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(&EXT4_I(inode)->jinode,
53 static void ext4_invalidatepage(struct page *page, unsigned long offset);
56 * Test whether an inode is a fast symlink.
58 static int ext4_inode_is_fast_symlink(struct inode *inode)
60 int ea_blocks = EXT4_I(inode)->i_file_acl ?
61 (inode->i_sb->s_blocksize >> 9) : 0;
63 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
67 * The ext4 forget function must perform a revoke if we are freeing data
68 * which has been journaled. Metadata (eg. indirect blocks) must be
69 * revoked in all cases.
71 * "bh" may be NULL: a metadata block may have been freed from memory
72 * but there may still be a record of it in the journal, and that record
73 * still needs to be revoked.
75 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
76 struct buffer_head *bh, ext4_fsblk_t blocknr)
82 BUFFER_TRACE(bh, "enter");
84 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
86 bh, is_metadata, inode->i_mode,
87 test_opt(inode->i_sb, DATA_FLAGS));
89 /* Never use the revoke function if we are doing full data
90 * journaling: there is no need to, and a V1 superblock won't
91 * support it. Otherwise, only skip the revoke on un-journaled
94 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
95 (!is_metadata && !ext4_should_journal_data(inode))) {
97 BUFFER_TRACE(bh, "call jbd2_journal_forget");
98 return ext4_journal_forget(handle, bh);
104 * data!=journal && (is_metadata || should_journal_data(inode))
106 BUFFER_TRACE(bh, "call ext4_journal_revoke");
107 err = ext4_journal_revoke(handle, blocknr, bh);
109 ext4_abort(inode->i_sb, __func__,
110 "error %d when attempting revoke", err);
111 BUFFER_TRACE(bh, "exit");
116 * Work out how many blocks we need to proceed with the next chunk of a
117 * truncate transaction.
119 static unsigned long blocks_for_truncate(struct inode *inode)
123 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
125 /* Give ourselves just enough room to cope with inodes in which
126 * i_blocks is corrupt: we've seen disk corruptions in the past
127 * which resulted in random data in an inode which looked enough
128 * like a regular file for ext4 to try to delete it. Things
129 * will go a bit crazy if that happens, but at least we should
130 * try not to panic the whole kernel. */
134 /* But we need to bound the transaction so we don't overflow the
136 if (needed > EXT4_MAX_TRANS_DATA)
137 needed = EXT4_MAX_TRANS_DATA;
139 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
143 * Truncate transactions can be complex and absolutely huge. So we need to
144 * be able to restart the transaction at a conventient checkpoint to make
145 * sure we don't overflow the journal.
147 * start_transaction gets us a new handle for a truncate transaction,
148 * and extend_transaction tries to extend the existing one a bit. If
149 * extend fails, we need to propagate the failure up and restart the
150 * transaction in the top-level truncate loop. --sct
152 static handle_t *start_transaction(struct inode *inode)
156 result = ext4_journal_start(inode, blocks_for_truncate(inode));
160 ext4_std_error(inode->i_sb, PTR_ERR(result));
165 * Try to extend this transaction for the purposes of truncation.
167 * Returns 0 if we managed to create more room. If we can't create more
168 * room, and the transaction must be restarted we return 1.
170 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
172 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
174 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
180 * Restart the transaction associated with *handle. This does a commit,
181 * so before we call here everything must be consistently dirtied against
184 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
186 jbd_debug(2, "restarting handle %p\n", handle);
187 return ext4_journal_restart(handle, blocks_for_truncate(inode));
191 * Called at the last iput() if i_nlink is zero.
193 void ext4_delete_inode (struct inode * inode)
198 if (ext4_should_order_data(inode))
199 ext4_begin_ordered_truncate(inode, 0);
200 truncate_inode_pages(&inode->i_data, 0);
202 if (is_bad_inode(inode))
205 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
206 if (IS_ERR(handle)) {
207 ext4_std_error(inode->i_sb, PTR_ERR(handle));
209 * If we're going to skip the normal cleanup, we still need to
210 * make sure that the in-core orphan linked list is properly
213 ext4_orphan_del(NULL, inode);
220 err = ext4_mark_inode_dirty(handle, inode);
222 ext4_warning(inode->i_sb, __func__,
223 "couldn't mark inode dirty (err %d)", err);
227 ext4_truncate(inode);
230 * ext4_ext_truncate() doesn't reserve any slop when it
231 * restarts journal transactions; therefore there may not be
232 * enough credits left in the handle to remove the inode from
233 * the orphan list and set the dtime field.
235 if (handle->h_buffer_credits < 3) {
236 err = ext4_journal_extend(handle, 3);
238 err = ext4_journal_restart(handle, 3);
240 ext4_warning(inode->i_sb, __func__,
241 "couldn't extend journal (err %d)", err);
243 ext4_journal_stop(handle);
249 * Kill off the orphan record which ext4_truncate created.
250 * AKPM: I think this can be inside the above `if'.
251 * Note that ext4_orphan_del() has to be able to cope with the
252 * deletion of a non-existent orphan - this is because we don't
253 * know if ext4_truncate() actually created an orphan record.
254 * (Well, we could do this if we need to, but heck - it works)
256 ext4_orphan_del(handle, inode);
257 EXT4_I(inode)->i_dtime = get_seconds();
260 * One subtle ordering requirement: if anything has gone wrong
261 * (transaction abort, IO errors, whatever), then we can still
262 * do these next steps (the fs will already have been marked as
263 * having errors), but we can't free the inode if the mark_dirty
266 if (ext4_mark_inode_dirty(handle, inode))
267 /* If that failed, just do the required in-core inode clear. */
270 ext4_free_inode(handle, inode);
271 ext4_journal_stop(handle);
274 clear_inode(inode); /* We must guarantee clearing of inode... */
280 struct buffer_head *bh;
283 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
285 p->key = *(p->p = v);
290 * ext4_block_to_path - parse the block number into array of offsets
291 * @inode: inode in question (we are only interested in its superblock)
292 * @i_block: block number to be parsed
293 * @offsets: array to store the offsets in
294 * @boundary: set this non-zero if the referred-to block is likely to be
295 * followed (on disk) by an indirect block.
297 * To store the locations of file's data ext4 uses a data structure common
298 * for UNIX filesystems - tree of pointers anchored in the inode, with
299 * data blocks at leaves and indirect blocks in intermediate nodes.
300 * This function translates the block number into path in that tree -
301 * return value is the path length and @offsets[n] is the offset of
302 * pointer to (n+1)th node in the nth one. If @block is out of range
303 * (negative or too large) warning is printed and zero returned.
305 * Note: function doesn't find node addresses, so no IO is needed. All
306 * we need to know is the capacity of indirect blocks (taken from the
311 * Portability note: the last comparison (check that we fit into triple
312 * indirect block) is spelled differently, because otherwise on an
313 * architecture with 32-bit longs and 8Kb pages we might get into trouble
314 * if our filesystem had 8Kb blocks. We might use long long, but that would
315 * kill us on x86. Oh, well, at least the sign propagation does not matter -
316 * i_block would have to be negative in the very beginning, so we would not
320 static int ext4_block_to_path(struct inode *inode,
322 ext4_lblk_t offsets[4], int *boundary)
324 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
325 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
326 const long direct_blocks = EXT4_NDIR_BLOCKS,
327 indirect_blocks = ptrs,
328 double_blocks = (1 << (ptrs_bits * 2));
333 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
334 } else if (i_block < direct_blocks) {
335 offsets[n++] = i_block;
336 final = direct_blocks;
337 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
338 offsets[n++] = EXT4_IND_BLOCK;
339 offsets[n++] = i_block;
341 } else if ((i_block -= indirect_blocks) < double_blocks) {
342 offsets[n++] = EXT4_DIND_BLOCK;
343 offsets[n++] = i_block >> ptrs_bits;
344 offsets[n++] = i_block & (ptrs - 1);
346 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
347 offsets[n++] = EXT4_TIND_BLOCK;
348 offsets[n++] = i_block >> (ptrs_bits * 2);
349 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
350 offsets[n++] = i_block & (ptrs - 1);
353 ext4_warning(inode->i_sb, "ext4_block_to_path",
355 i_block + direct_blocks +
356 indirect_blocks + double_blocks);
359 *boundary = final - 1 - (i_block & (ptrs - 1));
364 * ext4_get_branch - read the chain of indirect blocks leading to data
365 * @inode: inode in question
366 * @depth: depth of the chain (1 - direct pointer, etc.)
367 * @offsets: offsets of pointers in inode/indirect blocks
368 * @chain: place to store the result
369 * @err: here we store the error value
371 * Function fills the array of triples <key, p, bh> and returns %NULL
372 * if everything went OK or the pointer to the last filled triple
373 * (incomplete one) otherwise. Upon the return chain[i].key contains
374 * the number of (i+1)-th block in the chain (as it is stored in memory,
375 * i.e. little-endian 32-bit), chain[i].p contains the address of that
376 * number (it points into struct inode for i==0 and into the bh->b_data
377 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
378 * block for i>0 and NULL for i==0. In other words, it holds the block
379 * numbers of the chain, addresses they were taken from (and where we can
380 * verify that chain did not change) and buffer_heads hosting these
383 * Function stops when it stumbles upon zero pointer (absent block)
384 * (pointer to last triple returned, *@err == 0)
385 * or when it gets an IO error reading an indirect block
386 * (ditto, *@err == -EIO)
387 * or when it reads all @depth-1 indirect blocks successfully and finds
388 * the whole chain, all way to the data (returns %NULL, *err == 0).
390 * Need to be called with
391 * down_read(&EXT4_I(inode)->i_data_sem)
393 static Indirect *ext4_get_branch(struct inode *inode, int depth,
394 ext4_lblk_t *offsets,
395 Indirect chain[4], int *err)
397 struct super_block *sb = inode->i_sb;
399 struct buffer_head *bh;
402 /* i_data is not going away, no lock needed */
403 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
407 bh = sb_bread(sb, le32_to_cpu(p->key));
410 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
424 * ext4_find_near - find a place for allocation with sufficient locality
426 * @ind: descriptor of indirect block.
428 * This function returns the preferred place for block allocation.
429 * It is used when heuristic for sequential allocation fails.
431 * + if there is a block to the left of our position - allocate near it.
432 * + if pointer will live in indirect block - allocate near that block.
433 * + if pointer will live in inode - allocate in the same
436 * In the latter case we colour the starting block by the callers PID to
437 * prevent it from clashing with concurrent allocations for a different inode
438 * in the same block group. The PID is used here so that functionally related
439 * files will be close-by on-disk.
441 * Caller must make sure that @ind is valid and will stay that way.
443 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
445 struct ext4_inode_info *ei = EXT4_I(inode);
446 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
448 ext4_fsblk_t bg_start;
449 ext4_fsblk_t last_block;
450 ext4_grpblk_t colour;
452 /* Try to find previous block */
453 for (p = ind->p - 1; p >= start; p--) {
455 return le32_to_cpu(*p);
458 /* No such thing, so let's try location of indirect block */
460 return ind->bh->b_blocknr;
463 * It is going to be referred to from the inode itself? OK, just put it
464 * into the same cylinder group then.
466 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
467 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
469 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
470 colour = (current->pid % 16) *
471 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
473 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
474 return bg_start + colour;
478 * ext4_find_goal - find a preferred place for allocation.
480 * @block: block we want
481 * @partial: pointer to the last triple within a chain
483 * Normally this function find the preferred place for block allocation,
486 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
489 struct ext4_block_alloc_info *block_i;
491 block_i = EXT4_I(inode)->i_block_alloc_info;
494 * try the heuristic for sequential allocation,
495 * failing that at least try to get decent locality.
497 if (block_i && (block == block_i->last_alloc_logical_block + 1)
498 && (block_i->last_alloc_physical_block != 0)) {
499 return block_i->last_alloc_physical_block + 1;
502 return ext4_find_near(inode, partial);
506 * ext4_blks_to_allocate: Look up the block map and count the number
507 * of direct blocks need to be allocated for the given branch.
509 * @branch: chain of indirect blocks
510 * @k: number of blocks need for indirect blocks
511 * @blks: number of data blocks to be mapped.
512 * @blocks_to_boundary: the offset in the indirect block
514 * return the total number of blocks to be allocate, including the
515 * direct and indirect blocks.
517 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
518 int blocks_to_boundary)
520 unsigned long count = 0;
523 * Simple case, [t,d]Indirect block(s) has not allocated yet
524 * then it's clear blocks on that path have not allocated
527 /* right now we don't handle cross boundary allocation */
528 if (blks < blocks_to_boundary + 1)
531 count += blocks_to_boundary + 1;
536 while (count < blks && count <= blocks_to_boundary &&
537 le32_to_cpu(*(branch[0].p + count)) == 0) {
544 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
545 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @new_blocks: on return it will store the new block numbers for
549 * the indirect blocks(if needed) and the first direct block,
550 * @blks: on return it will store the total number of allocated
553 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
554 ext4_lblk_t iblock, ext4_fsblk_t goal,
555 int indirect_blks, int blks,
556 ext4_fsblk_t new_blocks[4], int *err)
559 unsigned long count = 0, blk_allocated = 0;
561 ext4_fsblk_t current_block = 0;
565 * Here we try to allocate the requested multiple blocks at once,
566 * on a best-effort basis.
567 * To build a branch, we should allocate blocks for
568 * the indirect blocks(if not allocated yet), and at least
569 * the first direct block of this branch. That's the
570 * minimum number of blocks need to allocate(required)
572 /* first we try to allocate the indirect blocks */
573 target = indirect_blks;
576 /* allocating blocks for indirect blocks and direct blocks */
577 current_block = ext4_new_meta_blocks(handle, inode,
583 /* allocate blocks for indirect blocks */
584 while (index < indirect_blks && count) {
585 new_blocks[index++] = current_block++;
590 * save the new block number
591 * for the first direct block
593 new_blocks[index] = current_block;
594 printk(KERN_INFO "%s returned more blocks than "
595 "requested\n", __func__);
601 target = blks - count ;
602 blk_allocated = count;
605 /* Now allocate data blocks */
607 /* allocating blocks for data blocks */
608 current_block = ext4_new_blocks(handle, inode, iblock,
610 if (*err && (target == blks)) {
612 * if the allocation failed and we didn't allocate
618 if (target == blks) {
620 * save the new block number
621 * for the first direct block
623 new_blocks[index] = current_block;
625 blk_allocated += count;
628 /* total number of blocks allocated for direct blocks */
633 for (i = 0; i <index; i++)
634 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
639 * ext4_alloc_branch - allocate and set up a chain of blocks.
641 * @indirect_blks: number of allocated indirect blocks
642 * @blks: number of allocated direct blocks
643 * @offsets: offsets (in the blocks) to store the pointers to next.
644 * @branch: place to store the chain in.
646 * This function allocates blocks, zeroes out all but the last one,
647 * links them into chain and (if we are synchronous) writes them to disk.
648 * In other words, it prepares a branch that can be spliced onto the
649 * inode. It stores the information about that chain in the branch[], in
650 * the same format as ext4_get_branch() would do. We are calling it after
651 * we had read the existing part of chain and partial points to the last
652 * triple of that (one with zero ->key). Upon the exit we have the same
653 * picture as after the successful ext4_get_block(), except that in one
654 * place chain is disconnected - *branch->p is still zero (we did not
655 * set the last link), but branch->key contains the number that should
656 * be placed into *branch->p to fill that gap.
658 * If allocation fails we free all blocks we've allocated (and forget
659 * their buffer_heads) and return the error value the from failed
660 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
661 * as described above and return 0.
663 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
664 ext4_lblk_t iblock, int indirect_blks,
665 int *blks, ext4_fsblk_t goal,
666 ext4_lblk_t *offsets, Indirect *branch)
668 int blocksize = inode->i_sb->s_blocksize;
671 struct buffer_head *bh;
673 ext4_fsblk_t new_blocks[4];
674 ext4_fsblk_t current_block;
676 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
677 *blks, new_blocks, &err);
681 branch[0].key = cpu_to_le32(new_blocks[0]);
683 * metadata blocks and data blocks are allocated.
685 for (n = 1; n <= indirect_blks; n++) {
687 * Get buffer_head for parent block, zero it out
688 * and set the pointer to new one, then send
691 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
694 BUFFER_TRACE(bh, "call get_create_access");
695 err = ext4_journal_get_create_access(handle, bh);
702 memset(bh->b_data, 0, blocksize);
703 branch[n].p = (__le32 *) bh->b_data + offsets[n];
704 branch[n].key = cpu_to_le32(new_blocks[n]);
705 *branch[n].p = branch[n].key;
706 if ( n == indirect_blks) {
707 current_block = new_blocks[n];
709 * End of chain, update the last new metablock of
710 * the chain to point to the new allocated
711 * data blocks numbers
713 for (i=1; i < num; i++)
714 *(branch[n].p + i) = cpu_to_le32(++current_block);
716 BUFFER_TRACE(bh, "marking uptodate");
717 set_buffer_uptodate(bh);
720 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
721 err = ext4_journal_dirty_metadata(handle, bh);
728 /* Allocation failed, free what we already allocated */
729 for (i = 1; i <= n ; i++) {
730 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
731 ext4_journal_forget(handle, branch[i].bh);
733 for (i = 0; i <indirect_blks; i++)
734 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
736 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
742 * ext4_splice_branch - splice the allocated branch onto inode.
744 * @block: (logical) number of block we are adding
745 * @chain: chain of indirect blocks (with a missing link - see
747 * @where: location of missing link
748 * @num: number of indirect blocks we are adding
749 * @blks: number of direct blocks we are adding
751 * This function fills the missing link and does all housekeeping needed in
752 * inode (->i_blocks, etc.). In case of success we end up with the full
753 * chain to new block and return 0.
755 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
756 ext4_lblk_t block, Indirect *where, int num, int blks)
760 struct ext4_block_alloc_info *block_i;
761 ext4_fsblk_t current_block;
763 block_i = EXT4_I(inode)->i_block_alloc_info;
765 * If we're splicing into a [td]indirect block (as opposed to the
766 * inode) then we need to get write access to the [td]indirect block
770 BUFFER_TRACE(where->bh, "get_write_access");
771 err = ext4_journal_get_write_access(handle, where->bh);
777 *where->p = where->key;
780 * Update the host buffer_head or inode to point to more just allocated
781 * direct blocks blocks
783 if (num == 0 && blks > 1) {
784 current_block = le32_to_cpu(where->key) + 1;
785 for (i = 1; i < blks; i++)
786 *(where->p + i ) = cpu_to_le32(current_block++);
790 * update the most recently allocated logical & physical block
791 * in i_block_alloc_info, to assist find the proper goal block for next
795 block_i->last_alloc_logical_block = block + blks - 1;
796 block_i->last_alloc_physical_block =
797 le32_to_cpu(where[num].key) + blks - 1;
800 /* We are done with atomic stuff, now do the rest of housekeeping */
802 inode->i_ctime = ext4_current_time(inode);
803 ext4_mark_inode_dirty(handle, inode);
805 /* had we spliced it onto indirect block? */
808 * If we spliced it onto an indirect block, we haven't
809 * altered the inode. Note however that if it is being spliced
810 * onto an indirect block at the very end of the file (the
811 * file is growing) then we *will* alter the inode to reflect
812 * the new i_size. But that is not done here - it is done in
813 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
815 jbd_debug(5, "splicing indirect only\n");
816 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
817 err = ext4_journal_dirty_metadata(handle, where->bh);
822 * OK, we spliced it into the inode itself on a direct block.
823 * Inode was dirtied above.
825 jbd_debug(5, "splicing direct\n");
830 for (i = 1; i <= num; i++) {
831 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
832 ext4_journal_forget(handle, where[i].bh);
833 ext4_free_blocks(handle, inode,
834 le32_to_cpu(where[i-1].key), 1, 0);
836 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
842 * Allocation strategy is simple: if we have to allocate something, we will
843 * have to go the whole way to leaf. So let's do it before attaching anything
844 * to tree, set linkage between the newborn blocks, write them if sync is
845 * required, recheck the path, free and repeat if check fails, otherwise
846 * set the last missing link (that will protect us from any truncate-generated
847 * removals - all blocks on the path are immune now) and possibly force the
848 * write on the parent block.
849 * That has a nice additional property: no special recovery from the failed
850 * allocations is needed - we simply release blocks and do not touch anything
851 * reachable from inode.
853 * `handle' can be NULL if create == 0.
855 * return > 0, # of blocks mapped or allocated.
856 * return = 0, if plain lookup failed.
857 * return < 0, error case.
860 * Need to be called with
861 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
862 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
864 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
865 ext4_lblk_t iblock, unsigned long maxblocks,
866 struct buffer_head *bh_result,
867 int create, int extend_disksize)
870 ext4_lblk_t offsets[4];
875 int blocks_to_boundary = 0;
877 struct ext4_inode_info *ei = EXT4_I(inode);
879 ext4_fsblk_t first_block = 0;
883 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
884 J_ASSERT(handle != NULL || create == 0);
885 depth = ext4_block_to_path(inode, iblock, offsets,
886 &blocks_to_boundary);
891 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
893 /* Simplest case - block found, no allocation needed */
895 first_block = le32_to_cpu(chain[depth - 1].key);
896 clear_buffer_new(bh_result);
899 while (count < maxblocks && count <= blocks_to_boundary) {
902 blk = le32_to_cpu(*(chain[depth-1].p + count));
904 if (blk == first_block + count)
912 /* Next simple case - plain lookup or failed read of indirect block */
913 if (!create || err == -EIO)
917 * Okay, we need to do block allocation. Lazily initialize the block
918 * allocation info here if necessary
920 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
921 ext4_init_block_alloc_info(inode);
923 goal = ext4_find_goal(inode, iblock, partial);
925 /* the number of blocks need to allocate for [d,t]indirect blocks */
926 indirect_blks = (chain + depth) - partial - 1;
929 * Next look up the indirect map to count the totoal number of
930 * direct blocks to allocate for this branch.
932 count = ext4_blks_to_allocate(partial, indirect_blks,
933 maxblocks, blocks_to_boundary);
935 * Block out ext4_truncate while we alter the tree
937 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
939 offsets + (partial - chain), partial);
942 * The ext4_splice_branch call will free and forget any buffers
943 * on the new chain if there is a failure, but that risks using
944 * up transaction credits, especially for bitmaps where the
945 * credits cannot be returned. Can we handle this somehow? We
946 * may need to return -EAGAIN upwards in the worst case. --sct
949 err = ext4_splice_branch(handle, inode, iblock,
950 partial, indirect_blks, count);
952 * i_disksize growing is protected by i_data_sem. Don't forget to
953 * protect it if you're about to implement concurrent
954 * ext4_get_block() -bzzz
956 if (!err && extend_disksize) {
957 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
958 if (disksize > i_size_read(inode))
959 disksize = i_size_read(inode);
960 if (disksize > ei->i_disksize)
961 ei->i_disksize = disksize;
966 set_buffer_new(bh_result);
968 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
969 if (count > blocks_to_boundary)
970 set_buffer_boundary(bh_result);
972 /* Clean up and exit */
973 partial = chain + depth - 1; /* the whole chain */
975 while (partial > chain) {
976 BUFFER_TRACE(partial->bh, "call brelse");
980 BUFFER_TRACE(bh_result, "returned");
986 * Calculate the number of metadata blocks need to reserve
987 * to allocate @blocks for non extent file based file
989 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
991 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
992 int ind_blks, dind_blks, tind_blks;
994 /* number of new indirect blocks needed */
995 ind_blks = (blocks + icap - 1) / icap;
997 dind_blks = (ind_blks + icap - 1) / icap;
1001 return ind_blks + dind_blks + tind_blks;
1005 * Calculate the number of metadata blocks need to reserve
1006 * to allocate given number of blocks
1008 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1013 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1014 return ext4_ext_calc_metadata_amount(inode, blocks);
1016 return ext4_indirect_calc_metadata_amount(inode, blocks);
1019 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1021 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1022 int total, mdb, mdb_free;
1024 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1025 /* recalculate the number of metablocks still need to be reserved */
1026 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1027 mdb = ext4_calc_metadata_amount(inode, total);
1029 /* figure out how many metablocks to release */
1030 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1031 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1033 /* Account for allocated meta_blocks */
1034 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1036 /* update fs free blocks counter for truncate case */
1037 percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
1039 /* update per-inode reservations */
1040 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1041 EXT4_I(inode)->i_reserved_data_blocks -= used;
1043 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1044 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1045 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1046 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1050 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1051 * and returns if the blocks are already mapped.
1053 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1054 * and store the allocated blocks in the result buffer head and mark it
1057 * If file type is extents based, it will call ext4_ext_get_blocks(),
1058 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1061 * On success, it returns the number of blocks being mapped or allocate.
1062 * if create==0 and the blocks are pre-allocated and uninitialized block,
1063 * the result buffer head is unmapped. If the create ==1, it will make sure
1064 * the buffer head is mapped.
1066 * It returns 0 if plain look up failed (blocks have not been allocated), in
1067 * that casem, buffer head is unmapped
1069 * It returns the error in case of allocation failure.
1071 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1072 unsigned long max_blocks, struct buffer_head *bh,
1073 int create, int extend_disksize, int flag)
1077 clear_buffer_mapped(bh);
1080 * Try to see if we can get the block without requesting
1081 * for new file system block.
1083 down_read((&EXT4_I(inode)->i_data_sem));
1084 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1085 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1088 retval = ext4_get_blocks_handle(handle,
1089 inode, block, max_blocks, bh, 0, 0);
1091 up_read((&EXT4_I(inode)->i_data_sem));
1093 /* If it is only a block(s) look up */
1098 * Returns if the blocks have already allocated
1100 * Note that if blocks have been preallocated
1101 * ext4_ext_get_block() returns th create = 0
1102 * with buffer head unmapped.
1104 if (retval > 0 && buffer_mapped(bh))
1108 * New blocks allocate and/or writing to uninitialized extent
1109 * will possibly result in updating i_data, so we take
1110 * the write lock of i_data_sem, and call get_blocks()
1111 * with create == 1 flag.
1113 down_write((&EXT4_I(inode)->i_data_sem));
1116 * if the caller is from delayed allocation writeout path
1117 * we have already reserved fs blocks for allocation
1118 * let the underlying get_block() function know to
1119 * avoid double accounting
1122 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1124 * We need to check for EXT4 here because migrate
1125 * could have changed the inode type in between
1127 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1128 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1129 bh, create, extend_disksize);
1131 retval = ext4_get_blocks_handle(handle, inode, block,
1132 max_blocks, bh, create, extend_disksize);
1134 if (retval > 0 && buffer_new(bh)) {
1136 * We allocated new blocks which will result in
1137 * i_data's format changing. Force the migrate
1138 * to fail by clearing migrate flags
1140 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1146 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1148 * Update reserved blocks/metadata blocks
1149 * after successful block allocation
1150 * which were deferred till now
1152 if ((retval > 0) && buffer_delay(bh))
1153 ext4_da_update_reserve_space(inode, retval);
1156 up_write((&EXT4_I(inode)->i_data_sem));
1160 /* Maximum number of blocks we map for direct IO at once. */
1161 #define DIO_MAX_BLOCKS 4096
1163 static int ext4_get_block(struct inode *inode, sector_t iblock,
1164 struct buffer_head *bh_result, int create)
1166 handle_t *handle = ext4_journal_current_handle();
1167 int ret = 0, started = 0;
1168 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1171 if (create && !handle) {
1172 /* Direct IO write... */
1173 if (max_blocks > DIO_MAX_BLOCKS)
1174 max_blocks = DIO_MAX_BLOCKS;
1175 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1176 handle = ext4_journal_start(inode, dio_credits);
1177 if (IS_ERR(handle)) {
1178 ret = PTR_ERR(handle);
1184 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1185 max_blocks, bh_result, create, 0, 0);
1187 bh_result->b_size = (ret << inode->i_blkbits);
1191 ext4_journal_stop(handle);
1197 * `handle' can be NULL if create is zero
1199 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1200 ext4_lblk_t block, int create, int *errp)
1202 struct buffer_head dummy;
1205 J_ASSERT(handle != NULL || create == 0);
1208 dummy.b_blocknr = -1000;
1209 buffer_trace_init(&dummy.b_history);
1210 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1211 &dummy, create, 1, 0);
1213 * ext4_get_blocks_handle() returns number of blocks
1214 * mapped. 0 in case of a HOLE.
1222 if (!err && buffer_mapped(&dummy)) {
1223 struct buffer_head *bh;
1224 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1229 if (buffer_new(&dummy)) {
1230 J_ASSERT(create != 0);
1231 J_ASSERT(handle != NULL);
1234 * Now that we do not always journal data, we should
1235 * keep in mind whether this should always journal the
1236 * new buffer as metadata. For now, regular file
1237 * writes use ext4_get_block instead, so it's not a
1241 BUFFER_TRACE(bh, "call get_create_access");
1242 fatal = ext4_journal_get_create_access(handle, bh);
1243 if (!fatal && !buffer_uptodate(bh)) {
1244 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1245 set_buffer_uptodate(bh);
1248 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1249 err = ext4_journal_dirty_metadata(handle, bh);
1253 BUFFER_TRACE(bh, "not a new buffer");
1266 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1267 ext4_lblk_t block, int create, int *err)
1269 struct buffer_head * bh;
1271 bh = ext4_getblk(handle, inode, block, create, err);
1274 if (buffer_uptodate(bh))
1276 ll_rw_block(READ_META, 1, &bh);
1278 if (buffer_uptodate(bh))
1285 static int walk_page_buffers( handle_t *handle,
1286 struct buffer_head *head,
1290 int (*fn)( handle_t *handle,
1291 struct buffer_head *bh))
1293 struct buffer_head *bh;
1294 unsigned block_start, block_end;
1295 unsigned blocksize = head->b_size;
1297 struct buffer_head *next;
1299 for ( bh = head, block_start = 0;
1300 ret == 0 && (bh != head || !block_start);
1301 block_start = block_end, bh = next)
1303 next = bh->b_this_page;
1304 block_end = block_start + blocksize;
1305 if (block_end <= from || block_start >= to) {
1306 if (partial && !buffer_uptodate(bh))
1310 err = (*fn)(handle, bh);
1318 * To preserve ordering, it is essential that the hole instantiation and
1319 * the data write be encapsulated in a single transaction. We cannot
1320 * close off a transaction and start a new one between the ext4_get_block()
1321 * and the commit_write(). So doing the jbd2_journal_start at the start of
1322 * prepare_write() is the right place.
1324 * Also, this function can nest inside ext4_writepage() ->
1325 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1326 * has generated enough buffer credits to do the whole page. So we won't
1327 * block on the journal in that case, which is good, because the caller may
1330 * By accident, ext4 can be reentered when a transaction is open via
1331 * quota file writes. If we were to commit the transaction while thus
1332 * reentered, there can be a deadlock - we would be holding a quota
1333 * lock, and the commit would never complete if another thread had a
1334 * transaction open and was blocking on the quota lock - a ranking
1337 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1338 * will _not_ run commit under these circumstances because handle->h_ref
1339 * is elevated. We'll still have enough credits for the tiny quotafile
1342 static int do_journal_get_write_access(handle_t *handle,
1343 struct buffer_head *bh)
1345 if (!buffer_mapped(bh) || buffer_freed(bh))
1347 return ext4_journal_get_write_access(handle, bh);
1350 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1351 loff_t pos, unsigned len, unsigned flags,
1352 struct page **pagep, void **fsdata)
1354 struct inode *inode = mapping->host;
1355 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1362 index = pos >> PAGE_CACHE_SHIFT;
1363 from = pos & (PAGE_CACHE_SIZE - 1);
1367 handle = ext4_journal_start(inode, needed_blocks);
1368 if (IS_ERR(handle)) {
1369 ret = PTR_ERR(handle);
1373 page = __grab_cache_page(mapping, index);
1375 ext4_journal_stop(handle);
1381 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1384 if (!ret && ext4_should_journal_data(inode)) {
1385 ret = walk_page_buffers(handle, page_buffers(page),
1386 from, to, NULL, do_journal_get_write_access);
1391 ext4_journal_stop(handle);
1392 page_cache_release(page);
1395 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1401 /* For write_end() in data=journal mode */
1402 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1404 if (!buffer_mapped(bh) || buffer_freed(bh))
1406 set_buffer_uptodate(bh);
1407 return ext4_journal_dirty_metadata(handle, bh);
1411 * We need to pick up the new inode size which generic_commit_write gave us
1412 * `file' can be NULL - eg, when called from page_symlink().
1414 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1415 * buffers are managed internally.
1417 static int ext4_ordered_write_end(struct file *file,
1418 struct address_space *mapping,
1419 loff_t pos, unsigned len, unsigned copied,
1420 struct page *page, void *fsdata)
1422 handle_t *handle = ext4_journal_current_handle();
1423 struct inode *inode = mapping->host;
1426 ret = ext4_jbd2_file_inode(handle, inode);
1430 * generic_write_end() will run mark_inode_dirty() if i_size
1431 * changes. So let's piggyback the i_disksize mark_inode_dirty
1436 new_i_size = pos + copied;
1437 if (new_i_size > EXT4_I(inode)->i_disksize)
1438 EXT4_I(inode)->i_disksize = new_i_size;
1439 ret2 = generic_write_end(file, mapping, pos, len, copied,
1445 ret2 = ext4_journal_stop(handle);
1449 return ret ? ret : copied;
1452 static int ext4_writeback_write_end(struct file *file,
1453 struct address_space *mapping,
1454 loff_t pos, unsigned len, unsigned copied,
1455 struct page *page, void *fsdata)
1457 handle_t *handle = ext4_journal_current_handle();
1458 struct inode *inode = mapping->host;
1462 new_i_size = pos + copied;
1463 if (new_i_size > EXT4_I(inode)->i_disksize)
1464 EXT4_I(inode)->i_disksize = new_i_size;
1466 ret2 = generic_write_end(file, mapping, pos, len, copied,
1472 ret2 = ext4_journal_stop(handle);
1476 return ret ? ret : copied;
1479 static int ext4_journalled_write_end(struct file *file,
1480 struct address_space *mapping,
1481 loff_t pos, unsigned len, unsigned copied,
1482 struct page *page, void *fsdata)
1484 handle_t *handle = ext4_journal_current_handle();
1485 struct inode *inode = mapping->host;
1490 from = pos & (PAGE_CACHE_SIZE - 1);
1494 if (!PageUptodate(page))
1496 page_zero_new_buffers(page, from+copied, to);
1499 ret = walk_page_buffers(handle, page_buffers(page), from,
1500 to, &partial, write_end_fn);
1502 SetPageUptodate(page);
1503 if (pos+copied > inode->i_size)
1504 i_size_write(inode, pos+copied);
1505 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1506 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1507 EXT4_I(inode)->i_disksize = inode->i_size;
1508 ret2 = ext4_mark_inode_dirty(handle, inode);
1514 ret2 = ext4_journal_stop(handle);
1517 page_cache_release(page);
1519 return ret ? ret : copied;
1522 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1524 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1525 unsigned long md_needed, mdblocks, total = 0;
1528 * recalculate the amount of metadata blocks to reserve
1529 * in order to allocate nrblocks
1530 * worse case is one extent per block
1532 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1533 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1534 mdblocks = ext4_calc_metadata_amount(inode, total);
1535 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1537 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1538 total = md_needed + nrblocks;
1540 if (ext4_has_free_blocks(sbi, total) < total) {
1541 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1544 /* reduce fs free blocks counter */
1545 percpu_counter_sub(&sbi->s_freeblocks_counter, total);
1547 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1548 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1550 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1551 return 0; /* success */
1554 static void ext4_da_release_space(struct inode *inode, int to_free)
1556 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1557 int total, mdb, mdb_free, release;
1560 return; /* Nothing to release, exit */
1562 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1564 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1566 * if there is no reserved blocks, but we try to free some
1567 * then the counter is messed up somewhere.
1568 * but since this function is called from invalidate
1569 * page, it's harmless to return without any action
1571 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1572 "blocks for inode %lu, but there is no reserved "
1573 "data blocks\n", to_free, inode->i_ino);
1574 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1578 /* recalculate the number of metablocks still need to be reserved */
1579 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1580 mdb = ext4_calc_metadata_amount(inode, total);
1582 /* figure out how many metablocks to release */
1583 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1584 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1586 release = to_free + mdb_free;
1588 /* update fs free blocks counter for truncate case */
1589 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1591 /* update per-inode reservations */
1592 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1593 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1595 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1596 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1597 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1600 static void ext4_da_page_release_reservation(struct page *page,
1601 unsigned long offset)
1604 struct buffer_head *head, *bh;
1605 unsigned int curr_off = 0;
1607 head = page_buffers(page);
1610 unsigned int next_off = curr_off + bh->b_size;
1612 if ((offset <= curr_off) && (buffer_delay(bh))) {
1614 clear_buffer_delay(bh);
1616 curr_off = next_off;
1617 } while ((bh = bh->b_this_page) != head);
1618 ext4_da_release_space(page->mapping->host, to_release);
1622 * Delayed allocation stuff
1625 struct mpage_da_data {
1626 struct inode *inode;
1627 struct buffer_head lbh; /* extent of blocks */
1628 unsigned long first_page, next_page; /* extent of pages */
1629 get_block_t *get_block;
1630 struct writeback_control *wbc;
1636 * mpage_da_submit_io - walks through extent of pages and try to write
1637 * them with writepage() call back
1639 * @mpd->inode: inode
1640 * @mpd->first_page: first page of the extent
1641 * @mpd->next_page: page after the last page of the extent
1642 * @mpd->get_block: the filesystem's block mapper function
1644 * By the time mpage_da_submit_io() is called we expect all blocks
1645 * to be allocated. this may be wrong if allocation failed.
1647 * As pages are already locked by write_cache_pages(), we can't use it
1649 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1651 struct address_space *mapping = mpd->inode->i_mapping;
1652 int ret = 0, err, nr_pages, i;
1653 unsigned long index, end;
1654 struct pagevec pvec;
1656 BUG_ON(mpd->next_page <= mpd->first_page);
1657 pagevec_init(&pvec, 0);
1658 index = mpd->first_page;
1659 end = mpd->next_page - 1;
1661 while (index <= end) {
1662 /* XXX: optimize tail */
1663 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1666 for (i = 0; i < nr_pages; i++) {
1667 struct page *page = pvec.pages[i];
1669 index = page->index;
1674 err = mapping->a_ops->writepage(page, mpd->wbc);
1676 mpd->pages_written++;
1678 * In error case, we have to continue because
1679 * remaining pages are still locked
1680 * XXX: unlock and re-dirty them?
1685 pagevec_release(&pvec);
1691 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1693 * @mpd->inode - inode to walk through
1694 * @exbh->b_blocknr - first block on a disk
1695 * @exbh->b_size - amount of space in bytes
1696 * @logical - first logical block to start assignment with
1698 * the function goes through all passed space and put actual disk
1699 * block numbers into buffer heads, dropping BH_Delay
1701 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1702 struct buffer_head *exbh)
1704 struct inode *inode = mpd->inode;
1705 struct address_space *mapping = inode->i_mapping;
1706 int blocks = exbh->b_size >> inode->i_blkbits;
1707 sector_t pblock = exbh->b_blocknr, cur_logical;
1708 struct buffer_head *head, *bh;
1710 struct pagevec pvec;
1713 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1714 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1715 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1717 pagevec_init(&pvec, 0);
1719 while (index <= end) {
1720 /* XXX: optimize tail */
1721 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1724 for (i = 0; i < nr_pages; i++) {
1725 struct page *page = pvec.pages[i];
1727 index = page->index;
1732 BUG_ON(!PageLocked(page));
1733 BUG_ON(PageWriteback(page));
1734 BUG_ON(!page_has_buffers(page));
1736 bh = page_buffers(page);
1739 /* skip blocks out of the range */
1741 if (cur_logical >= logical)
1744 } while ((bh = bh->b_this_page) != head);
1747 if (cur_logical >= logical + blocks)
1749 if (buffer_delay(bh)) {
1750 bh->b_blocknr = pblock;
1751 clear_buffer_delay(bh);
1752 bh->b_bdev = inode->i_sb->s_bdev;
1753 } else if (buffer_unwritten(bh)) {
1754 bh->b_blocknr = pblock;
1755 clear_buffer_unwritten(bh);
1756 set_buffer_mapped(bh);
1758 bh->b_bdev = inode->i_sb->s_bdev;
1759 } else if (buffer_mapped(bh))
1760 BUG_ON(bh->b_blocknr != pblock);
1764 } while ((bh = bh->b_this_page) != head);
1766 pagevec_release(&pvec);
1772 * __unmap_underlying_blocks - just a helper function to unmap
1773 * set of blocks described by @bh
1775 static inline void __unmap_underlying_blocks(struct inode *inode,
1776 struct buffer_head *bh)
1778 struct block_device *bdev = inode->i_sb->s_bdev;
1781 blocks = bh->b_size >> inode->i_blkbits;
1782 for (i = 0; i < blocks; i++)
1783 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1787 * mpage_da_map_blocks - go through given space
1789 * @mpd->lbh - bh describing space
1790 * @mpd->get_block - the filesystem's block mapper function
1792 * The function skips space we know is already mapped to disk blocks.
1795 static void mpage_da_map_blocks(struct mpage_da_data *mpd)
1798 struct buffer_head *lbh = &mpd->lbh;
1799 sector_t next = lbh->b_blocknr;
1800 struct buffer_head new;
1803 * We consider only non-mapped and non-allocated blocks
1805 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1808 new.b_state = lbh->b_state;
1810 new.b_size = lbh->b_size;
1813 * If we didn't accumulate anything
1814 * to write simply return
1818 err = mpd->get_block(mpd->inode, next, &new, 1);
1821 BUG_ON(new.b_size == 0);
1823 if (buffer_new(&new))
1824 __unmap_underlying_blocks(mpd->inode, &new);
1827 * If blocks are delayed marked, we need to
1828 * put actual blocknr and drop delayed bit
1830 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1831 mpage_put_bnr_to_bhs(mpd, next, &new);
1836 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1837 (1 << BH_Delay) | (1 << BH_Unwritten))
1840 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1842 * @mpd->lbh - extent of blocks
1843 * @logical - logical number of the block in the file
1844 * @bh - bh of the block (used to access block's state)
1846 * the function is used to collect contig. blocks in same state
1848 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1849 sector_t logical, struct buffer_head *bh)
1852 size_t b_size = bh->b_size;
1853 struct buffer_head *lbh = &mpd->lbh;
1854 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1856 /* check if thereserved journal credits might overflow */
1857 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1858 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1860 * With non-extent format we are limited by the journal
1861 * credit available. Total credit needed to insert
1862 * nrblocks contiguous blocks is dependent on the
1863 * nrblocks. So limit nrblocks.
1866 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1867 EXT4_MAX_TRANS_DATA) {
1869 * Adding the new buffer_head would make it cross the
1870 * allowed limit for which we have journal credit
1871 * reserved. So limit the new bh->b_size
1873 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1874 mpd->inode->i_blkbits;
1875 /* we will do mpage_da_submit_io in the next loop */
1879 * First block in the extent
1881 if (lbh->b_size == 0) {
1882 lbh->b_blocknr = logical;
1883 lbh->b_size = b_size;
1884 lbh->b_state = bh->b_state & BH_FLAGS;
1888 next = lbh->b_blocknr + nrblocks;
1890 * Can we merge the block to our big extent?
1892 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1893 lbh->b_size += b_size;
1899 * We couldn't merge the block to our extent, so we
1900 * need to flush current extent and start new one
1902 mpage_da_map_blocks(mpd);
1903 mpage_da_submit_io(mpd);
1909 * __mpage_da_writepage - finds extent of pages and blocks
1911 * @page: page to consider
1912 * @wbc: not used, we just follow rules
1915 * The function finds extents of pages and scan them for all blocks.
1917 static int __mpage_da_writepage(struct page *page,
1918 struct writeback_control *wbc, void *data)
1920 struct mpage_da_data *mpd = data;
1921 struct inode *inode = mpd->inode;
1922 struct buffer_head *bh, *head, fake;
1927 * Rest of the page in the page_vec
1928 * redirty then and skip then. We will
1929 * try to to write them again after
1930 * starting a new transaction
1932 redirty_page_for_writepage(wbc, page);
1934 return MPAGE_DA_EXTENT_TAIL;
1937 * Can we merge this page to current extent?
1939 if (mpd->next_page != page->index) {
1941 * Nope, we can't. So, we map non-allocated blocks
1942 * and start IO on them using writepage()
1944 if (mpd->next_page != mpd->first_page) {
1945 mpage_da_map_blocks(mpd);
1946 mpage_da_submit_io(mpd);
1948 * skip rest of the page in the page_vec
1951 redirty_page_for_writepage(wbc, page);
1953 return MPAGE_DA_EXTENT_TAIL;
1957 * Start next extent of pages ...
1959 mpd->first_page = page->index;
1964 mpd->lbh.b_size = 0;
1965 mpd->lbh.b_state = 0;
1966 mpd->lbh.b_blocknr = 0;
1969 mpd->next_page = page->index + 1;
1970 logical = (sector_t) page->index <<
1971 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1973 if (!page_has_buffers(page)) {
1975 * There is no attached buffer heads yet (mmap?)
1976 * we treat the page asfull of dirty blocks
1979 bh->b_size = PAGE_CACHE_SIZE;
1981 set_buffer_dirty(bh);
1982 set_buffer_uptodate(bh);
1983 mpage_add_bh_to_extent(mpd, logical, bh);
1985 return MPAGE_DA_EXTENT_TAIL;
1988 * Page with regular buffer heads, just add all dirty ones
1990 head = page_buffers(page);
1993 BUG_ON(buffer_locked(bh));
1994 if (buffer_dirty(bh) &&
1995 (!buffer_mapped(bh) || buffer_delay(bh))) {
1996 mpage_add_bh_to_extent(mpd, logical, bh);
1998 return MPAGE_DA_EXTENT_TAIL;
2001 } while ((bh = bh->b_this_page) != head);
2008 * mpage_da_writepages - walk the list of dirty pages of the given
2009 * address space, allocates non-allocated blocks, maps newly-allocated
2010 * blocks to existing bhs and issue IO them
2012 * @mapping: address space structure to write
2013 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2014 * @get_block: the filesystem's block mapper function.
2016 * This is a library function, which implements the writepages()
2017 * address_space_operation.
2019 static int mpage_da_writepages(struct address_space *mapping,
2020 struct writeback_control *wbc,
2021 get_block_t get_block)
2023 struct mpage_da_data mpd;
2028 return generic_writepages(mapping, wbc);
2031 mpd.inode = mapping->host;
2033 mpd.lbh.b_state = 0;
2034 mpd.lbh.b_blocknr = 0;
2037 mpd.get_block = get_block;
2039 mpd.pages_written = 0;
2041 to_write = wbc->nr_to_write;
2043 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
2046 * Handle last extent of pages
2048 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2049 mpage_da_map_blocks(&mpd);
2050 mpage_da_submit_io(&mpd);
2053 wbc->nr_to_write = to_write - mpd.pages_written;
2058 * this is a special callback for ->write_begin() only
2059 * it's intention is to return mapped block or reserve space
2061 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2062 struct buffer_head *bh_result, int create)
2066 BUG_ON(create == 0);
2067 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2070 * first, we need to know whether the block is allocated already
2071 * preallocated blocks are unmapped but should treated
2072 * the same as allocated blocks.
2074 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2075 if ((ret == 0) && !buffer_delay(bh_result)) {
2076 /* the block isn't (pre)allocated yet, let's reserve space */
2078 * XXX: __block_prepare_write() unmaps passed block,
2081 ret = ext4_da_reserve_space(inode, 1);
2083 /* not enough space to reserve */
2086 map_bh(bh_result, inode->i_sb, 0);
2087 set_buffer_new(bh_result);
2088 set_buffer_delay(bh_result);
2089 } else if (ret > 0) {
2090 bh_result->b_size = (ret << inode->i_blkbits);
2096 #define EXT4_DELALLOC_RSVED 1
2097 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2098 struct buffer_head *bh_result, int create)
2101 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2102 loff_t disksize = EXT4_I(inode)->i_disksize;
2103 handle_t *handle = NULL;
2105 handle = ext4_journal_current_handle();
2107 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2108 bh_result, 0, 0, 0);
2111 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2112 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2116 bh_result->b_size = (ret << inode->i_blkbits);
2119 * Update on-disk size along with block allocation
2120 * we don't use 'extend_disksize' as size may change
2121 * within already allocated block -bzzz
2123 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2124 if (disksize > i_size_read(inode))
2125 disksize = i_size_read(inode);
2126 if (disksize > EXT4_I(inode)->i_disksize) {
2128 * XXX: replace with spinlock if seen contended -bzzz
2130 down_write(&EXT4_I(inode)->i_data_sem);
2131 if (disksize > EXT4_I(inode)->i_disksize)
2132 EXT4_I(inode)->i_disksize = disksize;
2133 up_write(&EXT4_I(inode)->i_data_sem);
2135 if (EXT4_I(inode)->i_disksize == disksize) {
2136 ret = ext4_mark_inode_dirty(handle, inode);
2145 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2148 * unmapped buffer is possible for holes.
2149 * delay buffer is possible with delayed allocation
2151 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2154 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2155 struct buffer_head *bh_result, int create)
2158 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2161 * we don't want to do block allocation in writepage
2162 * so call get_block_wrap with create = 0
2164 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2165 bh_result, 0, 0, 0);
2167 bh_result->b_size = (ret << inode->i_blkbits);
2174 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2175 * get called via journal_submit_inode_data_buffers (no journal handle)
2176 * get called via shrink_page_list via pdflush (no journal handle)
2177 * or grab_page_cache when doing write_begin (have journal handle)
2179 static int ext4_da_writepage(struct page *page,
2180 struct writeback_control *wbc)
2185 struct buffer_head *page_bufs;
2186 struct inode *inode = page->mapping->host;
2188 size = i_size_read(inode);
2189 if (page->index == size >> PAGE_CACHE_SHIFT)
2190 len = size & ~PAGE_CACHE_MASK;
2192 len = PAGE_CACHE_SIZE;
2194 if (page_has_buffers(page)) {
2195 page_bufs = page_buffers(page);
2196 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2197 ext4_bh_unmapped_or_delay)) {
2199 * We don't want to do block allocation
2200 * So redirty the page and return
2201 * We may reach here when we do a journal commit
2202 * via journal_submit_inode_data_buffers.
2203 * If we don't have mapping block we just ignore
2204 * them. We can also reach here via shrink_page_list
2206 redirty_page_for_writepage(wbc, page);
2212 * The test for page_has_buffers() is subtle:
2213 * We know the page is dirty but it lost buffers. That means
2214 * that at some moment in time after write_begin()/write_end()
2215 * has been called all buffers have been clean and thus they
2216 * must have been written at least once. So they are all
2217 * mapped and we can happily proceed with mapping them
2218 * and writing the page.
2220 * Try to initialize the buffer_heads and check whether
2221 * all are mapped and non delay. We don't want to
2222 * do block allocation here.
2224 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2225 ext4_normal_get_block_write);
2227 page_bufs = page_buffers(page);
2228 /* check whether all are mapped and non delay */
2229 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2230 ext4_bh_unmapped_or_delay)) {
2231 redirty_page_for_writepage(wbc, page);
2237 * We can't do block allocation here
2238 * so just redity the page and unlock
2241 redirty_page_for_writepage(wbc, page);
2247 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2248 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2250 ret = block_write_full_page(page,
2251 ext4_normal_get_block_write,
2258 * This is called via ext4_da_writepages() to
2259 * calulate the total number of credits to reserve to fit
2260 * a single extent allocation into a single transaction,
2261 * ext4_da_writpeages() will loop calling this before
2262 * the block allocation.
2265 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2267 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2270 * With non-extent format the journal credit needed to
2271 * insert nrblocks contiguous block is dependent on
2272 * number of contiguous block. So we will limit
2273 * number of contiguous block to a sane value
2275 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2276 (max_blocks > EXT4_MAX_TRANS_DATA))
2277 max_blocks = EXT4_MAX_TRANS_DATA;
2279 return ext4_chunk_trans_blocks(inode, max_blocks);
2282 static int ext4_da_writepages(struct address_space *mapping,
2283 struct writeback_control *wbc)
2285 struct inode *inode = mapping->host;
2286 handle_t *handle = NULL;
2290 loff_t range_start = 0;
2291 long pages_skipped = 0;
2294 * No pages to write? This is mainly a kludge to avoid starting
2295 * a transaction for special inodes like journal inode on last iput()
2296 * because that could violate lock ordering on umount
2298 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2301 if (!wbc->range_cyclic)
2303 * If range_cyclic is not set force range_cont
2304 * and save the old writeback_index
2306 wbc->range_cont = 1;
2308 range_start = wbc->range_start;
2309 pages_skipped = wbc->pages_skipped;
2312 to_write = wbc->nr_to_write;
2313 while (!ret && to_write > 0) {
2316 * we insert one extent at a time. So we need
2317 * credit needed for single extent allocation.
2318 * journalled mode is currently not supported
2321 BUG_ON(ext4_should_journal_data(inode));
2322 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2324 /* start a new transaction*/
2325 handle = ext4_journal_start(inode, needed_blocks);
2326 if (IS_ERR(handle)) {
2327 ret = PTR_ERR(handle);
2328 printk(KERN_EMERG "%s: jbd2_start: "
2329 "%ld pages, ino %lu; err %d\n", __func__,
2330 wbc->nr_to_write, inode->i_ino, ret);
2332 goto out_writepages;
2334 if (ext4_should_order_data(inode)) {
2336 * With ordered mode we need to add
2337 * the inode to the journal handl
2338 * when we do block allocation.
2340 ret = ext4_jbd2_file_inode(handle, inode);
2342 ext4_journal_stop(handle);
2343 goto out_writepages;
2347 to_write -= wbc->nr_to_write;
2348 ret = mpage_da_writepages(mapping, wbc,
2349 ext4_da_get_block_write);
2350 ext4_journal_stop(handle);
2351 if (ret == MPAGE_DA_EXTENT_TAIL) {
2353 * got one extent now try with
2356 to_write += wbc->nr_to_write;
2358 } else if (wbc->nr_to_write) {
2360 * There is no more writeout needed
2361 * or we requested for a noblocking writeout
2362 * and we found the device congested
2364 to_write += wbc->nr_to_write;
2367 wbc->nr_to_write = to_write;
2370 if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
2371 /* We skipped pages in this loop */
2372 wbc->range_start = range_start;
2373 wbc->nr_to_write = to_write +
2374 wbc->pages_skipped - pages_skipped;
2375 wbc->pages_skipped = pages_skipped;
2380 wbc->nr_to_write = to_write;
2381 wbc->range_start = range_start;
2385 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2386 loff_t pos, unsigned len, unsigned flags,
2387 struct page **pagep, void **fsdata)
2389 int ret, retries = 0;
2393 struct inode *inode = mapping->host;
2396 index = pos >> PAGE_CACHE_SHIFT;
2397 from = pos & (PAGE_CACHE_SIZE - 1);
2402 * With delayed allocation, we don't log the i_disksize update
2403 * if there is delayed block allocation. But we still need
2404 * to journalling the i_disksize update if writes to the end
2405 * of file which has an already mapped buffer.
2407 handle = ext4_journal_start(inode, 1);
2408 if (IS_ERR(handle)) {
2409 ret = PTR_ERR(handle);
2413 page = __grab_cache_page(mapping, index);
2415 ext4_journal_stop(handle);
2421 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2422 ext4_da_get_block_prep);
2425 ext4_journal_stop(handle);
2426 page_cache_release(page);
2429 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2436 * Check if we should update i_disksize
2437 * when write to the end of file but not require block allocation
2439 static int ext4_da_should_update_i_disksize(struct page *page,
2440 unsigned long offset)
2442 struct buffer_head *bh;
2443 struct inode *inode = page->mapping->host;
2447 bh = page_buffers(page);
2448 idx = offset >> inode->i_blkbits;
2450 for (i=0; i < idx; i++)
2451 bh = bh->b_this_page;
2453 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2458 static int ext4_da_write_end(struct file *file,
2459 struct address_space *mapping,
2460 loff_t pos, unsigned len, unsigned copied,
2461 struct page *page, void *fsdata)
2463 struct inode *inode = mapping->host;
2465 handle_t *handle = ext4_journal_current_handle();
2467 unsigned long start, end;
2469 start = pos & (PAGE_CACHE_SIZE - 1);
2470 end = start + copied -1;
2473 * generic_write_end() will run mark_inode_dirty() if i_size
2474 * changes. So let's piggyback the i_disksize mark_inode_dirty
2478 new_i_size = pos + copied;
2479 if (new_i_size > EXT4_I(inode)->i_disksize) {
2480 if (ext4_da_should_update_i_disksize(page, end)) {
2481 down_write(&EXT4_I(inode)->i_data_sem);
2482 if (new_i_size > EXT4_I(inode)->i_disksize) {
2484 * Updating i_disksize when extending file
2485 * without needing block allocation
2487 if (ext4_should_order_data(inode))
2488 ret = ext4_jbd2_file_inode(handle,
2491 EXT4_I(inode)->i_disksize = new_i_size;
2493 up_write(&EXT4_I(inode)->i_data_sem);
2496 ret2 = generic_write_end(file, mapping, pos, len, copied,
2501 ret2 = ext4_journal_stop(handle);
2505 return ret ? ret : copied;
2508 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2511 * Drop reserved blocks
2513 BUG_ON(!PageLocked(page));
2514 if (!page_has_buffers(page))
2517 ext4_da_page_release_reservation(page, offset);
2520 ext4_invalidatepage(page, offset);
2527 * bmap() is special. It gets used by applications such as lilo and by
2528 * the swapper to find the on-disk block of a specific piece of data.
2530 * Naturally, this is dangerous if the block concerned is still in the
2531 * journal. If somebody makes a swapfile on an ext4 data-journaling
2532 * filesystem and enables swap, then they may get a nasty shock when the
2533 * data getting swapped to that swapfile suddenly gets overwritten by
2534 * the original zero's written out previously to the journal and
2535 * awaiting writeback in the kernel's buffer cache.
2537 * So, if we see any bmap calls here on a modified, data-journaled file,
2538 * take extra steps to flush any blocks which might be in the cache.
2540 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2542 struct inode *inode = mapping->host;
2546 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2547 test_opt(inode->i_sb, DELALLOC)) {
2549 * With delalloc we want to sync the file
2550 * so that we can make sure we allocate
2553 filemap_write_and_wait(mapping);
2556 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2558 * This is a REALLY heavyweight approach, but the use of
2559 * bmap on dirty files is expected to be extremely rare:
2560 * only if we run lilo or swapon on a freshly made file
2561 * do we expect this to happen.
2563 * (bmap requires CAP_SYS_RAWIO so this does not
2564 * represent an unprivileged user DOS attack --- we'd be
2565 * in trouble if mortal users could trigger this path at
2568 * NB. EXT4_STATE_JDATA is not set on files other than
2569 * regular files. If somebody wants to bmap a directory
2570 * or symlink and gets confused because the buffer
2571 * hasn't yet been flushed to disk, they deserve
2572 * everything they get.
2575 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2576 journal = EXT4_JOURNAL(inode);
2577 jbd2_journal_lock_updates(journal);
2578 err = jbd2_journal_flush(journal);
2579 jbd2_journal_unlock_updates(journal);
2585 return generic_block_bmap(mapping,block,ext4_get_block);
2588 static int bget_one(handle_t *handle, struct buffer_head *bh)
2594 static int bput_one(handle_t *handle, struct buffer_head *bh)
2601 * Note that we don't need to start a transaction unless we're journaling data
2602 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2603 * need to file the inode to the transaction's list in ordered mode because if
2604 * we are writing back data added by write(), the inode is already there and if
2605 * we are writing back data modified via mmap(), noone guarantees in which
2606 * transaction the data will hit the disk. In case we are journaling data, we
2607 * cannot start transaction directly because transaction start ranks above page
2608 * lock so we have to do some magic.
2610 * In all journaling modes block_write_full_page() will start the I/O.
2614 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2619 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2621 * Same applies to ext4_get_block(). We will deadlock on various things like
2622 * lock_journal and i_data_sem
2624 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2627 * 16May01: If we're reentered then journal_current_handle() will be
2628 * non-zero. We simply *return*.
2630 * 1 July 2001: @@@ FIXME:
2631 * In journalled data mode, a data buffer may be metadata against the
2632 * current transaction. But the same file is part of a shared mapping
2633 * and someone does a writepage() on it.
2635 * We will move the buffer onto the async_data list, but *after* it has
2636 * been dirtied. So there's a small window where we have dirty data on
2639 * Note that this only applies to the last partial page in the file. The
2640 * bit which block_write_full_page() uses prepare/commit for. (That's
2641 * broken code anyway: it's wrong for msync()).
2643 * It's a rare case: affects the final partial page, for journalled data
2644 * where the file is subject to bith write() and writepage() in the same
2645 * transction. To fix it we'll need a custom block_write_full_page().
2646 * We'll probably need that anyway for journalling writepage() output.
2648 * We don't honour synchronous mounts for writepage(). That would be
2649 * disastrous. Any write() or metadata operation will sync the fs for
2653 static int __ext4_normal_writepage(struct page *page,
2654 struct writeback_control *wbc)
2656 struct inode *inode = page->mapping->host;
2658 if (test_opt(inode->i_sb, NOBH))
2659 return nobh_writepage(page,
2660 ext4_normal_get_block_write, wbc);
2662 return block_write_full_page(page,
2663 ext4_normal_get_block_write,
2667 static int ext4_normal_writepage(struct page *page,
2668 struct writeback_control *wbc)
2670 struct inode *inode = page->mapping->host;
2671 loff_t size = i_size_read(inode);
2674 J_ASSERT(PageLocked(page));
2675 if (page->index == size >> PAGE_CACHE_SHIFT)
2676 len = size & ~PAGE_CACHE_MASK;
2678 len = PAGE_CACHE_SIZE;
2680 if (page_has_buffers(page)) {
2681 /* if page has buffers it should all be mapped
2682 * and allocated. If there are not buffers attached
2683 * to the page we know the page is dirty but it lost
2684 * buffers. That means that at some moment in time
2685 * after write_begin() / write_end() has been called
2686 * all buffers have been clean and thus they must have been
2687 * written at least once. So they are all mapped and we can
2688 * happily proceed with mapping them and writing the page.
2690 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2691 ext4_bh_unmapped_or_delay));
2694 if (!ext4_journal_current_handle())
2695 return __ext4_normal_writepage(page, wbc);
2697 redirty_page_for_writepage(wbc, page);
2702 static int __ext4_journalled_writepage(struct page *page,
2703 struct writeback_control *wbc)
2705 struct address_space *mapping = page->mapping;
2706 struct inode *inode = mapping->host;
2707 struct buffer_head *page_bufs;
2708 handle_t *handle = NULL;
2712 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2713 ext4_normal_get_block_write);
2717 page_bufs = page_buffers(page);
2718 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2720 /* As soon as we unlock the page, it can go away, but we have
2721 * references to buffers so we are safe */
2724 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2725 if (IS_ERR(handle)) {
2726 ret = PTR_ERR(handle);
2730 ret = walk_page_buffers(handle, page_bufs, 0,
2731 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2733 err = walk_page_buffers(handle, page_bufs, 0,
2734 PAGE_CACHE_SIZE, NULL, write_end_fn);
2737 err = ext4_journal_stop(handle);
2741 walk_page_buffers(handle, page_bufs, 0,
2742 PAGE_CACHE_SIZE, NULL, bput_one);
2743 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2752 static int ext4_journalled_writepage(struct page *page,
2753 struct writeback_control *wbc)
2755 struct inode *inode = page->mapping->host;
2756 loff_t size = i_size_read(inode);
2759 J_ASSERT(PageLocked(page));
2760 if (page->index == size >> PAGE_CACHE_SHIFT)
2761 len = size & ~PAGE_CACHE_MASK;
2763 len = PAGE_CACHE_SIZE;
2765 if (page_has_buffers(page)) {
2766 /* if page has buffers it should all be mapped
2767 * and allocated. If there are not buffers attached
2768 * to the page we know the page is dirty but it lost
2769 * buffers. That means that at some moment in time
2770 * after write_begin() / write_end() has been called
2771 * all buffers have been clean and thus they must have been
2772 * written at least once. So they are all mapped and we can
2773 * happily proceed with mapping them and writing the page.
2775 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2776 ext4_bh_unmapped_or_delay));
2779 if (ext4_journal_current_handle())
2782 if (PageChecked(page)) {
2784 * It's mmapped pagecache. Add buffers and journal it. There
2785 * doesn't seem much point in redirtying the page here.
2787 ClearPageChecked(page);
2788 return __ext4_journalled_writepage(page, wbc);
2791 * It may be a page full of checkpoint-mode buffers. We don't
2792 * really know unless we go poke around in the buffer_heads.
2793 * But block_write_full_page will do the right thing.
2795 return block_write_full_page(page,
2796 ext4_normal_get_block_write,
2800 redirty_page_for_writepage(wbc, page);
2805 static int ext4_readpage(struct file *file, struct page *page)
2807 return mpage_readpage(page, ext4_get_block);
2811 ext4_readpages(struct file *file, struct address_space *mapping,
2812 struct list_head *pages, unsigned nr_pages)
2814 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2817 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2819 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2822 * If it's a full truncate we just forget about the pending dirtying
2825 ClearPageChecked(page);
2827 jbd2_journal_invalidatepage(journal, page, offset);
2830 static int ext4_releasepage(struct page *page, gfp_t wait)
2832 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2834 WARN_ON(PageChecked(page));
2835 if (!page_has_buffers(page))
2837 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2841 * If the O_DIRECT write will extend the file then add this inode to the
2842 * orphan list. So recovery will truncate it back to the original size
2843 * if the machine crashes during the write.
2845 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2846 * crashes then stale disk data _may_ be exposed inside the file. But current
2847 * VFS code falls back into buffered path in that case so we are safe.
2849 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2850 const struct iovec *iov, loff_t offset,
2851 unsigned long nr_segs)
2853 struct file *file = iocb->ki_filp;
2854 struct inode *inode = file->f_mapping->host;
2855 struct ext4_inode_info *ei = EXT4_I(inode);
2859 size_t count = iov_length(iov, nr_segs);
2862 loff_t final_size = offset + count;
2864 if (final_size > inode->i_size) {
2865 /* Credits for sb + inode write */
2866 handle = ext4_journal_start(inode, 2);
2867 if (IS_ERR(handle)) {
2868 ret = PTR_ERR(handle);
2871 ret = ext4_orphan_add(handle, inode);
2873 ext4_journal_stop(handle);
2877 ei->i_disksize = inode->i_size;
2878 ext4_journal_stop(handle);
2882 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2884 ext4_get_block, NULL);
2889 /* Credits for sb + inode write */
2890 handle = ext4_journal_start(inode, 2);
2891 if (IS_ERR(handle)) {
2892 /* This is really bad luck. We've written the data
2893 * but cannot extend i_size. Bail out and pretend
2894 * the write failed... */
2895 ret = PTR_ERR(handle);
2899 ext4_orphan_del(handle, inode);
2901 loff_t end = offset + ret;
2902 if (end > inode->i_size) {
2903 ei->i_disksize = end;
2904 i_size_write(inode, end);
2906 * We're going to return a positive `ret'
2907 * here due to non-zero-length I/O, so there's
2908 * no way of reporting error returns from
2909 * ext4_mark_inode_dirty() to userspace. So
2912 ext4_mark_inode_dirty(handle, inode);
2915 err = ext4_journal_stop(handle);
2924 * Pages can be marked dirty completely asynchronously from ext4's journalling
2925 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2926 * much here because ->set_page_dirty is called under VFS locks. The page is
2927 * not necessarily locked.
2929 * We cannot just dirty the page and leave attached buffers clean, because the
2930 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2931 * or jbddirty because all the journalling code will explode.
2933 * So what we do is to mark the page "pending dirty" and next time writepage
2934 * is called, propagate that into the buffers appropriately.
2936 static int ext4_journalled_set_page_dirty(struct page *page)
2938 SetPageChecked(page);
2939 return __set_page_dirty_nobuffers(page);
2942 static const struct address_space_operations ext4_ordered_aops = {
2943 .readpage = ext4_readpage,
2944 .readpages = ext4_readpages,
2945 .writepage = ext4_normal_writepage,
2946 .sync_page = block_sync_page,
2947 .write_begin = ext4_write_begin,
2948 .write_end = ext4_ordered_write_end,
2950 .invalidatepage = ext4_invalidatepage,
2951 .releasepage = ext4_releasepage,
2952 .direct_IO = ext4_direct_IO,
2953 .migratepage = buffer_migrate_page,
2954 .is_partially_uptodate = block_is_partially_uptodate,
2957 static const struct address_space_operations ext4_writeback_aops = {
2958 .readpage = ext4_readpage,
2959 .readpages = ext4_readpages,
2960 .writepage = ext4_normal_writepage,
2961 .sync_page = block_sync_page,
2962 .write_begin = ext4_write_begin,
2963 .write_end = ext4_writeback_write_end,
2965 .invalidatepage = ext4_invalidatepage,
2966 .releasepage = ext4_releasepage,
2967 .direct_IO = ext4_direct_IO,
2968 .migratepage = buffer_migrate_page,
2969 .is_partially_uptodate = block_is_partially_uptodate,
2972 static const struct address_space_operations ext4_journalled_aops = {
2973 .readpage = ext4_readpage,
2974 .readpages = ext4_readpages,
2975 .writepage = ext4_journalled_writepage,
2976 .sync_page = block_sync_page,
2977 .write_begin = ext4_write_begin,
2978 .write_end = ext4_journalled_write_end,
2979 .set_page_dirty = ext4_journalled_set_page_dirty,
2981 .invalidatepage = ext4_invalidatepage,
2982 .releasepage = ext4_releasepage,
2983 .is_partially_uptodate = block_is_partially_uptodate,
2986 static const struct address_space_operations ext4_da_aops = {
2987 .readpage = ext4_readpage,
2988 .readpages = ext4_readpages,
2989 .writepage = ext4_da_writepage,
2990 .writepages = ext4_da_writepages,
2991 .sync_page = block_sync_page,
2992 .write_begin = ext4_da_write_begin,
2993 .write_end = ext4_da_write_end,
2995 .invalidatepage = ext4_da_invalidatepage,
2996 .releasepage = ext4_releasepage,
2997 .direct_IO = ext4_direct_IO,
2998 .migratepage = buffer_migrate_page,
2999 .is_partially_uptodate = block_is_partially_uptodate,
3002 void ext4_set_aops(struct inode *inode)
3004 if (ext4_should_order_data(inode) &&
3005 test_opt(inode->i_sb, DELALLOC))
3006 inode->i_mapping->a_ops = &ext4_da_aops;
3007 else if (ext4_should_order_data(inode))
3008 inode->i_mapping->a_ops = &ext4_ordered_aops;
3009 else if (ext4_should_writeback_data(inode) &&
3010 test_opt(inode->i_sb, DELALLOC))
3011 inode->i_mapping->a_ops = &ext4_da_aops;
3012 else if (ext4_should_writeback_data(inode))
3013 inode->i_mapping->a_ops = &ext4_writeback_aops;
3015 inode->i_mapping->a_ops = &ext4_journalled_aops;
3019 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3020 * up to the end of the block which corresponds to `from'.
3021 * This required during truncate. We need to physically zero the tail end
3022 * of that block so it doesn't yield old data if the file is later grown.
3024 int ext4_block_truncate_page(handle_t *handle,
3025 struct address_space *mapping, loff_t from)
3027 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3028 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3029 unsigned blocksize, length, pos;
3031 struct inode *inode = mapping->host;
3032 struct buffer_head *bh;
3036 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3040 blocksize = inode->i_sb->s_blocksize;
3041 length = blocksize - (offset & (blocksize - 1));
3042 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3045 * For "nobh" option, we can only work if we don't need to
3046 * read-in the page - otherwise we create buffers to do the IO.
3048 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3049 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3050 zero_user(page, offset, length);
3051 set_page_dirty(page);
3055 if (!page_has_buffers(page))
3056 create_empty_buffers(page, blocksize, 0);
3058 /* Find the buffer that contains "offset" */
3059 bh = page_buffers(page);
3061 while (offset >= pos) {
3062 bh = bh->b_this_page;
3068 if (buffer_freed(bh)) {
3069 BUFFER_TRACE(bh, "freed: skip");
3073 if (!buffer_mapped(bh)) {
3074 BUFFER_TRACE(bh, "unmapped");
3075 ext4_get_block(inode, iblock, bh, 0);
3076 /* unmapped? It's a hole - nothing to do */
3077 if (!buffer_mapped(bh)) {
3078 BUFFER_TRACE(bh, "still unmapped");
3083 /* Ok, it's mapped. Make sure it's up-to-date */
3084 if (PageUptodate(page))
3085 set_buffer_uptodate(bh);
3087 if (!buffer_uptodate(bh)) {
3089 ll_rw_block(READ, 1, &bh);
3091 /* Uhhuh. Read error. Complain and punt. */
3092 if (!buffer_uptodate(bh))
3096 if (ext4_should_journal_data(inode)) {
3097 BUFFER_TRACE(bh, "get write access");
3098 err = ext4_journal_get_write_access(handle, bh);
3103 zero_user(page, offset, length);
3105 BUFFER_TRACE(bh, "zeroed end of block");
3108 if (ext4_should_journal_data(inode)) {
3109 err = ext4_journal_dirty_metadata(handle, bh);
3111 if (ext4_should_order_data(inode))
3112 err = ext4_jbd2_file_inode(handle, inode);
3113 mark_buffer_dirty(bh);
3118 page_cache_release(page);
3123 * Probably it should be a library function... search for first non-zero word
3124 * or memcmp with zero_page, whatever is better for particular architecture.
3127 static inline int all_zeroes(__le32 *p, __le32 *q)
3136 * ext4_find_shared - find the indirect blocks for partial truncation.
3137 * @inode: inode in question
3138 * @depth: depth of the affected branch
3139 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3140 * @chain: place to store the pointers to partial indirect blocks
3141 * @top: place to the (detached) top of branch
3143 * This is a helper function used by ext4_truncate().
3145 * When we do truncate() we may have to clean the ends of several
3146 * indirect blocks but leave the blocks themselves alive. Block is
3147 * partially truncated if some data below the new i_size is refered
3148 * from it (and it is on the path to the first completely truncated
3149 * data block, indeed). We have to free the top of that path along
3150 * with everything to the right of the path. Since no allocation
3151 * past the truncation point is possible until ext4_truncate()
3152 * finishes, we may safely do the latter, but top of branch may
3153 * require special attention - pageout below the truncation point
3154 * might try to populate it.
3156 * We atomically detach the top of branch from the tree, store the
3157 * block number of its root in *@top, pointers to buffer_heads of
3158 * partially truncated blocks - in @chain[].bh and pointers to
3159 * their last elements that should not be removed - in
3160 * @chain[].p. Return value is the pointer to last filled element
3163 * The work left to caller to do the actual freeing of subtrees:
3164 * a) free the subtree starting from *@top
3165 * b) free the subtrees whose roots are stored in
3166 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3167 * c) free the subtrees growing from the inode past the @chain[0].
3168 * (no partially truncated stuff there). */
3170 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3171 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3173 Indirect *partial, *p;
3177 /* Make k index the deepest non-null offest + 1 */
3178 for (k = depth; k > 1 && !offsets[k-1]; k--)
3180 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3181 /* Writer: pointers */
3183 partial = chain + k-1;
3185 * If the branch acquired continuation since we've looked at it -
3186 * fine, it should all survive and (new) top doesn't belong to us.
3188 if (!partial->key && *partial->p)
3191 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
3194 * OK, we've found the last block that must survive. The rest of our
3195 * branch should be detached before unlocking. However, if that rest
3196 * of branch is all ours and does not grow immediately from the inode
3197 * it's easier to cheat and just decrement partial->p.
3199 if (p == chain + k - 1 && p > chain) {
3203 /* Nope, don't do this in ext4. Must leave the tree intact */
3210 while(partial > p) {
3211 brelse(partial->bh);
3219 * Zero a number of block pointers in either an inode or an indirect block.
3220 * If we restart the transaction we must again get write access to the
3221 * indirect block for further modification.
3223 * We release `count' blocks on disk, but (last - first) may be greater
3224 * than `count' because there can be holes in there.
3226 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3227 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3228 unsigned long count, __le32 *first, __le32 *last)
3231 if (try_to_extend_transaction(handle, inode)) {
3233 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3234 ext4_journal_dirty_metadata(handle, bh);
3236 ext4_mark_inode_dirty(handle, inode);
3237 ext4_journal_test_restart(handle, inode);
3239 BUFFER_TRACE(bh, "retaking write access");
3240 ext4_journal_get_write_access(handle, bh);
3245 * Any buffers which are on the journal will be in memory. We find
3246 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3247 * on them. We've already detached each block from the file, so
3248 * bforget() in jbd2_journal_forget() should be safe.
3250 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3252 for (p = first; p < last; p++) {
3253 u32 nr = le32_to_cpu(*p);
3255 struct buffer_head *tbh;
3258 tbh = sb_find_get_block(inode->i_sb, nr);
3259 ext4_forget(handle, 0, inode, tbh, nr);
3263 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3267 * ext4_free_data - free a list of data blocks
3268 * @handle: handle for this transaction
3269 * @inode: inode we are dealing with
3270 * @this_bh: indirect buffer_head which contains *@first and *@last
3271 * @first: array of block numbers
3272 * @last: points immediately past the end of array
3274 * We are freeing all blocks refered from that array (numbers are stored as
3275 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3277 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3278 * blocks are contiguous then releasing them at one time will only affect one
3279 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3280 * actually use a lot of journal space.
3282 * @this_bh will be %NULL if @first and @last point into the inode's direct
3285 static void ext4_free_data(handle_t *handle, struct inode *inode,
3286 struct buffer_head *this_bh,
3287 __le32 *first, __le32 *last)
3289 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3290 unsigned long count = 0; /* Number of blocks in the run */
3291 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3294 ext4_fsblk_t nr; /* Current block # */
3295 __le32 *p; /* Pointer into inode/ind
3296 for current block */
3299 if (this_bh) { /* For indirect block */
3300 BUFFER_TRACE(this_bh, "get_write_access");
3301 err = ext4_journal_get_write_access(handle, this_bh);
3302 /* Important: if we can't update the indirect pointers
3303 * to the blocks, we can't free them. */
3308 for (p = first; p < last; p++) {
3309 nr = le32_to_cpu(*p);
3311 /* accumulate blocks to free if they're contiguous */
3314 block_to_free_p = p;
3316 } else if (nr == block_to_free + count) {
3319 ext4_clear_blocks(handle, inode, this_bh,
3321 count, block_to_free_p, p);
3323 block_to_free_p = p;
3330 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3331 count, block_to_free_p, p);
3334 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3337 * The buffer head should have an attached journal head at this
3338 * point. However, if the data is corrupted and an indirect
3339 * block pointed to itself, it would have been detached when
3340 * the block was cleared. Check for this instead of OOPSing.
3343 ext4_journal_dirty_metadata(handle, this_bh);
3345 ext4_error(inode->i_sb, __func__,
3346 "circular indirect block detected, "
3347 "inode=%lu, block=%llu",
3349 (unsigned long long) this_bh->b_blocknr);
3354 * ext4_free_branches - free an array of branches
3355 * @handle: JBD handle for this transaction
3356 * @inode: inode we are dealing with
3357 * @parent_bh: the buffer_head which contains *@first and *@last
3358 * @first: array of block numbers
3359 * @last: pointer immediately past the end of array
3360 * @depth: depth of the branches to free
3362 * We are freeing all blocks refered from these branches (numbers are
3363 * stored as little-endian 32-bit) and updating @inode->i_blocks
3366 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3367 struct buffer_head *parent_bh,
3368 __le32 *first, __le32 *last, int depth)
3373 if (is_handle_aborted(handle))
3377 struct buffer_head *bh;
3378 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3380 while (--p >= first) {
3381 nr = le32_to_cpu(*p);
3383 continue; /* A hole */
3385 /* Go read the buffer for the next level down */
3386 bh = sb_bread(inode->i_sb, nr);
3389 * A read failure? Report error and clear slot
3393 ext4_error(inode->i_sb, "ext4_free_branches",
3394 "Read failure, inode=%lu, block=%llu",
3399 /* This zaps the entire block. Bottom up. */
3400 BUFFER_TRACE(bh, "free child branches");
3401 ext4_free_branches(handle, inode, bh,
3402 (__le32*)bh->b_data,
3403 (__le32*)bh->b_data + addr_per_block,
3407 * We've probably journalled the indirect block several
3408 * times during the truncate. But it's no longer
3409 * needed and we now drop it from the transaction via
3410 * jbd2_journal_revoke().
3412 * That's easy if it's exclusively part of this
3413 * transaction. But if it's part of the committing
3414 * transaction then jbd2_journal_forget() will simply
3415 * brelse() it. That means that if the underlying
3416 * block is reallocated in ext4_get_block(),
3417 * unmap_underlying_metadata() will find this block
3418 * and will try to get rid of it. damn, damn.
3420 * If this block has already been committed to the
3421 * journal, a revoke record will be written. And
3422 * revoke records must be emitted *before* clearing
3423 * this block's bit in the bitmaps.
3425 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3428 * Everything below this this pointer has been
3429 * released. Now let this top-of-subtree go.
3431 * We want the freeing of this indirect block to be
3432 * atomic in the journal with the updating of the
3433 * bitmap block which owns it. So make some room in
3436 * We zero the parent pointer *after* freeing its
3437 * pointee in the bitmaps, so if extend_transaction()
3438 * for some reason fails to put the bitmap changes and
3439 * the release into the same transaction, recovery
3440 * will merely complain about releasing a free block,
3441 * rather than leaking blocks.
3443 if (is_handle_aborted(handle))
3445 if (try_to_extend_transaction(handle, inode)) {
3446 ext4_mark_inode_dirty(handle, inode);
3447 ext4_journal_test_restart(handle, inode);
3450 ext4_free_blocks(handle, inode, nr, 1, 1);
3454 * The block which we have just freed is
3455 * pointed to by an indirect block: journal it
3457 BUFFER_TRACE(parent_bh, "get_write_access");
3458 if (!ext4_journal_get_write_access(handle,
3461 BUFFER_TRACE(parent_bh,
3462 "call ext4_journal_dirty_metadata");
3463 ext4_journal_dirty_metadata(handle,
3469 /* We have reached the bottom of the tree. */
3470 BUFFER_TRACE(parent_bh, "free data blocks");
3471 ext4_free_data(handle, inode, parent_bh, first, last);
3475 int ext4_can_truncate(struct inode *inode)
3477 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3479 if (S_ISREG(inode->i_mode))
3481 if (S_ISDIR(inode->i_mode))
3483 if (S_ISLNK(inode->i_mode))
3484 return !ext4_inode_is_fast_symlink(inode);
3491 * We block out ext4_get_block() block instantiations across the entire
3492 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3493 * simultaneously on behalf of the same inode.
3495 * As we work through the truncate and commmit bits of it to the journal there
3496 * is one core, guiding principle: the file's tree must always be consistent on
3497 * disk. We must be able to restart the truncate after a crash.
3499 * The file's tree may be transiently inconsistent in memory (although it
3500 * probably isn't), but whenever we close off and commit a journal transaction,
3501 * the contents of (the filesystem + the journal) must be consistent and
3502 * restartable. It's pretty simple, really: bottom up, right to left (although
3503 * left-to-right works OK too).
3505 * Note that at recovery time, journal replay occurs *before* the restart of
3506 * truncate against the orphan inode list.
3508 * The committed inode has the new, desired i_size (which is the same as
3509 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3510 * that this inode's truncate did not complete and it will again call
3511 * ext4_truncate() to have another go. So there will be instantiated blocks
3512 * to the right of the truncation point in a crashed ext4 filesystem. But
3513 * that's fine - as long as they are linked from the inode, the post-crash
3514 * ext4_truncate() run will find them and release them.
3516 void ext4_truncate(struct inode *inode)
3519 struct ext4_inode_info *ei = EXT4_I(inode);
3520 __le32 *i_data = ei->i_data;
3521 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3522 struct address_space *mapping = inode->i_mapping;
3523 ext4_lblk_t offsets[4];
3528 ext4_lblk_t last_block;
3529 unsigned blocksize = inode->i_sb->s_blocksize;
3531 if (!ext4_can_truncate(inode))
3534 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3535 ext4_ext_truncate(inode);
3539 handle = start_transaction(inode);
3541 return; /* AKPM: return what? */
3543 last_block = (inode->i_size + blocksize-1)
3544 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3546 if (inode->i_size & (blocksize - 1))
3547 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3550 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3552 goto out_stop; /* error */
3555 * OK. This truncate is going to happen. We add the inode to the
3556 * orphan list, so that if this truncate spans multiple transactions,
3557 * and we crash, we will resume the truncate when the filesystem
3558 * recovers. It also marks the inode dirty, to catch the new size.
3560 * Implication: the file must always be in a sane, consistent
3561 * truncatable state while each transaction commits.
3563 if (ext4_orphan_add(handle, inode))
3567 * From here we block out all ext4_get_block() callers who want to
3568 * modify the block allocation tree.
3570 down_write(&ei->i_data_sem);
3572 ext4_discard_reservation(inode);
3575 * The orphan list entry will now protect us from any crash which
3576 * occurs before the truncate completes, so it is now safe to propagate
3577 * the new, shorter inode size (held for now in i_size) into the
3578 * on-disk inode. We do this via i_disksize, which is the value which
3579 * ext4 *really* writes onto the disk inode.
3581 ei->i_disksize = inode->i_size;
3583 if (n == 1) { /* direct blocks */
3584 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3585 i_data + EXT4_NDIR_BLOCKS);
3589 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3590 /* Kill the top of shared branch (not detached) */
3592 if (partial == chain) {
3593 /* Shared branch grows from the inode */
3594 ext4_free_branches(handle, inode, NULL,
3595 &nr, &nr+1, (chain+n-1) - partial);
3598 * We mark the inode dirty prior to restart,
3599 * and prior to stop. No need for it here.
3602 /* Shared branch grows from an indirect block */
3603 BUFFER_TRACE(partial->bh, "get_write_access");
3604 ext4_free_branches(handle, inode, partial->bh,
3606 partial->p+1, (chain+n-1) - partial);
3609 /* Clear the ends of indirect blocks on the shared branch */
3610 while (partial > chain) {
3611 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3612 (__le32*)partial->bh->b_data+addr_per_block,
3613 (chain+n-1) - partial);
3614 BUFFER_TRACE(partial->bh, "call brelse");
3615 brelse (partial->bh);
3619 /* Kill the remaining (whole) subtrees */
3620 switch (offsets[0]) {
3622 nr = i_data[EXT4_IND_BLOCK];
3624 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3625 i_data[EXT4_IND_BLOCK] = 0;
3627 case EXT4_IND_BLOCK:
3628 nr = i_data[EXT4_DIND_BLOCK];
3630 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3631 i_data[EXT4_DIND_BLOCK] = 0;
3633 case EXT4_DIND_BLOCK:
3634 nr = i_data[EXT4_TIND_BLOCK];
3636 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3637 i_data[EXT4_TIND_BLOCK] = 0;
3639 case EXT4_TIND_BLOCK:
3643 up_write(&ei->i_data_sem);
3644 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3645 ext4_mark_inode_dirty(handle, inode);
3648 * In a multi-transaction truncate, we only make the final transaction
3655 * If this was a simple ftruncate(), and the file will remain alive
3656 * then we need to clear up the orphan record which we created above.
3657 * However, if this was a real unlink then we were called by
3658 * ext4_delete_inode(), and we allow that function to clean up the
3659 * orphan info for us.
3662 ext4_orphan_del(handle, inode);
3664 ext4_journal_stop(handle);
3667 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3668 unsigned long ino, struct ext4_iloc *iloc)
3670 ext4_group_t block_group;
3671 unsigned long offset;
3673 struct ext4_group_desc *gdp;
3675 if (!ext4_valid_inum(sb, ino)) {
3677 * This error is already checked for in namei.c unless we are
3678 * looking at an NFS filehandle, in which case no error
3684 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3685 gdp = ext4_get_group_desc(sb, block_group, NULL);
3690 * Figure out the offset within the block group inode table
3692 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3693 EXT4_INODE_SIZE(sb);
3694 block = ext4_inode_table(sb, gdp) +
3695 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3697 iloc->block_group = block_group;
3698 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3703 * ext4_get_inode_loc returns with an extra refcount against the inode's
3704 * underlying buffer_head on success. If 'in_mem' is true, we have all
3705 * data in memory that is needed to recreate the on-disk version of this
3708 static int __ext4_get_inode_loc(struct inode *inode,
3709 struct ext4_iloc *iloc, int in_mem)
3712 struct buffer_head *bh;
3714 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3718 bh = sb_getblk(inode->i_sb, block);
3720 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3721 "unable to read inode block - "
3722 "inode=%lu, block=%llu",
3723 inode->i_ino, block);
3726 if (!buffer_uptodate(bh)) {
3730 * If the buffer has the write error flag, we have failed
3731 * to write out another inode in the same block. In this
3732 * case, we don't have to read the block because we may
3733 * read the old inode data successfully.
3735 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3736 set_buffer_uptodate(bh);
3738 if (buffer_uptodate(bh)) {
3739 /* someone brought it uptodate while we waited */
3745 * If we have all information of the inode in memory and this
3746 * is the only valid inode in the block, we need not read the
3750 struct buffer_head *bitmap_bh;
3751 struct ext4_group_desc *desc;
3752 int inodes_per_buffer;
3753 int inode_offset, i;
3754 ext4_group_t block_group;
3757 block_group = (inode->i_ino - 1) /
3758 EXT4_INODES_PER_GROUP(inode->i_sb);
3759 inodes_per_buffer = bh->b_size /
3760 EXT4_INODE_SIZE(inode->i_sb);
3761 inode_offset = ((inode->i_ino - 1) %
3762 EXT4_INODES_PER_GROUP(inode->i_sb));
3763 start = inode_offset & ~(inodes_per_buffer - 1);
3765 /* Is the inode bitmap in cache? */
3766 desc = ext4_get_group_desc(inode->i_sb,
3771 bitmap_bh = sb_getblk(inode->i_sb,
3772 ext4_inode_bitmap(inode->i_sb, desc));
3777 * If the inode bitmap isn't in cache then the
3778 * optimisation may end up performing two reads instead
3779 * of one, so skip it.
3781 if (!buffer_uptodate(bitmap_bh)) {
3785 for (i = start; i < start + inodes_per_buffer; i++) {
3786 if (i == inode_offset)
3788 if (ext4_test_bit(i, bitmap_bh->b_data))
3792 if (i == start + inodes_per_buffer) {
3793 /* all other inodes are free, so skip I/O */
3794 memset(bh->b_data, 0, bh->b_size);
3795 set_buffer_uptodate(bh);
3803 * There are other valid inodes in the buffer, this inode
3804 * has in-inode xattrs, or we don't have this inode in memory.
3805 * Read the block from disk.
3808 bh->b_end_io = end_buffer_read_sync;
3809 submit_bh(READ_META, bh);
3811 if (!buffer_uptodate(bh)) {
3812 ext4_error(inode->i_sb, "ext4_get_inode_loc",
3813 "unable to read inode block - "
3814 "inode=%lu, block=%llu",
3815 inode->i_ino, block);
3825 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3827 /* We have all inode data except xattrs in memory here. */
3828 return __ext4_get_inode_loc(inode, iloc,
3829 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3832 void ext4_set_inode_flags(struct inode *inode)
3834 unsigned int flags = EXT4_I(inode)->i_flags;
3836 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3837 if (flags & EXT4_SYNC_FL)
3838 inode->i_flags |= S_SYNC;
3839 if (flags & EXT4_APPEND_FL)
3840 inode->i_flags |= S_APPEND;
3841 if (flags & EXT4_IMMUTABLE_FL)
3842 inode->i_flags |= S_IMMUTABLE;
3843 if (flags & EXT4_NOATIME_FL)
3844 inode->i_flags |= S_NOATIME;
3845 if (flags & EXT4_DIRSYNC_FL)
3846 inode->i_flags |= S_DIRSYNC;
3849 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3850 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3852 unsigned int flags = ei->vfs_inode.i_flags;
3854 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3855 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
3857 ei->i_flags |= EXT4_SYNC_FL;
3858 if (flags & S_APPEND)
3859 ei->i_flags |= EXT4_APPEND_FL;
3860 if (flags & S_IMMUTABLE)
3861 ei->i_flags |= EXT4_IMMUTABLE_FL;
3862 if (flags & S_NOATIME)
3863 ei->i_flags |= EXT4_NOATIME_FL;
3864 if (flags & S_DIRSYNC)
3865 ei->i_flags |= EXT4_DIRSYNC_FL;
3867 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3868 struct ext4_inode_info *ei)
3871 struct inode *inode = &(ei->vfs_inode);
3872 struct super_block *sb = inode->i_sb;
3874 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3875 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3876 /* we are using combined 48 bit field */
3877 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3878 le32_to_cpu(raw_inode->i_blocks_lo);
3879 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
3880 /* i_blocks represent file system block size */
3881 return i_blocks << (inode->i_blkbits - 9);
3886 return le32_to_cpu(raw_inode->i_blocks_lo);
3890 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3892 struct ext4_iloc iloc;
3893 struct ext4_inode *raw_inode;
3894 struct ext4_inode_info *ei;
3895 struct buffer_head *bh;
3896 struct inode *inode;
3900 inode = iget_locked(sb, ino);
3902 return ERR_PTR(-ENOMEM);
3903 if (!(inode->i_state & I_NEW))
3907 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3908 ei->i_acl = EXT4_ACL_NOT_CACHED;
3909 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
3911 ei->i_block_alloc_info = NULL;
3913 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3917 raw_inode = ext4_raw_inode(&iloc);
3918 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3919 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3920 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3921 if(!(test_opt (inode->i_sb, NO_UID32))) {
3922 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3923 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3925 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3928 ei->i_dir_start_lookup = 0;
3929 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3930 /* We now have enough fields to check if the inode was active or not.
3931 * This is needed because nfsd might try to access dead inodes
3932 * the test is that same one that e2fsck uses
3933 * NeilBrown 1999oct15
3935 if (inode->i_nlink == 0) {
3936 if (inode->i_mode == 0 ||
3937 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3938 /* this inode is deleted */
3943 /* The only unlinked inodes we let through here have
3944 * valid i_mode and are being read by the orphan
3945 * recovery code: that's fine, we're about to complete
3946 * the process of deleting those. */
3948 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3949 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3950 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3951 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3952 cpu_to_le32(EXT4_OS_HURD)) {
3954 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3956 inode->i_size = ext4_isize(raw_inode);
3957 ei->i_disksize = inode->i_size;
3958 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3959 ei->i_block_group = iloc.block_group;
3961 * NOTE! The in-memory inode i_data array is in little-endian order
3962 * even on big-endian machines: we do NOT byteswap the block numbers!
3964 for (block = 0; block < EXT4_N_BLOCKS; block++)
3965 ei->i_data[block] = raw_inode->i_block[block];
3966 INIT_LIST_HEAD(&ei->i_orphan);
3968 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3969 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3970 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3971 EXT4_INODE_SIZE(inode->i_sb)) {
3976 if (ei->i_extra_isize == 0) {
3977 /* The extra space is currently unused. Use it. */
3978 ei->i_extra_isize = sizeof(struct ext4_inode) -
3979 EXT4_GOOD_OLD_INODE_SIZE;
3981 __le32 *magic = (void *)raw_inode +
3982 EXT4_GOOD_OLD_INODE_SIZE +
3984 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3985 ei->i_state |= EXT4_STATE_XATTR;
3988 ei->i_extra_isize = 0;
3990 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3991 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3992 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3993 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3995 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3996 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3997 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3999 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4002 if (S_ISREG(inode->i_mode)) {
4003 inode->i_op = &ext4_file_inode_operations;
4004 inode->i_fop = &ext4_file_operations;
4005 ext4_set_aops(inode);
4006 } else if (S_ISDIR(inode->i_mode)) {
4007 inode->i_op = &ext4_dir_inode_operations;
4008 inode->i_fop = &ext4_dir_operations;
4009 } else if (S_ISLNK(inode->i_mode)) {
4010 if (ext4_inode_is_fast_symlink(inode))
4011 inode->i_op = &ext4_fast_symlink_inode_operations;
4013 inode->i_op = &ext4_symlink_inode_operations;
4014 ext4_set_aops(inode);
4017 inode->i_op = &ext4_special_inode_operations;
4018 if (raw_inode->i_block[0])
4019 init_special_inode(inode, inode->i_mode,
4020 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4022 init_special_inode(inode, inode->i_mode,
4023 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4026 ext4_set_inode_flags(inode);
4027 unlock_new_inode(inode);
4032 return ERR_PTR(ret);
4035 static int ext4_inode_blocks_set(handle_t *handle,
4036 struct ext4_inode *raw_inode,
4037 struct ext4_inode_info *ei)
4039 struct inode *inode = &(ei->vfs_inode);
4040 u64 i_blocks = inode->i_blocks;
4041 struct super_block *sb = inode->i_sb;
4044 if (i_blocks <= ~0U) {
4046 * i_blocks can be represnted in a 32 bit variable
4047 * as multiple of 512 bytes
4049 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4050 raw_inode->i_blocks_high = 0;
4051 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4052 } else if (i_blocks <= 0xffffffffffffULL) {
4054 * i_blocks can be represented in a 48 bit variable
4055 * as multiple of 512 bytes
4057 err = ext4_update_rocompat_feature(handle, sb,
4058 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4061 /* i_block is stored in the split 48 bit fields */
4062 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4063 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4064 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4067 * i_blocks should be represented in a 48 bit variable
4068 * as multiple of file system block size
4070 err = ext4_update_rocompat_feature(handle, sb,
4071 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4074 ei->i_flags |= EXT4_HUGE_FILE_FL;
4075 /* i_block is stored in file system block size */
4076 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4077 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4078 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4085 * Post the struct inode info into an on-disk inode location in the
4086 * buffer-cache. This gobbles the caller's reference to the
4087 * buffer_head in the inode location struct.
4089 * The caller must have write access to iloc->bh.
4091 static int ext4_do_update_inode(handle_t *handle,
4092 struct inode *inode,
4093 struct ext4_iloc *iloc)
4095 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4096 struct ext4_inode_info *ei = EXT4_I(inode);
4097 struct buffer_head *bh = iloc->bh;
4098 int err = 0, rc, block;
4100 /* For fields not not tracking in the in-memory inode,
4101 * initialise them to zero for new inodes. */
4102 if (ei->i_state & EXT4_STATE_NEW)
4103 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4105 ext4_get_inode_flags(ei);
4106 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4107 if(!(test_opt(inode->i_sb, NO_UID32))) {
4108 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4109 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4111 * Fix up interoperability with old kernels. Otherwise, old inodes get
4112 * re-used with the upper 16 bits of the uid/gid intact
4115 raw_inode->i_uid_high =
4116 cpu_to_le16(high_16_bits(inode->i_uid));
4117 raw_inode->i_gid_high =
4118 cpu_to_le16(high_16_bits(inode->i_gid));
4120 raw_inode->i_uid_high = 0;
4121 raw_inode->i_gid_high = 0;
4124 raw_inode->i_uid_low =
4125 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4126 raw_inode->i_gid_low =
4127 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4128 raw_inode->i_uid_high = 0;
4129 raw_inode->i_gid_high = 0;
4131 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4133 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4134 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4135 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4136 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4138 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4140 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4141 /* clear the migrate flag in the raw_inode */
4142 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4143 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4144 cpu_to_le32(EXT4_OS_HURD))
4145 raw_inode->i_file_acl_high =
4146 cpu_to_le16(ei->i_file_acl >> 32);
4147 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4148 ext4_isize_set(raw_inode, ei->i_disksize);
4149 if (ei->i_disksize > 0x7fffffffULL) {
4150 struct super_block *sb = inode->i_sb;
4151 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4152 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4153 EXT4_SB(sb)->s_es->s_rev_level ==
4154 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4155 /* If this is the first large file
4156 * created, add a flag to the superblock.
4158 err = ext4_journal_get_write_access(handle,
4159 EXT4_SB(sb)->s_sbh);
4162 ext4_update_dynamic_rev(sb);
4163 EXT4_SET_RO_COMPAT_FEATURE(sb,
4164 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4167 err = ext4_journal_dirty_metadata(handle,
4168 EXT4_SB(sb)->s_sbh);
4171 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4172 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4173 if (old_valid_dev(inode->i_rdev)) {
4174 raw_inode->i_block[0] =
4175 cpu_to_le32(old_encode_dev(inode->i_rdev));
4176 raw_inode->i_block[1] = 0;
4178 raw_inode->i_block[0] = 0;
4179 raw_inode->i_block[1] =
4180 cpu_to_le32(new_encode_dev(inode->i_rdev));
4181 raw_inode->i_block[2] = 0;
4183 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4184 raw_inode->i_block[block] = ei->i_data[block];
4186 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4187 if (ei->i_extra_isize) {
4188 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4189 raw_inode->i_version_hi =
4190 cpu_to_le32(inode->i_version >> 32);
4191 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4195 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4196 rc = ext4_journal_dirty_metadata(handle, bh);
4199 ei->i_state &= ~EXT4_STATE_NEW;
4203 ext4_std_error(inode->i_sb, err);
4208 * ext4_write_inode()
4210 * We are called from a few places:
4212 * - Within generic_file_write() for O_SYNC files.
4213 * Here, there will be no transaction running. We wait for any running
4214 * trasnaction to commit.
4216 * - Within sys_sync(), kupdate and such.
4217 * We wait on commit, if tol to.
4219 * - Within prune_icache() (PF_MEMALLOC == true)
4220 * Here we simply return. We can't afford to block kswapd on the
4223 * In all cases it is actually safe for us to return without doing anything,
4224 * because the inode has been copied into a raw inode buffer in
4225 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4228 * Note that we are absolutely dependent upon all inode dirtiers doing the
4229 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4230 * which we are interested.
4232 * It would be a bug for them to not do this. The code:
4234 * mark_inode_dirty(inode)
4236 * inode->i_size = expr;
4238 * is in error because a kswapd-driven write_inode() could occur while
4239 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4240 * will no longer be on the superblock's dirty inode list.
4242 int ext4_write_inode(struct inode *inode, int wait)
4244 if (current->flags & PF_MEMALLOC)
4247 if (ext4_journal_current_handle()) {
4248 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4256 return ext4_force_commit(inode->i_sb);
4262 * Called from notify_change.
4264 * We want to trap VFS attempts to truncate the file as soon as
4265 * possible. In particular, we want to make sure that when the VFS
4266 * shrinks i_size, we put the inode on the orphan list and modify
4267 * i_disksize immediately, so that during the subsequent flushing of
4268 * dirty pages and freeing of disk blocks, we can guarantee that any
4269 * commit will leave the blocks being flushed in an unused state on
4270 * disk. (On recovery, the inode will get truncated and the blocks will
4271 * be freed, so we have a strong guarantee that no future commit will
4272 * leave these blocks visible to the user.)
4274 * Another thing we have to assure is that if we are in ordered mode
4275 * and inode is still attached to the committing transaction, we must
4276 * we start writeout of all the dirty pages which are being truncated.
4277 * This way we are sure that all the data written in the previous
4278 * transaction are already on disk (truncate waits for pages under
4281 * Called with inode->i_mutex down.
4283 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4285 struct inode *inode = dentry->d_inode;
4287 const unsigned int ia_valid = attr->ia_valid;
4289 error = inode_change_ok(inode, attr);
4293 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4294 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4297 /* (user+group)*(old+new) structure, inode write (sb,
4298 * inode block, ? - but truncate inode update has it) */
4299 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4300 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4301 if (IS_ERR(handle)) {
4302 error = PTR_ERR(handle);
4305 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4307 ext4_journal_stop(handle);
4310 /* Update corresponding info in inode so that everything is in
4311 * one transaction */
4312 if (attr->ia_valid & ATTR_UID)
4313 inode->i_uid = attr->ia_uid;
4314 if (attr->ia_valid & ATTR_GID)
4315 inode->i_gid = attr->ia_gid;
4316 error = ext4_mark_inode_dirty(handle, inode);
4317 ext4_journal_stop(handle);
4320 if (attr->ia_valid & ATTR_SIZE) {
4321 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4322 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4324 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4331 if (S_ISREG(inode->i_mode) &&
4332 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4335 handle = ext4_journal_start(inode, 3);
4336 if (IS_ERR(handle)) {
4337 error = PTR_ERR(handle);
4341 error = ext4_orphan_add(handle, inode);
4342 EXT4_I(inode)->i_disksize = attr->ia_size;
4343 rc = ext4_mark_inode_dirty(handle, inode);
4346 ext4_journal_stop(handle);
4348 if (ext4_should_order_data(inode)) {
4349 error = ext4_begin_ordered_truncate(inode,
4352 /* Do as much error cleanup as possible */
4353 handle = ext4_journal_start(inode, 3);
4354 if (IS_ERR(handle)) {
4355 ext4_orphan_del(NULL, inode);
4358 ext4_orphan_del(handle, inode);
4359 ext4_journal_stop(handle);
4365 rc = inode_setattr(inode, attr);
4367 /* If inode_setattr's call to ext4_truncate failed to get a
4368 * transaction handle at all, we need to clean up the in-core
4369 * orphan list manually. */
4371 ext4_orphan_del(NULL, inode);
4373 if (!rc && (ia_valid & ATTR_MODE))
4374 rc = ext4_acl_chmod(inode);
4377 ext4_std_error(inode->i_sb, error);
4383 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4386 struct inode *inode;
4387 unsigned long delalloc_blocks;
4389 inode = dentry->d_inode;
4390 generic_fillattr(inode, stat);
4393 * We can't update i_blocks if the block allocation is delayed
4394 * otherwise in the case of system crash before the real block
4395 * allocation is done, we will have i_blocks inconsistent with
4396 * on-disk file blocks.
4397 * We always keep i_blocks updated together with real
4398 * allocation. But to not confuse with user, stat
4399 * will return the blocks that include the delayed allocation
4400 * blocks for this file.
4402 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4403 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4404 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4406 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4410 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4415 /* if nrblocks are contiguous */
4418 * With N contiguous data blocks, it need at most
4419 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4420 * 2 dindirect blocks
4423 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4424 return indirects + 3;
4427 * if nrblocks are not contiguous, worse case, each block touch
4428 * a indirect block, and each indirect block touch a double indirect
4429 * block, plus a triple indirect block
4431 indirects = nrblocks * 2 + 1;
4435 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4437 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4438 return ext4_indirect_trans_blocks(inode, nrblocks, 0);
4439 return ext4_ext_index_trans_blocks(inode, nrblocks, 0);
4442 * Account for index blocks, block groups bitmaps and block group
4443 * descriptor blocks if modify datablocks and index blocks
4444 * worse case, the indexs blocks spread over different block groups
4446 * If datablocks are discontiguous, they are possible to spread over
4447 * different block groups too. If they are contiugous, with flexbg,
4448 * they could still across block group boundary.
4450 * Also account for superblock, inode, quota and xattr blocks
4452 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4454 int groups, gdpblocks;
4459 * How many index blocks need to touch to modify nrblocks?
4460 * The "Chunk" flag indicating whether the nrblocks is
4461 * physically contiguous on disk
4463 * For Direct IO and fallocate, they calls get_block to allocate
4464 * one single extent at a time, so they could set the "Chunk" flag
4466 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4471 * Now let's see how many group bitmaps and group descriptors need
4481 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4482 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4483 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4484 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4486 /* bitmaps and block group descriptor blocks */
4487 ret += groups + gdpblocks;
4489 /* Blocks for super block, inode, quota and xattr blocks */
4490 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4496 * Calulate the total number of credits to reserve to fit
4497 * the modification of a single pages into a single transaction,
4498 * which may include multiple chunks of block allocations.
4500 * This could be called via ext4_write_begin()
4502 * We need to consider the worse case, when
4503 * one new block per extent.
4505 int ext4_writepage_trans_blocks(struct inode *inode)
4507 int bpp = ext4_journal_blocks_per_page(inode);
4510 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4512 /* Account for data blocks for journalled mode */
4513 if (ext4_should_journal_data(inode))
4519 * Calculate the journal credits for a chunk of data modification.
4521 * This is called from DIO, fallocate or whoever calling
4522 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4524 * journal buffers for data blocks are not included here, as DIO
4525 * and fallocate do no need to journal data buffers.
4527 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4529 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4533 * The caller must have previously called ext4_reserve_inode_write().
4534 * Give this, we know that the caller already has write access to iloc->bh.
4536 int ext4_mark_iloc_dirty(handle_t *handle,
4537 struct inode *inode, struct ext4_iloc *iloc)
4541 if (test_opt(inode->i_sb, I_VERSION))
4542 inode_inc_iversion(inode);
4544 /* the do_update_inode consumes one bh->b_count */
4547 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4548 err = ext4_do_update_inode(handle, inode, iloc);
4554 * On success, We end up with an outstanding reference count against
4555 * iloc->bh. This _must_ be cleaned up later.
4559 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4560 struct ext4_iloc *iloc)
4564 err = ext4_get_inode_loc(inode, iloc);
4566 BUFFER_TRACE(iloc->bh, "get_write_access");
4567 err = ext4_journal_get_write_access(handle, iloc->bh);
4574 ext4_std_error(inode->i_sb, err);
4579 * Expand an inode by new_extra_isize bytes.
4580 * Returns 0 on success or negative error number on failure.
4582 static int ext4_expand_extra_isize(struct inode *inode,
4583 unsigned int new_extra_isize,
4584 struct ext4_iloc iloc,
4587 struct ext4_inode *raw_inode;
4588 struct ext4_xattr_ibody_header *header;
4589 struct ext4_xattr_entry *entry;
4591 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4594 raw_inode = ext4_raw_inode(&iloc);
4596 header = IHDR(inode, raw_inode);
4597 entry = IFIRST(header);
4599 /* No extended attributes present */
4600 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4601 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4602 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4604 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4608 /* try to expand with EAs present */
4609 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4614 * What we do here is to mark the in-core inode as clean with respect to inode
4615 * dirtiness (it may still be data-dirty).
4616 * This means that the in-core inode may be reaped by prune_icache
4617 * without having to perform any I/O. This is a very good thing,
4618 * because *any* task may call prune_icache - even ones which
4619 * have a transaction open against a different journal.
4621 * Is this cheating? Not really. Sure, we haven't written the
4622 * inode out, but prune_icache isn't a user-visible syncing function.
4623 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4624 * we start and wait on commits.
4626 * Is this efficient/effective? Well, we're being nice to the system
4627 * by cleaning up our inodes proactively so they can be reaped
4628 * without I/O. But we are potentially leaving up to five seconds'
4629 * worth of inodes floating about which prune_icache wants us to
4630 * write out. One way to fix that would be to get prune_icache()
4631 * to do a write_super() to free up some memory. It has the desired
4634 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4636 struct ext4_iloc iloc;
4637 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4638 static unsigned int mnt_count;
4642 err = ext4_reserve_inode_write(handle, inode, &iloc);
4643 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4644 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4646 * We need extra buffer credits since we may write into EA block
4647 * with this same handle. If journal_extend fails, then it will
4648 * only result in a minor loss of functionality for that inode.
4649 * If this is felt to be critical, then e2fsck should be run to
4650 * force a large enough s_min_extra_isize.
4652 if ((jbd2_journal_extend(handle,
4653 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4654 ret = ext4_expand_extra_isize(inode,
4655 sbi->s_want_extra_isize,
4658 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4660 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4661 ext4_warning(inode->i_sb, __func__,
4662 "Unable to expand inode %lu. Delete"
4663 " some EAs or run e2fsck.",
4666 le16_to_cpu(sbi->s_es->s_mnt_count);
4672 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4677 * ext4_dirty_inode() is called from __mark_inode_dirty()
4679 * We're really interested in the case where a file is being extended.
4680 * i_size has been changed by generic_commit_write() and we thus need
4681 * to include the updated inode in the current transaction.
4683 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4684 * are allocated to the file.
4686 * If the inode is marked synchronous, we don't honour that here - doing
4687 * so would cause a commit on atime updates, which we don't bother doing.
4688 * We handle synchronous inodes at the highest possible level.
4690 void ext4_dirty_inode(struct inode *inode)
4692 handle_t *current_handle = ext4_journal_current_handle();
4695 handle = ext4_journal_start(inode, 2);
4698 if (current_handle &&
4699 current_handle->h_transaction != handle->h_transaction) {
4700 /* This task has a transaction open against a different fs */
4701 printk(KERN_EMERG "%s: transactions do not match!\n",
4704 jbd_debug(5, "marking dirty. outer handle=%p\n",
4706 ext4_mark_inode_dirty(handle, inode);
4708 ext4_journal_stop(handle);
4715 * Bind an inode's backing buffer_head into this transaction, to prevent
4716 * it from being flushed to disk early. Unlike
4717 * ext4_reserve_inode_write, this leaves behind no bh reference and
4718 * returns no iloc structure, so the caller needs to repeat the iloc
4719 * lookup to mark the inode dirty later.
4721 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4723 struct ext4_iloc iloc;
4727 err = ext4_get_inode_loc(inode, &iloc);
4729 BUFFER_TRACE(iloc.bh, "get_write_access");
4730 err = jbd2_journal_get_write_access(handle, iloc.bh);
4732 err = ext4_journal_dirty_metadata(handle,
4737 ext4_std_error(inode->i_sb, err);
4742 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4749 * We have to be very careful here: changing a data block's
4750 * journaling status dynamically is dangerous. If we write a
4751 * data block to the journal, change the status and then delete
4752 * that block, we risk forgetting to revoke the old log record
4753 * from the journal and so a subsequent replay can corrupt data.
4754 * So, first we make sure that the journal is empty and that
4755 * nobody is changing anything.
4758 journal = EXT4_JOURNAL(inode);
4759 if (is_journal_aborted(journal))
4762 jbd2_journal_lock_updates(journal);
4763 jbd2_journal_flush(journal);
4766 * OK, there are no updates running now, and all cached data is
4767 * synced to disk. We are now in a completely consistent state
4768 * which doesn't have anything in the journal, and we know that
4769 * no filesystem updates are running, so it is safe to modify
4770 * the inode's in-core data-journaling state flag now.
4774 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4776 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4777 ext4_set_aops(inode);
4779 jbd2_journal_unlock_updates(journal);
4781 /* Finally we can mark the inode as dirty. */
4783 handle = ext4_journal_start(inode, 1);
4785 return PTR_ERR(handle);
4787 err = ext4_mark_inode_dirty(handle, inode);
4789 ext4_journal_stop(handle);
4790 ext4_std_error(inode->i_sb, err);
4795 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4797 return !buffer_mapped(bh);
4800 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4805 struct file *file = vma->vm_file;
4806 struct inode *inode = file->f_path.dentry->d_inode;
4807 struct address_space *mapping = inode->i_mapping;
4810 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4811 * get i_mutex because we are already holding mmap_sem.
4813 down_read(&inode->i_alloc_sem);
4814 size = i_size_read(inode);
4815 if (page->mapping != mapping || size <= page_offset(page)
4816 || !PageUptodate(page)) {
4817 /* page got truncated from under us? */
4821 if (PageMappedToDisk(page))
4824 if (page->index == size >> PAGE_CACHE_SHIFT)
4825 len = size & ~PAGE_CACHE_MASK;
4827 len = PAGE_CACHE_SIZE;
4829 if (page_has_buffers(page)) {
4830 /* return if we have all the buffers mapped */
4831 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4836 * OK, we need to fill the hole... Do write_begin write_end
4837 * to do block allocation/reservation.We are not holding
4838 * inode.i__mutex here. That allow * parallel write_begin,
4839 * write_end call. lock_page prevent this from happening
4840 * on the same page though
4842 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4843 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4846 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4847 len, len, page, NULL);
4852 up_read(&inode->i_alloc_sem);