
587 new_blocks[index++] = current_block++;
592 * save the new block number
593 * for the first direct block
595 new_blocks[index] = current_block;
596 printk(KERN_INFO "%s returned more blocks than "
597 "requested\n", __func__);
603 target = blks - count ;
604 blk_allocated = count;
607 /* Now allocate data blocks */
609 /* allocating blocks for data blocks */
610 current_block = ext4_new_blocks(handle, inode, iblock,
612 if (*err && (target == blks)) {
614 * if the allocation failed and we didn't allocate
620 if (target == blks) {
622 * save the new block number
623 * for the first direct block
625 new_blocks[index] = current_block;
627 blk_allocated += count;
630 /* total number of blocks allocated for direct blocks */
635 for (i = 0; i <index; i++)
636 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
641 * ext4_alloc_branch - allocate and set up a chain of blocks.
643 * @indirect_blks: number of allocated indirect blocks
644 * @blks: number of allocated direct blocks
645 * @offsets: offsets (in the blocks) to store the pointers to next.
646 * @branch: place to store the chain in.
648 * This function allocates blocks, zeroes out all but the last one,
649 * links them into chain and (if we are synchronous) writes them to disk.
650 * In other words, it prepares a branch that can be spliced onto the
651 * inode. It stores the information about that chain in the branch[], in
652 * the same format as ext4_get_branch() would do. We are calling it after
653 * we had read the existing part of chain and partial points to the last
654 * triple of that (one with zero ->key). Upon the exit we have the same
655 * picture as after the successful ext4_get_block(), except that in one
656 * place chain is disconnected - *branch->p is still zero (we did not
657 * set the last link), but branch->key contains the number that should
658 * be placed into *branch->p to fill that gap.
660 * If allocation fails we free all blocks we've allocated (and forget
661 * their buffer_heads) and return the error value the from failed
662 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
663 * as described above and return 0.
665 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
666 ext4_lblk_t iblock, int indirect_blks,
667 int *blks, ext4_fsblk_t goal,
668 ext4_lblk_t *offsets, Indirect *branch)
670 int blocksize = inode->i_sb->s_blocksize;
673 struct buffer_head *bh;
675 ext4_fsblk_t new_blocks[4];
676 ext4_fsblk_t current_block;
678 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
679 *blks, new_blocks, &err);
683 branch[0].key = cpu_to_le32(new_blocks[0]);
685 * metadata blocks and data blocks are allocated.
687 for (n = 1; n <= indirect_blks; n++) {
689 * Get buffer_head for parent block, zero it out
690 * and set the pointer to new one, then send
693 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
696 BUFFER_TRACE(bh, "call get_create_access");
697 err = ext4_journal_get_create_access(handle, bh);
704 memset(bh->b_data, 0, blocksize);
705 branch[n].p = (__le32 *) bh->b_data + offsets[n];
706 branch[n].key = cpu_to_le32(new_blocks[n]);
707 *branch[n].p = branch[n].key;
708 if ( n == indirect_blks) {
709 current_block = new_blocks[n];
711 * End of chain, update the last new metablock of
712 * the chain to point to the new allocated
713 * data blocks numbers
715 for (i=1; i < num; i++)
716 *(branch[n].p + i) = cpu_to_le32(++current_block);
718 BUFFER_TRACE(bh, "marking uptodate");
719 set_buffer_uptodate(bh);
722 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
723 err = ext4_journal_dirty_metadata(handle, bh);
730 /* Allocation failed, free what we already allocated */
731 for (i = 1; i <= n ; i++) {
732 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
733 ext4_journal_forget(handle, branch[i].bh);
735 for (i = 0; i <indirect_blks; i++)
736 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
738 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
744 * ext4_splice_branch - splice the allocated branch onto inode.
746 * @block: (logical) number of block we are adding
747 * @chain: chain of indirect blocks (with a missing link - see
749 * @where: location of missing link
750 * @num: number of indirect blocks we are adding
751 * @blks: number of direct blocks we are adding
753 * This function fills the missing link and does all housekeeping needed in
754 * inode (->i_blocks, etc.). In case of success we end up with the full
755 * chain to new block and return 0.
757 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
758 ext4_lblk_t block, Indirect *where, int num, int blks)
762 struct ext4_block_alloc_info *block_i;
763 ext4_fsblk_t current_block;
765 block_i = EXT4_I(inode)->i_block_alloc_info;
767 * If we're splicing into a [td]indirect block (as opposed to the
768 * inode) then we need to get write access to the [td]indirect block
772 BUFFER_TRACE(where->bh, "get_write_access");
773 err = ext4_journal_get_write_access(handle, where->bh);
779 *where->p = where->key;
782 * Update the host buffer_head or inode to point to more just allocated
783 * direct blocks blocks
785 if (num == 0 && blks > 1) {
786 current_block = le32_to_cpu(where->key) + 1;
787 for (i = 1; i < blks; i++)
788 *(where->p + i ) = cpu_to_le32(current_block++);
792 * update the most recently allocated logical & physical block
793 * in i_block_alloc_info, to assist find the proper goal block for next
797 block_i->last_alloc_logical_block = block + blks - 1;
798 block_i->last_alloc_physical_block =
799 le32_to_cpu(where[num].key) + blks - 1;
802 /* We are done with atomic stuff, now do the rest of housekeeping */
804 inode->i_ctime = ext4_current_time(inode);
805 ext4_mark_inode_dirty(handle, inode);
807 /* had we spliced it onto indirect block? */
810 * If we spliced it onto an indirect block, we haven't
811 * altered the inode. Note however that if it is being spliced
812 * onto an indirect block at the very end of the file (the
813 * file is growing) then we *will* alter the inode to reflect
814 * the new i_size. But that is not done here - it is done in
815 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
817 jbd_debug(5, "splicing indirect only\n");
818 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
819 err = ext4_journal_dirty_metadata(handle, where->bh);
824 * OK, we spliced it into the inode itself on a direct block.
825 * Inode was dirtied above.
827 jbd_debug(5, "splicing direct\n");
832 for (i = 1; i <= num; i++) {
833 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
834 ext4_journal_forget(handle, where[i].bh);
835 ext4_free_blocks(handle, inode,
836 le32_to_cpu(where[i-1].key), 1, 0);
838 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
844 * Allocation strategy is simple: if we have to allocate something, we will
845 * have to go the whole way to leaf. So let's do it before attaching anything
846 * to tree, set linkage between the newborn blocks, write them if sync is
847 * required, recheck the path, free and repeat if check fails, otherwise
848 * set the last missing link (that will protect us from any truncate-generated
849 * removals - all blocks on the path are immune now) and possibly force the
850 * write on the parent block.
851 * That has a nice additional property: no special recovery from the failed
852 * allocations is needed - we simply release blocks and do not touch anything
853 * reachable from inode.
855 * `handle' can be NULL if create == 0.
857 * return > 0, # of blocks mapped or allocated.
858 * return = 0, if plain lookup failed.
859 * return < 0, error case.
862 * Need to be called with
863 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
864 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
866 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
867 ext4_lblk_t iblock, unsigned long maxblocks,
868 struct buffer_head *bh_result,
869 int create, int extend_disksize)
872 ext4_lblk_t offsets[4];
877 int blocks_to_boundary = 0;
879 struct ext4_inode_info *ei = EXT4_I(inode);
881 ext4_fsblk_t first_block = 0;
885 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
886 J_ASSERT(handle != NULL || create == 0);
887 depth = ext4_block_to_path(inode, iblock, offsets,
888 &blocks_to_boundary);
893 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
895 /* Simplest case - block found, no allocation needed */
897 first_block = le32_to_cpu(chain[depth - 1].key);
898 clear_buffer_new(bh_result);
901 while (count < maxblocks && count <= blocks_to_boundary) {
904 blk = le32_to_cpu(*(chain[depth-1].p + count));
906 if (blk == first_block + count)
914 /* Next simple case - plain lookup or failed read of indirect block */
915 if (!create || err == -EIO)
919 * Okay, we need to do block allocation. Lazily initialize the block
920 * allocation info here if necessary
922 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
923 ext4_init_block_alloc_info(inode);
925 goal = ext4_find_goal(inode, iblock, partial);
927 /* the number of blocks need to allocate for [d,t]indirect blocks */
928 indirect_blks = (chain + depth) - partial - 1;
931 * Next look up the indirect map to count the totoal number of
932 * direct blocks to allocate for this branch.
934 count = ext4_blks_to_allocate(partial, indirect_blks,
935 maxblocks, blocks_to_boundary);
937 * Block out ext4_truncate while we alter the tree
939 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
941 offsets + (partial - chain), partial);
944 * The ext4_splice_branch call will free and forget any buffers
945 * on the new chain if there is a failure, but that risks using
946 * up transaction credits, especially for bitmaps where the
947 * credits cannot be returned. Can we handle this somehow? We
948 * may need to return -EAGAIN upwards in the worst case. --sct
951 err = ext4_splice_branch(handle, inode, iblock,
952 partial, indirect_blks, count);
954 * i_disksize growing is protected by i_data_sem. Don't forget to
955 * protect it if you're about to implement concurrent
956 * ext4_get_block() -bzzz
958 if (!err && extend_disksize) {
959 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
960 if (disksize > i_size_read(inode))
961 disksize = i_size_read(inode);
962 if (disksize > ei->i_disksize)
963 ei->i_disksize = disksize;
968 set_buffer_new(bh_result);
970 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
971 if (count > blocks_to_boundary)
972 set_buffer_boundary(bh_result);
974 /* Clean up and exit */
975 partial = chain + depth - 1; /* the whole chain */
977 while (partial > chain) {
978 BUFFER_TRACE(partial->bh, "call brelse");
982 BUFFER_TRACE(bh_result, "returned");
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate @blocks for non extent file based file
991 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
993 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
994 int ind_blks, dind_blks, tind_blks;
996 /* number of new indirect blocks needed */
997 ind_blks = (blocks + icap - 1) / icap;
999 dind_blks = (ind_blks + icap - 1) / icap;
1003 return ind_blks + dind_blks + tind_blks;
1007 * Calculate the number of metadata blocks need to reserve
1008 * to allocate given number of blocks
1010 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1015 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1016 return ext4_ext_calc_metadata_amount(inode, blocks);
1018 return ext4_indirect_calc_metadata_amount(inode, blocks);
1021 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1023 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1024 int total, mdb, mdb_free;
1026 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1027 /* recalculate the number of metablocks still need to be reserved */
1028 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1029 mdb = ext4_calc_metadata_amount(inode, total);
1031 /* figure out how many metablocks to release */
1032 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1033 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1035 /* Account for allocated meta_blocks */
1036 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1038 /* update fs free blocks counter for truncate case */
1039 percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
1041 /* update per-inode reservations */
1042 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1043 EXT4_I(inode)->i_reserved_data_blocks -= used;
1045 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1046 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1047 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1048 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1052 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1053 * and returns if the blocks are already mapped.
1055 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1056 * and store the allocated blocks in the result buffer head and mark it
1059 * If file type is extents based, it will call ext4_ext_get_blocks(),
1060 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1063 * On success, it returns the number of blocks being mapped or allocate.
1064 * if create==0 and the blocks are pre-allocated and uninitialized block,
1065 * the result buffer head is unmapped. If the create ==1, it will make sure
1066 * the buffer head is mapped.
1068 * It returns 0 if plain look up failed (blocks have not been allocated), in
1069 * that casem, buffer head is unmapped
1071 * It returns the error in case of allocation failure.
1073 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1074 unsigned long max_blocks, struct buffer_head *bh,
1075 int create, int extend_disksize, int flag)
1079 clear_buffer_mapped(bh);
1082 * Try to see if we can get the block without requesting
1083 * for new file system block.
1085 down_read((&EXT4_I(inode)->i_data_sem));
1086 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1087 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1090 retval = ext4_get_blocks_handle(handle,
1091 inode, block, max_blocks, bh, 0, 0);
1093 up_read((&EXT4_I(inode)->i_data_sem));
1095 /* If it is only a block(s) look up */
1100 * Returns if the blocks have already allocated
1102 * Note that if blocks have been preallocated
1103 * ext4_ext_get_block() returns th create = 0
1104 * with buffer head unmapped.
1106 if (retval > 0 && buffer_mapped(bh))
1110 * New blocks allocate and/or writing to uninitialized extent
1111 * will possibly result in updating i_data, so we take
1112 * the write lock of i_data_sem, and call get_blocks()
1113 * with create == 1 flag.
1115 down_write((&EXT4_I(inode)->i_data_sem));
1118 * if the caller is from delayed allocation writeout path
1119 * we have already reserved fs blocks for allocation
1120 * let the underlying get_block() function know to
1121 * avoid double accounting
1124 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1126 * We need to check for EXT4 here because migrate
1127 * could have changed the inode type in between
1129 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1130 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1131 bh, create, extend_disksize);
1133 retval = ext4_get_blocks_handle(handle, inode, block,
1134 max_blocks, bh, create, extend_disksize);
1136 if (retval > 0 && buffer_new(bh)) {
1138 * We allocated new blocks which will result in
1139 * i_data's format changing. Force the migrate
1140 * to fail by clearing migrate flags
1142 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1148 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1150 * Update reserved blocks/metadata blocks
1151 * after successful block allocation
1152 * which were deferred till now
1154 if ((retval > 0) && buffer_delay(bh))
1155 ext4_da_update_reserve_space(inode, retval);
1158 up_write((&EXT4_I(inode)->i_data_sem));
1162 /* Maximum number of blocks we map for direct IO at once. */
1163 #define DIO_MAX_BLOCKS 4096
1165 static int ext4_get_block(struct inode *inode, sector_t iblock,
1166 struct buffer_head *bh_result, int create)
1168 handle_t *handle = ext4_journal_current_handle();
1169 int ret = 0, started = 0;
1170 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1173 if (create && !handle) {
1174 /* Direct IO write... */
1175 if (max_blocks > DIO_MAX_BLOCKS)
1176 max_blocks = DIO_MAX_BLOCKS;
1177 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1178 handle = ext4_journal_start(inode, dio_credits);
1179 if (IS_ERR(handle)) {
1180 ret = PTR_ERR(handle);
1186 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1187 max_blocks, bh_result, create, 0, 0);
1189 bh_result->b_size = (ret << inode->i_blkbits);
1193 ext4_journal_stop(handle);
1199 * `handle' can be NULL if create is zero
1201 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1202 ext4_lblk_t block, int create, int *errp)
1204 struct buffer_head dummy;
1207 J_ASSERT(handle != NULL || create == 0);
1210 dummy.b_blocknr = -1000;
1211 buffer_trace_init(&dummy.b_history);
1212 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1213 &dummy, create, 1, 0);
1215 * ext4_get_blocks_handle() returns number of blocks
1216 * mapped. 0 in case of a HOLE.
1224 if (!err && buffer_mapped(&dummy)) {
1225 struct buffer_head *bh;
1226 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1231 if (buffer_new(&dummy)) {
1232 J_ASSERT(create != 0);
1233 J_ASSERT(handle != NULL);
1236 * Now that we do not always journal data, we should
1237 * keep in mind whether this should always journal the
1238 * new buffer as metadata. For now, regular file
1239 * writes use ext4_get_block instead, so it's not a
1243 BUFFER_TRACE(bh, "call get_create_access");
1244 fatal = ext4_journal_get_create_access(handle, bh);
1245 if (!fatal && !buffer_uptodate(bh)) {
1246 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1247 set_buffer_uptodate(bh);
1250 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1251 err = ext4_journal_dirty_metadata(handle, bh);
1255 BUFFER_TRACE(bh, "not a new buffer");
1268 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1269 ext4_lblk_t block, int create, int *err)
1271 struct buffer_head * bh;
1273 bh = ext4_getblk(handle, inode, block, create, err);
1276 if (buffer_uptodate(bh))
1278 ll_rw_block(READ_META, 1, &bh);
1280 if (buffer_uptodate(bh))
1287 static int walk_page_buffers( handle_t *handle,
1288 struct buffer_head *head,
1292 int (*fn)( handle_t *handle,
1293 struct buffer_head *bh))
1295 struct buffer_head *bh;
1296 unsigned block_start, block_end;
1297 unsigned blocksize = head->b_size;
1299 struct buffer_head *next;
1301 for ( bh = head, block_start = 0;
1302 ret == 0 && (bh != head || !block_start);
1303 block_start = block_end, bh = next)
1305 next = bh->b_this_page;
1306 block_end = block_start + blocksize;
1307 if (block_end <= from || block_start >= to) {
1308 if (partial && !buffer_uptodate(bh))
1312 err = (*fn)(handle, bh);
1320 * To preserve ordering, it is essential that the hole instantiation and
1321 * the data write be encapsulated in a single transaction. We cannot
1322 * close off a transaction and start a new one between the ext4_get_block()
1323 * and the commit_write(). So doing the jbd2_journal_start at the start of
1324 * prepare_write() is the right place.
1326 * Also, this function can nest inside ext4_writepage() ->
1327 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1328 * has generated enough buffer credits to do the whole page. So we won't
1329 * block on the journal in that case, which is good, because the caller may
1332 * By accident, ext4 can be reentered when a transaction is open via
1333 * quota file writes. If we were to commit the transaction while thus
1334 * reentered, there can be a deadlock - we would be holding a quota
1335 * lock, and the commit would never complete if another thread had a
1336 * transaction open and was blocking on the quota lock - a ranking
1339 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1340 * will _not_ run commit under these circumstances because handle->h_ref
1341 * is elevated. We'll still have enough credits for the tiny quotafile
1344 static int do_journal_get_write_access(handle_t *handle,
1345 struct buffer_head *bh)
1347 if (!buffer_mapped(bh) || buffer_freed(bh))
1349 return ext4_journal_get_write_access(handle, bh);
1352 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1353 loff_t pos, unsigned len, unsigned flags,
1354 struct page **pagep, void **fsdata)
1356 struct inode *inode = mapping->host;
1357 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1364 index = pos >> PAGE_CACHE_SHIFT;
1365 from = pos & (PAGE_CACHE_SIZE - 1);
1369 handle = ext4_journal_start(inode, needed_blocks);
1370 if (IS_ERR(handle)) {
1371 ret = PTR_ERR(handle);
1375 page = grab_cache_page_write_begin(mapping, index, flags);
1377 ext4_journal_stop(handle);
1383 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1386 if (!ret && ext4_should_journal_data(inode)) {
1387 ret = walk_page_buffers(handle, page_buffers(page),
1388 from, to, NULL, do_journal_get_write_access);
1393 ext4_journal_stop(handle);
1394 page_cache_release(page);
1397 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1403 /* For write_end() in data=journal mode */
1404 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1406 if (!buffer_mapped(bh) || buffer_freed(bh))
1408 set_buffer_uptodate(bh);
1409 return ext4_journal_dirty_metadata(handle, bh);
1413 * We need to pick up the new inode size which generic_commit_write gave us
1414 * `file' can be NULL - eg, when called from page_symlink().
1416 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1417 * buffers are managed internally.
1419 static int ext4_ordered_write_end(struct file *file,
1420 struct address_space *mapping,
1421 loff_t pos, unsigned len, unsigned copied,
1422 struct page *page, void *fsdata)
1424 handle_t *handle = ext4_journal_current_handle();
1425 struct inode *inode = mapping->host;
1428 ret = ext4_jbd2_file_inode(handle, inode);
1432 * generic_write_end() will run mark_inode_dirty() if i_size
1433 * changes. So let's piggyback the i_disksize mark_inode_dirty
1438 new_i_size = pos + copied;
1439 if (new_i_size > EXT4_I(inode)->i_disksize)
1440 EXT4_I(inode)->i_disksize = new_i_size;
1441 ret2 = generic_write_end(file, mapping, pos, len, copied,
1447 ret2 = ext4_journal_stop(handle);
1451 return ret ? ret : copied;
1454 static int ext4_writeback_write_end(struct file *file,
1455 struct address_space *mapping,
1456 loff_t pos, unsigned len, unsigned copied,
1457 struct page *page, void *fsdata)
1459 handle_t *handle = ext4_journal_current_handle();
1460 struct inode *inode = mapping->host;
1464 new_i_size = pos + copied;
1465 if (new_i_size > EXT4_I(inode)->i_disksize)
1466 EXT4_I(inode)->i_disksize = new_i_size;
1468 ret2 = generic_write_end(file, mapping, pos, len, copied,
1474 ret2 = ext4_journal_stop(handle);
1478 return ret ? ret : copied;
1481 static int ext4_journalled_write_end(struct file *file,
1482 struct address_space *mapping,
1483 loff_t pos, unsigned len, unsigned copied,
1484 struct page *page, void *fsdata)
1486 handle_t *handle = ext4_journal_current_handle();
1487 struct inode *inode = mapping->host;
1492 from = pos & (PAGE_CACHE_SIZE - 1);
1496 if (!PageUptodate(page))
1498 page_zero_new_buffers(page, from+copied, to);
1501 ret = walk_page_buffers(handle, page_buffers(page), from,
1502 to, &partial, write_end_fn);
1504 SetPageUptodate(page);
1505 if (pos+copied > inode->i_size)
1506 i_size_write(inode, pos+copied);
1507 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1508 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1509 EXT4_I(inode)->i_disksize = inode->i_size;
1510 ret2 = ext4_mark_inode_dirty(handle, inode);
1516 ret2 = ext4_journal_stop(handle);
1519 page_cache_release(page);
1521 return ret ? ret : copied;
1524 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1526 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1527 unsigned long md_needed, mdblocks, total = 0;
1530 * recalculate the amount of metadata blocks to reserve
1531 * in order to allocate nrblocks
1532 * worse case is one extent per block
1534 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1535 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1536 mdblocks = ext4_calc_metadata_amount(inode, total);
1537 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1539 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1540 total = md_needed + nrblocks;
1542 if (ext4_has_free_blocks(sbi, total) < total) {
1543 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1546 /* reduce fs free blocks counter */
1547 percpu_counter_sub(&sbi->s_freeblocks_counter, total);
1549 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1550 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1552 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1553 return 0; /* success */
1556 static void ext4_da_release_space(struct inode *inode, int to_free)
1558 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1559 int total, mdb, mdb_free, release;
1562 return; /* Nothing to release, exit */
1564 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1566 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1568 * if there is no reserved blocks, but we try to free some
1569 * then the counter is messed up somewhere.
1570 * but since this function is called from invalidate
1571 * page, it's harmless to return without any action
1573 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1574 "blocks for inode %lu, but there is no reserved "
1575 "data blocks\n", to_free, inode->i_ino);
1576 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1580 /* recalculate the number of metablocks still need to be reserved */
1581 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1582 mdb = ext4_calc_metadata_amount(inode, total);
1584 /* figure out how many metablocks to release */
1585 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1586 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1588 release = to_free + mdb_free;
1590 /* update fs free blocks counter for truncate case */
1591 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1593 /* update per-inode reservations */
1594 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1595 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1597 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1598 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1599 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1602 static void ext4_da_page_release_reservation(struct page *page,
1603 unsigned long offset)
1606 struct buffer_head *head, *bh;
1607 unsigned int curr_off = 0;
1609 head = page_buffers(page);
1612 unsigned int next_off = curr_off + bh->b_size;
1614 if ((offset <= curr_off) && (buffer_delay(bh))) {
1616 clear_buffer_delay(bh);
1618 curr_off = next_off;
1619 } while ((bh = bh->b_this_page) != head);
1620 ext4_da_release_space(page->mapping->host, to_release);
1624 * Delayed allocation stuff
1627 struct mpage_da_data {
1628 struct inode *inode;
1629 struct buffer_head lbh; /* extent of blocks */
1630 unsigned long first_page, next_page; /* extent of pages */
1631 get_block_t *get_block;
1632 struct writeback_control *wbc;
1638 * mpage_da_submit_io - walks through extent of pages and try to write
1639 * them with writepage() call back
1641 * @mpd->inode: inode
1642 * @mpd->first_page: first page of the extent
1643 * @mpd->next_page: page after the last page of the extent
1644 * @mpd->get_block: the filesystem's block mapper function
1646 * By the time mpage_da_submit_io() is called we expect all blocks
1647 * to be allocated. this may be wrong if allocation failed.
1649 * As pages are already locked by write_cache_pages(), we can't use it
1651 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1654 struct pagevec pvec;
1655 unsigned long index, end;
1656 int ret = 0, err, nr_pages, i;
1657 struct inode *inode = mpd->inode;
1658 struct address_space *mapping = inode->i_mapping;
1660 BUG_ON(mpd->next_page <= mpd->first_page);
1662 * We need to start from the first_page to the next_page - 1
1663 * to make sure we also write the mapped dirty buffer_heads.
1664 * If we look at mpd->lbh.b_blocknr we would only be looking
1665 * at the currently mapped buffer_heads.
1667 index = mpd->first_page;
1668 end = mpd->next_page - 1;
1670 pagevec_init(&pvec, 0);
1671 while (index <= end) {
1672 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1675 for (i = 0; i < nr_pages; i++) {
1676 struct page *page = pvec.pages[i];
1678 index = page->index;
1683 BUG_ON(!PageLocked(page));
1684 BUG_ON(PageWriteback(page));
1686 pages_skipped = mpd->wbc->pages_skipped;
1687 err = mapping->a_ops->writepage(page, mpd->wbc);
1689 mpd->pages_written++;
1691 * In error case, we have to continue because
1692 * remaining pages are still locked
1693 * XXX: unlock and re-dirty them?
1698 pagevec_release(&pvec);
1704 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1706 * @mpd->inode - inode to walk through
1707 * @exbh->b_blocknr - first block on a disk
1708 * @exbh->b_size - amount of space in bytes
1709 * @logical - first logical block to start assignment with
1711 * the function goes through all passed space and put actual disk
1712 * block numbers into buffer heads, dropping BH_Delay
1714 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1715 struct buffer_head *exbh)
1717 struct inode *inode = mpd->inode;
1718 struct address_space *mapping = inode->i_mapping;
1719 int blocks = exbh->b_size >> inode->i_blkbits;
1720 sector_t pblock = exbh->b_blocknr, cur_logical;
1721 struct buffer_head *head, *bh;
1723 struct pagevec pvec;
1726 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1727 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1728 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1730 pagevec_init(&pvec, 0);
1732 while (index <= end) {
1733 /* XXX: optimize tail */
1734 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1737 for (i = 0; i < nr_pages; i++) {
1738 struct page *page = pvec.pages[i];
1740 index = page->index;
1745 BUG_ON(!PageLocked(page));
1746 BUG_ON(PageWriteback(page));
1747 BUG_ON(!page_has_buffers(page));
1749 bh = page_buffers(page);
1752 /* skip blocks out of the range */
1754 if (cur_logical >= logical)
1757 } while ((bh = bh->b_this_page) != head);
1760 if (cur_logical >= logical + blocks)
1762 if (buffer_delay(bh)) {
1763 bh->b_blocknr = pblock;
1764 clear_buffer_delay(bh);
1765 bh->b_bdev = inode->i_sb->s_bdev;
1766 } else if (buffer_unwritten(bh)) {
1767 bh->b_blocknr = pblock;
1768 clear_buffer_unwritten(bh);
1769 set_buffer_mapped(bh);
1771 bh->b_bdev = inode->i_sb->s_bdev;
1772 } else if (buffer_mapped(bh))
1773 BUG_ON(bh->b_blocknr != pblock);
1777 } while ((bh = bh->b_this_page) != head);
1779 pagevec_release(&pvec);
1785 * __unmap_underlying_blocks - just a helper function to unmap
1786 * set of blocks described by @bh
1788 static inline void __unmap_underlying_blocks(struct inode *inode,
1789 struct buffer_head *bh)
1791 struct block_device *bdev = inode->i_sb->s_bdev;
1794 blocks = bh->b_size >> inode->i_blkbits;
1795 for (i = 0; i < blocks; i++)
1796 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1800 * mpage_da_map_blocks - go through given space
1802 * @mpd->lbh - bh describing space
1803 * @mpd->get_block - the filesystem's block mapper function
1805 * The function skips space we know is already mapped to disk blocks.
1808 static void mpage_da_map_blocks(struct mpage_da_data *mpd)
1811 struct buffer_head *lbh = &mpd->lbh;
1812 sector_t next = lbh->b_blocknr;
1813 struct buffer_head new;
1816 * We consider only non-mapped and non-allocated blocks
1818 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1821 new.b_state = lbh->b_state;
1823 new.b_size = lbh->b_size;
1826 * If we didn't accumulate anything
1827 * to write simply return
1831 err = mpd->get_block(mpd->inode, next, &new, 1);
1834 BUG_ON(new.b_size == 0);
1836 if (buffer_new(&new))
1837 __unmap_underlying_blocks(mpd->inode, &new);
1840 * If blocks are delayed marked, we need to
1841 * put actual blocknr and drop delayed bit
1843 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1844 mpage_put_bnr_to_bhs(mpd, next, &new);
1849 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1850 (1 << BH_Delay) | (1 << BH_Unwritten))
1853 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1855 * @mpd->lbh - extent of blocks
1856 * @logical - logical number of the block in the file
1857 * @bh - bh of the block (used to access block's state)
1859 * the function is used to collect contig. blocks in same state
1861 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1862 sector_t logical, struct buffer_head *bh)
1865 size_t b_size = bh->b_size;
1866 struct buffer_head *lbh = &mpd->lbh;
1867 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1869 /* check if thereserved journal credits might overflow */
1870 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1871 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1873 * With non-extent format we are limited by the journal
1874 * credit available. Total credit needed to insert
1875 * nrblocks contiguous blocks is dependent on the
1876 * nrblocks. So limit nrblocks.
1879 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1880 EXT4_MAX_TRANS_DATA) {
1882 * Adding the new buffer_head would make it cross the
1883 * allowed limit for which we have journal credit
1884 * reserved. So limit the new bh->b_size
1886 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1887 mpd->inode->i_blkbits;
1888 /* we will do mpage_da_submit_io in the next loop */
1892 * First block in the extent
1894 if (lbh->b_size == 0) {
1895 lbh->b_blocknr = logical;
1896 lbh->b_size = b_size;
1897 lbh->b_state = bh->b_state & BH_FLAGS;
1901 next = lbh->b_blocknr + nrblocks;
1903 * Can we merge the block to our big extent?
1905 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1906 lbh->b_size += b_size;
1912 * We couldn't merge the block to our extent, so we
1913 * need to flush current extent and start new one
1915 mpage_da_map_blocks(mpd);
1916 mpage_da_submit_io(mpd);
1922 * __mpage_da_writepage - finds extent of pages and blocks
1924 * @page: page to consider
1925 * @wbc: not used, we just follow rules
1928 * The function finds extents of pages and scan them for all blocks.
1930 static int __mpage_da_writepage(struct page *page,
1931 struct writeback_control *wbc, void *data)
1933 struct mpage_da_data *mpd = data;
1934 struct inode *inode = mpd->inode;
1935 struct buffer_head *bh, *head, fake;
1940 * Rest of the page in the page_vec
1941 * redirty then and skip then. We will
1942 * try to to write them again after
1943 * starting a new transaction
1945 redirty_page_for_writepage(wbc, page);
1947 return MPAGE_DA_EXTENT_TAIL;
1950 * Can we merge this page to current extent?
1952 if (mpd->next_page != page->index) {
1954 * Nope, we can't. So, we map non-allocated blocks
1955 * and start IO on them using writepage()
1957 if (mpd->next_page != mpd->first_page) {
1958 mpage_da_map_blocks(mpd);
1959 mpage_da_submit_io(mpd);
1961 * skip rest of the page in the page_vec
1964 redirty_page_for_writepage(wbc, page);
1966 return MPAGE_DA_EXTENT_TAIL;
1970 * Start next extent of pages ...
1972 mpd->first_page = page->index;
1977 mpd->lbh.b_size = 0;
1978 mpd->lbh.b_state = 0;
1979 mpd->lbh.b_blocknr = 0;
1982 mpd->next_page = page->index + 1;
1983 logical = (sector_t) page->index <<
1984 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1986 if (!page_has_buffers(page)) {
1988 * There is no attached buffer heads yet (mmap?)
1989 * we treat the page asfull of dirty blocks
1992 bh->b_size = PAGE_CACHE_SIZE;
1994 set_buffer_dirty(bh);
1995 set_buffer_uptodate(bh);
1996 mpage_add_bh_to_extent(mpd, logical, bh);
1998 return MPAGE_DA_EXTENT_TAIL;
2001 * Page with regular buffer heads, just add all dirty ones
2003 head = page_buffers(page);
2006 BUG_ON(buffer_locked(bh));
2008 * We need to try to allocate
2009 * unmapped blocks in the same page.
2010 * Otherwise we won't make progress
2011 * with the page in ext4_da_writepage
2013 if (buffer_dirty(bh) &&
2014 (!buffer_mapped(bh) || buffer_delay(bh))) {
2015 mpage_add_bh_to_extent(mpd, logical, bh);
2017 return MPAGE_DA_EXTENT_TAIL;
2018 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2020 * mapped dirty buffer. We need to update
2021 * the b_state because we look at
2022 * b_state in mpage_da_map_blocks. We don't
2023 * update b_size because if we find an
2024 * unmapped buffer_head later we need to
2025 * use the b_state flag of that buffer_head.
2027 if (mpd->lbh.b_size == 0)
2029 bh->b_state & BH_FLAGS;
2032 } while ((bh = bh->b_this_page) != head);
2039 * mpage_da_writepages - walk the list of dirty pages of the given
2040 * address space, allocates non-allocated blocks, maps newly-allocated
2041 * blocks to existing bhs and issue IO them
2043 * @mapping: address space structure to write
2044 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2045 * @get_block: the filesystem's block mapper function.
2047 * This is a library function, which implements the writepages()
2048 * address_space_operation.
2050 static int mpage_da_writepages(struct address_space *mapping,
2051 struct writeback_control *wbc,
2052 get_block_t get_block)
2054 struct mpage_da_data mpd;
2059 return generic_writepages(mapping, wbc);
2062 mpd.inode = mapping->host;
2064 mpd.lbh.b_state = 0;
2065 mpd.lbh.b_blocknr = 0;
2068 mpd.get_block = get_block;
2070 mpd.pages_written = 0;
2072 to_write = wbc->nr_to_write;
2074 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
2077 * Handle last extent of pages
2079 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2080 mpage_da_map_blocks(&mpd);
2081 mpage_da_submit_io(&mpd);
2084 wbc->nr_to_write = to_write - mpd.pages_written;
2089 * this is a special callback for ->write_begin() only
2090 * it's intention is to return mapped block or reserve space
2092 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2093 struct buffer_head *bh_result, int create)
2097 BUG_ON(create == 0);
2098 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2101 * first, we need to know whether the block is allocated already
2102 * preallocated blocks are unmapped but should treated
2103 * the same as allocated blocks.
2105 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2106 if ((ret == 0) && !buffer_delay(bh_result)) {
2107 /* the block isn't (pre)allocated yet, let's reserve space */
2109 * XXX: __block_prepare_write() unmaps passed block,
2112 ret = ext4_da_reserve_space(inode, 1);
2114 /* not enough space to reserve */
2117 map_bh(bh_result, inode->i_sb, 0);
2118 set_buffer_new(bh_result);
2119 set_buffer_delay(bh_result);
2120 } else if (ret > 0) {
2121 bh_result->b_size = (ret << inode->i_blkbits);
2127 #define EXT4_DELALLOC_RSVED 1
2128 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2129 struct buffer_head *bh_result, int create)
2132 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2133 loff_t disksize = EXT4_I(inode)->i_disksize;
2134 handle_t *handle = NULL;
2136 handle = ext4_journal_current_handle();
2138 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2139 bh_result, 0, 0, 0);
2142 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2143 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2147 bh_result->b_size = (ret << inode->i_blkbits);
2150 * Update on-disk size along with block allocation
2151 * we don't use 'extend_disksize' as size may change
2152 * within already allocated block -bzzz
2154 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2155 if (disksize > i_size_read(inode))
2156 disksize = i_size_read(inode);
2157 if (disksize > EXT4_I(inode)->i_disksize) {
2159 * XXX: replace with spinlock if seen contended -bzzz
2161 down_write(&EXT4_I(inode)->i_data_sem);
2162 if (disksize > EXT4_I(inode)->i_disksize)
2163 EXT4_I(inode)->i_disksize = disksize;
2164 up_write(&EXT4_I(inode)->i_data_sem);
2166 if (EXT4_I(inode)->i_disksize == disksize) {
2167 ret = ext4_mark_inode_dirty(handle, inode);
2176 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2179 * unmapped buffer is possible for holes.
2180 * delay buffer is possible with delayed allocation
2182 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2185 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2186 struct buffer_head *bh_result, int create)
2189 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2192 * we don't want to do block allocation in writepage
2193 * so call get_block_wrap with create = 0
2195 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2196 bh_result, 0, 0, 0);
2198 bh_result->b_size = (ret << inode->i_blkbits);
2205 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2206 * get called via journal_submit_inode_data_buffers (no journal handle)
2207 * get called via shrink_page_list via pdflush (no journal handle)
2208 * or grab_page_cache when doing write_begin (have journal handle)
2210 static int ext4_da_writepage(struct page *page,
2211 struct writeback_control *wbc)
2216 struct buffer_head *page_bufs;
2217 struct inode *inode = page->mapping->host;
2219 size = i_size_read(inode);
2220 if (page->index == size >> PAGE_CACHE_SHIFT)
2221 len = size & ~PAGE_CACHE_MASK;
2223 len = PAGE_CACHE_SIZE;
2225 if (page_has_buffers(page)) {
2226 page_bufs = page_buffers(page);
2227 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2228 ext4_bh_unmapped_or_delay)) {
2230 * We don't want to do block allocation
2231 * So redirty the page and return
2232 * We may reach here when we do a journal commit
2233 * via journal_submit_inode_data_buffers.
2234 * If we don't have mapping block we just ignore
2235 * them. We can also reach here via shrink_page_list
2237 redirty_page_for_writepage(wbc, page);
2243 * The test for page_has_buffers() is subtle:
2244 * We know the page is dirty but it lost buffers. That means
2245 * that at some moment in time after write_begin()/write_end()
2246 * has been called all buffers have been clean and thus they
2247 * must have been written at least once. So they are all
2248 * mapped and we can happily proceed with mapping them
2249 * and writing the page.
2251 * Try to initialize the buffer_heads and check whether
2252 * all are mapped and non delay. We don't want to
2253 * do block allocation here.
2255 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2256 ext4_normal_get_block_write);
2258 page_bufs = page_buffers(page);
2259 /* check whether all are mapped and non delay */
2260 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2261 ext4_bh_unmapped_or_delay)) {
2262 redirty_page_for_writepage(wbc, page);
2268 * We can't do block allocation here
2269 * so just redity the page and unlock
2272 redirty_page_for_writepage(wbc, page);
2276 /* now mark the buffer_heads as dirty and uptodate */
2277 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2280 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2281 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2283 ret = block_write_full_page(page,
2284 ext4_normal_get_block_write,
2291 * This is called via ext4_da_writepages() to
2292 * calulate the total number of credits to reserve to fit
2293 * a single extent allocation into a single transaction,
2294 * ext4_da_writpeages() will loop calling this before
2295 * the block allocation.
2298 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2300 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2303 * With non-extent format the journal credit needed to
2304 * insert nrblocks contiguous block is dependent on
2305 * number of contiguous block. So we will limit
2306 * number of contiguous block to a sane value
2308 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2309 (max_blocks > EXT4_MAX_TRANS_DATA))
2310 max_blocks = EXT4_MAX_TRANS_DATA;
2312 return ext4_chunk_trans_blocks(inode, max_blocks);
2315 static int ext4_da_writepages(struct address_space *mapping,
2316 struct writeback_control *wbc)
2318 handle_t *handle = NULL;
2319 loff_t range_start = 0;
2320 struct inode *inode = mapping->host;
2321 int needed_blocks, ret = 0, nr_to_writebump = 0;
2322 long to_write, pages_skipped = 0;
2323 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2326 * No pages to write? This is mainly a kludge to avoid starting
2327 * a transaction for special inodes like journal inode on last iput()
2328 * because that could violate lock ordering on umount
2330 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2334 * If the filesystem has aborted, it is read-only, so return
2335 * right away instead of dumping stack traces later on that
2336 * will obscure the real source of the problem. We test
2337 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2338 * the latter could be true if the filesystem is mounted
2339 * read-only, and in that case, ext4_da_writepages should
2340 * *never* be called, so if that ever happens, we would want
2343 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2347 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2348 * This make sure small files blocks are allocated in
2349 * single attempt. This ensure that small files
2350 * get less fragmented.
2352 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2353 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2354 wbc->nr_to_write = sbi->s_mb_stream_request;
2357 if (!wbc->range_cyclic)
2359 * If range_cyclic is not set force range_cont
2360 * and save the old writeback_index
2362 wbc->range_cont = 1;
2364 range_start = wbc->range_start;
2365 pages_skipped = wbc->pages_skipped;
2368 to_write = wbc->nr_to_write;
2369 while (!ret && to_write > 0) {
2372 * we insert one extent at a time. So we need
2373 * credit needed for single extent allocation.
2374 * journalled mode is currently not supported
2377 BUG_ON(ext4_should_journal_data(inode));
2378 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2380 /* start a new transaction*/
2381 handle = ext4_journal_start(inode, needed_blocks);
2382 if (IS_ERR(handle)) {
2383 ret = PTR_ERR(handle);
2384 printk(KERN_CRIT "%s: jbd2_start: "
2385 "%ld pages, ino %lu; err %d\n", __func__,
2386 wbc->nr_to_write, inode->i_ino, ret);
2388 goto out_writepages;
2390 if (ext4_should_order_data(inode)) {
2392 * With ordered mode we need to add
2393 * the inode to the journal handl
2394 * when we do block allocation.
2396 ret = ext4_jbd2_file_inode(handle, inode);
2398 ext4_journal_stop(handle);
2399 goto out_writepages;
2403 to_write -= wbc->nr_to_write;
2404 ret = mpage_da_writepages(mapping, wbc,
2405 ext4_da_get_block_write);
2406 ext4_journal_stop(handle);
2407 if (ret == MPAGE_DA_EXTENT_TAIL) {
2409 * got one extent now try with
2412 to_write += wbc->nr_to_write;
2414 } else if (wbc->nr_to_write) {
2416 * There is no more writeout needed
2417 * or we requested for a noblocking writeout
2418 * and we found the device congested
2420 to_write += wbc->nr_to_write;
2423 wbc->nr_to_write = to_write;
2426 if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
2427 /* We skipped pages in this loop */
2428 wbc->range_start = range_start;
2429 wbc->nr_to_write = to_write +
2430 wbc->pages_skipped - pages_skipped;
2431 wbc->pages_skipped = pages_skipped;
2436 wbc->nr_to_write = to_write - nr_to_writebump;
2437 wbc->range_start = range_start;
2441 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2442 loff_t pos, unsigned len, unsigned flags,
2443 struct page **pagep, void **fsdata)
2445 int ret, retries = 0;
2449 struct inode *inode = mapping->host;
2452 index = pos >> PAGE_CACHE_SHIFT;
2453 from = pos & (PAGE_CACHE_SIZE - 1);
2458 * With delayed allocation, we don't log the i_disksize update
2459 * if there is delayed block allocation. But we still need
2460 * to journalling the i_disksize update if writes to the end
2461 * of file which has an already mapped buffer.
2463 handle = ext4_journal_start(inode, 1);
2464 if (IS_ERR(handle)) {
2465 ret = PTR_ERR(handle);
2469 page = grab_cache_page_write_begin(mapping, index, flags);
2471 ext4_journal_stop(handle);
2477 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2478 ext4_da_get_block_prep);
2481 ext4_journal_stop(handle);
2482 page_cache_release(page);
2485 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2492 * Check if we should update i_disksize
2493 * when write to the end of file but not require block allocation
2495 static int ext4_da_should_update_i_disksize(struct page *page,
2496 unsigned long offset)
2498 struct buffer_head *bh;
2499 struct inode *inode = page->mapping->host;
2503 bh = page_buffers(page);
2504 idx = offset >> inode->i_blkbits;
2506 for (i=0; i < idx; i++)
2507 bh = bh->b_this_page;
2509 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2514 static int ext4_da_write_end(struct file *file,
2515 struct address_space *mapping,
2516 loff_t pos, unsigned len, unsigned copied,
2517 struct page *page, void *fsdata)
2519 struct inode *inode = mapping->host;
2521 handle_t *handle = ext4_journal_current_handle();
2523 unsigned long start, end;
2525 start = pos & (PAGE_CACHE_SIZE - 1);
2526 end = start + copied -1;
2529 * generic_write_end() will run mark_inode_dirty() if i_size
2530 * changes. So let's piggyback the i_disksize mark_inode_dirty
2534 new_i_size = pos + copied;
2535 if (new_i_size > EXT4_I(inode)->i_disksize) {
2536 if (ext4_da_should_update_i_disksize(page, end)) {
2537 down_write(&EXT4_I(inode)->i_data_sem);
2538 if (new_i_size > EXT4_I(inode)->i_disksize) {
2540 * Updating i_disksize when extending file
2541 * without needing block allocation
2543 if (ext4_should_order_data(inode))
2544 ret = ext4_jbd2_file_inode(handle,
2547 EXT4_I(inode)->i_disksize = new_i_size;
2549 up_write(&EXT4_I(inode)->i_data_sem);
2552 ret2 = generic_write_end(file, mapping, pos, len, copied,
2557 ret2 = ext4_journal_stop(handle);
2561 return ret ? ret : copied;
2564 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2567 * Drop reserved blocks
2569 BUG_ON(!PageLocked(page));
2570 if (!page_has_buffers(page))
2573 ext4_da_page_release_reservation(page, offset);
2576 ext4_invalidatepage(page, offset);
2583 * bmap() is special. It gets used by applications such as lilo and by
2584 * the swapper to find the on-disk block of a specific piece of data.
2586 * Naturally, this is dangerous if the block concerned is still in the
2587 * journal. If somebody makes a swapfile on an ext4 data-journaling
2588 * filesystem and enables swap, then they may get a nasty shock when the
2589 * data getting swapped to that swapfile suddenly gets overwritten by
2590 * the original zero's written out previously to the journal and
2591 * awaiting writeback in the kernel's buffer cache.
2593 * So, if we see any bmap calls here on a modified, data-journaled file,
2594 * take extra steps to flush any blocks which might be in the cache.
2596 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2598 struct inode *inode = mapping->host;
2602 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2603 test_opt(inode->i_sb, DELALLOC)) {
2605 * With delalloc we want to sync the file
2606 * so that we can make sure we allocate
2609 filemap_write_and_wait(mapping);
2612 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2614 * This is a REALLY heavyweight approach, but the use of
2615 * bmap on dirty files is expected to be extremely rare:
2616 * only if we run lilo or swapon on a freshly made file
2617 * do we expect this to happen.
2619 * (bmap requires CAP_SYS_RAWIO so this does not
2620 * represent an unprivileged user DOS attack --- we'd be
2621 * in trouble if mortal users could trigger this path at
2624 * NB. EXT4_STATE_JDATA is not set on files other than
2625 * regular files. If somebody wants to bmap a directory
2626 * or symlink and gets confused because the buffer
2627 * hasn't yet been flushed to disk, they deserve
2628 * everything they get.
2631 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2632 journal = EXT4_JOURNAL(inode);
2633 jbd2_journal_lock_updates(journal);
2634 err = jbd2_journal_flush(journal);
2635 jbd2_journal_unlock_updates(journal);
2641 return generic_block_bmap(mapping,block,ext4_get_block);
2644 static int bget_one(handle_t *handle, struct buffer_head *bh)
2650 static int bput_one(handle_t *handle, struct buffer_head *bh)
2657 * Note that we don't need to start a transaction unless we're journaling data
2658 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2659 * need to file the inode to the transaction's list in ordered mode because if
2660 * we are writing back data added by write(), the inode is already there and if
2661 * we are writing back data modified via mmap(), noone guarantees in which
2662 * transaction the data will hit the disk. In case we are journaling data, we
2663 * cannot start transaction directly because transaction start ranks above page
2664 * lock so we have to do some magic.
2666 * In all journaling modes block_write_full_page() will start the I/O.
2670 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2675 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2677 * Same applies to ext4_get_block(). We will deadlock on various things like
2678 * lock_journal and i_data_sem
2680 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2683 * 16May01: If we're reentered then journal_current_handle() will be
2684 * non-zero. We simply *return*.
2686 * 1 July 2001: @@@ FIXME:
2687 * In journalled data mode, a data buffer may be metadata against the
2688 * current transaction. But the same file is part of a shared mapping
2689 * and someone does a writepage() on it.
2691 * We will move the buffer onto the async_data list, but *after* it has
2692 * been dirtied. So there's a small window where we have dirty data on
2695 * Note that this only applies to the last partial page in the file. The
2696 * bit which block_write_full_page() uses prepare/commit for. (That's
2697 * broken code anyway: it's wrong for msync()).
2699 * It's a rare case: affects the final partial page, for journalled data
2700 * where the file is subject to bith write() and writepage() in the same
2701 * transction. To fix it we'll need a custom block_write_full_page().
2702 * We'll probably need that anyway for journalling writepage() output.
2704 * We don't honour synchronous mounts for writepage(). That would be
2705 * disastrous. Any write() or metadata operation will sync the fs for
2709 static int __ext4_normal_writepage(struct page *page,
2710 struct writeback_control *wbc)
2712 struct inode *inode = page->mapping->host;
2714 if (test_opt(inode->i_sb, NOBH))
2715 return nobh_writepage(page,
2716 ext4_normal_get_block_write, wbc);
2718 return block_write_full_page(page,
2719 ext4_normal_get_block_write,
2723 static int ext4_normal_writepage(struct page *page,
2724 struct writeback_control *wbc)
2726 struct inode *inode = page->mapping->host;
2727 loff_t size = i_size_read(inode);
2730 J_ASSERT(PageLocked(page));
2731 if (page->index == size >> PAGE_CACHE_SHIFT)
2732 len = size & ~PAGE_CACHE_MASK;
2734 len = PAGE_CACHE_SIZE;
2736 if (page_has_buffers(page)) {
2737 /* if page has buffers it should all be mapped
2738 * and allocated. If there are not buffers attached
2739 * to the page we know the page is dirty but it lost
2740 * buffers. That means that at some moment in time
2741 * after write_begin() / write_end() has been called
2742 * all buffers have been clean and thus they must have been
2743 * written at least once. So they are all mapped and we can
2744 * happily proceed with mapping them and writing the page.
2746 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2747 ext4_bh_unmapped_or_delay));
2750 if (!ext4_journal_current_handle())
2751 return __ext4_normal_writepage(page, wbc);
2753 redirty_page_for_writepage(wbc, page);
2758 static int __ext4_journalled_writepage(struct page *page,
2759 struct writeback_control *wbc)
2761 struct address_space *mapping = page->mapping;
2762 struct inode *inode = mapping->host;
2763 struct buffer_head *page_bufs;
2764 handle_t *handle = NULL;
2768 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2769 ext4_normal_get_block_write);
2773 page_bufs = page_buffers(page);
2774 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2776 /* As soon as we unlock the page, it can go away, but we have
2777 * references to buffers so we are safe */
2780 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2781 if (IS_ERR(handle)) {
2782 ret = PTR_ERR(handle);
2786 ret = walk_page_buffers(handle, page_bufs, 0,
2787 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2789 err = walk_page_buffers(handle, page_bufs, 0,
2790 PAGE_CACHE_SIZE, NULL, write_end_fn);
2793 err = ext4_journal_stop(handle);
2797 walk_page_buffers(handle, page_bufs, 0,
2798 PAGE_CACHE_SIZE, NULL, bput_one);
2799 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2808 static int ext4_journalled_writepage(struct page *page,
2809 struct writeback_control *wbc)
2811 struct inode *inode = page->mapping->host;
2812 loff_t size = i_size_read(inode);
2815 J_ASSERT(PageLocked(page));
2816 if (page->index == size >> PAGE_CACHE_SHIFT)
2817 len = size & ~PAGE_CACHE_MASK;
2819 len = PAGE_CACHE_SIZE;
2821 if (page_has_buffers(page)) {
2822 /* if page has buffers it should all be mapped
2823 * and allocated. If there are not buffers attached
2824 * to the page we know the page is dirty but it lost
2825 * buffers. That means that at some moment in time
2826 * after write_begin() / write_end() has been called
2827 * all buffers have been clean and thus they must have been
2828 * written at least once. So they are all mapped and we can
2829 * happily proceed with mapping them and writing the page.
2831 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2832 ext4_bh_unmapped_or_delay));
2835 if (ext4_journal_current_handle())
2838 if (PageChecked(page)) {
2840 * It's mmapped pagecache. Add buffers and journal it. There
2841 * doesn't seem much point in redirtying the page here.
2843 ClearPageChecked(page);
2844 return __ext4_journalled_writepage(page, wbc);
2847 * It may be a page full of checkpoint-mode buffers. We don't
2848 * really know unless we go poke around in the buffer_heads.
2849 * But block_write_full_page will do the right thing.
2851 return block_write_full_page(page,
2852 ext4_normal_get_block_write,
2856 redirty_page_for_writepage(wbc, page);
2861 static int ext4_readpage(struct file *file, struct page *page)
2863 return mpage_readpage(page, ext4_get_block);
2867 ext4_readpages(struct file *file, struct address_space *mapping,
2868 struct list_head *pages, unsigned nr_pages)
2870 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2873 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2875 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2878 * If it's a full truncate we just forget about the pending dirtying
2881 ClearPageChecked(page);
2883 jbd2_journal_invalidatepage(journal, page, offset);
2886 static int ext4_releasepage(struct page *page, gfp_t wait)
2888 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2890 WARN_ON(PageChecked(page));
2891 if (!page_has_buffers(page))
2893 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2897 * If the O_DIRECT write will extend the file then add this inode to the
2898 * orphan list. So recovery will truncate it back to the original size
2899 * if the machine crashes during the write.
2901 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2902 * crashes then stale disk data _may_ be exposed inside the file. But current
2903 * VFS code falls back into buffered path in that case so we are safe.
2905 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2906 const struct iovec *iov, loff_t offset,
2907 unsigned long nr_segs)
2909 struct file *file = iocb->ki_filp;
2910 struct inode *inode = file->f_mapping->host;
2911 struct ext4_inode_info *ei = EXT4_I(inode);
2915 size_t count = iov_length(iov, nr_segs);
2918 loff_t final_size = offset + count;
2920 if (final_size > inode->i_size) {
2921 /* Credits for sb + inode write */
2922 handle = ext4_journal_start(inode, 2);
2923 if (IS_ERR(handle)) {
2924 ret = PTR_ERR(handle);
2927 ret = ext4_orphan_add(handle, inode);
2929 ext4_journal_stop(handle);
2933 ei->i_disksize = inode->i_size;
2934 ext4_journal_stop(handle);
2938 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2940 ext4_get_block, NULL);
2945 /* Credits for sb + inode write */
2946 handle = ext4_journal_start(inode, 2);
2947 if (IS_ERR(handle)) {
2948 /* This is really bad luck. We've written the data
2949 * but cannot extend i_size. Bail out and pretend
2950 * the write failed... */
2951 ret = PTR_ERR(handle);
2955 ext4_orphan_del(handle, inode);
2957 loff_t end = offset + ret;
2958 if (end > inode->i_size) {
2959 ei->i_disksize = end;
2960 i_size_write(inode, end);
2962 * We're going to return a positive `ret'
2963 * here due to non-zero-length I/O, so there's
2964 * no way of reporting error returns from
2965 * ext4_mark_inode_dirty() to userspace. So
2968 ext4_mark_inode_dirty(handle, inode);
2971 err = ext4_journal_stop(handle);
2980 * Pages can be marked dirty completely asynchronously from ext4's journalling
2981 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2982 * much here because ->set_page_dirty is called under VFS locks. The page is
2983 * not necessarily locked.
2985 * We cannot just dirty the page and leave attached buffers clean, because the
2986 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2987 * or jbddirty because all the journalling code will explode.
2989 * So what we do is to mark the page "pending dirty" and next time writepage
2990 * is called, propagate that into the buffers appropriately.
2992 static int ext4_journalled_set_page_dirty(struct page *page)
2994 SetPageChecked(page);
2995 return __set_page_dirty_nobuffers(page);
2998 static const struct address_space_operations ext4_ordered_aops = {
2999 .readpage = ext4_readpage,
3000 .readpages = ext4_readpages,
3001 .writepage = ext4_normal_writepage,
3002 .sync_page = block_sync_page,
3003 .write_begin = ext4_write_begin,
3004 .write_end = ext4_ordered_write_end,
3006 .invalidatepage = ext4_invalidatepage,
3007 .releasepage = ext4_releasepage,
3008 .direct_IO = ext4_direct_IO,
3009 .migratepage = buffer_migrate_page,
3010 .is_partially_uptodate = block_is_partially_uptodate,
3013 static const struct address_space_operations ext4_writeback_aops = {
3014 .readpage = ext4_readpage,
3015 .readpages = ext4_readpages,
3016 .writepage = ext4_normal_writepage,
3017 .sync_page = block_sync_page,
3018 .write_begin = ext4_write_begin,
3019 .write_end = ext4_writeback_write_end,
3021 .invalidatepage = ext4_invalidatepage,
3022 .releasepage = ext4_releasepage,
3023 .direct_IO = ext4_direct_IO,
3024 .migratepage = buffer_migrate_page,
3025 .is_partially_uptodate = block_is_partially_uptodate,
3028 static const struct address_space_operations ext4_journalled_aops = {
3029 .readpage = ext4_readpage,
3030 .readpages = ext4_readpages,
3031 .writepage = ext4_journalled_writepage,
3032 .sync_page = block_sync_page,
3033 .write_begin = ext4_write_begin,
3034 .write_end = ext4_journalled_write_end,
3035 .set_page_dirty = ext4_journalled_set_page_dirty,
3037 .invalidatepage = ext4_invalidatepage,
3038 .releasepage = ext4_releasepage,
3039 .is_partially_uptodate = block_is_partially_uptodate,
3042 static const struct address_space_operations ext4_da_aops = {
3043 .readpage = ext4_readpage,
3044 .readpages = ext4_readpages,
3045 .writepage = ext4_da_writepage,
3046 .writepages = ext4_da_writepages,
3047 .sync_page = block_sync_page,
3048 .write_begin = ext4_da_write_begin,
3049 .write_end = ext4_da_write_end,
3051 .invalidatepage = ext4_da_invalidatepage,
3052 .releasepage = ext4_releasepage,
3053 .direct_IO = ext4_direct_IO,
3054 .migratepage = buffer_migrate_page,
3055 .is_partially_uptodate = block_is_partially_uptodate,
3058 void ext4_set_aops(struct inode *inode)
3060 if (ext4_should_order_data(inode) &&
3061 test_opt(inode->i_sb, DELALLOC))
3062 inode->i_mapping->a_ops = &ext4_da_aops;
3063 else if (ext4_should_order_data(inode))
3064 inode->i_mapping->a_ops = &ext4_ordered_aops;
3065 else if (ext4_should_writeback_data(inode) &&
3066 test_opt(inode->i_sb, DELALLOC))
3067 inode->i_mapping->a_ops = &ext4_da_aops;
3068 else if (ext4_should_writeback_data(inode))
3069 inode->i_mapping->a_ops = &ext4_writeback_aops;
3071 inode->i_mapping->a_ops = &ext4_journalled_aops;
3075 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3076 * up to the end of the block which corresponds to `from'.
3077 * This required during truncate. We need to physically zero the tail end
3078 * of that block so it doesn't yield old data if the file is later grown.
3080 int ext4_block_truncate_page(handle_t *handle,
3081 struct address_space *mapping, loff_t from)
3083 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3084 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3085 unsigned blocksize, length, pos;
3087 struct inode *inode = mapping->host;
3088 struct buffer_head *bh;
3092 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3096 blocksize = inode->i_sb->s_blocksize;
3097 length = blocksize - (offset & (blocksize - 1));
3098 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3101 * For "nobh" option, we can only work if we don't need to
3102 * read-in the page - otherwise we create buffers to do the IO.
3104 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3105 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3106 zero_user(page, offset, length);
3107 set_page_dirty(page);
3111 if (!page_has_buffers(page))
3112 create_empty_buffers(page, blocksize, 0);
3114 /* Find the buffer that contains "offset" */
3115 bh = page_buffers(page);
3117 while (offset >= pos) {
3118 bh = bh->b_this_page;
3124 if (buffer_freed(bh)) {
3125 BUFFER_TRACE(bh, "freed: skip");
3129 if (!buffer_mapped(bh)) {
3130 BUFFER_TRACE(bh, "unmapped");
3131 ext4_get_block(inode, iblock, bh, 0);
3132 /* unmapped? It's a hole - nothing to do */
3133 if (!buffer_mapped(bh)) {
3134 BUFFER_TRACE(bh, "still unmapped");
3139 /* Ok, it's mapped. Make sure it's up-to-date */
3140 if (PageUptodate(page))
3141 set_buffer_uptodate(bh);
3143 if (!buffer_uptodate(bh)) {
3145 ll_rw_block(READ, 1, &bh);
3147 /* Uhhuh. Read error. Complain and punt. */
3148 if (!buffer_uptodate(bh))
3152 if (ext4_should_journal_data(inode)) {
3153 BUFFER_TRACE(bh, "get write access");
3154 err = ext4_journal_get_write_access(handle, bh);
3159 zero_user(page, offset, length);
3161 BUFFER_TRACE(bh, "zeroed end of block");
3164 if (ext4_should_journal_data(inode)) {
3165 err = ext4_journal_dirty_metadata(handle, bh);
3167 if (ext4_should_order_data(inode))
3168 err = ext4_jbd2_file_inode(handle, inode);
3169 mark_buffer_dirty(bh);
3174 page_cache_release(page);
3179 * Probably it should be a library function... search for first non-zero word
3180 * or memcmp with zero_page, whatever is better for particular architecture.
3183 static inline int all_zeroes(__le32 *p, __le32 *q)
3192 * ext4_find_shared - find the indirect blocks for partial truncation.
3193 * @inode: inode in question
3194 * @depth: depth of the affected branch
3195 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3196 * @chain: place to store the pointers to partial indirect blocks
3197 * @top: place to the (detached) top of branch
3199 * This is a helper function used by ext4_truncate().
3201 * When we do truncate() we may have to clean the ends of several
3202 * indirect blocks but leave the blocks themselves alive. Block is
3203 * partially truncated if some data below the new i_size is refered
3204 * from it (and it is on the path to the first completely truncated
3205 * data block, indeed). We have to free the top of that path along
3206 * with everything to the right of the path. Since no allocation
3207 * past the truncation point is possible until ext4_truncate()
3208 * finishes, we may safely do the latter, but top of branch may
3209 * require special attention - pageout below the truncation point
3210 * might try to populate it.
3212 * We atomically detach the top of branch from the tree, store the
3213 * block number of its root in *@top, pointers to buffer_heads of
3214 * partially truncated blocks - in @chain[].bh and pointers to
3215 * their last elements that should not be removed - in
3216 * @chain[].p. Return value is the pointer to last filled element
3219 * The work left to caller to do the actual freeing of subtrees:
3220 * a) free the subtree starting from *@top
3221 * b) free the subtrees whose roots are stored in
3222 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3223 * c) free the subtrees growing from the inode past the @chain[0].
3224 * (no partially truncated stuff there). */
3226 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3227 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3229 Indirect *partial, *p;
3233 /* Make k index the deepest non-null offest + 1 */
3234 for (k = depth; k > 1 && !offsets[k-1]; k--)
3236 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3237 /* Writer: pointers */
3239 partial = chain + k-1;
3241 * If the branch acquired continuation since we've looked at it -
3242 * fine, it should all survive and (new) top doesn't belong to us.
3244 if (!partial->key && *partial->p)
3247 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
3250 * OK, we've found the last block that must survive. The rest of our
3251 * branch should be detached before unlocking. However, if that rest
3252 * of branch is all ours and does not grow immediately from the inode
3253 * it's easier to cheat and just decrement partial->p.
3255 if (p == chain + k - 1 && p > chain) {
3259 /* Nope, don't do this in ext4. Must leave the tree intact */
3266 while(partial > p) {
3267 brelse(partial->bh);
3275 * Zero a number of block pointers in either an inode or an indirect block.
3276 * If we restart the transaction we must again get write access to the
3277 * indirect block for further modification.
3279 * We release `count' blocks on disk, but (last - first) may be greater
3280 * than `count' because there can be holes in there.
3282 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3283 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3284 unsigned long count, __le32 *first, __le32 *last)
3287 if (try_to_extend_transaction(handle, inode)) {
3289 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3290 ext4_journal_dirty_metadata(handle, bh);
3292 ext4_mark_inode_dirty(handle, inode);
3293 ext4_journal_test_restart(handle, inode);
3295 BUFFER_TRACE(bh, "retaking write access");
3296 ext4_journal_get_write_access(handle, bh);
3301 * Any buffers which are on the journal will be in memory. We find
3302 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3303 * on them. We've already detached each block from the file, so
3304 * bforget() in jbd2_journal_forget() should be safe.
3306 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3308 for (p = first; p < last; p++) {
3309 u32 nr = le32_to_cpu(*p);
3311 struct buffer_head *tbh;
3314 tbh = sb_find_get_block(inode->i_sb, nr);
3315 ext4_forget(handle, 0, inode, tbh, nr);
3319 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3323 * ext4_free_data - free a list of data blocks
3324 * @handle: handle for this transaction
3325 * @inode: inode we are dealing with
3326 * @this_bh: indirect buffer_head which contains *@first and *@last
3327 * @first: array of block numbers
3328 * @last: points immediately past the end of array
3330 * We are freeing all blocks refered from that array (numbers are stored as
3331 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3333 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3334 * blocks are contiguous then releasing them at one time will only affect one
3335 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3336 * actually use a lot of journal space.
3338 * @this_bh will be %NULL if @first and @last point into the inode's direct
3341 static void ext4_free_data(handle_t *handle, struct inode *inode,
3342 struct buffer_head *this_bh,
3343 __le32 *first, __le32 *last)
3345 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3346 unsigned long count = 0; /* Number of blocks in the run */
3347 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3350 ext4_fsblk_t nr; /* Current block # */
3351 __le32 *p; /* Pointer into inode/ind
3352 for current block */
3355 if (this_bh) { /* For indirect block */
3356 BUFFER_TRACE(this_bh, "get_write_access");
3357 err = ext4_journal_get_write_access(handle, this_bh);
3358 /* Important: if we can't update the indirect pointers
3359 * to the blocks, we can't free them. */
3364 for (p = first; p < last; p++) {
3365 nr = le32_to_cpu(*p);
3367 /* accumulate blocks to free if they're contiguous */
3370 block_to_free_p = p;
3372 } else if (nr == block_to_free + count) {
3375 ext4_clear_blocks(handle, inode, this_bh,
3377 count, block_to_free_p, p);
3379 block_to_free_p = p;
3386 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3387 count, block_to_free_p, p);
3390 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3393 * The buffer head should have an attached journal head at this
3394 * point. However, if the data is corrupted and an indirect
3395 * block pointed to itself, it would have been detached when
3396 * the block was cleared. Check for this instead of OOPSing.
3399 ext4_journal_dirty_metadata(handle, this_bh);
3401 ext4_error(inode->i_sb, __func__,
3402 "circular indirect block detected, "
3403 "inode=%lu, block=%llu",
3405 (unsigned long long) this_bh->b_blocknr);
3410 * ext4_free_branches - free an array of branches
3411 * @handle: JBD handle for this transaction
3412 * @inode: inode we are dealing with
3413 * @parent_bh: the buffer_head which contains *@first and *@last
3414 * @first: array of block numbers
3415 * @last: pointer immediately past the end of array
3416 * @depth: depth of the branches to free
3418 * We are freeing all blocks refered from these branches (numbers are
3419 * stored as little-endian 32-bit) and updating @inode->i_blocks
3422 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3423 struct buffer_head *parent_bh,
3424 __le32 *first, __le32 *last, int depth)
3429 if (is_handle_aborted(handle))
3433 struct buffer_head *bh;
3434 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3436 while (--p >= first) {
3437 nr = le32_to_cpu(*p);
3439 continue; /* A hole */
3441 /* Go read the buffer for the next level down */
3442 bh = sb_bread(inode->i_sb, nr);
3445 * A read failure? Report error and clear slot
3449 ext4_error(inode->i_sb, "ext4_free_branches",
3450 "Read failure, inode=%lu, block=%llu",
3455 /* This zaps the entire block. Bottom up. */
3456 BUFFER_TRACE(bh, "free child branches");
3457 ext4_free_branches(handle, inode, bh,
3458 (__le32*)bh->b_data,
3459 (__le32*)bh->b_data + addr_per_block,
3463 * We've probably journalled the indirect block several
3464 * times during the truncate. But it's no longer
3465 * needed and we now drop it from the transaction via
3466 * jbd2_journal_revoke().
3468 * That's easy if it's exclusively part of this
3469 * transaction. But if it's part of the committing
3470 * transaction then jbd2_journal_forget() will simply
3471 * brelse() it. That means that if the underlying
3472 * block is reallocated in ext4_get_block(),
3473 * unmap_underlying_metadata() will find this block
3474 * and will try to get rid of it. damn, damn.
3476 * If this block has already been committed to the
3477 * journal, a revoke record will be written. And
3478 * revoke records must be emitted *before* clearing
3479 * this block's bit in the bitmaps.
3481 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3484 * Everything below this this pointer has been
3485 * released. Now let this top-of-subtree go.
3487 * We want the freeing of this indirect block to be
3488 * atomic in the journal with the updating of the
3489 * bitmap block which owns it. So make some room in
3492 * We zero the parent pointer *after* freeing its
3493 * pointee in the bitmaps, so if extend_transaction()
3494 * for some reason fails to put the bitmap changes and
3495 * the release into the same transaction, recovery
3496 * will merely complain about releasing a free block,
3497 * rather than leaking blocks.
3499 if (is_handle_aborted(handle))
3501 if (try_to_extend_transaction(handle, inode)) {
3502 ext4_mark_inode_dirty(handle, inode);
3503 ext4_journal_test_restart(handle, inode);
3506 ext4_free_blocks(handle, inode, nr, 1, 1);
3510 * The block which we have just freed is
3511 * pointed to by an indirect block: journal it
3513 BUFFER_TRACE(parent_bh, "get_write_access");
3514 if (!ext4_journal_get_write_access(handle,
3517 BUFFER_TRACE(parent_bh,
3518 "call ext4_journal_dirty_metadata");
3519 ext4_journal_dirty_metadata(handle,
3525 /* We have reached the bottom of the tree. */
3526 BUFFER_TRACE(parent_bh, "free data blocks");
3527 ext4_free_data(handle, inode, parent_bh, first, last);
3531 int ext4_can_truncate(struct inode *inode)
3533 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3535 if (S_ISREG(inode->i_mode))
3537 if (S_ISDIR(inode->i_mode))
3539 if (S_ISLNK(inode->i_mode))
3540 return !ext4_inode_is_fast_symlink(inode);
3547 * We block out ext4_get_block() block instantiations across the entire
3548 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3549 * simultaneously on behalf of the same inode.
3551 * As we work through the truncate and commmit bits of it to the journal there
3552 * is one core, guiding principle: the file's tree must always be consistent on
3553 * disk. We must be able to restart the truncate after a crash.
3555 * The file's tree may be transiently inconsistent in memory (although it
3556 * probably isn't), but whenever we close off and commit a journal transaction,
3557 * the contents of (the filesystem + the journal) must be consistent and
3558 * restartable. It's pretty simple, really: bottom up, right to left (although
3559 * left-to-right works OK too).
3561 * Note that at recovery time, journal replay occurs *before* the restart of
3562 * truncate against the orphan inode list.
3564 * The committed inode has the new, desired i_size (which is the same as
3565 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3566 * that this inode's truncate did not complete and it will again call
3567 * ext4_truncate() to have another go. So there will be instantiated blocks
3568 * to the right of the truncation point in a crashed ext4 filesystem. But
3569 * that's fine - as long as they are linked from the inode, the post-crash
3570 * ext4_truncate() run will find them and release them.
3572 void ext4_truncate(struct inode *inode)
3575 struct ext4_inode_info *ei = EXT4_I(inode);
3576 __le32 *i_data = ei->i_data;
3577 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3578 struct address_space *mapping = inode->i_mapping;
3579 ext4_lblk_t offsets[4];
3584 ext4_lblk_t last_block;
3585 unsigned blocksize = inode->i_sb->s_blocksize;
3587 if (!ext4_can_truncate(inode))
3590 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3591 ext4_ext_truncate(inode);
3595 handle = start_transaction(inode);
3597 return; /* AKPM: return what? */
3599 last_block = (inode->i_size + blocksize-1)
3600 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3602 if (inode->i_size & (blocksize - 1))
3603 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3606 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3608 goto out_stop; /* error */
3611 * OK. This truncate is going to happen. We add the inode to the
3612 * orphan list, so that if this truncate spans multiple transactions,
3613 * and we crash, we will resume the truncate when the filesystem
3614 * recovers. It also marks the inode dirty, to catch the new size.
3616 * Implication: the file must always be in a sane, consistent
3617 * truncatable state while each transaction commits.
3619 if (ext4_orphan_add(handle, inode))
3623 * From here we block out all ext4_get_block() callers who want to
3624 * modify the block allocation tree.
3626 down_write(&ei->i_data_sem);
3628 ext4_discard_reservation(inode);
3631 * The orphan list entry will now protect us from any crash which
3632 * occurs before the truncate completes, so it is now safe to propagate
3633 * the new, shorter inode size (held for now in i_size) into the
3634 * on-disk inode. We do this via i_disksize, which is the value which
3635 * ext4 *really* writes onto the disk inode.
3637 ei->i_disksize = inode->i_size;
3639 if (n == 1) { /* direct blocks */
3640 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3641 i_data + EXT4_NDIR_BLOCKS);
3645 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3646 /* Kill the top of shared branch (not detached) */
3648 if (partial == chain) {
3649 /* Shared branch grows from the inode */
3650 ext4_free_branches(handle, inode, NULL,
3651 &nr, &nr+1, (chain+n-1) - partial);
3654 * We mark the inode dirty prior to restart,
3655 * and prior to stop. No need for it here.
3658 /* Shared branch grows from an indirect block */
3659 BUFFER_TRACE(partial->bh, "get_write_access");
3660 ext4_free_branches(handle, inode, partial->bh,
3662 partial->p+1, (chain+n-1) - partial);
3665 /* Clear the ends of indirect blocks on the shared branch */
3666 while (partial > chain) {
3667 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3668 (__le32*)partial->bh->b_data+addr_per_block,
3669 (chain+n-1) - partial);
3670 BUFFER_TRACE(partial->bh, "call brelse");
3671 brelse (partial->bh);
3675 /* Kill the remaining (whole) subtrees */
3676 switch (offsets[0]) {
3678 nr = i_data[EXT4_IND_BLOCK];
3680 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3681 i_data[EXT4_IND_BLOCK] = 0;
3683 case EXT4_IND_BLOCK:
3684 nr = i_data[EXT4_DIND_BLOCK];
3686 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3687 i_data[EXT4_DIND_BLOCK] = 0;
3689 case EXT4_DIND_BLOCK:
3690 nr = i_data[EXT4_TIND_BLOCK];
3692 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3693 i_data[EXT4_TIND_BLOCK] = 0;
3695 case EXT4_TIND_BLOCK:
3699 up_write(&ei->i_data_sem);
3700 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3701 ext4_mark_inode_dirty(handle, inode);
3704 * In a multi-transaction truncate, we only make the final transaction
3711 * If this was a simple ftruncate(), and the file will remain alive
3712 * then we need to clear up the orphan record which we created above.
3713 * However, if this was a real unlink then we were called by
3714 * ext4_delete_inode(), and we allow that function to clean up the
3715 * orphan info for us.
3718 ext4_orphan_del(handle, inode);
3720 ext4_journal_stop(handle);
3723 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3724 unsigned long ino, struct ext4_iloc *iloc)
3726 ext4_group_t block_group;
3727 unsigned long offset;
3729 struct ext4_group_desc *gdp;
3731 if (!ext4_valid_inum(sb, ino)) {
3733 * This error is already checked for in namei.c unless we are
3734 * looking at an NFS filehandle, in which case no error
3740 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3741 gdp = ext4_get_group_desc(sb, block_group, NULL);
3746 * Figure out the offset within the block group inode table
3748 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3749 EXT4_INODE_SIZE(sb);
3750 block = ext4_inode_table(sb, gdp) +
3751 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3753 iloc->block_group = block_group;
3754 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3759 * ext4_get_inode_loc returns with an extra refcount against the inode's
3760 * underlying buffer_head on success. If 'in_mem' is true, we have all
3761 * data in memory that is needed to recreate the on-disk version of this
3764 static int __ext4_get_inode_loc(struct inode *inode,
3765 struct ext4_iloc *iloc, int in_mem)
3768 struct buffer_head *bh;
3770 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3774 bh = sb_getblk(inode->i_sb, block);
3776 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3777 "unable to read inode block - "
3778 "inode=%lu, block=%llu",
3779 inode->i_ino, block);
3782 if (!buffer_uptodate(bh)) {
3786 * If the buffer has the write error flag, we have failed
3787 * to write out another inode in the same block. In this
3788 * case, we don't have to read the block because we may
3789 * read the old inode data successfully.
3791 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3792 set_buffer_uptodate(bh);
3794 if (buffer_uptodate(bh)) {
3795 /* someone brought it uptodate while we waited */
3801 * If we have all information of the inode in memory and this
3802 * is the only valid inode in the block, we need not read the
3806 struct buffer_head *bitmap_bh;
3807 struct ext4_group_desc *desc;
3808 int inodes_per_buffer;
3809 int inode_offset, i;
3810 ext4_group_t block_group;
3813 block_group = (inode->i_ino - 1) /
3814 EXT4_INODES_PER_GROUP(inode->i_sb);
3815 inodes_per_buffer = bh->b_size /
3816 EXT4_INODE_SIZE(inode->i_sb);
3817 inode_offset = ((inode->i_ino - 1) %
3818 EXT4_INODES_PER_GROUP(inode->i_sb));
3819 start = inode_offset & ~(inodes_per_buffer - 1);
3821 /* Is the inode bitmap in cache? */
3822 desc = ext4_get_group_desc(inode->i_sb,
3827 bitmap_bh = sb_getblk(inode->i_sb,
3828 ext4_inode_bitmap(inode->i_sb, desc));
3833 * If the inode bitmap isn't in cache then the
3834 * optimisation may end up performing two reads instead
3835 * of one, so skip it.
3837 if (!buffer_uptodate(bitmap_bh)) {
3841 for (i = start; i < start + inodes_per_buffer; i++) {
3842 if (i == inode_offset)
3844 if (ext4_test_bit(i, bitmap_bh->b_data))
3848 if (i == start + inodes_per_buffer) {
3849 /* all other inodes are free, so skip I/O */
3850 memset(bh->b_data, 0, bh->b_size);
3851 set_buffer_uptodate(bh);
3859 * There are other valid inodes in the buffer, this inode
3860 * has in-inode xattrs, or we don't have this inode in memory.
3861 * Read the block from disk.
3864 bh->b_end_io = end_buffer_read_sync;
3865 submit_bh(READ_META, bh);
3867 if (!buffer_uptodate(bh)) {
3868 ext4_error(inode->i_sb, "ext4_get_inode_loc",
3869 "unable to read inode block - "
3870 "inode=%lu, block=%llu",
3871 inode->i_ino, block);
3881 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3883 /* We have all inode data except xattrs in memory here. */
3884 return __ext4_get_inode_loc(inode, iloc,
3885 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3888 void ext4_set_inode_flags(struct inode *inode)
3890 unsigned int flags = EXT4_I(inode)->i_flags;
3892 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3893 if (flags & EXT4_SYNC_FL)
3894 inode->i_flags |= S_SYNC;
3895 if (flags & EXT4_APPEND_FL)
3896 inode->i_flags |= S_APPEND;
3897 if (flags & EXT4_IMMUTABLE_FL)
3898 inode->i_flags |= S_IMMUTABLE;
3899 if (flags & EXT4_NOATIME_FL)
3900 inode->i_flags |= S_NOATIME;
3901 if (flags & EXT4_DIRSYNC_FL)
3902 inode->i_flags |= S_DIRSYNC;
3905 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3906 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3908 unsigned int flags = ei->vfs_inode.i_flags;
3910 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3911 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
3913 ei->i_flags |= EXT4_SYNC_FL;
3914 if (flags & S_APPEND)
3915 ei->i_flags |= EXT4_APPEND_FL;
3916 if (flags & S_IMMUTABLE)
3917 ei->i_flags |= EXT4_IMMUTABLE_FL;
3918 if (flags & S_NOATIME)
3919 ei->i_flags |= EXT4_NOATIME_FL;
3920 if (flags & S_DIRSYNC)
3921 ei->i_flags |= EXT4_DIRSYNC_FL;
3923 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3924 struct ext4_inode_info *ei)
3927 struct inode *inode = &(ei->vfs_inode);
3928 struct super_block *sb = inode->i_sb;
3930 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3931 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3932 /* we are using combined 48 bit field */
3933 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3934 le32_to_cpu(raw_inode->i_blocks_lo);
3935 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
3936 /* i_blocks represent file system block size */
3937 return i_blocks << (inode->i_blkbits - 9);
3942 return le32_to_cpu(raw_inode->i_blocks_lo);
3946 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3948 struct ext4_iloc iloc;
3949 struct ext4_inode *raw_inode;
3950 struct ext4_inode_info *ei;
3951 struct buffer_head *bh;
3952 struct inode *inode;
3956 inode = iget_locked(sb, ino);
3958 return ERR_PTR(-ENOMEM);
3959 if (!(inode->i_state & I_NEW))
3963 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3964 ei->i_acl = EXT4_ACL_NOT_CACHED;
3965 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
3967 ei->i_block_alloc_info = NULL;
3969 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3973 raw_inode = ext4_raw_inode(&iloc);
3974 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3975 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3976 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3977 if(!(test_opt (inode->i_sb, NO_UID32))) {
3978 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3979 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3981 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3984 ei->i_dir_start_lookup = 0;
3985 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3986 /* We now have enough fields to check if the inode was active or not.
3987 * This is needed because nfsd might try to access dead inodes
3988 * the test is that same one that e2fsck uses
3989 * NeilBrown 1999oct15
3991 if (inode->i_nlink == 0) {
3992 if (inode->i_mode == 0 ||
3993 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3994 /* this inode is deleted */
3999 /* The only unlinked inodes we let through here have
4000 * valid i_mode and are being read by the orphan
4001 * recovery code: that's fine, we're about to complete
4002 * the process of deleting those. */
4004 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4005 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4006 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4007 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4008 cpu_to_le32(EXT4_OS_HURD)) {
4010 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4012 inode->i_size = ext4_isize(raw_inode);
4013 ei->i_disksize = inode->i_size;
4014 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4015 ei->i_block_group = iloc.block_group;
4017 * NOTE! The in-memory inode i_data array is in little-endian order
4018 * even on big-endian machines: we do NOT byteswap the block numbers!
4020 for (block = 0; block < EXT4_N_BLOCKS; block++)
4021 ei->i_data[block] = raw_inode->i_block[block];
4022 INIT_LIST_HEAD(&ei->i_orphan);
4024 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4025 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4026 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4027 EXT4_INODE_SIZE(inode->i_sb)) {
4032 if (ei->i_extra_isize == 0) {
4033 /* The extra space is currently unused. Use it. */
4034 ei->i_extra_isize = sizeof(struct ext4_inode) -
4035 EXT4_GOOD_OLD_INODE_SIZE;
4037 __le32 *magic = (void *)raw_inode +
4038 EXT4_GOOD_OLD_INODE_SIZE +
4040 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4041 ei->i_state |= EXT4_STATE_XATTR;
4044 ei->i_extra_isize = 0;
4046 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4047 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4048 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4049 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4051 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4052 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4053 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4055 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4058 if (S_ISREG(inode->i_mode)) {
4059 inode->i_op = &ext4_file_inode_operations;
4060 inode->i_fop = &ext4_file_operations;
4061 ext4_set_aops(inode);
4062 } else if (S_ISDIR(inode->i_mode)) {
4063 inode->i_op = &ext4_dir_inode_operations;
4064 inode->i_fop = &ext4_dir_operations;
4065 } else if (S_ISLNK(inode->i_mode)) {
4066 if (ext4_inode_is_fast_symlink(inode))
4067 inode->i_op = &ext4_fast_symlink_inode_operations;
4069 inode->i_op = &ext4_symlink_inode_operations;
4070 ext4_set_aops(inode);
4073 inode->i_op = &ext4_special_inode_operations;
4074 if (raw_inode->i_block[0])
4075 init_special_inode(inode, inode->i_mode,
4076 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4078 init_special_inode(inode, inode->i_mode,
4079 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4082 ext4_set_inode_flags(inode);
4083 unlock_new_inode(inode);
4088 return ERR_PTR(ret);
4091 static int ext4_inode_blocks_set(handle_t *handle,
4092 struct ext4_inode *raw_inode,
4093 struct ext4_inode_info *ei)
4095 struct inode *inode = &(ei->vfs_inode);
4096 u64 i_blocks = inode->i_blocks;
4097 struct super_block *sb = inode->i_sb;
4100 if (i_blocks <= ~0U) {
4102 * i_blocks can be represnted in a 32 bit variable
4103 * as multiple of 512 bytes
4105 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4106 raw_inode->i_blocks_high = 0;
4107 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4108 } else if (i_blocks <= 0xffffffffffffULL) {
4110 * i_blocks can be represented in a 48 bit variable
4111 * as multiple of 512 bytes
4113 err = ext4_update_rocompat_feature(handle, sb,
4114 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4117 /* i_block is stored in the split 48 bit fields */
4118 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4119 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4120 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4123 * i_blocks should be represented in a 48 bit variable
4124 * as multiple of file system block size
4126 err = ext4_update_rocompat_feature(handle, sb,
4127 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4130 ei->i_flags |= EXT4_HUGE_FILE_FL;
4131 /* i_block is stored in file system block size */
4132 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4133 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4134 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4141 * Post the struct inode info into an on-disk inode location in the
4142 * buffer-cache. This gobbles the caller's reference to the
4143 * buffer_head in the inode location struct.
4145 * The caller must have write access to iloc->bh.
4147 static int ext4_do_update_inode(handle_t *handle,
4148 struct inode *inode,
4149 struct ext4_iloc *iloc)
4151 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4152 struct ext4_inode_info *ei = EXT4_I(inode);
4153 struct buffer_head *bh = iloc->bh;
4154 int err = 0, rc, block;
4156 /* For fields not not tracking in the in-memory inode,
4157 * initialise them to zero for new inodes. */
4158 if (ei->i_state & EXT4_STATE_NEW)
4159 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4161 ext4_get_inode_flags(ei);
4162 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4163 if(!(test_opt(inode->i_sb, NO_UID32))) {
4164 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4165 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4167 * Fix up interoperability with old kernels. Otherwise, old inodes get
4168 * re-used with the upper 16 bits of the uid/gid intact
4171 raw_inode->i_uid_high =
4172 cpu_to_le16(high_16_bits(inode->i_uid));
4173 raw_inode->i_gid_high =
4174 cpu_to_le16(high_16_bits(inode->i_gid));
4176 raw_inode->i_uid_high = 0;
4177 raw_inode->i_gid_high = 0;
4180 raw_inode->i_uid_low =
4181 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4182 raw_inode->i_gid_low =
4183 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4184 raw_inode->i_uid_high = 0;
4185 raw_inode->i_gid_high = 0;
4187 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4189 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4190 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4191 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4192 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4194 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4196 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4197 /* clear the migrate flag in the raw_inode */
4198 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4199 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4200 cpu_to_le32(EXT4_OS_HURD))
4201 raw_inode->i_file_acl_high =
4202 cpu_to_le16(ei->i_file_acl >> 32);
4203 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4204 ext4_isize_set(raw_inode, ei->i_disksize);
4205 if (ei->i_disksize > 0x7fffffffULL) {
4206 struct super_block *sb = inode->i_sb;
4207 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4208 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4209 EXT4_SB(sb)->s_es->s_rev_level ==
4210 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4211 /* If this is the first large file
4212 * created, add a flag to the superblock.
4214 err = ext4_journal_get_write_access(handle,
4215 EXT4_SB(sb)->s_sbh);
4218 ext4_update_dynamic_rev(sb);
4219 EXT4_SET_RO_COMPAT_FEATURE(sb,
4220 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4223 err = ext4_journal_dirty_metadata(handle,
4224 EXT4_SB(sb)->s_sbh);
4227 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4228 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4229 if (old_valid_dev(inode->i_rdev)) {
4230 raw_inode->i_block[0] =
4231 cpu_to_le32(old_encode_dev(inode->i_rdev));
4232 raw_inode->i_block[1] = 0;
4234 raw_inode->i_block[0] = 0;
4235 raw_inode->i_block[1] =
4236 cpu_to_le32(new_encode_dev(inode->i_rdev));
4237 raw_inode->i_block[2] = 0;
4239 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4240 raw_inode->i_block[block] = ei->i_data[block];
4242 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4243 if (ei->i_extra_isize) {
4244 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4245 raw_inode->i_version_hi =
4246 cpu_to_le32(inode->i_version >> 32);
4247 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4251 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4252 rc = ext4_journal_dirty_metadata(handle, bh);
4255 ei->i_state &= ~EXT4_STATE_NEW;
4259 ext4_std_error(inode->i_sb, err);
4264 * ext4_write_inode()
4266 * We are called from a few places:
4268 * - Within generic_file_write() for O_SYNC files.
4269 * Here, there will be no transaction running. We wait for any running
4270 * trasnaction to commit.
4272 * - Within sys_sync(), kupdate and such.
4273 * We wait on commit, if tol to.
4275 * - Within prune_icache() (PF_MEMALLOC == true)
4276 * Here we simply return. We can't afford to block kswapd on the
4279 * In all cases it is actually safe for us to return without doing anything,
4280 * because the inode has been copied into a raw inode buffer in
4281 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4284 * Note that we are absolutely dependent upon all inode dirtiers doing the
4285 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4286 * which we are interested.
4288 * It would be a bug for them to not do this. The code:
4290 * mark_inode_dirty(inode)
4292 * inode->i_size = expr;
4294 * is in error because a kswapd-driven write_inode() could occur while
4295 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4296 * will no longer be on the superblock's dirty inode list.
4298 int ext4_write_inode(struct inode *inode, int wait)
4300 if (current->flags & PF_MEMALLOC)
4303 if (ext4_journal_current_handle()) {
4304 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4312 return ext4_force_commit(inode->i_sb);
4318 * Called from notify_change.
4320 * We want to trap VFS attempts to truncate the file as soon as
4321 * possible. In particular, we want to make sure that when the VFS
4322 * shrinks i_size, we put the inode on the orphan list and modify
4323 * i_disksize immediately, so that during the subsequent flushing of
4324 * dirty pages and freeing of disk blocks, we can guarantee that any
4325 * commit will leave the blocks being flushed in an unused state on
4326 * disk. (On recovery, the inode will get truncated and the blocks will
4327 * be freed, so we have a strong guarantee that no future commit will
4328 * leave these blocks visible to the user.)
4330 * Another thing we have to assure is that if we are in ordered mode
4331 * and inode is still attached to the committing transaction, we must
4332 * we start writeout of all the dirty pages which are being truncated.
4333 * This way we are sure that all the data written in the previous
4334 * transaction are already on disk (truncate waits for pages under
4337 * Called with inode->i_mutex down.
4339 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4341 struct inode *inode = dentry->d_inode;
4343 const unsigned int ia_valid = attr->ia_valid;
4345 error = inode_change_ok(inode, attr);
4349 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4350 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4353 /* (user+group)*(old+new) structure, inode write (sb,
4354 * inode block, ? - but truncate inode update has it) */
4355 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4356 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4357 if (IS_ERR(handle)) {
4358 error = PTR_ERR(handle);
4361 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4363 ext4_journal_stop(handle);
4366 /* Update corresponding info in inode so that everything is in
4367 * one transaction */
4368 if (attr->ia_valid & ATTR_UID)
4369 inode->i_uid = attr->ia_uid;
4370 if (attr->ia_valid & ATTR_GID)
4371 inode->i_gid = attr->ia_gid;
4372 error = ext4_mark_inode_dirty(handle, inode);
4373 ext4_journal_stop(handle);
4376 if (attr->ia_valid & ATTR_SIZE) {
4377 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4378 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4380 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4387 if (S_ISREG(inode->i_mode) &&
4388 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4391 handle = ext4_journal_start(inode, 3);
4392 if (IS_ERR(handle)) {
4393 error = PTR_ERR(handle);
4397 error = ext4_orphan_add(handle, inode);
4398 EXT4_I(inode)->i_disksize = attr->ia_size;
4399 rc = ext4_mark_inode_dirty(handle, inode);
4402 ext4_journal_stop(handle);
4404 if (ext4_should_order_data(inode)) {
4405 error = ext4_begin_ordered_truncate(inode,
4408 /* Do as much error cleanup as possible */
4409 handle = ext4_journal_start(inode, 3);
4410 if (IS_ERR(handle)) {
4411 ext4_orphan_del(NULL, inode);
4414 ext4_orphan_del(handle, inode);
4415 ext4_journal_stop(handle);
4421 rc = inode_setattr(inode, attr);
4423 /* If inode_setattr's call to ext4_truncate failed to get a
4424 * transaction handle at all, we need to clean up the in-core
4425 * orphan list manually. */
4427 ext4_orphan_del(NULL, inode);
4429 if (!rc && (ia_valid & ATTR_MODE))
4430 rc = ext4_acl_chmod(inode);
4433 ext4_std_error(inode->i_sb, error);
4439 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4442 struct inode *inode;
4443 unsigned long delalloc_blocks;
4445 inode = dentry->d_inode;
4446 generic_fillattr(inode, stat);
4449 * We can't update i_blocks if the block allocation is delayed
4450 * otherwise in the case of system crash before the real block
4451 * allocation is done, we will have i_blocks inconsistent with
4452 * on-disk file blocks.
4453 * We always keep i_blocks updated together with real
4454 * allocation. But to not confuse with user, stat
4455 * will return the blocks that include the delayed allocation
4456 * blocks for this file.
4458 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4459 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4460 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4462 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4466 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4471 /* if nrblocks are contiguous */
4474 * With N contiguous data blocks, it need at most
4475 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4476 * 2 dindirect blocks
4479 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4480 return indirects + 3;
4483 * if nrblocks are not contiguous, worse case, each block touch
4484 * a indirect block, and each indirect block touch a double indirect
4485 * block, plus a triple indirect block
4487 indirects = nrblocks * 2 + 1;
4491 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4493 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4494 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4495 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4499 * Account for index blocks, block groups bitmaps and block group
4500 * descriptor blocks if modify datablocks and index blocks
4501 * worse case, the indexs blocks spread over different block groups
4503 * If datablocks are discontiguous, they are possible to spread over
4504 * different block groups too. If they are contiugous, with flexbg,
4505 * they could still across block group boundary.
4507 * Also account for superblock, inode, quota and xattr blocks
4509 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4511 int groups, gdpblocks;
4516 * How many index blocks need to touch to modify nrblocks?
4517 * The "Chunk" flag indicating whether the nrblocks is
4518 * physically contiguous on disk
4520 * For Direct IO and fallocate, they calls get_block to allocate
4521 * one single extent at a time, so they could set the "Chunk" flag
4523 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4528 * Now let's see how many group bitmaps and group descriptors need
4538 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4539 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4540 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4541 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4543 /* bitmaps and block group descriptor blocks */
4544 ret += groups + gdpblocks;
4546 /* Blocks for super block, inode, quota and xattr blocks */
4547 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4553 * Calulate the total number of credits to reserve to fit
4554 * the modification of a single pages into a single transaction,
4555 * which may include multiple chunks of block allocations.
4557 * This could be called via ext4_write_begin()
4559 * We need to consider the worse case, when
4560 * one new block per extent.
4562 int ext4_writepage_trans_blocks(struct inode *inode)
4564 int bpp = ext4_journal_blocks_per_page(inode);
4567 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4569 /* Account for data blocks for journalled mode */
4570 if (ext4_should_journal_data(inode))
4576 * Calculate the journal credits for a chunk of data modification.
4578 * This is called from DIO, fallocate or whoever calling
4579 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4581 * journal buffers for data blocks are not included here, as DIO
4582 * and fallocate do no need to journal data buffers.
4584 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4586 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4590 * The caller must have previously called ext4_reserve_inode_write().
4591 * Give this, we know that the caller already has write access to iloc->bh.
4593 int ext4_mark_iloc_dirty(handle_t *handle,
4594 struct inode *inode, struct ext4_iloc *iloc)
4598 if (test_opt(inode->i_sb, I_VERSION))
4599 inode_inc_iversion(inode);
4601 /* the do_update_inode consumes one bh->b_count */
4604 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4605 err = ext4_do_update_inode(handle, inode, iloc);
4611 * On success, We end up with an outstanding reference count against
4612 * iloc->bh. This _must_ be cleaned up later.
4616 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4617 struct ext4_iloc *iloc)
4621 err = ext4_get_inode_loc(inode, iloc);
4623 BUFFER_TRACE(iloc->bh, "get_write_access");
4624 err = ext4_journal_get_write_access(handle, iloc->bh);
4631 ext4_std_error(inode->i_sb, err);
4636 * Expand an inode by new_extra_isize bytes.
4637 * Returns 0 on success or negative error number on failure.
4639 static int ext4_expand_extra_isize(struct inode *inode,
4640 unsigned int new_extra_isize,
4641 struct ext4_iloc iloc,
4644 struct ext4_inode *raw_inode;
4645 struct ext4_xattr_ibody_header *header;
4646 struct ext4_xattr_entry *entry;
4648 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4651 raw_inode = ext4_raw_inode(&iloc);
4653 header = IHDR(inode, raw_inode);
4654 entry = IFIRST(header);
4656 /* No extended attributes present */
4657 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4658 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4659 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4661 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4665 /* try to expand with EAs present */
4666 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4671 * What we do here is to mark the in-core inode as clean with respect to inode
4672 * dirtiness (it may still be data-dirty).
4673 * This means that the in-core inode may be reaped by prune_icache
4674 * without having to perform any I/O. This is a very good thing,
4675 * because *any* task may call prune_icache - even ones which
4676 * have a transaction open against a different journal.
4678 * Is this cheating? Not really. Sure, we haven't written the
4679 * inode out, but prune_icache isn't a user-visible syncing function.
4680 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4681 * we start and wait on commits.
4683 * Is this efficient/effective? Well, we're being nice to the system
4684 * by cleaning up our inodes proactively so they can be reaped
4685 * without I/O. But we are potentially leaving up to five seconds'
4686 * worth of inodes floating about which prune_icache wants us to
4687 * write out. One way to fix that would be to get prune_icache()
4688 * to do a write_super() to free up some memory. It has the desired
4691 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4693 struct ext4_iloc iloc;
4694 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4695 static unsigned int mnt_count;
4699 err = ext4_reserve_inode_write(handle, inode, &iloc);
4700 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4701 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4703 * We need extra buffer credits since we may write into EA block
4704 * with this same handle. If journal_extend fails, then it will
4705 * only result in a minor loss of functionality for that inode.
4706 * If this is felt to be critical, then e2fsck should be run to
4707 * force a large enough s_min_extra_isize.
4709 if ((jbd2_journal_extend(handle,
4710 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4711 ret = ext4_expand_extra_isize(inode,
4712 sbi->s_want_extra_isize,
4715 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4717 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4718 ext4_warning(inode->i_sb, __func__,
4719 "Unable to expand inode %lu. Delete"
4720 " some EAs or run e2fsck.",
4723 le16_to_cpu(sbi->s_es->s_mnt_count);
4729 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4734 * ext4_dirty_inode() is called from __mark_inode_dirty()
4736 * We're really interested in the case where a file is being extended.
4737 * i_size has been changed by generic_commit_write() and we thus need
4738 * to include the updated inode in the current transaction.
4740 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4741 * are allocated to the file.
4743 * If the inode is marked synchronous, we don't honour that here - doing
4744 * so would cause a commit on atime updates, which we don't bother doing.
4745 * We handle synchronous inodes at the highest possible level.
4747 void ext4_dirty_inode(struct inode *inode)
4749 handle_t *current_handle = ext4_journal_current_handle();
4752 handle = ext4_journal_start(inode, 2);
4755 if (current_handle &&
4756 current_handle->h_transaction != handle->h_transaction) {
4757 /* This task has a transaction open against a different fs */
4758 printk(KERN_EMERG "%s: transactions do not match!\n",
4761 jbd_debug(5, "marking dirty. outer handle=%p\n",
4763 ext4_mark_inode_dirty(handle, inode);
4765 ext4_journal_stop(handle);
4772 * Bind an inode's backing buffer_head into this transaction, to prevent
4773 * it from being flushed to disk early. Unlike
4774 * ext4_reserve_inode_write, this leaves behind no bh reference and
4775 * returns no iloc structure, so the caller needs to repeat the iloc
4776 * lookup to mark the inode dirty later.
4778 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4780 struct ext4_iloc iloc;
4784 err = ext4_get_inode_loc(inode, &iloc);
4786 BUFFER_TRACE(iloc.bh, "get_write_access");
4787 err = jbd2_journal_get_write_access(handle, iloc.bh);
4789 err = ext4_journal_dirty_metadata(handle,
4794 ext4_std_error(inode->i_sb, err);
4799 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4806 * We have to be very careful here: changing a data block's
4807 * journaling status dynamically is dangerous. If we write a
4808 * data block to the journal, change the status and then delete
4809 * that block, we risk forgetting to revoke the old log record
4810 * from the journal and so a subsequent replay can corrupt data.
4811 * So, first we make sure that the journal is empty and that
4812 * nobody is changing anything.
4815 journal = EXT4_JOURNAL(inode);
4816 if (is_journal_aborted(journal))
4819 jbd2_journal_lock_updates(journal);
4820 jbd2_journal_flush(journal);
4823 * OK, there are no updates running now, and all cached data is
4824 * synced to disk. We are now in a completely consistent state
4825 * which doesn't have anything in the journal, and we know that
4826 * no filesystem updates are running, so it is safe to modify
4827 * the inode's in-core data-journaling state flag now.
4831 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4833 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4834 ext4_set_aops(inode);
4836 jbd2_journal_unlock_updates(journal);
4838 /* Finally we can mark the inode as dirty. */
4840 handle = ext4_journal_start(inode, 1);
4842 return PTR_ERR(handle);
4844 err = ext4_mark_inode_dirty(handle, inode);
4846 ext4_journal_stop(handle);
4847 ext4_std_error(inode->i_sb, err);
4852 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4854 return !buffer_mapped(bh);
4857 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4862 struct file *file = vma->vm_file;
4863 struct inode *inode = file->f_path.dentry->d_inode;
4864 struct address_space *mapping = inode->i_mapping;
4867 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4868 * get i_mutex because we are already holding mmap_sem.
4870 down_read(&inode->i_alloc_sem);
4871 size = i_size_read(inode);
4872 if (page->mapping != mapping || size <= page_offset(page)
4873 || !PageUptodate(page)) {
4874 /* page got truncated from under us? */
4878 if (PageMappedToDisk(page))
4881 if (page->index == size >> PAGE_CACHE_SHIFT)
4882 len = size & ~PAGE_CACHE_MASK;
4884 len = PAGE_CACHE_SIZE;
4886 if (page_has_buffers(page)) {
4887 /* return if we have all the buffers mapped */
4888 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4893 * OK, we need to fill the hole... Do write_begin write_end
4894 * to do block allocation/reservation.We are not holding
4895 * inode.i__mutex here. That allow * parallel write_begin,
4896 * write_end call. lock_page prevent this from happening
4897 * on the same page though
4899 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4900 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4903 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4904 len, len, page, NULL);
4909 up_read(&inode->i_alloc_sem);