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,
490 * XXX need to get goal block from mballoc's data structures
493 return ext4_find_near(inode, partial);
497 * ext4_blks_to_allocate: Look up the block map and count the number
498 * of direct blocks need to be allocated for the given branch.
500 * @branch: chain of indirect blocks
501 * @k: number of blocks need for indirect blocks
502 * @blks: number of data blocks to be mapped.
503 * @blocks_to_boundary: the offset in the indirect block
505 * return the total number of blocks to be allocate, including the
506 * direct and indirect blocks.
508 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
509 int blocks_to_boundary)
511 unsigned long count = 0;
514 * Simple case, [t,d]Indirect block(s) has not allocated yet
515 * then it's clear blocks on that path have not allocated
518 /* right now we don't handle cross boundary allocation */
519 if (blks < blocks_to_boundary + 1)
522 count += blocks_to_boundary + 1;
527 while (count < blks && count <= blocks_to_boundary &&
528 le32_to_cpu(*(branch[0].p + count)) == 0) {
535 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
536 * @indirect_blks: the number of blocks need to allocate for indirect
539 * @new_blocks: on return it will store the new block numbers for
540 * the indirect blocks(if needed) and the first direct block,
541 * @blks: on return it will store the total number of allocated
544 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
545 ext4_lblk_t iblock, ext4_fsblk_t goal,
546 int indirect_blks, int blks,
547 ext4_fsblk_t new_blocks[4], int *err)
550 unsigned long count = 0, blk_allocated = 0;
552 ext4_fsblk_t current_block = 0;
556 * Here we try to allocate the requested multiple blocks at once,
557 * on a best-effort basis.
558 * To build a branch, we should allocate blocks for
559 * the indirect blocks(if not allocated yet), and at least
560 * the first direct block of this branch. That's the
561 * minimum number of blocks need to allocate(required)
563 /* first we try to allocate the indirect blocks */
564 target = indirect_blks;
567 /* allocating blocks for indirect blocks and direct blocks */
568 current_block = ext4_new_meta_blocks(handle, inode,
574 /* allocate blocks for indirect blocks */
575 while (index < indirect_blks && count) {
576 new_blocks[index++] = current_block++;
581 * save the new block number
582 * for the first direct block
584 new_blocks[index] = current_block;
585 printk(KERN_INFO "%s returned more blocks than "
586 "requested\n", __func__);
592 target = blks - count ;
593 blk_allocated = count;
596 /* Now allocate data blocks */
598 /* allocating blocks for data blocks */
599 current_block = ext4_new_blocks(handle, inode, iblock,
601 if (*err && (target == blks)) {
603 * if the allocation failed and we didn't allocate
609 if (target == blks) {
611 * save the new block number
612 * for the first direct block
614 new_blocks[index] = current_block;
616 blk_allocated += count;
619 /* total number of blocks allocated for direct blocks */
624 for (i = 0; i < index; i++)
625 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
630 * ext4_alloc_branch - allocate and set up a chain of blocks.
632 * @indirect_blks: number of allocated indirect blocks
633 * @blks: number of allocated direct blocks
634 * @offsets: offsets (in the blocks) to store the pointers to next.
635 * @branch: place to store the chain in.
637 * This function allocates blocks, zeroes out all but the last one,
638 * links them into chain and (if we are synchronous) writes them to disk.
639 * In other words, it prepares a branch that can be spliced onto the
640 * inode. It stores the information about that chain in the branch[], in
641 * the same format as ext4_get_branch() would do. We are calling it after
642 * we had read the existing part of chain and partial points to the last
643 * triple of that (one with zero ->key). Upon the exit we have the same
644 * picture as after the successful ext4_get_block(), except that in one
645 * place chain is disconnected - *branch->p is still zero (we did not
646 * set the last link), but branch->key contains the number that should
647 * be placed into *branch->p to fill that gap.
649 * If allocation fails we free all blocks we've allocated (and forget
650 * their buffer_heads) and return the error value the from failed
651 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
652 * as described above and return 0.
654 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
655 ext4_lblk_t iblock, int indirect_blks,
656 int *blks, ext4_fsblk_t goal,
657 ext4_lblk_t *offsets, Indirect *branch)
659 int blocksize = inode->i_sb->s_blocksize;
662 struct buffer_head *bh;
664 ext4_fsblk_t new_blocks[4];
665 ext4_fsblk_t current_block;
667 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
668 *blks, new_blocks, &err);
672 branch[0].key = cpu_to_le32(new_blocks[0]);
674 * metadata blocks and data blocks are allocated.
676 for (n = 1; n <= indirect_blks; n++) {
678 * Get buffer_head for parent block, zero it out
679 * and set the pointer to new one, then send
682 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
685 BUFFER_TRACE(bh, "call get_create_access");
686 err = ext4_journal_get_create_access(handle, bh);
693 memset(bh->b_data, 0, blocksize);
694 branch[n].p = (__le32 *) bh->b_data + offsets[n];
695 branch[n].key = cpu_to_le32(new_blocks[n]);
696 *branch[n].p = branch[n].key;
697 if (n == indirect_blks) {
698 current_block = new_blocks[n];
700 * End of chain, update the last new metablock of
701 * the chain to point to the new allocated
702 * data blocks numbers
704 for (i=1; i < num; i++)
705 *(branch[n].p + i) = cpu_to_le32(++current_block);
707 BUFFER_TRACE(bh, "marking uptodate");
708 set_buffer_uptodate(bh);
711 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
712 err = ext4_journal_dirty_metadata(handle, bh);
719 /* Allocation failed, free what we already allocated */
720 for (i = 1; i <= n ; i++) {
721 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
722 ext4_journal_forget(handle, branch[i].bh);
724 for (i = 0; i < indirect_blks; i++)
725 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
727 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
733 * ext4_splice_branch - splice the allocated branch onto inode.
735 * @block: (logical) number of block we are adding
736 * @chain: chain of indirect blocks (with a missing link - see
738 * @where: location of missing link
739 * @num: number of indirect blocks we are adding
740 * @blks: number of direct blocks we are adding
742 * This function fills the missing link and does all housekeeping needed in
743 * inode (->i_blocks, etc.). In case of success we end up with the full
744 * chain to new block and return 0.
746 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
747 ext4_lblk_t block, Indirect *where, int num, int blks)
751 ext4_fsblk_t current_block;
754 * If we're splicing into a [td]indirect block (as opposed to the
755 * inode) then we need to get write access to the [td]indirect block
759 BUFFER_TRACE(where->bh, "get_write_access");
760 err = ext4_journal_get_write_access(handle, where->bh);
766 *where->p = where->key;
769 * Update the host buffer_head or inode to point to more just allocated
770 * direct blocks blocks
772 if (num == 0 && blks > 1) {
773 current_block = le32_to_cpu(where->key) + 1;
774 for (i = 1; i < blks; i++)
775 *(where->p + i) = cpu_to_le32(current_block++);
778 /* We are done with atomic stuff, now do the rest of housekeeping */
780 inode->i_ctime = ext4_current_time(inode);
781 ext4_mark_inode_dirty(handle, inode);
783 /* had we spliced it onto indirect block? */
786 * If we spliced it onto an indirect block, we haven't
787 * altered the inode. Note however that if it is being spliced
788 * onto an indirect block at the very end of the file (the
789 * file is growing) then we *will* alter the inode to reflect
790 * the new i_size. But that is not done here - it is done in
791 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
793 jbd_debug(5, "splicing indirect only\n");
794 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
795 err = ext4_journal_dirty_metadata(handle, where->bh);
800 * OK, we spliced it into the inode itself on a direct block.
801 * Inode was dirtied above.
803 jbd_debug(5, "splicing direct\n");
808 for (i = 1; i <= num; i++) {
809 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
810 ext4_journal_forget(handle, where[i].bh);
811 ext4_free_blocks(handle, inode,
812 le32_to_cpu(where[i-1].key), 1, 0);
814 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
820 * Allocation strategy is simple: if we have to allocate something, we will
821 * have to go the whole way to leaf. So let's do it before attaching anything
822 * to tree, set linkage between the newborn blocks, write them if sync is
823 * required, recheck the path, free and repeat if check fails, otherwise
824 * set the last missing link (that will protect us from any truncate-generated
825 * removals - all blocks on the path are immune now) and possibly force the
826 * write on the parent block.
827 * That has a nice additional property: no special recovery from the failed
828 * allocations is needed - we simply release blocks and do not touch anything
829 * reachable from inode.
831 * `handle' can be NULL if create == 0.
833 * return > 0, # of blocks mapped or allocated.
834 * return = 0, if plain lookup failed.
835 * return < 0, error case.
838 * Need to be called with
839 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
840 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
842 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
843 ext4_lblk_t iblock, unsigned long maxblocks,
844 struct buffer_head *bh_result,
845 int create, int extend_disksize)
848 ext4_lblk_t offsets[4];
853 int blocks_to_boundary = 0;
855 struct ext4_inode_info *ei = EXT4_I(inode);
857 ext4_fsblk_t first_block = 0;
861 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
862 J_ASSERT(handle != NULL || create == 0);
863 depth = ext4_block_to_path(inode, iblock, offsets,
864 &blocks_to_boundary);
869 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
871 /* Simplest case - block found, no allocation needed */
873 first_block = le32_to_cpu(chain[depth - 1].key);
874 clear_buffer_new(bh_result);
877 while (count < maxblocks && count <= blocks_to_boundary) {
880 blk = le32_to_cpu(*(chain[depth-1].p + count));
882 if (blk == first_block + count)
890 /* Next simple case - plain lookup or failed read of indirect block */
891 if (!create || err == -EIO)
895 * Okay, we need to do block allocation.
897 goal = ext4_find_goal(inode, iblock, partial);
899 /* the number of blocks need to allocate for [d,t]indirect blocks */
900 indirect_blks = (chain + depth) - partial - 1;
903 * Next look up the indirect map to count the totoal number of
904 * direct blocks to allocate for this branch.
906 count = ext4_blks_to_allocate(partial, indirect_blks,
907 maxblocks, blocks_to_boundary);
909 * Block out ext4_truncate while we alter the tree
911 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
913 offsets + (partial - chain), partial);
916 * The ext4_splice_branch call will free and forget any buffers
917 * on the new chain if there is a failure, but that risks using
918 * up transaction credits, especially for bitmaps where the
919 * credits cannot be returned. Can we handle this somehow? We
920 * may need to return -EAGAIN upwards in the worst case. --sct
923 err = ext4_splice_branch(handle, inode, iblock,
924 partial, indirect_blks, count);
926 * i_disksize growing is protected by i_data_sem. Don't forget to
927 * protect it if you're about to implement concurrent
928 * ext4_get_block() -bzzz
930 if (!err && extend_disksize) {
931 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
932 if (disksize > i_size_read(inode))
933 disksize = i_size_read(inode);
934 if (disksize > ei->i_disksize)
935 ei->i_disksize = disksize;
940 set_buffer_new(bh_result);
942 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
943 if (count > blocks_to_boundary)
944 set_buffer_boundary(bh_result);
946 /* Clean up and exit */
947 partial = chain + depth - 1; /* the whole chain */
949 while (partial > chain) {
950 BUFFER_TRACE(partial->bh, "call brelse");
954 BUFFER_TRACE(bh_result, "returned");
960 * Calculate the number of metadata blocks need to reserve
961 * to allocate @blocks for non extent file based file
963 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
965 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
966 int ind_blks, dind_blks, tind_blks;
968 /* number of new indirect blocks needed */
969 ind_blks = (blocks + icap - 1) / icap;
971 dind_blks = (ind_blks + icap - 1) / icap;
975 return ind_blks + dind_blks + tind_blks;
979 * Calculate the number of metadata blocks need to reserve
980 * to allocate given number of blocks
982 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
987 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
988 return ext4_ext_calc_metadata_amount(inode, blocks);
990 return ext4_indirect_calc_metadata_amount(inode, blocks);
993 static void ext4_da_update_reserve_space(struct inode *inode, int used)
995 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
996 int total, mdb, mdb_free;
998 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
999 /* recalculate the number of metablocks still need to be reserved */
1000 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1001 mdb = ext4_calc_metadata_amount(inode, total);
1003 /* figure out how many metablocks to release */
1004 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1005 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1008 /* Account for allocated meta_blocks */
1009 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1011 /* update fs dirty blocks counter */
1012 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1013 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1014 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1017 /* update per-inode reservations */
1018 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1019 EXT4_I(inode)->i_reserved_data_blocks -= used;
1021 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1025 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1026 * and returns if the blocks are already mapped.
1028 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1029 * and store the allocated blocks in the result buffer head and mark it
1032 * If file type is extents based, it will call ext4_ext_get_blocks(),
1033 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1036 * On success, it returns the number of blocks being mapped or allocate.
1037 * if create==0 and the blocks are pre-allocated and uninitialized block,
1038 * the result buffer head is unmapped. If the create ==1, it will make sure
1039 * the buffer head is mapped.
1041 * It returns 0 if plain look up failed (blocks have not been allocated), in
1042 * that casem, buffer head is unmapped
1044 * It returns the error in case of allocation failure.
1046 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1047 unsigned long max_blocks, struct buffer_head *bh,
1048 int create, int extend_disksize, int flag)
1052 clear_buffer_mapped(bh);
1055 * Try to see if we can get the block without requesting
1056 * for new file system block.
1058 down_read((&EXT4_I(inode)->i_data_sem));
1059 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1060 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1063 retval = ext4_get_blocks_handle(handle,
1064 inode, block, max_blocks, bh, 0, 0);
1066 up_read((&EXT4_I(inode)->i_data_sem));
1068 /* If it is only a block(s) look up */
1073 * Returns if the blocks have already allocated
1075 * Note that if blocks have been preallocated
1076 * ext4_ext_get_block() returns th create = 0
1077 * with buffer head unmapped.
1079 if (retval > 0 && buffer_mapped(bh))
1083 * New blocks allocate and/or writing to uninitialized extent
1084 * will possibly result in updating i_data, so we take
1085 * the write lock of i_data_sem, and call get_blocks()
1086 * with create == 1 flag.
1088 down_write((&EXT4_I(inode)->i_data_sem));
1091 * if the caller is from delayed allocation writeout path
1092 * we have already reserved fs blocks for allocation
1093 * let the underlying get_block() function know to
1094 * avoid double accounting
1097 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1099 * We need to check for EXT4 here because migrate
1100 * could have changed the inode type in between
1102 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1103 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1104 bh, create, extend_disksize);
1106 retval = ext4_get_blocks_handle(handle, inode, block,
1107 max_blocks, bh, create, extend_disksize);
1109 if (retval > 0 && buffer_new(bh)) {
1111 * We allocated new blocks which will result in
1112 * i_data's format changing. Force the migrate
1113 * to fail by clearing migrate flags
1115 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1121 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1123 * Update reserved blocks/metadata blocks
1124 * after successful block allocation
1125 * which were deferred till now
1127 if ((retval > 0) && buffer_delay(bh))
1128 ext4_da_update_reserve_space(inode, retval);
1131 up_write((&EXT4_I(inode)->i_data_sem));
1135 /* Maximum number of blocks we map for direct IO at once. */
1136 #define DIO_MAX_BLOCKS 4096
1138 int ext4_get_block(struct inode *inode, sector_t iblock,
1139 struct buffer_head *bh_result, int create)
1141 handle_t *handle = ext4_journal_current_handle();
1142 int ret = 0, started = 0;
1143 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1146 if (create && !handle) {
1147 /* Direct IO write... */
1148 if (max_blocks > DIO_MAX_BLOCKS)
1149 max_blocks = DIO_MAX_BLOCKS;
1150 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1151 handle = ext4_journal_start(inode, dio_credits);
1152 if (IS_ERR(handle)) {
1153 ret = PTR_ERR(handle);
1159 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1160 max_blocks, bh_result, create, 0, 0);
1162 bh_result->b_size = (ret << inode->i_blkbits);
1166 ext4_journal_stop(handle);
1172 * `handle' can be NULL if create is zero
1174 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1175 ext4_lblk_t block, int create, int *errp)
1177 struct buffer_head dummy;
1180 J_ASSERT(handle != NULL || create == 0);
1183 dummy.b_blocknr = -1000;
1184 buffer_trace_init(&dummy.b_history);
1185 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1186 &dummy, create, 1, 0);
1188 * ext4_get_blocks_handle() returns number of blocks
1189 * mapped. 0 in case of a HOLE.
1197 if (!err && buffer_mapped(&dummy)) {
1198 struct buffer_head *bh;
1199 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1204 if (buffer_new(&dummy)) {
1205 J_ASSERT(create != 0);
1206 J_ASSERT(handle != NULL);
1209 * Now that we do not always journal data, we should
1210 * keep in mind whether this should always journal the
1211 * new buffer as metadata. For now, regular file
1212 * writes use ext4_get_block instead, so it's not a
1216 BUFFER_TRACE(bh, "call get_create_access");
1217 fatal = ext4_journal_get_create_access(handle, bh);
1218 if (!fatal && !buffer_uptodate(bh)) {
1219 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1220 set_buffer_uptodate(bh);
1223 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1224 err = ext4_journal_dirty_metadata(handle, bh);
1228 BUFFER_TRACE(bh, "not a new buffer");
1241 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1242 ext4_lblk_t block, int create, int *err)
1244 struct buffer_head *bh;
1246 bh = ext4_getblk(handle, inode, block, create, err);
1249 if (buffer_uptodate(bh))
1251 ll_rw_block(READ_META, 1, &bh);
1253 if (buffer_uptodate(bh))
1260 static int walk_page_buffers(handle_t *handle,
1261 struct buffer_head *head,
1265 int (*fn)(handle_t *handle,
1266 struct buffer_head *bh))
1268 struct buffer_head *bh;
1269 unsigned block_start, block_end;
1270 unsigned blocksize = head->b_size;
1272 struct buffer_head *next;
1274 for (bh = head, block_start = 0;
1275 ret == 0 && (bh != head || !block_start);
1276 block_start = block_end, bh = next)
1278 next = bh->b_this_page;
1279 block_end = block_start + blocksize;
1280 if (block_end <= from || block_start >= to) {
1281 if (partial && !buffer_uptodate(bh))
1285 err = (*fn)(handle, bh);
1293 * To preserve ordering, it is essential that the hole instantiation and
1294 * the data write be encapsulated in a single transaction. We cannot
1295 * close off a transaction and start a new one between the ext4_get_block()
1296 * and the commit_write(). So doing the jbd2_journal_start at the start of
1297 * prepare_write() is the right place.
1299 * Also, this function can nest inside ext4_writepage() ->
1300 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1301 * has generated enough buffer credits to do the whole page. So we won't
1302 * block on the journal in that case, which is good, because the caller may
1305 * By accident, ext4 can be reentered when a transaction is open via
1306 * quota file writes. If we were to commit the transaction while thus
1307 * reentered, there can be a deadlock - we would be holding a quota
1308 * lock, and the commit would never complete if another thread had a
1309 * transaction open and was blocking on the quota lock - a ranking
1312 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1313 * will _not_ run commit under these circumstances because handle->h_ref
1314 * is elevated. We'll still have enough credits for the tiny quotafile
1317 static int do_journal_get_write_access(handle_t *handle,
1318 struct buffer_head *bh)
1320 if (!buffer_mapped(bh) || buffer_freed(bh))
1322 return ext4_journal_get_write_access(handle, bh);
1325 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1326 loff_t pos, unsigned len, unsigned flags,
1327 struct page **pagep, void **fsdata)
1329 struct inode *inode = mapping->host;
1330 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1337 index = pos >> PAGE_CACHE_SHIFT;
1338 from = pos & (PAGE_CACHE_SIZE - 1);
1342 handle = ext4_journal_start(inode, needed_blocks);
1343 if (IS_ERR(handle)) {
1344 ret = PTR_ERR(handle);
1348 page = grab_cache_page_write_begin(mapping, index, flags);
1350 ext4_journal_stop(handle);
1356 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1359 if (!ret && ext4_should_journal_data(inode)) {
1360 ret = walk_page_buffers(handle, page_buffers(page),
1361 from, to, NULL, do_journal_get_write_access);
1366 ext4_journal_stop(handle);
1367 page_cache_release(page);
1369 * block_write_begin may have instantiated a few blocks
1370 * outside i_size. Trim these off again. Don't need
1371 * i_size_read because we hold i_mutex.
1373 if (pos + len > inode->i_size)
1374 vmtruncate(inode, inode->i_size);
1377 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1383 /* For write_end() in data=journal mode */
1384 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1386 if (!buffer_mapped(bh) || buffer_freed(bh))
1388 set_buffer_uptodate(bh);
1389 return ext4_journal_dirty_metadata(handle, bh);
1393 * We need to pick up the new inode size which generic_commit_write gave us
1394 * `file' can be NULL - eg, when called from page_symlink().
1396 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1397 * buffers are managed internally.
1399 static int ext4_ordered_write_end(struct file *file,
1400 struct address_space *mapping,
1401 loff_t pos, unsigned len, unsigned copied,
1402 struct page *page, void *fsdata)
1404 handle_t *handle = ext4_journal_current_handle();
1405 struct inode *inode = mapping->host;
1408 ret = ext4_jbd2_file_inode(handle, inode);
1413 new_i_size = pos + copied;
1414 if (new_i_size > EXT4_I(inode)->i_disksize) {
1415 ext4_update_i_disksize(inode, new_i_size);
1416 /* We need to mark inode dirty even if
1417 * new_i_size is less that inode->i_size
1418 * bu greater than i_disksize.(hint delalloc)
1420 ext4_mark_inode_dirty(handle, inode);
1423 ret2 = generic_write_end(file, mapping, pos, len, copied,
1429 ret2 = ext4_journal_stop(handle);
1433 return ret ? ret : copied;
1436 static int ext4_writeback_write_end(struct file *file,
1437 struct address_space *mapping,
1438 loff_t pos, unsigned len, unsigned copied,
1439 struct page *page, void *fsdata)
1441 handle_t *handle = ext4_journal_current_handle();
1442 struct inode *inode = mapping->host;
1446 new_i_size = pos + copied;
1447 if (new_i_size > EXT4_I(inode)->i_disksize) {
1448 ext4_update_i_disksize(inode, new_i_size);
1449 /* We need to mark inode dirty even if
1450 * new_i_size is less that inode->i_size
1451 * bu greater than i_disksize.(hint delalloc)
1453 ext4_mark_inode_dirty(handle, inode);
1456 ret2 = generic_write_end(file, mapping, pos, len, copied,
1462 ret2 = ext4_journal_stop(handle);
1466 return ret ? ret : copied;
1469 static int ext4_journalled_write_end(struct file *file,
1470 struct address_space *mapping,
1471 loff_t pos, unsigned len, unsigned copied,
1472 struct page *page, void *fsdata)
1474 handle_t *handle = ext4_journal_current_handle();
1475 struct inode *inode = mapping->host;
1481 from = pos & (PAGE_CACHE_SIZE - 1);
1485 if (!PageUptodate(page))
1487 page_zero_new_buffers(page, from+copied, to);
1490 ret = walk_page_buffers(handle, page_buffers(page), from,
1491 to, &partial, write_end_fn);
1493 SetPageUptodate(page);
1494 new_i_size = pos + copied;
1495 if (new_i_size > inode->i_size)
1496 i_size_write(inode, pos+copied);
1497 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1498 if (new_i_size > EXT4_I(inode)->i_disksize) {
1499 ext4_update_i_disksize(inode, new_i_size);
1500 ret2 = ext4_mark_inode_dirty(handle, inode);
1506 ret2 = ext4_journal_stop(handle);
1509 page_cache_release(page);
1511 return ret ? ret : copied;
1514 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1517 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1518 unsigned long md_needed, mdblocks, total = 0;
1521 * recalculate the amount of metadata blocks to reserve
1522 * in order to allocate nrblocks
1523 * worse case is one extent per block
1526 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1527 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1528 mdblocks = ext4_calc_metadata_amount(inode, total);
1529 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1531 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1532 total = md_needed + nrblocks;
1534 if (ext4_claim_free_blocks(sbi, total)) {
1535 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1536 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1542 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1543 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1545 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1546 return 0; /* success */
1549 static void ext4_da_release_space(struct inode *inode, int to_free)
1551 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1552 int total, mdb, mdb_free, release;
1555 return; /* Nothing to release, exit */
1557 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1559 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1561 * if there is no reserved blocks, but we try to free some
1562 * then the counter is messed up somewhere.
1563 * but since this function is called from invalidate
1564 * page, it's harmless to return without any action
1566 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1567 "blocks for inode %lu, but there is no reserved "
1568 "data blocks\n", to_free, inode->i_ino);
1569 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1573 /* recalculate the number of metablocks still need to be reserved */
1574 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1575 mdb = ext4_calc_metadata_amount(inode, total);
1577 /* figure out how many metablocks to release */
1578 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1579 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1581 release = to_free + mdb_free;
1583 /* update fs dirty blocks counter for truncate case */
1584 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1586 /* update per-inode reservations */
1587 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1588 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1590 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1591 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1592 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1595 static void ext4_da_page_release_reservation(struct page *page,
1596 unsigned long offset)
1599 struct buffer_head *head, *bh;
1600 unsigned int curr_off = 0;
1602 head = page_buffers(page);
1605 unsigned int next_off = curr_off + bh->b_size;
1607 if ((offset <= curr_off) && (buffer_delay(bh))) {
1609 clear_buffer_delay(bh);
1611 curr_off = next_off;
1612 } while ((bh = bh->b_this_page) != head);
1613 ext4_da_release_space(page->mapping->host, to_release);
1617 * Delayed allocation stuff
1620 struct mpage_da_data {
1621 struct inode *inode;
1622 struct buffer_head lbh; /* extent of blocks */
1623 unsigned long first_page, next_page; /* extent of pages */
1624 get_block_t *get_block;
1625 struct writeback_control *wbc;
1632 * mpage_da_submit_io - walks through extent of pages and try to write
1633 * them with writepage() call back
1635 * @mpd->inode: inode
1636 * @mpd->first_page: first page of the extent
1637 * @mpd->next_page: page after the last page of the extent
1638 * @mpd->get_block: the filesystem's block mapper function
1640 * By the time mpage_da_submit_io() is called we expect all blocks
1641 * to be allocated. this may be wrong if allocation failed.
1643 * As pages are already locked by write_cache_pages(), we can't use it
1645 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1648 struct pagevec pvec;
1649 unsigned long index, end;
1650 int ret = 0, err, nr_pages, i;
1651 struct inode *inode = mpd->inode;
1652 struct address_space *mapping = inode->i_mapping;
1654 BUG_ON(mpd->next_page <= mpd->first_page);
1656 * We need to start from the first_page to the next_page - 1
1657 * to make sure we also write the mapped dirty buffer_heads.
1658 * If we look at mpd->lbh.b_blocknr we would only be looking
1659 * at the currently mapped buffer_heads.
1661 index = mpd->first_page;
1662 end = mpd->next_page - 1;
1664 pagevec_init(&pvec, 0);
1665 while (index <= end) {
1666 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1669 for (i = 0; i < nr_pages; i++) {
1670 struct page *page = pvec.pages[i];
1672 index = page->index;
1677 BUG_ON(!PageLocked(page));
1678 BUG_ON(PageWriteback(page));
1680 pages_skipped = mpd->wbc->pages_skipped;
1681 err = mapping->a_ops->writepage(page, mpd->wbc);
1682 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1684 * have successfully written the page
1685 * without skipping the same
1687 mpd->pages_written++;
1689 * In error case, we have to continue because
1690 * remaining pages are still locked
1691 * XXX: unlock and re-dirty them?
1696 pagevec_release(&pvec);
1702 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1704 * @mpd->inode - inode to walk through
1705 * @exbh->b_blocknr - first block on a disk
1706 * @exbh->b_size - amount of space in bytes
1707 * @logical - first logical block to start assignment with
1709 * the function goes through all passed space and put actual disk
1710 * block numbers into buffer heads, dropping BH_Delay
1712 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1713 struct buffer_head *exbh)
1715 struct inode *inode = mpd->inode;
1716 struct address_space *mapping = inode->i_mapping;
1717 int blocks = exbh->b_size >> inode->i_blkbits;
1718 sector_t pblock = exbh->b_blocknr, cur_logical;
1719 struct buffer_head *head, *bh;
1721 struct pagevec pvec;
1724 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1725 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1726 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1728 pagevec_init(&pvec, 0);
1730 while (index <= end) {
1731 /* XXX: optimize tail */
1732 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1735 for (i = 0; i < nr_pages; i++) {
1736 struct page *page = pvec.pages[i];
1738 index = page->index;
1743 BUG_ON(!PageLocked(page));
1744 BUG_ON(PageWriteback(page));
1745 BUG_ON(!page_has_buffers(page));
1747 bh = page_buffers(page);
1750 /* skip blocks out of the range */
1752 if (cur_logical >= logical)
1755 } while ((bh = bh->b_this_page) != head);
1758 if (cur_logical >= logical + blocks)
1760 if (buffer_delay(bh)) {
1761 bh->b_blocknr = pblock;
1762 clear_buffer_delay(bh);
1763 bh->b_bdev = inode->i_sb->s_bdev;
1764 } else if (buffer_unwritten(bh)) {
1765 bh->b_blocknr = pblock;
1766 clear_buffer_unwritten(bh);
1767 set_buffer_mapped(bh);
1769 bh->b_bdev = inode->i_sb->s_bdev;
1770 } else if (buffer_mapped(bh))
1771 BUG_ON(bh->b_blocknr != pblock);
1775 } while ((bh = bh->b_this_page) != head);
1777 pagevec_release(&pvec);
1783 * __unmap_underlying_blocks - just a helper function to unmap
1784 * set of blocks described by @bh
1786 static inline void __unmap_underlying_blocks(struct inode *inode,
1787 struct buffer_head *bh)
1789 struct block_device *bdev = inode->i_sb->s_bdev;
1792 blocks = bh->b_size >> inode->i_blkbits;
1793 for (i = 0; i < blocks; i++)
1794 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1797 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1798 sector_t logical, long blk_cnt)
1802 struct pagevec pvec;
1803 struct inode *inode = mpd->inode;
1804 struct address_space *mapping = inode->i_mapping;
1806 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1807 end = (logical + blk_cnt - 1) >>
1808 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1809 while (index <= end) {
1810 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1813 for (i = 0; i < nr_pages; i++) {
1814 struct page *page = pvec.pages[i];
1815 index = page->index;
1820 BUG_ON(!PageLocked(page));
1821 BUG_ON(PageWriteback(page));
1822 block_invalidatepage(page, 0);
1823 ClearPageUptodate(page);
1830 static void ext4_print_free_blocks(struct inode *inode)
1832 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1833 printk(KERN_EMERG "Total free blocks count %lld\n",
1834 ext4_count_free_blocks(inode->i_sb));
1835 printk(KERN_EMERG "Free/Dirty block details\n");
1836 printk(KERN_EMERG "free_blocks=%lld\n",
1837 percpu_counter_sum(&sbi->s_freeblocks_counter));
1838 printk(KERN_EMERG "dirty_blocks=%lld\n",
1839 percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1840 printk(KERN_EMERG "Block reservation details\n");
1841 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1842 EXT4_I(inode)->i_reserved_data_blocks);
1843 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1844 EXT4_I(inode)->i_reserved_meta_blocks);
1849 * mpage_da_map_blocks - go through given space
1851 * @mpd->lbh - bh describing space
1852 * @mpd->get_block - the filesystem's block mapper function
1854 * The function skips space we know is already mapped to disk blocks.
1857 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1860 struct buffer_head new;
1861 struct buffer_head *lbh = &mpd->lbh;
1865 * We consider only non-mapped and non-allocated blocks
1867 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1869 new.b_state = lbh->b_state;
1871 new.b_size = lbh->b_size;
1872 next = lbh->b_blocknr;
1874 * If we didn't accumulate anything
1875 * to write simply return
1879 err = mpd->get_block(mpd->inode, next, &new, 1);
1882 /* If get block returns with error
1883 * we simply return. Later writepage
1884 * will redirty the page and writepages
1885 * will find the dirty page again
1890 if (err == -ENOSPC &&
1891 ext4_count_free_blocks(mpd->inode->i_sb)) {
1897 * get block failure will cause us
1898 * to loop in writepages. Because
1899 * a_ops->writepage won't be able to
1900 * make progress. The page will be redirtied
1901 * by writepage and writepages will again
1902 * try to write the same.
1904 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1905 "at logical offset %llu with max blocks "
1906 "%zd with error %d\n",
1907 __func__, mpd->inode->i_ino,
1908 (unsigned long long)next,
1909 lbh->b_size >> mpd->inode->i_blkbits, err);
1910 printk(KERN_EMERG "This should not happen.!! "
1911 "Data will be lost\n");
1912 if (err == -ENOSPC) {
1913 ext4_print_free_blocks(mpd->inode);
1915 /* invlaidate all the pages */
1916 ext4_da_block_invalidatepages(mpd, next,
1917 lbh->b_size >> mpd->inode->i_blkbits);
1920 BUG_ON(new.b_size == 0);
1922 if (buffer_new(&new))
1923 __unmap_underlying_blocks(mpd->inode, &new);
1926 * If blocks are delayed marked, we need to
1927 * put actual blocknr and drop delayed bit
1929 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1930 mpage_put_bnr_to_bhs(mpd, next, &new);
1935 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1936 (1 << BH_Delay) | (1 << BH_Unwritten))
1939 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1941 * @mpd->lbh - extent of blocks
1942 * @logical - logical number of the block in the file
1943 * @bh - bh of the block (used to access block's state)
1945 * the function is used to collect contig. blocks in same state
1947 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1948 sector_t logical, struct buffer_head *bh)
1951 size_t b_size = bh->b_size;
1952 struct buffer_head *lbh = &mpd->lbh;
1953 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1955 /* check if thereserved journal credits might overflow */
1956 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1957 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1959 * With non-extent format we are limited by the journal
1960 * credit available. Total credit needed to insert
1961 * nrblocks contiguous blocks is dependent on the
1962 * nrblocks. So limit nrblocks.
1965 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1966 EXT4_MAX_TRANS_DATA) {
1968 * Adding the new buffer_head would make it cross the
1969 * allowed limit for which we have journal credit
1970 * reserved. So limit the new bh->b_size
1972 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1973 mpd->inode->i_blkbits;
1974 /* we will do mpage_da_submit_io in the next loop */
1978 * First block in the extent
1980 if (lbh->b_size == 0) {
1981 lbh->b_blocknr = logical;
1982 lbh->b_size = b_size;
1983 lbh->b_state = bh->b_state & BH_FLAGS;
1987 next = lbh->b_blocknr + nrblocks;
1989 * Can we merge the block to our big extent?
1991 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1992 lbh->b_size += b_size;
1998 * We couldn't merge the block to our extent, so we
1999 * need to flush current extent and start new one
2001 if (mpage_da_map_blocks(mpd) == 0)
2002 mpage_da_submit_io(mpd);
2008 * __mpage_da_writepage - finds extent of pages and blocks
2010 * @page: page to consider
2011 * @wbc: not used, we just follow rules
2014 * The function finds extents of pages and scan them for all blocks.
2016 static int __mpage_da_writepage(struct page *page,
2017 struct writeback_control *wbc, void *data)
2019 struct mpage_da_data *mpd = data;
2020 struct inode *inode = mpd->inode;
2021 struct buffer_head *bh, *head, fake;
2026 * Rest of the page in the page_vec
2027 * redirty then and skip then. We will
2028 * try to to write them again after
2029 * starting a new transaction
2031 redirty_page_for_writepage(wbc, page);
2033 return MPAGE_DA_EXTENT_TAIL;
2036 * Can we merge this page to current extent?
2038 if (mpd->next_page != page->index) {
2040 * Nope, we can't. So, we map non-allocated blocks
2041 * and start IO on them using writepage()
2043 if (mpd->next_page != mpd->first_page) {
2044 if (mpage_da_map_blocks(mpd) == 0)
2045 mpage_da_submit_io(mpd);
2047 * skip rest of the page in the page_vec
2050 redirty_page_for_writepage(wbc, page);
2052 return MPAGE_DA_EXTENT_TAIL;
2056 * Start next extent of pages ...
2058 mpd->first_page = page->index;
2063 mpd->lbh.b_size = 0;
2064 mpd->lbh.b_state = 0;
2065 mpd->lbh.b_blocknr = 0;
2068 mpd->next_page = page->index + 1;
2069 logical = (sector_t) page->index <<
2070 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2072 if (!page_has_buffers(page)) {
2074 * There is no attached buffer heads yet (mmap?)
2075 * we treat the page asfull of dirty blocks
2078 bh->b_size = PAGE_CACHE_SIZE;
2080 set_buffer_dirty(bh);
2081 set_buffer_uptodate(bh);
2082 mpage_add_bh_to_extent(mpd, logical, bh);
2084 return MPAGE_DA_EXTENT_TAIL;
2087 * Page with regular buffer heads, just add all dirty ones
2089 head = page_buffers(page);
2092 BUG_ON(buffer_locked(bh));
2094 * We need to try to allocate
2095 * unmapped blocks in the same page.
2096 * Otherwise we won't make progress
2097 * with the page in ext4_da_writepage
2099 if (buffer_dirty(bh) &&
2100 (!buffer_mapped(bh) || buffer_delay(bh))) {
2101 mpage_add_bh_to_extent(mpd, logical, bh);
2103 return MPAGE_DA_EXTENT_TAIL;
2104 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2106 * mapped dirty buffer. We need to update
2107 * the b_state because we look at
2108 * b_state in mpage_da_map_blocks. We don't
2109 * update b_size because if we find an
2110 * unmapped buffer_head later we need to
2111 * use the b_state flag of that buffer_head.
2113 if (mpd->lbh.b_size == 0)
2115 bh->b_state & BH_FLAGS;
2118 } while ((bh = bh->b_this_page) != head);
2125 * mpage_da_writepages - walk the list of dirty pages of the given
2126 * address space, allocates non-allocated blocks, maps newly-allocated
2127 * blocks to existing bhs and issue IO them
2129 * @mapping: address space structure to write
2130 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2131 * @get_block: the filesystem's block mapper function.
2133 * This is a library function, which implements the writepages()
2134 * address_space_operation.
2136 static int mpage_da_writepages(struct address_space *mapping,
2137 struct writeback_control *wbc,
2138 struct mpage_da_data *mpd)
2142 if (!mpd->get_block)
2143 return generic_writepages(mapping, wbc);
2145 mpd->lbh.b_size = 0;
2146 mpd->lbh.b_state = 0;
2147 mpd->lbh.b_blocknr = 0;
2148 mpd->first_page = 0;
2151 mpd->pages_written = 0;
2154 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2156 * Handle last extent of pages
2158 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2159 if (mpage_da_map_blocks(mpd) == 0)
2160 mpage_da_submit_io(mpd);
2163 ret = MPAGE_DA_EXTENT_TAIL;
2165 wbc->nr_to_write -= mpd->pages_written;
2170 * this is a special callback for ->write_begin() only
2171 * it's intention is to return mapped block or reserve space
2173 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2174 struct buffer_head *bh_result, int create)
2178 BUG_ON(create == 0);
2179 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2182 * first, we need to know whether the block is allocated already
2183 * preallocated blocks are unmapped but should treated
2184 * the same as allocated blocks.
2186 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2187 if ((ret == 0) && !buffer_delay(bh_result)) {
2188 /* the block isn't (pre)allocated yet, let's reserve space */
2190 * XXX: __block_prepare_write() unmaps passed block,
2193 ret = ext4_da_reserve_space(inode, 1);
2195 /* not enough space to reserve */
2198 map_bh(bh_result, inode->i_sb, 0);
2199 set_buffer_new(bh_result);
2200 set_buffer_delay(bh_result);
2201 } else if (ret > 0) {
2202 bh_result->b_size = (ret << inode->i_blkbits);
2208 #define EXT4_DELALLOC_RSVED 1
2209 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2210 struct buffer_head *bh_result, int create)
2213 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2214 loff_t disksize = EXT4_I(inode)->i_disksize;
2215 handle_t *handle = NULL;
2217 handle = ext4_journal_current_handle();
2219 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2220 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2223 bh_result->b_size = (ret << inode->i_blkbits);
2225 if (ext4_should_order_data(inode)) {
2227 retval = ext4_jbd2_file_inode(handle, inode);
2230 * Failed to add inode for ordered
2231 * mode. Don't update file size
2237 * Update on-disk size along with block allocation
2238 * we don't use 'extend_disksize' as size may change
2239 * within already allocated block -bzzz
2241 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2242 if (disksize > i_size_read(inode))
2243 disksize = i_size_read(inode);
2244 if (disksize > EXT4_I(inode)->i_disksize) {
2245 ext4_update_i_disksize(inode, disksize);
2246 ret = ext4_mark_inode_dirty(handle, inode);
2254 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2257 * unmapped buffer is possible for holes.
2258 * delay buffer is possible with delayed allocation
2260 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2263 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2264 struct buffer_head *bh_result, int create)
2267 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2270 * we don't want to do block allocation in writepage
2271 * so call get_block_wrap with create = 0
2273 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2274 bh_result, 0, 0, 0);
2276 bh_result->b_size = (ret << inode->i_blkbits);
2283 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2284 * get called via journal_submit_inode_data_buffers (no journal handle)
2285 * get called via shrink_page_list via pdflush (no journal handle)
2286 * or grab_page_cache when doing write_begin (have journal handle)
2288 static int ext4_da_writepage(struct page *page,
2289 struct writeback_control *wbc)
2294 struct buffer_head *page_bufs;
2295 struct inode *inode = page->mapping->host;
2297 size = i_size_read(inode);
2298 if (page->index == size >> PAGE_CACHE_SHIFT)
2299 len = size & ~PAGE_CACHE_MASK;
2301 len = PAGE_CACHE_SIZE;
2303 if (page_has_buffers(page)) {
2304 page_bufs = page_buffers(page);
2305 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2306 ext4_bh_unmapped_or_delay)) {
2308 * We don't want to do block allocation
2309 * So redirty the page and return
2310 * We may reach here when we do a journal commit
2311 * via journal_submit_inode_data_buffers.
2312 * If we don't have mapping block we just ignore
2313 * them. We can also reach here via shrink_page_list
2315 redirty_page_for_writepage(wbc, page);
2321 * The test for page_has_buffers() is subtle:
2322 * We know the page is dirty but it lost buffers. That means
2323 * that at some moment in time after write_begin()/write_end()
2324 * has been called all buffers have been clean and thus they
2325 * must have been written at least once. So they are all
2326 * mapped and we can happily proceed with mapping them
2327 * and writing the page.
2329 * Try to initialize the buffer_heads and check whether
2330 * all are mapped and non delay. We don't want to
2331 * do block allocation here.
2333 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2334 ext4_normal_get_block_write);
2336 page_bufs = page_buffers(page);
2337 /* check whether all are mapped and non delay */
2338 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2339 ext4_bh_unmapped_or_delay)) {
2340 redirty_page_for_writepage(wbc, page);
2346 * We can't do block allocation here
2347 * so just redity the page and unlock
2350 redirty_page_for_writepage(wbc, page);
2354 /* now mark the buffer_heads as dirty and uptodate */
2355 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2358 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2359 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2361 ret = block_write_full_page(page,
2362 ext4_normal_get_block_write,
2369 * This is called via ext4_da_writepages() to
2370 * calulate the total number of credits to reserve to fit
2371 * a single extent allocation into a single transaction,
2372 * ext4_da_writpeages() will loop calling this before
2373 * the block allocation.
2376 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2378 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2381 * With non-extent format the journal credit needed to
2382 * insert nrblocks contiguous block is dependent on
2383 * number of contiguous block. So we will limit
2384 * number of contiguous block to a sane value
2386 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2387 (max_blocks > EXT4_MAX_TRANS_DATA))
2388 max_blocks = EXT4_MAX_TRANS_DATA;
2390 return ext4_chunk_trans_blocks(inode, max_blocks);
2393 static int ext4_da_writepages(struct address_space *mapping,
2394 struct writeback_control *wbc)
2397 int range_whole = 0;
2398 handle_t *handle = NULL;
2399 struct mpage_da_data mpd;
2400 struct inode *inode = mapping->host;
2401 int no_nrwrite_index_update;
2402 long pages_written = 0, pages_skipped;
2403 int needed_blocks, ret = 0, nr_to_writebump = 0;
2404 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2407 * No pages to write? This is mainly a kludge to avoid starting
2408 * a transaction for special inodes like journal inode on last iput()
2409 * because that could violate lock ordering on umount
2411 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2415 * If the filesystem has aborted, it is read-only, so return
2416 * right away instead of dumping stack traces later on that
2417 * will obscure the real source of the problem. We test
2418 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2419 * the latter could be true if the filesystem is mounted
2420 * read-only, and in that case, ext4_da_writepages should
2421 * *never* be called, so if that ever happens, we would want
2424 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2428 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2429 * This make sure small files blocks are allocated in
2430 * single attempt. This ensure that small files
2431 * get less fragmented.
2433 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2434 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2435 wbc->nr_to_write = sbi->s_mb_stream_request;
2437 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2440 if (wbc->range_cyclic)
2441 index = mapping->writeback_index;
2443 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2446 mpd.inode = mapping->host;
2449 * we don't want write_cache_pages to update
2450 * nr_to_write and writeback_index
2452 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2453 wbc->no_nrwrite_index_update = 1;
2454 pages_skipped = wbc->pages_skipped;
2456 while (!ret && wbc->nr_to_write > 0) {
2459 * we insert one extent at a time. So we need
2460 * credit needed for single extent allocation.
2461 * journalled mode is currently not supported
2464 BUG_ON(ext4_should_journal_data(inode));
2465 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2467 /* start a new transaction*/
2468 handle = ext4_journal_start(inode, needed_blocks);
2469 if (IS_ERR(handle)) {
2470 ret = PTR_ERR(handle);
2471 printk(KERN_CRIT "%s: jbd2_start: "
2472 "%ld pages, ino %lu; err %d\n", __func__,
2473 wbc->nr_to_write, inode->i_ino, ret);
2475 goto out_writepages;
2477 mpd.get_block = ext4_da_get_block_write;
2478 ret = mpage_da_writepages(mapping, wbc, &mpd);
2480 ext4_journal_stop(handle);
2482 if (mpd.retval == -ENOSPC) {
2483 /* commit the transaction which would
2484 * free blocks released in the transaction
2487 jbd2_journal_force_commit_nested(sbi->s_journal);
2488 wbc->pages_skipped = pages_skipped;
2490 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2492 * got one extent now try with
2495 pages_written += mpd.pages_written;
2496 wbc->pages_skipped = pages_skipped;
2498 } else if (wbc->nr_to_write)
2500 * There is no more writeout needed
2501 * or we requested for a noblocking writeout
2502 * and we found the device congested
2506 if (pages_skipped != wbc->pages_skipped)
2507 printk(KERN_EMERG "This should not happen leaving %s "
2508 "with nr_to_write = %ld ret = %d\n",
2509 __func__, wbc->nr_to_write, ret);
2512 index += pages_written;
2513 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2515 * set the writeback_index so that range_cyclic
2516 * mode will write it back later
2518 mapping->writeback_index = index;
2521 if (!no_nrwrite_index_update)
2522 wbc->no_nrwrite_index_update = 0;
2523 wbc->nr_to_write -= nr_to_writebump;
2527 #define FALL_BACK_TO_NONDELALLOC 1
2528 static int ext4_nonda_switch(struct super_block *sb)
2530 s64 free_blocks, dirty_blocks;
2531 struct ext4_sb_info *sbi = EXT4_SB(sb);
2534 * switch to non delalloc mode if we are running low
2535 * on free block. The free block accounting via percpu
2536 * counters can get slightly wrong with FBC_BATCH getting
2537 * accumulated on each CPU without updating global counters
2538 * Delalloc need an accurate free block accounting. So switch
2539 * to non delalloc when we are near to error range.
2541 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2542 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2543 if (2 * free_blocks < 3 * dirty_blocks ||
2544 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2546 * free block count is less that 150% of dirty blocks
2547 * or free blocks is less that watermark
2554 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2555 loff_t pos, unsigned len, unsigned flags,
2556 struct page **pagep, void **fsdata)
2558 int ret, retries = 0;
2562 struct inode *inode = mapping->host;
2565 index = pos >> PAGE_CACHE_SHIFT;
2566 from = pos & (PAGE_CACHE_SIZE - 1);
2569 if (ext4_nonda_switch(inode->i_sb)) {
2570 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2571 return ext4_write_begin(file, mapping, pos,
2572 len, flags, pagep, fsdata);
2574 *fsdata = (void *)0;
2577 * With delayed allocation, we don't log the i_disksize update
2578 * if there is delayed block allocation. But we still need
2579 * to journalling the i_disksize update if writes to the end
2580 * of file which has an already mapped buffer.
2582 handle = ext4_journal_start(inode, 1);
2583 if (IS_ERR(handle)) {
2584 ret = PTR_ERR(handle);
2588 page = grab_cache_page_write_begin(mapping, index, flags);
2590 ext4_journal_stop(handle);
2596 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2597 ext4_da_get_block_prep);
2600 ext4_journal_stop(handle);
2601 page_cache_release(page);
2603 * block_write_begin may have instantiated a few blocks
2604 * outside i_size. Trim these off again. Don't need
2605 * i_size_read because we hold i_mutex.
2607 if (pos + len > inode->i_size)
2608 vmtruncate(inode, inode->i_size);
2611 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2618 * Check if we should update i_disksize
2619 * when write to the end of file but not require block allocation
2621 static int ext4_da_should_update_i_disksize(struct page *page,
2622 unsigned long offset)
2624 struct buffer_head *bh;
2625 struct inode *inode = page->mapping->host;
2629 bh = page_buffers(page);
2630 idx = offset >> inode->i_blkbits;
2632 for (i = 0; i < idx; i++)
2633 bh = bh->b_this_page;
2635 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2640 static int ext4_da_write_end(struct file *file,
2641 struct address_space *mapping,
2642 loff_t pos, unsigned len, unsigned copied,
2643 struct page *page, void *fsdata)
2645 struct inode *inode = mapping->host;
2647 handle_t *handle = ext4_journal_current_handle();
2649 unsigned long start, end;
2650 int write_mode = (int)(unsigned long)fsdata;
2652 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2653 if (ext4_should_order_data(inode)) {
2654 return ext4_ordered_write_end(file, mapping, pos,
2655 len, copied, page, fsdata);
2656 } else if (ext4_should_writeback_data(inode)) {
2657 return ext4_writeback_write_end(file, mapping, pos,
2658 len, copied, page, fsdata);
2664 start = pos & (PAGE_CACHE_SIZE - 1);
2665 end = start + copied - 1;
2668 * generic_write_end() will run mark_inode_dirty() if i_size
2669 * changes. So let's piggyback the i_disksize mark_inode_dirty
2673 new_i_size = pos + copied;
2674 if (new_i_size > EXT4_I(inode)->i_disksize) {
2675 if (ext4_da_should_update_i_disksize(page, end)) {
2676 down_write(&EXT4_I(inode)->i_data_sem);
2677 if (new_i_size > EXT4_I(inode)->i_disksize) {
2679 * Updating i_disksize when extending file
2680 * without needing block allocation
2682 if (ext4_should_order_data(inode))
2683 ret = ext4_jbd2_file_inode(handle,
2686 EXT4_I(inode)->i_disksize = new_i_size;
2688 up_write(&EXT4_I(inode)->i_data_sem);
2689 /* We need to mark inode dirty even if
2690 * new_i_size is less that inode->i_size
2691 * bu greater than i_disksize.(hint delalloc)
2693 ext4_mark_inode_dirty(handle, inode);
2696 ret2 = generic_write_end(file, mapping, pos, len, copied,
2701 ret2 = ext4_journal_stop(handle);
2705 return ret ? ret : copied;
2708 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2711 * Drop reserved blocks
2713 BUG_ON(!PageLocked(page));
2714 if (!page_has_buffers(page))
2717 ext4_da_page_release_reservation(page, offset);
2720 ext4_invalidatepage(page, offset);
2727 * bmap() is special. It gets used by applications such as lilo and by
2728 * the swapper to find the on-disk block of a specific piece of data.
2730 * Naturally, this is dangerous if the block concerned is still in the
2731 * journal. If somebody makes a swapfile on an ext4 data-journaling
2732 * filesystem and enables swap, then they may get a nasty shock when the
2733 * data getting swapped to that swapfile suddenly gets overwritten by
2734 * the original zero's written out previously to the journal and
2735 * awaiting writeback in the kernel's buffer cache.
2737 * So, if we see any bmap calls here on a modified, data-journaled file,
2738 * take extra steps to flush any blocks which might be in the cache.
2740 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2742 struct inode *inode = mapping->host;
2746 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2747 test_opt(inode->i_sb, DELALLOC)) {
2749 * With delalloc we want to sync the file
2750 * so that we can make sure we allocate
2753 filemap_write_and_wait(mapping);
2756 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2758 * This is a REALLY heavyweight approach, but the use of
2759 * bmap on dirty files is expected to be extremely rare:
2760 * only if we run lilo or swapon on a freshly made file
2761 * do we expect this to happen.
2763 * (bmap requires CAP_SYS_RAWIO so this does not
2764 * represent an unprivileged user DOS attack --- we'd be
2765 * in trouble if mortal users could trigger this path at
2768 * NB. EXT4_STATE_JDATA is not set on files other than
2769 * regular files. If somebody wants to bmap a directory
2770 * or symlink and gets confused because the buffer
2771 * hasn't yet been flushed to disk, they deserve
2772 * everything they get.
2775 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2776 journal = EXT4_JOURNAL(inode);
2777 jbd2_journal_lock_updates(journal);
2778 err = jbd2_journal_flush(journal);
2779 jbd2_journal_unlock_updates(journal);
2785 return generic_block_bmap(mapping, block, ext4_get_block);
2788 static int bget_one(handle_t *handle, struct buffer_head *bh)
2794 static int bput_one(handle_t *handle, struct buffer_head *bh)
2801 * Note that we don't need to start a transaction unless we're journaling data
2802 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2803 * need to file the inode to the transaction's list in ordered mode because if
2804 * we are writing back data added by write(), the inode is already there and if
2805 * we are writing back data modified via mmap(), noone guarantees in which
2806 * transaction the data will hit the disk. In case we are journaling data, we
2807 * cannot start transaction directly because transaction start ranks above page
2808 * lock so we have to do some magic.
2810 * In all journaling modes block_write_full_page() will start the I/O.
2814 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2819 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2821 * Same applies to ext4_get_block(). We will deadlock on various things like
2822 * lock_journal and i_data_sem
2824 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2827 * 16May01: If we're reentered then journal_current_handle() will be
2828 * non-zero. We simply *return*.
2830 * 1 July 2001: @@@ FIXME:
2831 * In journalled data mode, a data buffer may be metadata against the
2832 * current transaction. But the same file is part of a shared mapping
2833 * and someone does a writepage() on it.
2835 * We will move the buffer onto the async_data list, but *after* it has
2836 * been dirtied. So there's a small window where we have dirty data on
2839 * Note that this only applies to the last partial page in the file. The
2840 * bit which block_write_full_page() uses prepare/commit for. (That's
2841 * broken code anyway: it's wrong for msync()).
2843 * It's a rare case: affects the final partial page, for journalled data
2844 * where the file is subject to bith write() and writepage() in the same
2845 * transction. To fix it we'll need a custom block_write_full_page().
2846 * We'll probably need that anyway for journalling writepage() output.
2848 * We don't honour synchronous mounts for writepage(). That would be
2849 * disastrous. Any write() or metadata operation will sync the fs for
2853 static int __ext4_normal_writepage(struct page *page,
2854 struct writeback_control *wbc)
2856 struct inode *inode = page->mapping->host;
2858 if (test_opt(inode->i_sb, NOBH))
2859 return nobh_writepage(page,
2860 ext4_normal_get_block_write, wbc);
2862 return block_write_full_page(page,
2863 ext4_normal_get_block_write,
2867 static int ext4_normal_writepage(struct page *page,
2868 struct writeback_control *wbc)
2870 struct inode *inode = page->mapping->host;
2871 loff_t size = i_size_read(inode);
2874 J_ASSERT(PageLocked(page));
2875 if (page->index == size >> PAGE_CACHE_SHIFT)
2876 len = size & ~PAGE_CACHE_MASK;
2878 len = PAGE_CACHE_SIZE;
2880 if (page_has_buffers(page)) {
2881 /* if page has buffers it should all be mapped
2882 * and allocated. If there are not buffers attached
2883 * to the page we know the page is dirty but it lost
2884 * buffers. That means that at some moment in time
2885 * after write_begin() / write_end() has been called
2886 * all buffers have been clean and thus they must have been
2887 * written at least once. So they are all mapped and we can
2888 * happily proceed with mapping them and writing the page.
2890 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2891 ext4_bh_unmapped_or_delay));
2894 if (!ext4_journal_current_handle())
2895 return __ext4_normal_writepage(page, wbc);
2897 redirty_page_for_writepage(wbc, page);
2902 static int __ext4_journalled_writepage(struct page *page,
2903 struct writeback_control *wbc)
2905 struct address_space *mapping = page->mapping;
2906 struct inode *inode = mapping->host;
2907 struct buffer_head *page_bufs;
2908 handle_t *handle = NULL;
2912 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2913 ext4_normal_get_block_write);
2917 page_bufs = page_buffers(page);
2918 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2920 /* As soon as we unlock the page, it can go away, but we have
2921 * references to buffers so we are safe */
2924 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2925 if (IS_ERR(handle)) {
2926 ret = PTR_ERR(handle);
2930 ret = walk_page_buffers(handle, page_bufs, 0,
2931 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2933 err = walk_page_buffers(handle, page_bufs, 0,
2934 PAGE_CACHE_SIZE, NULL, write_end_fn);
2937 err = ext4_journal_stop(handle);
2941 walk_page_buffers(handle, page_bufs, 0,
2942 PAGE_CACHE_SIZE, NULL, bput_one);
2943 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2952 static int ext4_journalled_writepage(struct page *page,
2953 struct writeback_control *wbc)
2955 struct inode *inode = page->mapping->host;
2956 loff_t size = i_size_read(inode);
2959 J_ASSERT(PageLocked(page));
2960 if (page->index == size >> PAGE_CACHE_SHIFT)
2961 len = size & ~PAGE_CACHE_MASK;
2963 len = PAGE_CACHE_SIZE;
2965 if (page_has_buffers(page)) {
2966 /* if page has buffers it should all be mapped
2967 * and allocated. If there are not buffers attached
2968 * to the page we know the page is dirty but it lost
2969 * buffers. That means that at some moment in time
2970 * after write_begin() / write_end() has been called
2971 * all buffers have been clean and thus they must have been
2972 * written at least once. So they are all mapped and we can
2973 * happily proceed with mapping them and writing the page.
2975 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2976 ext4_bh_unmapped_or_delay));
2979 if (ext4_journal_current_handle())
2982 if (PageChecked(page)) {
2984 * It's mmapped pagecache. Add buffers and journal it. There
2985 * doesn't seem much point in redirtying the page here.
2987 ClearPageChecked(page);
2988 return __ext4_journalled_writepage(page, wbc);
2991 * It may be a page full of checkpoint-mode buffers. We don't
2992 * really know unless we go poke around in the buffer_heads.
2993 * But block_write_full_page will do the right thing.
2995 return block_write_full_page(page,
2996 ext4_normal_get_block_write,
3000 redirty_page_for_writepage(wbc, page);
3005 static int ext4_readpage(struct file *file, struct page *page)
3007 return mpage_readpage(page, ext4_get_block);
3011 ext4_readpages(struct file *file, struct address_space *mapping,
3012 struct list_head *pages, unsigned nr_pages)
3014 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3017 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3019 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3022 * If it's a full truncate we just forget about the pending dirtying
3025 ClearPageChecked(page);
3027 jbd2_journal_invalidatepage(journal, page, offset);
3030 static int ext4_releasepage(struct page *page, gfp_t wait)
3032 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3034 WARN_ON(PageChecked(page));
3035 if (!page_has_buffers(page))
3037 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3041 * If the O_DIRECT write will extend the file then add this inode to the
3042 * orphan list. So recovery will truncate it back to the original size
3043 * if the machine crashes during the write.
3045 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3046 * crashes then stale disk data _may_ be exposed inside the file. But current
3047 * VFS code falls back into buffered path in that case so we are safe.
3049 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3050 const struct iovec *iov, loff_t offset,
3051 unsigned long nr_segs)
3053 struct file *file = iocb->ki_filp;
3054 struct inode *inode = file->f_mapping->host;
3055 struct ext4_inode_info *ei = EXT4_I(inode);
3059 size_t count = iov_length(iov, nr_segs);
3062 loff_t final_size = offset + count;
3064 if (final_size > inode->i_size) {
3065 /* Credits for sb + inode write */
3066 handle = ext4_journal_start(inode, 2);
3067 if (IS_ERR(handle)) {
3068 ret = PTR_ERR(handle);
3071 ret = ext4_orphan_add(handle, inode);
3073 ext4_journal_stop(handle);
3077 ei->i_disksize = inode->i_size;
3078 ext4_journal_stop(handle);
3082 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3084 ext4_get_block, NULL);
3089 /* Credits for sb + inode write */
3090 handle = ext4_journal_start(inode, 2);
3091 if (IS_ERR(handle)) {
3092 /* This is really bad luck. We've written the data
3093 * but cannot extend i_size. Bail out and pretend
3094 * the write failed... */
3095 ret = PTR_ERR(handle);
3099 ext4_orphan_del(handle, inode);
3101 loff_t end = offset + ret;
3102 if (end > inode->i_size) {
3103 ei->i_disksize = end;
3104 i_size_write(inode, end);
3106 * We're going to return a positive `ret'
3107 * here due to non-zero-length I/O, so there's
3108 * no way of reporting error returns from
3109 * ext4_mark_inode_dirty() to userspace. So
3112 ext4_mark_inode_dirty(handle, inode);
3115 err = ext4_journal_stop(handle);
3124 * Pages can be marked dirty completely asynchronously from ext4's journalling
3125 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3126 * much here because ->set_page_dirty is called under VFS locks. The page is
3127 * not necessarily locked.
3129 * We cannot just dirty the page and leave attached buffers clean, because the
3130 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3131 * or jbddirty because all the journalling code will explode.
3133 * So what we do is to mark the page "pending dirty" and next time writepage
3134 * is called, propagate that into the buffers appropriately.
3136 static int ext4_journalled_set_page_dirty(struct page *page)
3138 SetPageChecked(page);
3139 return __set_page_dirty_nobuffers(page);
3142 static const struct address_space_operations ext4_ordered_aops = {
3143 .readpage = ext4_readpage,
3144 .readpages = ext4_readpages,
3145 .writepage = ext4_normal_writepage,
3146 .sync_page = block_sync_page,
3147 .write_begin = ext4_write_begin,
3148 .write_end = ext4_ordered_write_end,
3150 .invalidatepage = ext4_invalidatepage,
3151 .releasepage = ext4_releasepage,
3152 .direct_IO = ext4_direct_IO,
3153 .migratepage = buffer_migrate_page,
3154 .is_partially_uptodate = block_is_partially_uptodate,
3157 static const struct address_space_operations ext4_writeback_aops = {
3158 .readpage = ext4_readpage,
3159 .readpages = ext4_readpages,
3160 .writepage = ext4_normal_writepage,
3161 .sync_page = block_sync_page,
3162 .write_begin = ext4_write_begin,
3163 .write_end = ext4_writeback_write_end,
3165 .invalidatepage = ext4_invalidatepage,
3166 .releasepage = ext4_releasepage,
3167 .direct_IO = ext4_direct_IO,
3168 .migratepage = buffer_migrate_page,
3169 .is_partially_uptodate = block_is_partially_uptodate,
3172 static const struct address_space_operations ext4_journalled_aops = {
3173 .readpage = ext4_readpage,
3174 .readpages = ext4_readpages,
3175 .writepage = ext4_journalled_writepage,
3176 .sync_page = block_sync_page,
3177 .write_begin = ext4_write_begin,
3178 .write_end = ext4_journalled_write_end,
3179 .set_page_dirty = ext4_journalled_set_page_dirty,
3181 .invalidatepage = ext4_invalidatepage,
3182 .releasepage = ext4_releasepage,
3183 .is_partially_uptodate = block_is_partially_uptodate,
3186 static const struct address_space_operations ext4_da_aops = {
3187 .readpage = ext4_readpage,
3188 .readpages = ext4_readpages,
3189 .writepage = ext4_da_writepage,
3190 .writepages = ext4_da_writepages,
3191 .sync_page = block_sync_page,
3192 .write_begin = ext4_da_write_begin,
3193 .write_end = ext4_da_write_end,
3195 .invalidatepage = ext4_da_invalidatepage,
3196 .releasepage = ext4_releasepage,
3197 .direct_IO = ext4_direct_IO,
3198 .migratepage = buffer_migrate_page,
3199 .is_partially_uptodate = block_is_partially_uptodate,
3202 void ext4_set_aops(struct inode *inode)
3204 if (ext4_should_order_data(inode) &&
3205 test_opt(inode->i_sb, DELALLOC))
3206 inode->i_mapping->a_ops = &ext4_da_aops;
3207 else if (ext4_should_order_data(inode))
3208 inode->i_mapping->a_ops = &ext4_ordered_aops;
3209 else if (ext4_should_writeback_data(inode) &&
3210 test_opt(inode->i_sb, DELALLOC))
3211 inode->i_mapping->a_ops = &ext4_da_aops;
3212 else if (ext4_should_writeback_data(inode))
3213 inode->i_mapping->a_ops = &ext4_writeback_aops;
3215 inode->i_mapping->a_ops = &ext4_journalled_aops;
3219 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3220 * up to the end of the block which corresponds to `from'.
3221 * This required during truncate. We need to physically zero the tail end
3222 * of that block so it doesn't yield old data if the file is later grown.
3224 int ext4_block_truncate_page(handle_t *handle,
3225 struct address_space *mapping, loff_t from)
3227 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3228 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3229 unsigned blocksize, length, pos;
3231 struct inode *inode = mapping->host;
3232 struct buffer_head *bh;
3236 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3240 blocksize = inode->i_sb->s_blocksize;
3241 length = blocksize - (offset & (blocksize - 1));
3242 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3245 * For "nobh" option, we can only work if we don't need to
3246 * read-in the page - otherwise we create buffers to do the IO.
3248 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3249 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3250 zero_user(page, offset, length);
3251 set_page_dirty(page);
3255 if (!page_has_buffers(page))
3256 create_empty_buffers(page, blocksize, 0);
3258 /* Find the buffer that contains "offset" */
3259 bh = page_buffers(page);
3261 while (offset >= pos) {
3262 bh = bh->b_this_page;
3268 if (buffer_freed(bh)) {
3269 BUFFER_TRACE(bh, "freed: skip");
3273 if (!buffer_mapped(bh)) {
3274 BUFFER_TRACE(bh, "unmapped");
3275 ext4_get_block(inode, iblock, bh, 0);
3276 /* unmapped? It's a hole - nothing to do */
3277 if (!buffer_mapped(bh)) {
3278 BUFFER_TRACE(bh, "still unmapped");
3283 /* Ok, it's mapped. Make sure it's up-to-date */
3284 if (PageUptodate(page))
3285 set_buffer_uptodate(bh);
3287 if (!buffer_uptodate(bh)) {
3289 ll_rw_block(READ, 1, &bh);
3291 /* Uhhuh. Read error. Complain and punt. */
3292 if (!buffer_uptodate(bh))
3296 if (ext4_should_journal_data(inode)) {
3297 BUFFER_TRACE(bh, "get write access");
3298 err = ext4_journal_get_write_access(handle, bh);
3303 zero_user(page, offset, length);
3305 BUFFER_TRACE(bh, "zeroed end of block");
3308 if (ext4_should_journal_data(inode)) {
3309 err = ext4_journal_dirty_metadata(handle, bh);
3311 if (ext4_should_order_data(inode))
3312 err = ext4_jbd2_file_inode(handle, inode);
3313 mark_buffer_dirty(bh);
3318 page_cache_release(page);
3323 * Probably it should be a library function... search for first non-zero word
3324 * or memcmp with zero_page, whatever is better for particular architecture.
3327 static inline int all_zeroes(__le32 *p, __le32 *q)
3336 * ext4_find_shared - find the indirect blocks for partial truncation.
3337 * @inode: inode in question
3338 * @depth: depth of the affected branch
3339 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3340 * @chain: place to store the pointers to partial indirect blocks
3341 * @top: place to the (detached) top of branch
3343 * This is a helper function used by ext4_truncate().
3345 * When we do truncate() we may have to clean the ends of several
3346 * indirect blocks but leave the blocks themselves alive. Block is
3347 * partially truncated if some data below the new i_size is refered
3348 * from it (and it is on the path to the first completely truncated
3349 * data block, indeed). We have to free the top of that path along
3350 * with everything to the right of the path. Since no allocation
3351 * past the truncation point is possible until ext4_truncate()
3352 * finishes, we may safely do the latter, but top of branch may
3353 * require special attention - pageout below the truncation point
3354 * might try to populate it.
3356 * We atomically detach the top of branch from the tree, store the
3357 * block number of its root in *@top, pointers to buffer_heads of
3358 * partially truncated blocks - in @chain[].bh and pointers to
3359 * their last elements that should not be removed - in
3360 * @chain[].p. Return value is the pointer to last filled element
3363 * The work left to caller to do the actual freeing of subtrees:
3364 * a) free the subtree starting from *@top
3365 * b) free the subtrees whose roots are stored in
3366 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3367 * c) free the subtrees growing from the inode past the @chain[0].
3368 * (no partially truncated stuff there). */
3370 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3371 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3373 Indirect *partial, *p;
3377 /* Make k index the deepest non-null offest + 1 */
3378 for (k = depth; k > 1 && !offsets[k-1]; k--)
3380 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3381 /* Writer: pointers */
3383 partial = chain + k-1;
3385 * If the branch acquired continuation since we've looked at it -
3386 * fine, it should all survive and (new) top doesn't belong to us.
3388 if (!partial->key && *partial->p)
3391 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3394 * OK, we've found the last block that must survive. The rest of our
3395 * branch should be detached before unlocking. However, if that rest
3396 * of branch is all ours and does not grow immediately from the inode
3397 * it's easier to cheat and just decrement partial->p.
3399 if (p == chain + k - 1 && p > chain) {
3403 /* Nope, don't do this in ext4. Must leave the tree intact */
3410 while (partial > p) {
3411 brelse(partial->bh);
3419 * Zero a number of block pointers in either an inode or an indirect block.
3420 * If we restart the transaction we must again get write access to the
3421 * indirect block for further modification.
3423 * We release `count' blocks on disk, but (last - first) may be greater
3424 * than `count' because there can be holes in there.
3426 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3427 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3428 unsigned long count, __le32 *first, __le32 *last)
3431 if (try_to_extend_transaction(handle, inode)) {
3433 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3434 ext4_journal_dirty_metadata(handle, bh);
3436 ext4_mark_inode_dirty(handle, inode);
3437 ext4_journal_test_restart(handle, inode);
3439 BUFFER_TRACE(bh, "retaking write access");
3440 ext4_journal_get_write_access(handle, bh);
3445 * Any buffers which are on the journal will be in memory. We find
3446 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3447 * on them. We've already detached each block from the file, so
3448 * bforget() in jbd2_journal_forget() should be safe.
3450 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3452 for (p = first; p < last; p++) {
3453 u32 nr = le32_to_cpu(*p);
3455 struct buffer_head *tbh;
3458 tbh = sb_find_get_block(inode->i_sb, nr);
3459 ext4_forget(handle, 0, inode, tbh, nr);
3463 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3467 * ext4_free_data - free a list of data blocks
3468 * @handle: handle for this transaction
3469 * @inode: inode we are dealing with
3470 * @this_bh: indirect buffer_head which contains *@first and *@last
3471 * @first: array of block numbers
3472 * @last: points immediately past the end of array
3474 * We are freeing all blocks refered from that array (numbers are stored as
3475 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3477 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3478 * blocks are contiguous then releasing them at one time will only affect one
3479 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3480 * actually use a lot of journal space.
3482 * @this_bh will be %NULL if @first and @last point into the inode's direct
3485 static void ext4_free_data(handle_t *handle, struct inode *inode,
3486 struct buffer_head *this_bh,
3487 __le32 *first, __le32 *last)
3489 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3490 unsigned long count = 0; /* Number of blocks in the run */
3491 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3494 ext4_fsblk_t nr; /* Current block # */
3495 __le32 *p; /* Pointer into inode/ind
3496 for current block */
3499 if (this_bh) { /* For indirect block */
3500 BUFFER_TRACE(this_bh, "get_write_access");
3501 err = ext4_journal_get_write_access(handle, this_bh);
3502 /* Important: if we can't update the indirect pointers
3503 * to the blocks, we can't free them. */
3508 for (p = first; p < last; p++) {
3509 nr = le32_to_cpu(*p);
3511 /* accumulate blocks to free if they're contiguous */
3514 block_to_free_p = p;
3516 } else if (nr == block_to_free + count) {
3519 ext4_clear_blocks(handle, inode, this_bh,
3521 count, block_to_free_p, p);
3523 block_to_free_p = p;
3530 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3531 count, block_to_free_p, p);
3534 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3537 * The buffer head should have an attached journal head at this
3538 * point. However, if the data is corrupted and an indirect
3539 * block pointed to itself, it would have been detached when
3540 * the block was cleared. Check for this instead of OOPSing.
3543 ext4_journal_dirty_metadata(handle, this_bh);
3545 ext4_error(inode->i_sb, __func__,
3546 "circular indirect block detected, "
3547 "inode=%lu, block=%llu",
3549 (unsigned long long) this_bh->b_blocknr);
3554 * ext4_free_branches - free an array of branches
3555 * @handle: JBD handle for this transaction
3556 * @inode: inode we are dealing with
3557 * @parent_bh: the buffer_head which contains *@first and *@last
3558 * @first: array of block numbers
3559 * @last: pointer immediately past the end of array
3560 * @depth: depth of the branches to free
3562 * We are freeing all blocks refered from these branches (numbers are
3563 * stored as little-endian 32-bit) and updating @inode->i_blocks
3566 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3567 struct buffer_head *parent_bh,
3568 __le32 *first, __le32 *last, int depth)
3573 if (is_handle_aborted(handle))
3577 struct buffer_head *bh;
3578 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3580 while (--p >= first) {
3581 nr = le32_to_cpu(*p);
3583 continue; /* A hole */
3585 /* Go read the buffer for the next level down */
3586 bh = sb_bread(inode->i_sb, nr);
3589 * A read failure? Report error and clear slot
3593 ext4_error(inode->i_sb, "ext4_free_branches",
3594 "Read failure, inode=%lu, block=%llu",
3599 /* This zaps the entire block. Bottom up. */
3600 BUFFER_TRACE(bh, "free child branches");
3601 ext4_free_branches(handle, inode, bh,
3602 (__le32 *) bh->b_data,
3603 (__le32 *) bh->b_data + addr_per_block,
3607 * We've probably journalled the indirect block several
3608 * times during the truncate. But it's no longer
3609 * needed and we now drop it from the transaction via
3610 * jbd2_journal_revoke().
3612 * That's easy if it's exclusively part of this
3613 * transaction. But if it's part of the committing
3614 * transaction then jbd2_journal_forget() will simply
3615 * brelse() it. That means that if the underlying
3616 * block is reallocated in ext4_get_block(),
3617 * unmap_underlying_metadata() will find this block
3618 * and will try to get rid of it. damn, damn.
3620 * If this block has already been committed to the
3621 * journal, a revoke record will be written. And
3622 * revoke records must be emitted *before* clearing
3623 * this block's bit in the bitmaps.
3625 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3628 * Everything below this this pointer has been
3629 * released. Now let this top-of-subtree go.
3631 * We want the freeing of this indirect block to be
3632 * atomic in the journal with the updating of the
3633 * bitmap block which owns it. So make some room in
3636 * We zero the parent pointer *after* freeing its
3637 * pointee in the bitmaps, so if extend_transaction()
3638 * for some reason fails to put the bitmap changes and
3639 * the release into the same transaction, recovery
3640 * will merely complain about releasing a free block,
3641 * rather than leaking blocks.
3643 if (is_handle_aborted(handle))
3645 if (try_to_extend_transaction(handle, inode)) {
3646 ext4_mark_inode_dirty(handle, inode);
3647 ext4_journal_test_restart(handle, inode);
3650 ext4_free_blocks(handle, inode, nr, 1, 1);
3654 * The block which we have just freed is
3655 * pointed to by an indirect block: journal it
3657 BUFFER_TRACE(parent_bh, "get_write_access");
3658 if (!ext4_journal_get_write_access(handle,
3661 BUFFER_TRACE(parent_bh,
3662 "call ext4_journal_dirty_metadata");
3663 ext4_journal_dirty_metadata(handle,
3669 /* We have reached the bottom of the tree. */
3670 BUFFER_TRACE(parent_bh, "free data blocks");
3671 ext4_free_data(handle, inode, parent_bh, first, last);
3675 int ext4_can_truncate(struct inode *inode)
3677 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3679 if (S_ISREG(inode->i_mode))
3681 if (S_ISDIR(inode->i_mode))
3683 if (S_ISLNK(inode->i_mode))
3684 return !ext4_inode_is_fast_symlink(inode);
3691 * We block out ext4_get_block() block instantiations across the entire
3692 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3693 * simultaneously on behalf of the same inode.
3695 * As we work through the truncate and commmit bits of it to the journal there
3696 * is one core, guiding principle: the file's tree must always be consistent on
3697 * disk. We must be able to restart the truncate after a crash.
3699 * The file's tree may be transiently inconsistent in memory (although it
3700 * probably isn't), but whenever we close off and commit a journal transaction,
3701 * the contents of (the filesystem + the journal) must be consistent and
3702 * restartable. It's pretty simple, really: bottom up, right to left (although
3703 * left-to-right works OK too).
3705 * Note that at recovery time, journal replay occurs *before* the restart of
3706 * truncate against the orphan inode list.
3708 * The committed inode has the new, desired i_size (which is the same as
3709 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3710 * that this inode's truncate did not complete and it will again call
3711 * ext4_truncate() to have another go. So there will be instantiated blocks
3712 * to the right of the truncation point in a crashed ext4 filesystem. But
3713 * that's fine - as long as they are linked from the inode, the post-crash
3714 * ext4_truncate() run will find them and release them.
3716 void ext4_truncate(struct inode *inode)
3719 struct ext4_inode_info *ei = EXT4_I(inode);
3720 __le32 *i_data = ei->i_data;
3721 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3722 struct address_space *mapping = inode->i_mapping;
3723 ext4_lblk_t offsets[4];
3728 ext4_lblk_t last_block;
3729 unsigned blocksize = inode->i_sb->s_blocksize;
3731 if (!ext4_can_truncate(inode))
3734 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3735 ext4_ext_truncate(inode);
3739 handle = start_transaction(inode);
3741 return; /* AKPM: return what? */
3743 last_block = (inode->i_size + blocksize-1)
3744 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3746 if (inode->i_size & (blocksize - 1))
3747 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3750 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3752 goto out_stop; /* error */
3755 * OK. This truncate is going to happen. We add the inode to the
3756 * orphan list, so that if this truncate spans multiple transactions,
3757 * and we crash, we will resume the truncate when the filesystem
3758 * recovers. It also marks the inode dirty, to catch the new size.
3760 * Implication: the file must always be in a sane, consistent
3761 * truncatable state while each transaction commits.
3763 if (ext4_orphan_add(handle, inode))
3767 * From here we block out all ext4_get_block() callers who want to
3768 * modify the block allocation tree.
3770 down_write(&ei->i_data_sem);
3772 ext4_discard_preallocations(inode);
3775 * The orphan list entry will now protect us from any crash which
3776 * occurs before the truncate completes, so it is now safe to propagate
3777 * the new, shorter inode size (held for now in i_size) into the
3778 * on-disk inode. We do this via i_disksize, which is the value which
3779 * ext4 *really* writes onto the disk inode.
3781 ei->i_disksize = inode->i_size;
3783 if (n == 1) { /* direct blocks */
3784 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3785 i_data + EXT4_NDIR_BLOCKS);
3789 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3790 /* Kill the top of shared branch (not detached) */
3792 if (partial == chain) {
3793 /* Shared branch grows from the inode */
3794 ext4_free_branches(handle, inode, NULL,
3795 &nr, &nr+1, (chain+n-1) - partial);
3798 * We mark the inode dirty prior to restart,
3799 * and prior to stop. No need for it here.
3802 /* Shared branch grows from an indirect block */
3803 BUFFER_TRACE(partial->bh, "get_write_access");
3804 ext4_free_branches(handle, inode, partial->bh,
3806 partial->p+1, (chain+n-1) - partial);
3809 /* Clear the ends of indirect blocks on the shared branch */
3810 while (partial > chain) {
3811 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3812 (__le32*)partial->bh->b_data+addr_per_block,
3813 (chain+n-1) - partial);
3814 BUFFER_TRACE(partial->bh, "call brelse");
3815 brelse (partial->bh);
3819 /* Kill the remaining (whole) subtrees */
3820 switch (offsets[0]) {
3822 nr = i_data[EXT4_IND_BLOCK];
3824 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3825 i_data[EXT4_IND_BLOCK] = 0;
3827 case EXT4_IND_BLOCK:
3828 nr = i_data[EXT4_DIND_BLOCK];
3830 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3831 i_data[EXT4_DIND_BLOCK] = 0;
3833 case EXT4_DIND_BLOCK:
3834 nr = i_data[EXT4_TIND_BLOCK];
3836 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3837 i_data[EXT4_TIND_BLOCK] = 0;
3839 case EXT4_TIND_BLOCK:
3843 up_write(&ei->i_data_sem);
3844 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3845 ext4_mark_inode_dirty(handle, inode);
3848 * In a multi-transaction truncate, we only make the final transaction
3855 * If this was a simple ftruncate(), and the file will remain alive
3856 * then we need to clear up the orphan record which we created above.
3857 * However, if this was a real unlink then we were called by
3858 * ext4_delete_inode(), and we allow that function to clean up the
3859 * orphan info for us.
3862 ext4_orphan_del(handle, inode);
3864 ext4_journal_stop(handle);
3868 * ext4_get_inode_loc returns with an extra refcount against the inode's
3869 * underlying buffer_head on success. If 'in_mem' is true, we have all
3870 * data in memory that is needed to recreate the on-disk version of this
3873 static int __ext4_get_inode_loc(struct inode *inode,
3874 struct ext4_iloc *iloc, int in_mem)
3876 struct ext4_group_desc *gdp;
3877 struct buffer_head *bh;
3878 struct super_block *sb = inode->i_sb;
3880 int inodes_per_block, inode_offset;
3883 if (!ext4_valid_inum(sb, inode->i_ino))
3886 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3887 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3892 * Figure out the offset within the block group inode table
3894 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3895 inode_offset = ((inode->i_ino - 1) %
3896 EXT4_INODES_PER_GROUP(sb));
3897 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3898 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3900 bh = sb_getblk(sb, block);
3902 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3903 "inode block - inode=%lu, block=%llu",
3904 inode->i_ino, block);
3907 if (!buffer_uptodate(bh)) {
3911 * If the buffer has the write error flag, we have failed
3912 * to write out another inode in the same block. In this
3913 * case, we don't have to read the block because we may
3914 * read the old inode data successfully.
3916 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3917 set_buffer_uptodate(bh);
3919 if (buffer_uptodate(bh)) {
3920 /* someone brought it uptodate while we waited */
3926 * If we have all information of the inode in memory and this
3927 * is the only valid inode in the block, we need not read the
3931 struct buffer_head *bitmap_bh;
3934 start = inode_offset & ~(inodes_per_block - 1);
3936 /* Is the inode bitmap in cache? */
3937 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3942 * If the inode bitmap isn't in cache then the
3943 * optimisation may end up performing two reads instead
3944 * of one, so skip it.
3946 if (!buffer_uptodate(bitmap_bh)) {
3950 for (i = start; i < start + inodes_per_block; i++) {
3951 if (i == inode_offset)
3953 if (ext4_test_bit(i, bitmap_bh->b_data))
3957 if (i == start + inodes_per_block) {
3958 /* all other inodes are free, so skip I/O */
3959 memset(bh->b_data, 0, bh->b_size);
3960 set_buffer_uptodate(bh);
3968 * If we need to do any I/O, try to pre-readahead extra
3969 * blocks from the inode table.
3971 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3972 ext4_fsblk_t b, end, table;
3975 table = ext4_inode_table(sb, gdp);
3976 /* Make sure s_inode_readahead_blks is a power of 2 */
3977 while (EXT4_SB(sb)->s_inode_readahead_blks &
3978 (EXT4_SB(sb)->s_inode_readahead_blks-1))
3979 EXT4_SB(sb)->s_inode_readahead_blks =
3980 (EXT4_SB(sb)->s_inode_readahead_blks &
3981 (EXT4_SB(sb)->s_inode_readahead_blks-1));
3982 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3985 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3986 num = EXT4_INODES_PER_GROUP(sb);
3987 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3988 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3989 num -= le16_to_cpu(gdp->bg_itable_unused);
3990 table += num / inodes_per_block;
3994 sb_breadahead(sb, b++);
3998 * There are other valid inodes in the buffer, this inode
3999 * has in-inode xattrs, or we don't have this inode in memory.
4000 * Read the block from disk.
4003 bh->b_end_io = end_buffer_read_sync;
4004 submit_bh(READ_META, bh);
4006 if (!buffer_uptodate(bh)) {
4007 ext4_error(sb, __func__,
4008 "unable to read inode block - inode=%lu, "
4009 "block=%llu", inode->i_ino, block);
4019 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4021 /* We have all inode data except xattrs in memory here. */
4022 return __ext4_get_inode_loc(inode, iloc,
4023 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4026 void ext4_set_inode_flags(struct inode *inode)
4028 unsigned int flags = EXT4_I(inode)->i_flags;
4030 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4031 if (flags & EXT4_SYNC_FL)
4032 inode->i_flags |= S_SYNC;
4033 if (flags & EXT4_APPEND_FL)
4034 inode->i_flags |= S_APPEND;
4035 if (flags & EXT4_IMMUTABLE_FL)
4036 inode->i_flags |= S_IMMUTABLE;
4037 if (flags & EXT4_NOATIME_FL)
4038 inode->i_flags |= S_NOATIME;
4039 if (flags & EXT4_DIRSYNC_FL)
4040 inode->i_flags |= S_DIRSYNC;
4043 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4044 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4046 unsigned int flags = ei->vfs_inode.i_flags;
4048 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4049 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4051 ei->i_flags |= EXT4_SYNC_FL;
4052 if (flags & S_APPEND)
4053 ei->i_flags |= EXT4_APPEND_FL;
4054 if (flags & S_IMMUTABLE)
4055 ei->i_flags |= EXT4_IMMUTABLE_FL;
4056 if (flags & S_NOATIME)
4057 ei->i_flags |= EXT4_NOATIME_FL;
4058 if (flags & S_DIRSYNC)
4059 ei->i_flags |= EXT4_DIRSYNC_FL;
4061 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4062 struct ext4_inode_info *ei)
4065 struct inode *inode = &(ei->vfs_inode);
4066 struct super_block *sb = inode->i_sb;
4068 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4069 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4070 /* we are using combined 48 bit field */
4071 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4072 le32_to_cpu(raw_inode->i_blocks_lo);
4073 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4074 /* i_blocks represent file system block size */
4075 return i_blocks << (inode->i_blkbits - 9);
4080 return le32_to_cpu(raw_inode->i_blocks_lo);
4084 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4086 struct ext4_iloc iloc;
4087 struct ext4_inode *raw_inode;
4088 struct ext4_inode_info *ei;
4089 struct buffer_head *bh;
4090 struct inode *inode;
4094 inode = iget_locked(sb, ino);
4096 return ERR_PTR(-ENOMEM);
4097 if (!(inode->i_state & I_NEW))
4101 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4102 ei->i_acl = EXT4_ACL_NOT_CACHED;
4103 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4106 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4110 raw_inode = ext4_raw_inode(&iloc);
4111 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4112 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4113 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4114 if (!(test_opt(inode->i_sb, NO_UID32))) {
4115 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4116 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4118 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4121 ei->i_dir_start_lookup = 0;
4122 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4123 /* We now have enough fields to check if the inode was active or not.
4124 * This is needed because nfsd might try to access dead inodes
4125 * the test is that same one that e2fsck uses
4126 * NeilBrown 1999oct15
4128 if (inode->i_nlink == 0) {
4129 if (inode->i_mode == 0 ||
4130 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4131 /* this inode is deleted */
4136 /* The only unlinked inodes we let through here have
4137 * valid i_mode and are being read by the orphan
4138 * recovery code: that's fine, we're about to complete
4139 * the process of deleting those. */
4141 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4142 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4143 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4144 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4145 cpu_to_le32(EXT4_OS_HURD)) {
4147 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4149 inode->i_size = ext4_isize(raw_inode);
4150 ei->i_disksize = inode->i_size;
4151 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4152 ei->i_block_group = iloc.block_group;
4154 * NOTE! The in-memory inode i_data array is in little-endian order
4155 * even on big-endian machines: we do NOT byteswap the block numbers!
4157 for (block = 0; block < EXT4_N_BLOCKS; block++)
4158 ei->i_data[block] = raw_inode->i_block[block];
4159 INIT_LIST_HEAD(&ei->i_orphan);
4161 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4162 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4163 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4164 EXT4_INODE_SIZE(inode->i_sb)) {
4169 if (ei->i_extra_isize == 0) {
4170 /* The extra space is currently unused. Use it. */
4171 ei->i_extra_isize = sizeof(struct ext4_inode) -
4172 EXT4_GOOD_OLD_INODE_SIZE;
4174 __le32 *magic = (void *)raw_inode +
4175 EXT4_GOOD_OLD_INODE_SIZE +
4177 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4178 ei->i_state |= EXT4_STATE_XATTR;
4181 ei->i_extra_isize = 0;
4183 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4184 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4185 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4186 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4188 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4189 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4190 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4192 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4195 if (S_ISREG(inode->i_mode)) {
4196 inode->i_op = &ext4_file_inode_operations;
4197 inode->i_fop = &ext4_file_operations;
4198 ext4_set_aops(inode);
4199 } else if (S_ISDIR(inode->i_mode)) {
4200 inode->i_op = &ext4_dir_inode_operations;
4201 inode->i_fop = &ext4_dir_operations;
4202 } else if (S_ISLNK(inode->i_mode)) {
4203 if (ext4_inode_is_fast_symlink(inode))
4204 inode->i_op = &ext4_fast_symlink_inode_operations;
4206 inode->i_op = &ext4_symlink_inode_operations;
4207 ext4_set_aops(inode);
4210 inode->i_op = &ext4_special_inode_operations;
4211 if (raw_inode->i_block[0])
4212 init_special_inode(inode, inode->i_mode,
4213 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4215 init_special_inode(inode, inode->i_mode,
4216 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4219 ext4_set_inode_flags(inode);
4220 unlock_new_inode(inode);
4225 return ERR_PTR(ret);
4228 static int ext4_inode_blocks_set(handle_t *handle,
4229 struct ext4_inode *raw_inode,
4230 struct ext4_inode_info *ei)
4232 struct inode *inode = &(ei->vfs_inode);
4233 u64 i_blocks = inode->i_blocks;
4234 struct super_block *sb = inode->i_sb;
4236 if (i_blocks <= ~0U) {
4238 * i_blocks can be represnted in a 32 bit variable
4239 * as multiple of 512 bytes
4241 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4242 raw_inode->i_blocks_high = 0;
4243 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4246 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4249 if (i_blocks <= 0xffffffffffffULL) {
4251 * i_blocks can be represented in a 48 bit variable
4252 * as multiple of 512 bytes
4254 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4255 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4256 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4258 ei->i_flags |= EXT4_HUGE_FILE_FL;
4259 /* i_block is stored in file system block size */
4260 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4261 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4262 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4268 * Post the struct inode info into an on-disk inode location in the
4269 * buffer-cache. This gobbles the caller's reference to the
4270 * buffer_head in the inode location struct.
4272 * The caller must have write access to iloc->bh.
4274 static int ext4_do_update_inode(handle_t *handle,
4275 struct inode *inode,
4276 struct ext4_iloc *iloc)
4278 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4279 struct ext4_inode_info *ei = EXT4_I(inode);
4280 struct buffer_head *bh = iloc->bh;
4281 int err = 0, rc, block;
4283 /* For fields not not tracking in the in-memory inode,
4284 * initialise them to zero for new inodes. */
4285 if (ei->i_state & EXT4_STATE_NEW)
4286 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4288 ext4_get_inode_flags(ei);
4289 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4290 if (!(test_opt(inode->i_sb, NO_UID32))) {
4291 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4292 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4294 * Fix up interoperability with old kernels. Otherwise, old inodes get
4295 * re-used with the upper 16 bits of the uid/gid intact
4298 raw_inode->i_uid_high =
4299 cpu_to_le16(high_16_bits(inode->i_uid));
4300 raw_inode->i_gid_high =
4301 cpu_to_le16(high_16_bits(inode->i_gid));
4303 raw_inode->i_uid_high = 0;
4304 raw_inode->i_gid_high = 0;
4307 raw_inode->i_uid_low =
4308 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4309 raw_inode->i_gid_low =
4310 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4311 raw_inode->i_uid_high = 0;
4312 raw_inode->i_gid_high = 0;
4314 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4316 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4317 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4318 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4319 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4321 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4323 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4324 /* clear the migrate flag in the raw_inode */
4325 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4326 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4327 cpu_to_le32(EXT4_OS_HURD))
4328 raw_inode->i_file_acl_high =
4329 cpu_to_le16(ei->i_file_acl >> 32);
4330 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4331 ext4_isize_set(raw_inode, ei->i_disksize);
4332 if (ei->i_disksize > 0x7fffffffULL) {
4333 struct super_block *sb = inode->i_sb;
4334 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4335 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4336 EXT4_SB(sb)->s_es->s_rev_level ==
4337 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4338 /* If this is the first large file
4339 * created, add a flag to the superblock.
4341 err = ext4_journal_get_write_access(handle,
4342 EXT4_SB(sb)->s_sbh);
4345 ext4_update_dynamic_rev(sb);
4346 EXT4_SET_RO_COMPAT_FEATURE(sb,
4347 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4350 err = ext4_journal_dirty_metadata(handle,
4351 EXT4_SB(sb)->s_sbh);
4354 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4355 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4356 if (old_valid_dev(inode->i_rdev)) {
4357 raw_inode->i_block[0] =
4358 cpu_to_le32(old_encode_dev(inode->i_rdev));
4359 raw_inode->i_block[1] = 0;
4361 raw_inode->i_block[0] = 0;
4362 raw_inode->i_block[1] =
4363 cpu_to_le32(new_encode_dev(inode->i_rdev));
4364 raw_inode->i_block[2] = 0;
4366 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4367 raw_inode->i_block[block] = ei->i_data[block];
4369 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4370 if (ei->i_extra_isize) {
4371 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4372 raw_inode->i_version_hi =
4373 cpu_to_le32(inode->i_version >> 32);
4374 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4378 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4379 rc = ext4_journal_dirty_metadata(handle, bh);
4382 ei->i_state &= ~EXT4_STATE_NEW;
4386 ext4_std_error(inode->i_sb, err);
4391 * ext4_write_inode()
4393 * We are called from a few places:
4395 * - Within generic_file_write() for O_SYNC files.
4396 * Here, there will be no transaction running. We wait for any running
4397 * trasnaction to commit.
4399 * - Within sys_sync(), kupdate and such.
4400 * We wait on commit, if tol to.
4402 * - Within prune_icache() (PF_MEMALLOC == true)
4403 * Here we simply return. We can't afford to block kswapd on the
4406 * In all cases it is actually safe for us to return without doing anything,
4407 * because the inode has been copied into a raw inode buffer in
4408 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4411 * Note that we are absolutely dependent upon all inode dirtiers doing the
4412 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4413 * which we are interested.
4415 * It would be a bug for them to not do this. The code:
4417 * mark_inode_dirty(inode)
4419 * inode->i_size = expr;
4421 * is in error because a kswapd-driven write_inode() could occur while
4422 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4423 * will no longer be on the superblock's dirty inode list.
4425 int ext4_write_inode(struct inode *inode, int wait)
4427 if (current->flags & PF_MEMALLOC)
4430 if (ext4_journal_current_handle()) {
4431 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4439 return ext4_force_commit(inode->i_sb);
4445 * Called from notify_change.
4447 * We want to trap VFS attempts to truncate the file as soon as
4448 * possible. In particular, we want to make sure that when the VFS
4449 * shrinks i_size, we put the inode on the orphan list and modify
4450 * i_disksize immediately, so that during the subsequent flushing of
4451 * dirty pages and freeing of disk blocks, we can guarantee that any
4452 * commit will leave the blocks being flushed in an unused state on
4453 * disk. (On recovery, the inode will get truncated and the blocks will
4454 * be freed, so we have a strong guarantee that no future commit will
4455 * leave these blocks visible to the user.)
4457 * Another thing we have to assure is that if we are in ordered mode
4458 * and inode is still attached to the committing transaction, we must
4459 * we start writeout of all the dirty pages which are being truncated.
4460 * This way we are sure that all the data written in the previous
4461 * transaction are already on disk (truncate waits for pages under
4464 * Called with inode->i_mutex down.
4466 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4468 struct inode *inode = dentry->d_inode;
4470 const unsigned int ia_valid = attr->ia_valid;
4472 error = inode_change_ok(inode, attr);
4476 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4477 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4480 /* (user+group)*(old+new) structure, inode write (sb,
4481 * inode block, ? - but truncate inode update has it) */
4482 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4483 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4484 if (IS_ERR(handle)) {
4485 error = PTR_ERR(handle);
4488 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4490 ext4_journal_stop(handle);
4493 /* Update corresponding info in inode so that everything is in
4494 * one transaction */
4495 if (attr->ia_valid & ATTR_UID)
4496 inode->i_uid = attr->ia_uid;
4497 if (attr->ia_valid & ATTR_GID)
4498 inode->i_gid = attr->ia_gid;
4499 error = ext4_mark_inode_dirty(handle, inode);
4500 ext4_journal_stop(handle);
4503 if (attr->ia_valid & ATTR_SIZE) {
4504 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4505 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4507 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4514 if (S_ISREG(inode->i_mode) &&
4515 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4518 handle = ext4_journal_start(inode, 3);
4519 if (IS_ERR(handle)) {
4520 error = PTR_ERR(handle);
4524 error = ext4_orphan_add(handle, inode);
4525 EXT4_I(inode)->i_disksize = attr->ia_size;
4526 rc = ext4_mark_inode_dirty(handle, inode);
4529 ext4_journal_stop(handle);
4531 if (ext4_should_order_data(inode)) {
4532 error = ext4_begin_ordered_truncate(inode,
4535 /* Do as much error cleanup as possible */
4536 handle = ext4_journal_start(inode, 3);
4537 if (IS_ERR(handle)) {
4538 ext4_orphan_del(NULL, inode);
4541 ext4_orphan_del(handle, inode);
4542 ext4_journal_stop(handle);
4548 rc = inode_setattr(inode, attr);
4550 /* If inode_setattr's call to ext4_truncate failed to get a
4551 * transaction handle at all, we need to clean up the in-core
4552 * orphan list manually. */
4554 ext4_orphan_del(NULL, inode);
4556 if (!rc && (ia_valid & ATTR_MODE))
4557 rc = ext4_acl_chmod(inode);
4560 ext4_std_error(inode->i_sb, error);
4566 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4569 struct inode *inode;
4570 unsigned long delalloc_blocks;
4572 inode = dentry->d_inode;
4573 generic_fillattr(inode, stat);
4576 * We can't update i_blocks if the block allocation is delayed
4577 * otherwise in the case of system crash before the real block
4578 * allocation is done, we will have i_blocks inconsistent with
4579 * on-disk file blocks.
4580 * We always keep i_blocks updated together with real
4581 * allocation. But to not confuse with user, stat
4582 * will return the blocks that include the delayed allocation
4583 * blocks for this file.
4585 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4586 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4587 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4589 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4593 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4598 /* if nrblocks are contiguous */
4601 * With N contiguous data blocks, it need at most
4602 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4603 * 2 dindirect blocks
4606 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4607 return indirects + 3;
4610 * if nrblocks are not contiguous, worse case, each block touch
4611 * a indirect block, and each indirect block touch a double indirect
4612 * block, plus a triple indirect block
4614 indirects = nrblocks * 2 + 1;
4618 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4620 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4621 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4622 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4626 * Account for index blocks, block groups bitmaps and block group
4627 * descriptor blocks if modify datablocks and index blocks
4628 * worse case, the indexs blocks spread over different block groups
4630 * If datablocks are discontiguous, they are possible to spread over
4631 * different block groups too. If they are contiugous, with flexbg,
4632 * they could still across block group boundary.
4634 * Also account for superblock, inode, quota and xattr blocks
4636 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4638 int groups, gdpblocks;
4643 * How many index blocks need to touch to modify nrblocks?
4644 * The "Chunk" flag indicating whether the nrblocks is
4645 * physically contiguous on disk
4647 * For Direct IO and fallocate, they calls get_block to allocate
4648 * one single extent at a time, so they could set the "Chunk" flag
4650 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4655 * Now let's see how many group bitmaps and group descriptors need
4665 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4666 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4667 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4668 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4670 /* bitmaps and block group descriptor blocks */
4671 ret += groups + gdpblocks;
4673 /* Blocks for super block, inode, quota and xattr blocks */
4674 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4680 * Calulate the total number of credits to reserve to fit
4681 * the modification of a single pages into a single transaction,
4682 * which may include multiple chunks of block allocations.
4684 * This could be called via ext4_write_begin()
4686 * We need to consider the worse case, when
4687 * one new block per extent.
4689 int ext4_writepage_trans_blocks(struct inode *inode)
4691 int bpp = ext4_journal_blocks_per_page(inode);
4694 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4696 /* Account for data blocks for journalled mode */
4697 if (ext4_should_journal_data(inode))
4703 * Calculate the journal credits for a chunk of data modification.
4705 * This is called from DIO, fallocate or whoever calling
4706 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4708 * journal buffers for data blocks are not included here, as DIO
4709 * and fallocate do no need to journal data buffers.
4711 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4713 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4717 * The caller must have previously called ext4_reserve_inode_write().
4718 * Give this, we know that the caller already has write access to iloc->bh.
4720 int ext4_mark_iloc_dirty(handle_t *handle,
4721 struct inode *inode, struct ext4_iloc *iloc)
4725 if (test_opt(inode->i_sb, I_VERSION))
4726 inode_inc_iversion(inode);
4728 /* the do_update_inode consumes one bh->b_count */
4731 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4732 err = ext4_do_update_inode(handle, inode, iloc);
4738 * On success, We end up with an outstanding reference count against
4739 * iloc->bh. This _must_ be cleaned up later.
4743 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4744 struct ext4_iloc *iloc)
4748 err = ext4_get_inode_loc(inode, iloc);
4750 BUFFER_TRACE(iloc->bh, "get_write_access");
4751 err = ext4_journal_get_write_access(handle, iloc->bh);
4758 ext4_std_error(inode->i_sb, err);
4763 * Expand an inode by new_extra_isize bytes.
4764 * Returns 0 on success or negative error number on failure.
4766 static int ext4_expand_extra_isize(struct inode *inode,
4767 unsigned int new_extra_isize,
4768 struct ext4_iloc iloc,
4771 struct ext4_inode *raw_inode;
4772 struct ext4_xattr_ibody_header *header;
4773 struct ext4_xattr_entry *entry;
4775 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4778 raw_inode = ext4_raw_inode(&iloc);
4780 header = IHDR(inode, raw_inode);
4781 entry = IFIRST(header);
4783 /* No extended attributes present */
4784 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4785 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4786 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4788 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4792 /* try to expand with EAs present */
4793 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4798 * What we do here is to mark the in-core inode as clean with respect to inode
4799 * dirtiness (it may still be data-dirty).
4800 * This means that the in-core inode may be reaped by prune_icache
4801 * without having to perform any I/O. This is a very good thing,
4802 * because *any* task may call prune_icache - even ones which
4803 * have a transaction open against a different journal.
4805 * Is this cheating? Not really. Sure, we haven't written the
4806 * inode out, but prune_icache isn't a user-visible syncing function.
4807 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4808 * we start and wait on commits.
4810 * Is this efficient/effective? Well, we're being nice to the system
4811 * by cleaning up our inodes proactively so they can be reaped
4812 * without I/O. But we are potentially leaving up to five seconds'
4813 * worth of inodes floating about which prune_icache wants us to
4814 * write out. One way to fix that would be to get prune_icache()
4815 * to do a write_super() to free up some memory. It has the desired
4818 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4820 struct ext4_iloc iloc;
4821 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4822 static unsigned int mnt_count;
4826 err = ext4_reserve_inode_write(handle, inode, &iloc);
4827 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4828 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4830 * We need extra buffer credits since we may write into EA block
4831 * with this same handle. If journal_extend fails, then it will
4832 * only result in a minor loss of functionality for that inode.
4833 * If this is felt to be critical, then e2fsck should be run to
4834 * force a large enough s_min_extra_isize.
4836 if ((jbd2_journal_extend(handle,
4837 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4838 ret = ext4_expand_extra_isize(inode,
4839 sbi->s_want_extra_isize,
4842 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4844 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4845 ext4_warning(inode->i_sb, __func__,
4846 "Unable to expand inode %lu. Delete"
4847 " some EAs or run e2fsck.",
4850 le16_to_cpu(sbi->s_es->s_mnt_count);
4856 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4861 * ext4_dirty_inode() is called from __mark_inode_dirty()
4863 * We're really interested in the case where a file is being extended.
4864 * i_size has been changed by generic_commit_write() and we thus need
4865 * to include the updated inode in the current transaction.
4867 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4868 * are allocated to the file.
4870 * If the inode is marked synchronous, we don't honour that here - doing
4871 * so would cause a commit on atime updates, which we don't bother doing.
4872 * We handle synchronous inodes at the highest possible level.
4874 void ext4_dirty_inode(struct inode *inode)
4876 handle_t *current_handle = ext4_journal_current_handle();
4879 handle = ext4_journal_start(inode, 2);
4882 if (current_handle &&
4883 current_handle->h_transaction != handle->h_transaction) {
4884 /* This task has a transaction open against a different fs */
4885 printk(KERN_EMERG "%s: transactions do not match!\n",
4888 jbd_debug(5, "marking dirty. outer handle=%p\n",
4890 ext4_mark_inode_dirty(handle, inode);
4892 ext4_journal_stop(handle);
4899 * Bind an inode's backing buffer_head into this transaction, to prevent
4900 * it from being flushed to disk early. Unlike
4901 * ext4_reserve_inode_write, this leaves behind no bh reference and
4902 * returns no iloc structure, so the caller needs to repeat the iloc
4903 * lookup to mark the inode dirty later.
4905 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4907 struct ext4_iloc iloc;
4911 err = ext4_get_inode_loc(inode, &iloc);
4913 BUFFER_TRACE(iloc.bh, "get_write_access");
4914 err = jbd2_journal_get_write_access(handle, iloc.bh);
4916 err = ext4_journal_dirty_metadata(handle,
4921 ext4_std_error(inode->i_sb, err);
4926 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4933 * We have to be very careful here: changing a data block's
4934 * journaling status dynamically is dangerous. If we write a
4935 * data block to the journal, change the status and then delete
4936 * that block, we risk forgetting to revoke the old log record
4937 * from the journal and so a subsequent replay can corrupt data.
4938 * So, first we make sure that the journal is empty and that
4939 * nobody is changing anything.
4942 journal = EXT4_JOURNAL(inode);
4943 if (is_journal_aborted(journal))
4946 jbd2_journal_lock_updates(journal);
4947 jbd2_journal_flush(journal);
4950 * OK, there are no updates running now, and all cached data is
4951 * synced to disk. We are now in a completely consistent state
4952 * which doesn't have anything in the journal, and we know that
4953 * no filesystem updates are running, so it is safe to modify
4954 * the inode's in-core data-journaling state flag now.
4958 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4960 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4961 ext4_set_aops(inode);
4963 jbd2_journal_unlock_updates(journal);
4965 /* Finally we can mark the inode as dirty. */
4967 handle = ext4_journal_start(inode, 1);
4969 return PTR_ERR(handle);
4971 err = ext4_mark_inode_dirty(handle, inode);
4973 ext4_journal_stop(handle);
4974 ext4_std_error(inode->i_sb, err);
4979 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4981 return !buffer_mapped(bh);
4984 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4990 struct file *file = vma->vm_file;
4991 struct inode *inode = file->f_path.dentry->d_inode;
4992 struct address_space *mapping = inode->i_mapping;
4995 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4996 * get i_mutex because we are already holding mmap_sem.
4998 down_read(&inode->i_alloc_sem);
4999 size = i_size_read(inode);
5000 if (page->mapping != mapping || size <= page_offset(page)
5001 || !PageUptodate(page)) {
5002 /* page got truncated from under us? */
5006 if (PageMappedToDisk(page))
5009 if (page->index == size >> PAGE_CACHE_SHIFT)
5010 len = size & ~PAGE_CACHE_MASK;
5012 len = PAGE_CACHE_SIZE;
5014 if (page_has_buffers(page)) {
5015 /* return if we have all the buffers mapped */
5016 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5021 * OK, we need to fill the hole... Do write_begin write_end
5022 * to do block allocation/reservation.We are not holding
5023 * inode.i__mutex here. That allow * parallel write_begin,
5024 * write_end call. lock_page prevent this from happening
5025 * on the same page though
5027 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5028 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5031 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5032 len, len, page, fsdata);
5037 up_read(&inode->i_alloc_sem);