2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode *inode,
49 return jbd2_journal_begin_ordered_truncate(
50 EXT4_SB(inode->i_sb)->s_journal,
51 &EXT4_I(inode)->jinode,
55 static void ext4_invalidatepage(struct page *page, unsigned long offset);
58 * Test whether an inode is a fast symlink.
60 static int ext4_inode_is_fast_symlink(struct inode *inode)
62 int ea_blocks = EXT4_I(inode)->i_file_acl ?
63 (inode->i_sb->s_blocksize >> 9) : 0;
65 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
69 * The ext4 forget function must perform a revoke if we are freeing data
70 * which has been journaled. Metadata (eg. indirect blocks) must be
71 * revoked in all cases.
73 * "bh" may be NULL: a metadata block may have been freed from memory
74 * but there may still be a record of it in the journal, and that record
75 * still needs to be revoked.
77 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
78 struct buffer_head *bh, ext4_fsblk_t blocknr)
84 BUFFER_TRACE(bh, "enter");
86 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
88 bh, is_metadata, inode->i_mode,
89 test_opt(inode->i_sb, DATA_FLAGS));
91 /* Never use the revoke function if we are doing full data
92 * journaling: there is no need to, and a V1 superblock won't
93 * support it. Otherwise, only skip the revoke on un-journaled
96 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
97 (!is_metadata && !ext4_should_journal_data(inode))) {
99 BUFFER_TRACE(bh, "call jbd2_journal_forget");
100 return ext4_journal_forget(handle, bh);
106 * data!=journal && (is_metadata || should_journal_data(inode))
108 BUFFER_TRACE(bh, "call ext4_journal_revoke");
109 err = ext4_journal_revoke(handle, blocknr, bh);
111 ext4_abort(inode->i_sb, __func__,
112 "error %d when attempting revoke", err);
113 BUFFER_TRACE(bh, "exit");
118 * Work out how many blocks we need to proceed with the next chunk of a
119 * truncate transaction.
121 static unsigned long blocks_for_truncate(struct inode *inode)
125 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
127 /* Give ourselves just enough room to cope with inodes in which
128 * i_blocks is corrupt: we've seen disk corruptions in the past
129 * which resulted in random data in an inode which looked enough
130 * like a regular file for ext4 to try to delete it. Things
131 * will go a bit crazy if that happens, but at least we should
132 * try not to panic the whole kernel. */
136 /* But we need to bound the transaction so we don't overflow the
138 if (needed > EXT4_MAX_TRANS_DATA)
139 needed = EXT4_MAX_TRANS_DATA;
141 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
145 * Truncate transactions can be complex and absolutely huge. So we need to
146 * be able to restart the transaction at a conventient checkpoint to make
147 * sure we don't overflow the journal.
149 * start_transaction gets us a new handle for a truncate transaction,
150 * and extend_transaction tries to extend the existing one a bit. If
151 * extend fails, we need to propagate the failure up and restart the
152 * transaction in the top-level truncate loop. --sct
154 static handle_t *start_transaction(struct inode *inode)
158 result = ext4_journal_start(inode, blocks_for_truncate(inode));
162 ext4_std_error(inode->i_sb, PTR_ERR(result));
167 * Try to extend this transaction for the purposes of truncation.
169 * Returns 0 if we managed to create more room. If we can't create more
170 * room, and the transaction must be restarted we return 1.
172 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
174 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
176 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
182 * Restart the transaction associated with *handle. This does a commit,
183 * so before we call here everything must be consistently dirtied against
186 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
188 jbd_debug(2, "restarting handle %p\n", handle);
189 return ext4_journal_restart(handle, blocks_for_truncate(inode));
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_delete_inode (struct inode * inode)
200 if (ext4_should_order_data(inode))
201 ext4_begin_ordered_truncate(inode, 0);
202 truncate_inode_pages(&inode->i_data, 0);
204 if (is_bad_inode(inode))
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL, inode);
222 err = ext4_mark_inode_dirty(handle, inode);
224 ext4_warning(inode->i_sb, __func__,
225 "couldn't mark inode dirty (err %d)", err);
229 ext4_truncate(inode);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (handle->h_buffer_credits < 3) {
238 err = ext4_journal_extend(handle, 3);
240 err = ext4_journal_restart(handle, 3);
242 ext4_warning(inode->i_sb, __func__,
243 "couldn't extend journal (err %d)", err);
245 ext4_journal_stop(handle);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle, inode);
259 EXT4_I(inode)->i_dtime = get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle, inode))
269 /* If that failed, just do the required in-core inode clear. */
272 ext4_free_inode(handle, inode);
273 ext4_journal_stop(handle);
276 clear_inode(inode); /* We must guarantee clearing of inode... */
282 struct buffer_head *bh;
285 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287 p->key = *(p->p = v);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t offsets[4], int *boundary)
326 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
327 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
328 const long direct_blocks = EXT4_NDIR_BLOCKS,
329 indirect_blocks = ptrs,
330 double_blocks = (1 << (ptrs_bits * 2));
335 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
336 } else if (i_block < direct_blocks) {
337 offsets[n++] = i_block;
338 final = direct_blocks;
339 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
340 offsets[n++] = EXT4_IND_BLOCK;
341 offsets[n++] = i_block;
343 } else if ((i_block -= indirect_blocks) < double_blocks) {
344 offsets[n++] = EXT4_DIND_BLOCK;
345 offsets[n++] = i_block >> ptrs_bits;
346 offsets[n++] = i_block & (ptrs - 1);
348 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
349 offsets[n++] = EXT4_TIND_BLOCK;
350 offsets[n++] = i_block >> (ptrs_bits * 2);
351 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
352 offsets[n++] = i_block & (ptrs - 1);
355 ext4_warning(inode->i_sb, "ext4_block_to_path",
356 "block %lu > max in inode %lu",
357 i_block + direct_blocks +
358 indirect_blocks + double_blocks, inode->i_ino);
361 *boundary = final - 1 - (i_block & (ptrs - 1));
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect *ext4_get_branch(struct inode *inode, int depth,
396 ext4_lblk_t *offsets,
397 Indirect chain[4], int *err)
399 struct super_block *sb = inode->i_sb;
401 struct buffer_head *bh;
404 /* i_data is not going away, no lock needed */
405 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
409 bh = sb_bread(sb, le32_to_cpu(p->key));
412 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
426 * ext4_find_near - find a place for allocation with sufficient locality
428 * @ind: descriptor of indirect block.
430 * This function returns the preferred place for block allocation.
431 * It is used when heuristic for sequential allocation fails.
433 * + if there is a block to the left of our position - allocate near it.
434 * + if pointer will live in indirect block - allocate near that block.
435 * + if pointer will live in inode - allocate in the same
438 * In the latter case we colour the starting block by the callers PID to
439 * prevent it from clashing with concurrent allocations for a different inode
440 * in the same block group. The PID is used here so that functionally related
441 * files will be close-by on-disk.
443 * Caller must make sure that @ind is valid and will stay that way.
445 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
447 struct ext4_inode_info *ei = EXT4_I(inode);
448 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
450 ext4_fsblk_t bg_start;
451 ext4_fsblk_t last_block;
452 ext4_grpblk_t colour;
454 /* Try to find previous block */
455 for (p = ind->p - 1; p >= start; p--) {
457 return le32_to_cpu(*p);
460 /* No such thing, so let's try location of indirect block */
462 return ind->bh->b_blocknr;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
469 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
471 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
472 colour = (current->pid % 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
475 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
476 return bg_start + colour;
480 * ext4_find_goal - find a preferred place for allocation.
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
485 * Normally this function find the preferred place for block allocation,
488 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
491 struct ext4_block_alloc_info *block_i;
493 block_i = EXT4_I(inode)->i_block_alloc_info;
496 * try the heuristic for sequential allocation,
497 * failing that at least try to get decent locality.
499 if (block_i && (block == block_i->last_alloc_logical_block + 1)
500 && (block_i->last_alloc_physical_block != 0)) {
501 return block_i->last_alloc_physical_block + 1;
504 return ext4_find_near(inode, partial);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
520 int blocks_to_boundary)
522 unsigned long count = 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks < blocks_to_boundary + 1)
533 count += blocks_to_boundary + 1;
538 while (count < blks && count <= blocks_to_boundary &&
539 le32_to_cpu(*(branch[0].p + count)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
556 ext4_lblk_t iblock, ext4_fsblk_t goal,
557 int indirect_blks, int blks,
558 ext4_fsblk_t new_blocks[4], int *err)
561 unsigned long count = 0, blk_allocated = 0;
563 ext4_fsblk_t current_block = 0;
567 * Here we try to allocate the requested multiple blocks at once,
568 * on a best-effort basis.
569 * To build a branch, we should allocate blocks for
570 * the indirect blocks(if not allocated yet), and at least
571 * the first direct block of this branch. That's the
572 * minimum number of blocks need to allocate(required)
574 /* first we try to allocate the indirect blocks */
575 target = indirect_blks;
578 /* allocating blocks for indirect blocks and direct blocks */
579 current_block = ext4_new_meta_blocks(handle, inode,
585 /* allocate blocks for indirect blocks */
586 while (index < indirect_blks && count) {
587 new_blocks[index++] = current_block++;
592 * save the new block number
593 * for the first direct block
595 new_blocks[index] = current_block;
596 printk(KERN_INFO "%s returned more blocks than "
597 "requested\n", __func__);
603 target = blks - count ;
604 blk_allocated = count;
607 /* Now allocate data blocks */
609 /* allocating blocks for data blocks */
610 current_block = ext4_new_blocks(handle, inode, iblock,
612 if (*err && (target == blks)) {
614 * if the allocation failed and we didn't allocate
620 if (target == blks) {
622 * save the new block number
623 * for the first direct block
625 new_blocks[index] = current_block;
627 blk_allocated += count;
630 /* total number of blocks allocated for direct blocks */
635 for (i = 0; i <index; i++)
636 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
641 * ext4_alloc_branch - allocate and set up a chain of blocks.
643 * @indirect_blks: number of allocated indirect blocks
644 * @blks: number of allocated direct blocks
645 * @offsets: offsets (in the blocks) to store the pointers to next.
646 * @branch: place to store the chain in.
648 * This function allocates blocks, zeroes out all but the last one,
649 * links them into chain and (if we are synchronous) writes them to disk.
650 * In other words, it prepares a branch that can be spliced onto the
651 * inode. It stores the information about that chain in the branch[], in
652 * the same format as ext4_get_branch() would do. We are calling it after
653 * we had read the existing part of chain and partial points to the last
654 * triple of that (one with zero ->key). Upon the exit we have the same
655 * picture as after the successful ext4_get_block(), except that in one
656 * place chain is disconnected - *branch->p is still zero (we did not
657 * set the last link), but branch->key contains the number that should
658 * be placed into *branch->p to fill that gap.
660 * If allocation fails we free all blocks we've allocated (and forget
661 * their buffer_heads) and return the error value the from failed
662 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
663 * as described above and return 0.
665 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
666 ext4_lblk_t iblock, int indirect_blks,
667 int *blks, ext4_fsblk_t goal,
668 ext4_lblk_t *offsets, Indirect *branch)
670 int blocksize = inode->i_sb->s_blocksize;
673 struct buffer_head *bh;
675 ext4_fsblk_t new_blocks[4];
676 ext4_fsblk_t current_block;
678 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
679 *blks, new_blocks, &err);
683 branch[0].key = cpu_to_le32(new_blocks[0]);
685 * metadata blocks and data blocks are allocated.
687 for (n = 1; n <= indirect_blks; n++) {
689 * Get buffer_head for parent block, zero it out
690 * and set the pointer to new one, then send
693 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
696 BUFFER_TRACE(bh, "call get_create_access");
697 err = ext4_journal_get_create_access(handle, bh);
704 memset(bh->b_data, 0, blocksize);
705 branch[n].p = (__le32 *) bh->b_data + offsets[n];
706 branch[n].key = cpu_to_le32(new_blocks[n]);
707 *branch[n].p = branch[n].key;
708 if ( n == indirect_blks) {
709 current_block = new_blocks[n];
711 * End of chain, update the last new metablock of
712 * the chain to point to the new allocated
713 * data blocks numbers
715 for (i=1; i < num; i++)
716 *(branch[n].p + i) = cpu_to_le32(++current_block);
718 BUFFER_TRACE(bh, "marking uptodate");
719 set_buffer_uptodate(bh);
722 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
723 err = ext4_journal_dirty_metadata(handle, bh);
730 /* Allocation failed, free what we already allocated */
731 for (i = 1; i <= n ; i++) {
732 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
733 ext4_journal_forget(handle, branch[i].bh);
735 for (i = 0; i <indirect_blks; i++)
736 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
738 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
744 * ext4_splice_branch - splice the allocated branch onto inode.
746 * @block: (logical) number of block we are adding
747 * @chain: chain of indirect blocks (with a missing link - see
749 * @where: location of missing link
750 * @num: number of indirect blocks we are adding
751 * @blks: number of direct blocks we are adding
753 * This function fills the missing link and does all housekeeping needed in
754 * inode (->i_blocks, etc.). In case of success we end up with the full
755 * chain to new block and return 0.
757 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
758 ext4_lblk_t block, Indirect *where, int num, int blks)
762 struct ext4_block_alloc_info *block_i;
763 ext4_fsblk_t current_block;
765 block_i = EXT4_I(inode)->i_block_alloc_info;
767 * If we're splicing into a [td]indirect block (as opposed to the
768 * inode) then we need to get write access to the [td]indirect block
772 BUFFER_TRACE(where->bh, "get_write_access");
773 err = ext4_journal_get_write_access(handle, where->bh);
779 *where->p = where->key;
782 * Update the host buffer_head or inode to point to more just allocated
783 * direct blocks blocks
785 if (num == 0 && blks > 1) {
786 current_block = le32_to_cpu(where->key) + 1;
787 for (i = 1; i < blks; i++)
788 *(where->p + i ) = cpu_to_le32(current_block++);
792 * update the most recently allocated logical & physical block
793 * in i_block_alloc_info, to assist find the proper goal block for next
797 block_i->last_alloc_logical_block = block + blks - 1;
798 block_i->last_alloc_physical_block =
799 le32_to_cpu(where[num].key) + blks - 1;
802 /* We are done with atomic stuff, now do the rest of housekeeping */
804 inode->i_ctime = ext4_current_time(inode);
805 ext4_mark_inode_dirty(handle, inode);
807 /* had we spliced it onto indirect block? */
810 * If we spliced it onto an indirect block, we haven't
811 * altered the inode. Note however that if it is being spliced
812 * onto an indirect block at the very end of the file (the
813 * file is growing) then we *will* alter the inode to reflect
814 * the new i_size. But that is not done here - it is done in
815 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
817 jbd_debug(5, "splicing indirect only\n");
818 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
819 err = ext4_journal_dirty_metadata(handle, where->bh);
824 * OK, we spliced it into the inode itself on a direct block.
825 * Inode was dirtied above.
827 jbd_debug(5, "splicing direct\n");
832 for (i = 1; i <= num; i++) {
833 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
834 ext4_journal_forget(handle, where[i].bh);
835 ext4_free_blocks(handle, inode,
836 le32_to_cpu(where[i-1].key), 1, 0);
838 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
844 * Allocation strategy is simple: if we have to allocate something, we will
845 * have to go the whole way to leaf. So let's do it before attaching anything
846 * to tree, set linkage between the newborn blocks, write them if sync is
847 * required, recheck the path, free and repeat if check fails, otherwise
848 * set the last missing link (that will protect us from any truncate-generated
849 * removals - all blocks on the path are immune now) and possibly force the
850 * write on the parent block.
851 * That has a nice additional property: no special recovery from the failed
852 * allocations is needed - we simply release blocks and do not touch anything
853 * reachable from inode.
855 * `handle' can be NULL if create == 0.
857 * return > 0, # of blocks mapped or allocated.
858 * return = 0, if plain lookup failed.
859 * return < 0, error case.
862 * Need to be called with
863 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
864 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
866 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
867 ext4_lblk_t iblock, unsigned long maxblocks,
868 struct buffer_head *bh_result,
869 int create, int extend_disksize)
872 ext4_lblk_t offsets[4];
877 int blocks_to_boundary = 0;
879 struct ext4_inode_info *ei = EXT4_I(inode);
881 ext4_fsblk_t first_block = 0;
885 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
886 J_ASSERT(handle != NULL || create == 0);
887 depth = ext4_block_to_path(inode, iblock, offsets,
888 &blocks_to_boundary);
893 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
895 /* Simplest case - block found, no allocation needed */
897 first_block = le32_to_cpu(chain[depth - 1].key);
898 clear_buffer_new(bh_result);
901 while (count < maxblocks && count <= blocks_to_boundary) {
904 blk = le32_to_cpu(*(chain[depth-1].p + count));
906 if (blk == first_block + count)
914 /* Next simple case - plain lookup or failed read of indirect block */
915 if (!create || err == -EIO)
919 * Okay, we need to do block allocation. Lazily initialize the block
920 * allocation info here if necessary
922 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
923 ext4_init_block_alloc_info(inode);
925 goal = ext4_find_goal(inode, iblock, partial);
927 /* the number of blocks need to allocate for [d,t]indirect blocks */
928 indirect_blks = (chain + depth) - partial - 1;
931 * Next look up the indirect map to count the totoal number of
932 * direct blocks to allocate for this branch.
934 count = ext4_blks_to_allocate(partial, indirect_blks,
935 maxblocks, blocks_to_boundary);
937 * Block out ext4_truncate while we alter the tree
939 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
941 offsets + (partial - chain), partial);
944 * The ext4_splice_branch call will free and forget any buffers
945 * on the new chain if there is a failure, but that risks using
946 * up transaction credits, especially for bitmaps where the
947 * credits cannot be returned. Can we handle this somehow? We
948 * may need to return -EAGAIN upwards in the worst case. --sct
951 err = ext4_splice_branch(handle, inode, iblock,
952 partial, indirect_blks, count);
954 * i_disksize growing is protected by i_data_sem. Don't forget to
955 * protect it if you're about to implement concurrent
956 * ext4_get_block() -bzzz
958 if (!err && extend_disksize) {
959 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
960 if (disksize > i_size_read(inode))
961 disksize = i_size_read(inode);
962 if (disksize > ei->i_disksize)
963 ei->i_disksize = disksize;
968 set_buffer_new(bh_result);
970 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
971 if (count > blocks_to_boundary)
972 set_buffer_boundary(bh_result);
974 /* Clean up and exit */
975 partial = chain + depth - 1; /* the whole chain */
977 while (partial > chain) {
978 BUFFER_TRACE(partial->bh, "call brelse");
982 BUFFER_TRACE(bh_result, "returned");
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate @blocks for non extent file based file
991 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
993 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
994 int ind_blks, dind_blks, tind_blks;
996 /* number of new indirect blocks needed */
997 ind_blks = (blocks + icap - 1) / icap;
999 dind_blks = (ind_blks + icap - 1) / icap;
1003 return ind_blks + dind_blks + tind_blks;
1007 * Calculate the number of metadata blocks need to reserve
1008 * to allocate given number of blocks
1010 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1015 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1016 return ext4_ext_calc_metadata_amount(inode, blocks);
1018 return ext4_indirect_calc_metadata_amount(inode, blocks);
1021 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1023 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1024 int total, mdb, mdb_free;
1026 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1027 /* recalculate the number of metablocks still need to be reserved */
1028 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1029 mdb = ext4_calc_metadata_amount(inode, total);
1031 /* figure out how many metablocks to release */
1032 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1033 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1036 /* Account for allocated meta_blocks */
1037 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1039 /* update fs dirty blocks counter */
1040 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1041 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1042 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1045 /* update per-inode reservations */
1046 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1047 EXT4_I(inode)->i_reserved_data_blocks -= used;
1049 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1052 * If we have done all the pending block allocations and if
1053 * there aren't any writers on the inode, we can discard the
1054 * inode's preallocations.
1056 if (!total && (atomic_read(&inode->i_writecount) == 0))
1057 ext4_discard_reservation(inode);
1061 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1062 * and returns if the blocks are already mapped.
1064 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1065 * and store the allocated blocks in the result buffer head and mark it
1068 * If file type is extents based, it will call ext4_ext_get_blocks(),
1069 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1072 * On success, it returns the number of blocks being mapped or allocate.
1073 * if create==0 and the blocks are pre-allocated and uninitialized block,
1074 * the result buffer head is unmapped. If the create ==1, it will make sure
1075 * the buffer head is mapped.
1077 * It returns 0 if plain look up failed (blocks have not been allocated), in
1078 * that casem, buffer head is unmapped
1080 * It returns the error in case of allocation failure.
1082 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1083 unsigned long max_blocks, struct buffer_head *bh,
1084 int create, int extend_disksize, int flag)
1088 clear_buffer_mapped(bh);
1089 clear_buffer_unwritten(bh);
1092 * Try to see if we can get the block without requesting
1093 * for new file system block.
1095 down_read((&EXT4_I(inode)->i_data_sem));
1096 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1097 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1100 retval = ext4_get_blocks_handle(handle,
1101 inode, block, max_blocks, bh, 0, 0);
1103 up_read((&EXT4_I(inode)->i_data_sem));
1105 /* If it is only a block(s) look up */
1110 * Returns if the blocks have already allocated
1112 * Note that if blocks have been preallocated
1113 * ext4_ext_get_block() returns th create = 0
1114 * with buffer head unmapped.
1116 if (retval > 0 && buffer_mapped(bh))
1120 * When we call get_blocks without the create flag, the
1121 * BH_Unwritten flag could have gotten set if the blocks
1122 * requested were part of a uninitialized extent. We need to
1123 * clear this flag now that we are committed to convert all or
1124 * part of the uninitialized extent to be an initialized
1125 * extent. This is because we need to avoid the combination
1126 * of BH_Unwritten and BH_Mapped flags being simultaneously
1127 * set on the buffer_head.
1129 clear_buffer_unwritten(bh);
1132 * New blocks allocate and/or writing to uninitialized extent
1133 * will possibly result in updating i_data, so we take
1134 * the write lock of i_data_sem, and call get_blocks()
1135 * with create == 1 flag.
1137 down_write((&EXT4_I(inode)->i_data_sem));
1140 * if the caller is from delayed allocation writeout path
1141 * we have already reserved fs blocks for allocation
1142 * let the underlying get_block() function know to
1143 * avoid double accounting
1146 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1148 * We need to check for EXT4 here because migrate
1149 * could have changed the inode type in between
1151 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1152 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1153 bh, create, extend_disksize);
1155 retval = ext4_get_blocks_handle(handle, inode, block,
1156 max_blocks, bh, create, extend_disksize);
1158 if (retval > 0 && buffer_new(bh)) {
1160 * We allocated new blocks which will result in
1161 * i_data's format changing. Force the migrate
1162 * to fail by clearing migrate flags
1164 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1170 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1172 * Update reserved blocks/metadata blocks
1173 * after successful block allocation
1174 * which were deferred till now
1176 if ((retval > 0) && buffer_delay(bh))
1177 ext4_da_update_reserve_space(inode, retval);
1180 up_write((&EXT4_I(inode)->i_data_sem));
1184 /* Maximum number of blocks we map for direct IO at once. */
1185 #define DIO_MAX_BLOCKS 4096
1187 static int ext4_get_block(struct inode *inode, sector_t iblock,
1188 struct buffer_head *bh_result, int create)
1190 handle_t *handle = ext4_journal_current_handle();
1191 int ret = 0, started = 0;
1192 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1195 if (create && !handle) {
1196 /* Direct IO write... */
1197 if (max_blocks > DIO_MAX_BLOCKS)
1198 max_blocks = DIO_MAX_BLOCKS;
1199 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1200 handle = ext4_journal_start(inode, dio_credits);
1201 if (IS_ERR(handle)) {
1202 ret = PTR_ERR(handle);
1208 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1209 max_blocks, bh_result, create, 0, 0);
1211 bh_result->b_size = (ret << inode->i_blkbits);
1215 ext4_journal_stop(handle);
1221 * `handle' can be NULL if create is zero
1223 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1224 ext4_lblk_t block, int create, int *errp)
1226 struct buffer_head dummy;
1229 J_ASSERT(handle != NULL || create == 0);
1232 dummy.b_blocknr = -1000;
1233 buffer_trace_init(&dummy.b_history);
1234 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1235 &dummy, create, 1, 0);
1237 * ext4_get_blocks_handle() returns number of blocks
1238 * mapped. 0 in case of a HOLE.
1246 if (!err && buffer_mapped(&dummy)) {
1247 struct buffer_head *bh;
1248 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1253 if (buffer_new(&dummy)) {
1254 J_ASSERT(create != 0);
1255 J_ASSERT(handle != NULL);
1258 * Now that we do not always journal data, we should
1259 * keep in mind whether this should always journal the
1260 * new buffer as metadata. For now, regular file
1261 * writes use ext4_get_block instead, so it's not a
1265 BUFFER_TRACE(bh, "call get_create_access");
1266 fatal = ext4_journal_get_create_access(handle, bh);
1267 if (!fatal && !buffer_uptodate(bh)) {
1268 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1269 set_buffer_uptodate(bh);
1272 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1273 err = ext4_journal_dirty_metadata(handle, bh);
1277 BUFFER_TRACE(bh, "not a new buffer");
1290 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1291 ext4_lblk_t block, int create, int *err)
1293 struct buffer_head * bh;
1295 bh = ext4_getblk(handle, inode, block, create, err);
1298 if (buffer_uptodate(bh))
1300 ll_rw_block(READ_META, 1, &bh);
1302 if (buffer_uptodate(bh))
1309 static int walk_page_buffers( handle_t *handle,
1310 struct buffer_head *head,
1314 int (*fn)( handle_t *handle,
1315 struct buffer_head *bh))
1317 struct buffer_head *bh;
1318 unsigned block_start, block_end;
1319 unsigned blocksize = head->b_size;
1321 struct buffer_head *next;
1323 for ( bh = head, block_start = 0;
1324 ret == 0 && (bh != head || !block_start);
1325 block_start = block_end, bh = next)
1327 next = bh->b_this_page;
1328 block_end = block_start + blocksize;
1329 if (block_end <= from || block_start >= to) {
1330 if (partial && !buffer_uptodate(bh))
1334 err = (*fn)(handle, bh);
1342 * To preserve ordering, it is essential that the hole instantiation and
1343 * the data write be encapsulated in a single transaction. We cannot
1344 * close off a transaction and start a new one between the ext4_get_block()
1345 * and the commit_write(). So doing the jbd2_journal_start at the start of
1346 * prepare_write() is the right place.
1348 * Also, this function can nest inside ext4_writepage() ->
1349 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1350 * has generated enough buffer credits to do the whole page. So we won't
1351 * block on the journal in that case, which is good, because the caller may
1354 * By accident, ext4 can be reentered when a transaction is open via
1355 * quota file writes. If we were to commit the transaction while thus
1356 * reentered, there can be a deadlock - we would be holding a quota
1357 * lock, and the commit would never complete if another thread had a
1358 * transaction open and was blocking on the quota lock - a ranking
1361 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1362 * will _not_ run commit under these circumstances because handle->h_ref
1363 * is elevated. We'll still have enough credits for the tiny quotafile
1366 static int do_journal_get_write_access(handle_t *handle,
1367 struct buffer_head *bh)
1369 if (!buffer_mapped(bh) || buffer_freed(bh))
1371 return ext4_journal_get_write_access(handle, bh);
1374 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1375 loff_t pos, unsigned len, unsigned flags,
1376 struct page **pagep, void **fsdata)
1378 struct inode *inode = mapping->host;
1379 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1386 index = pos >> PAGE_CACHE_SHIFT;
1387 from = pos & (PAGE_CACHE_SIZE - 1);
1391 handle = ext4_journal_start(inode, needed_blocks);
1392 if (IS_ERR(handle)) {
1393 ret = PTR_ERR(handle);
1397 /* We cannot recurse into the filesystem as the transaction is already
1399 flags |= AOP_FLAG_NOFS;
1401 page = grab_cache_page_write_begin(mapping, index, flags);
1403 ext4_journal_stop(handle);
1409 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1412 if (!ret && ext4_should_journal_data(inode)) {
1413 ret = walk_page_buffers(handle, page_buffers(page),
1414 from, to, NULL, do_journal_get_write_access);
1419 ext4_journal_stop(handle);
1420 page_cache_release(page);
1423 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1429 /* For write_end() in data=journal mode */
1430 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1432 if (!buffer_mapped(bh) || buffer_freed(bh))
1434 set_buffer_uptodate(bh);
1435 return ext4_journal_dirty_metadata(handle, bh);
1439 * We need to pick up the new inode size which generic_commit_write gave us
1440 * `file' can be NULL - eg, when called from page_symlink().
1442 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1443 * buffers are managed internally.
1445 static int ext4_ordered_write_end(struct file *file,
1446 struct address_space *mapping,
1447 loff_t pos, unsigned len, unsigned copied,
1448 struct page *page, void *fsdata)
1450 handle_t *handle = ext4_journal_current_handle();
1451 struct inode *inode = mapping->host;
1454 ret = ext4_jbd2_file_inode(handle, inode);
1458 * generic_write_end() will run mark_inode_dirty() if i_size
1459 * changes. So let's piggyback the i_disksize mark_inode_dirty
1464 new_i_size = pos + copied;
1465 if (new_i_size > EXT4_I(inode)->i_disksize)
1466 EXT4_I(inode)->i_disksize = new_i_size;
1467 ret2 = generic_write_end(file, mapping, pos, len, copied,
1473 ret2 = ext4_journal_stop(handle);
1477 return ret ? ret : copied;
1480 static int ext4_writeback_write_end(struct file *file,
1481 struct address_space *mapping,
1482 loff_t pos, unsigned len, unsigned copied,
1483 struct page *page, void *fsdata)
1485 handle_t *handle = ext4_journal_current_handle();
1486 struct inode *inode = mapping->host;
1490 new_i_size = pos + copied;
1491 if (new_i_size > EXT4_I(inode)->i_disksize)
1492 EXT4_I(inode)->i_disksize = new_i_size;
1494 ret2 = generic_write_end(file, mapping, pos, len, copied,
1500 ret2 = ext4_journal_stop(handle);
1504 return ret ? ret : copied;
1507 static int ext4_journalled_write_end(struct file *file,
1508 struct address_space *mapping,
1509 loff_t pos, unsigned len, unsigned copied,
1510 struct page *page, void *fsdata)
1512 handle_t *handle = ext4_journal_current_handle();
1513 struct inode *inode = mapping->host;
1518 from = pos & (PAGE_CACHE_SIZE - 1);
1522 if (!PageUptodate(page))
1524 page_zero_new_buffers(page, from+copied, to);
1527 ret = walk_page_buffers(handle, page_buffers(page), from,
1528 to, &partial, write_end_fn);
1530 SetPageUptodate(page);
1531 if (pos+copied > inode->i_size)
1532 i_size_write(inode, pos+copied);
1533 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1534 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1535 EXT4_I(inode)->i_disksize = inode->i_size;
1536 ret2 = ext4_mark_inode_dirty(handle, inode);
1542 ret2 = ext4_journal_stop(handle);
1545 page_cache_release(page);
1547 return ret ? ret : copied;
1550 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1553 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1554 unsigned long md_needed, mdblocks, total = 0;
1557 * recalculate the amount of metadata blocks to reserve
1558 * in order to allocate nrblocks
1559 * worse case is one extent per block
1562 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1563 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1564 mdblocks = ext4_calc_metadata_amount(inode, total);
1565 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1567 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1568 total = md_needed + nrblocks;
1570 if (ext4_claim_free_blocks(sbi, total)) {
1571 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1572 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1578 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1579 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1581 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1582 return 0; /* success */
1585 static void ext4_da_release_space(struct inode *inode, int to_free)
1587 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1588 int total, mdb, mdb_free, release;
1591 return; /* Nothing to release, exit */
1593 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1595 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1597 * if there is no reserved blocks, but we try to free some
1598 * then the counter is messed up somewhere.
1599 * but since this function is called from invalidate
1600 * page, it's harmless to return without any action
1602 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1603 "blocks for inode %lu, but there is no reserved "
1604 "data blocks\n", to_free, inode->i_ino);
1605 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1609 /* recalculate the number of metablocks still need to be reserved */
1610 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1611 mdb = ext4_calc_metadata_amount(inode, total);
1613 /* figure out how many metablocks to release */
1614 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1615 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1617 release = to_free + mdb_free;
1619 /* update fs dirty blocks counter for truncate case */
1620 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1622 /* update per-inode reservations */
1623 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1624 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1626 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1627 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1628 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1631 static void ext4_da_page_release_reservation(struct page *page,
1632 unsigned long offset)
1635 struct buffer_head *head, *bh;
1636 unsigned int curr_off = 0;
1638 head = page_buffers(page);
1641 unsigned int next_off = curr_off + bh->b_size;
1643 if ((offset <= curr_off) && (buffer_delay(bh))) {
1645 clear_buffer_delay(bh);
1647 curr_off = next_off;
1648 } while ((bh = bh->b_this_page) != head);
1649 ext4_da_release_space(page->mapping->host, to_release);
1653 * Delayed allocation stuff
1656 struct mpage_da_data {
1657 struct inode *inode;
1658 struct buffer_head lbh; /* extent of blocks */
1659 unsigned long first_page, next_page; /* extent of pages */
1660 get_block_t *get_block;
1661 struct writeback_control *wbc;
1668 * mpage_da_submit_io - walks through extent of pages and try to write
1669 * them with writepage() call back
1671 * @mpd->inode: inode
1672 * @mpd->first_page: first page of the extent
1673 * @mpd->next_page: page after the last page of the extent
1674 * @mpd->get_block: the filesystem's block mapper function
1676 * By the time mpage_da_submit_io() is called we expect all blocks
1677 * to be allocated. this may be wrong if allocation failed.
1679 * As pages are already locked by write_cache_pages(), we can't use it
1681 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1684 struct pagevec pvec;
1685 unsigned long index, end;
1686 int ret = 0, err, nr_pages, i;
1687 struct inode *inode = mpd->inode;
1688 struct address_space *mapping = inode->i_mapping;
1690 BUG_ON(mpd->next_page <= mpd->first_page);
1692 * We need to start from the first_page to the next_page - 1
1693 * to make sure we also write the mapped dirty buffer_heads.
1694 * If we look at mpd->lbh.b_blocknr we would only be looking
1695 * at the currently mapped buffer_heads.
1697 index = mpd->first_page;
1698 end = mpd->next_page - 1;
1700 pagevec_init(&pvec, 0);
1701 while (index <= end) {
1703 * We can use PAGECACHE_TAG_DIRTY lookup here because
1704 * even though we have cleared the dirty flag on the page
1705 * We still keep the page in the radix tree with tag
1706 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1707 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1708 * which is called via the below writepage callback.
1710 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1711 PAGECACHE_TAG_DIRTY,
1713 (pgoff_t)PAGEVEC_SIZE-1) + 1);
1716 for (i = 0; i < nr_pages; i++) {
1717 struct page *page = pvec.pages[i];
1719 BUG_ON(!PageLocked(page));
1720 BUG_ON(PageWriteback(page));
1722 pages_skipped = mpd->wbc->pages_skipped;
1723 err = mapping->a_ops->writepage(page, mpd->wbc);
1725 mpd->pages_written++;
1727 * In error case, we have to continue because
1728 * remaining pages are still locked
1729 * XXX: unlock and re-dirty them?
1734 pagevec_release(&pvec);
1740 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1742 * @mpd->inode - inode to walk through
1743 * @exbh->b_blocknr - first block on a disk
1744 * @exbh->b_size - amount of space in bytes
1745 * @logical - first logical block to start assignment with
1747 * the function goes through all passed space and put actual disk
1748 * block numbers into buffer heads, dropping BH_Delay
1750 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1751 struct buffer_head *exbh)
1753 struct inode *inode = mpd->inode;
1754 struct address_space *mapping = inode->i_mapping;
1755 int blocks = exbh->b_size >> inode->i_blkbits;
1756 sector_t pblock = exbh->b_blocknr, cur_logical;
1757 struct buffer_head *head, *bh;
1759 struct pagevec pvec;
1762 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1763 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1764 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1766 pagevec_init(&pvec, 0);
1768 while (index <= end) {
1769 /* XXX: optimize tail */
1770 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1773 for (i = 0; i < nr_pages; i++) {
1774 struct page *page = pvec.pages[i];
1776 index = page->index;
1781 BUG_ON(!PageLocked(page));
1782 BUG_ON(PageWriteback(page));
1783 BUG_ON(!page_has_buffers(page));
1785 bh = page_buffers(page);
1788 /* skip blocks out of the range */
1790 if (cur_logical >= logical)
1793 } while ((bh = bh->b_this_page) != head);
1796 if (cur_logical >= logical + blocks)
1798 if (buffer_delay(bh)) {
1799 bh->b_blocknr = pblock;
1800 clear_buffer_delay(bh);
1801 bh->b_bdev = inode->i_sb->s_bdev;
1802 } else if (buffer_unwritten(bh)) {
1803 bh->b_blocknr = pblock;
1804 clear_buffer_unwritten(bh);
1805 set_buffer_mapped(bh);
1807 bh->b_bdev = inode->i_sb->s_bdev;
1808 } else if (buffer_mapped(bh))
1809 BUG_ON(bh->b_blocknr != pblock);
1813 } while ((bh = bh->b_this_page) != head);
1815 pagevec_release(&pvec);
1821 * __unmap_underlying_blocks - just a helper function to unmap
1822 * set of blocks described by @bh
1824 static inline void __unmap_underlying_blocks(struct inode *inode,
1825 struct buffer_head *bh)
1827 struct block_device *bdev = inode->i_sb->s_bdev;
1830 blocks = bh->b_size >> inode->i_blkbits;
1831 for (i = 0; i < blocks; i++)
1832 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1835 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1836 sector_t logical, long blk_cnt)
1840 struct pagevec pvec;
1841 struct inode *inode = mpd->inode;
1842 struct address_space *mapping = inode->i_mapping;
1844 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1845 end = (logical + blk_cnt - 1) >>
1846 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1847 while (index <= end) {
1848 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1851 for (i = 0; i < nr_pages; i++) {
1852 struct page *page = pvec.pages[i];
1853 index = page->index;
1858 BUG_ON(!PageLocked(page));
1859 BUG_ON(PageWriteback(page));
1860 block_invalidatepage(page, 0);
1861 ClearPageUptodate(page);
1868 static void ext4_print_free_blocks(struct inode *inode)
1870 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1871 printk(KERN_EMERG "Total free blocks count %lld\n",
1872 ext4_count_free_blocks(inode->i_sb));
1873 printk(KERN_EMERG "Free/Dirty block details\n");
1874 printk(KERN_EMERG "free_blocks=%lld\n",
1875 percpu_counter_sum(&sbi->s_freeblocks_counter));
1876 printk(KERN_EMERG "dirty_blocks=%lld\n",
1877 percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1878 printk(KERN_EMERG "Block reservation details\n");
1879 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1880 EXT4_I(inode)->i_reserved_data_blocks);
1881 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1882 EXT4_I(inode)->i_reserved_meta_blocks);
1887 * mpage_da_map_blocks - go through given space
1889 * @mpd->lbh - bh describing space
1890 * @mpd->get_block - the filesystem's block mapper function
1892 * The function skips space we know is already mapped to disk blocks.
1895 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1898 struct buffer_head new;
1899 struct buffer_head *lbh = &mpd->lbh;
1903 * We consider only non-mapped and non-allocated blocks
1905 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1907 new.b_state = lbh->b_state;
1909 new.b_size = lbh->b_size;
1910 next = lbh->b_blocknr;
1912 * If we didn't accumulate anything
1913 * to write simply return
1917 err = mpd->get_block(mpd->inode, next, &new, 1);
1920 /* If get block returns with error
1921 * we simply return. Later writepage
1922 * will redirty the page and writepages
1923 * will find the dirty page again
1928 if (err == -ENOSPC &&
1929 ext4_count_free_blocks(mpd->inode->i_sb)) {
1935 * get block failure will cause us
1936 * to loop in writepages. Because
1937 * a_ops->writepage won't be able to
1938 * make progress. The page will be redirtied
1939 * by writepage and writepages will again
1940 * try to write the same.
1942 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1943 "at logical offset %llu with max blocks "
1944 "%zd with error %d\n",
1945 __func__, mpd->inode->i_ino,
1946 (unsigned long long)next,
1947 lbh->b_size >> mpd->inode->i_blkbits, err);
1948 printk(KERN_EMERG "This should not happen.!! "
1949 "Data will be lost\n");
1950 if (err == -ENOSPC) {
1951 ext4_print_free_blocks(mpd->inode);
1953 /* invlaidate all the pages */
1954 ext4_da_block_invalidatepages(mpd, next,
1955 lbh->b_size >> mpd->inode->i_blkbits);
1958 BUG_ON(new.b_size == 0);
1960 if (buffer_new(&new))
1961 __unmap_underlying_blocks(mpd->inode, &new);
1964 * If blocks are delayed marked, we need to
1965 * put actual blocknr and drop delayed bit
1967 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1968 mpage_put_bnr_to_bhs(mpd, next, &new);
1973 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1974 (1 << BH_Delay) | (1 << BH_Unwritten))
1977 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1979 * @mpd->lbh - extent of blocks
1980 * @logical - logical number of the block in the file
1981 * @bh - bh of the block (used to access block's state)
1983 * the function is used to collect contig. blocks in same state
1985 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1986 sector_t logical, struct buffer_head *bh)
1989 size_t b_size = bh->b_size;
1990 struct buffer_head *lbh = &mpd->lbh;
1991 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1993 /* check if thereserved journal credits might overflow */
1994 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1995 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1997 * With non-extent format we are limited by the journal
1998 * credit available. Total credit needed to insert
1999 * nrblocks contiguous blocks is dependent on the
2000 * nrblocks. So limit nrblocks.
2003 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2004 EXT4_MAX_TRANS_DATA) {
2006 * Adding the new buffer_head would make it cross the
2007 * allowed limit for which we have journal credit
2008 * reserved. So limit the new bh->b_size
2010 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2011 mpd->inode->i_blkbits;
2012 /* we will do mpage_da_submit_io in the next loop */
2016 * First block in the extent
2018 if (lbh->b_size == 0) {
2019 lbh->b_blocknr = logical;
2020 lbh->b_size = b_size;
2021 lbh->b_state = bh->b_state & BH_FLAGS;
2025 next = lbh->b_blocknr + nrblocks;
2027 * Can we merge the block to our big extent?
2029 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
2030 lbh->b_size += b_size;
2036 * We couldn't merge the block to our extent, so we
2037 * need to flush current extent and start new one
2039 if (mpage_da_map_blocks(mpd) == 0)
2040 mpage_da_submit_io(mpd);
2046 * __mpage_da_writepage - finds extent of pages and blocks
2048 * @page: page to consider
2049 * @wbc: not used, we just follow rules
2052 * The function finds extents of pages and scan them for all blocks.
2054 static int __mpage_da_writepage(struct page *page,
2055 struct writeback_control *wbc, void *data)
2057 struct mpage_da_data *mpd = data;
2058 struct inode *inode = mpd->inode;
2059 struct buffer_head *bh, *head, fake;
2064 * Rest of the page in the page_vec
2065 * redirty then and skip then. We will
2066 * try to to write them again after
2067 * starting a new transaction
2069 redirty_page_for_writepage(wbc, page);
2071 return MPAGE_DA_EXTENT_TAIL;
2074 * Can we merge this page to current extent?
2076 if (mpd->next_page != page->index) {
2078 * Nope, we can't. So, we map non-allocated blocks
2079 * and start IO on them using writepage()
2081 if (mpd->next_page != mpd->first_page) {
2082 if (mpage_da_map_blocks(mpd) == 0)
2083 mpage_da_submit_io(mpd);
2085 * skip rest of the page in the page_vec
2088 redirty_page_for_writepage(wbc, page);
2090 return MPAGE_DA_EXTENT_TAIL;
2094 * Start next extent of pages ...
2096 mpd->first_page = page->index;
2101 mpd->lbh.b_size = 0;
2102 mpd->lbh.b_state = 0;
2103 mpd->lbh.b_blocknr = 0;
2106 mpd->next_page = page->index + 1;
2107 logical = (sector_t) page->index <<
2108 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2110 if (!page_has_buffers(page)) {
2112 * There is no attached buffer heads yet (mmap?)
2113 * we treat the page asfull of dirty blocks
2116 bh->b_size = PAGE_CACHE_SIZE;
2118 set_buffer_dirty(bh);
2119 set_buffer_uptodate(bh);
2120 mpage_add_bh_to_extent(mpd, logical, bh);
2122 return MPAGE_DA_EXTENT_TAIL;
2125 * Page with regular buffer heads, just add all dirty ones
2127 head = page_buffers(page);
2130 BUG_ON(buffer_locked(bh));
2132 * We need to try to allocate
2133 * unmapped blocks in the same page.
2134 * Otherwise we won't make progress
2135 * with the page in ext4_da_writepage
2137 if (buffer_dirty(bh) &&
2138 (!buffer_mapped(bh) || buffer_delay(bh))) {
2139 mpage_add_bh_to_extent(mpd, logical, bh);
2141 return MPAGE_DA_EXTENT_TAIL;
2142 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2144 * mapped dirty buffer. We need to update
2145 * the b_state because we look at
2146 * b_state in mpage_da_map_blocks. We don't
2147 * update b_size because if we find an
2148 * unmapped buffer_head later we need to
2149 * use the b_state flag of that buffer_head.
2151 if (mpd->lbh.b_size == 0)
2153 bh->b_state & BH_FLAGS;
2156 } while ((bh = bh->b_this_page) != head);
2163 * mpage_da_writepages - walk the list of dirty pages of the given
2164 * address space, allocates non-allocated blocks, maps newly-allocated
2165 * blocks to existing bhs and issue IO them
2167 * @mapping: address space structure to write
2168 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2169 * @get_block: the filesystem's block mapper function.
2171 * This is a library function, which implements the writepages()
2172 * address_space_operation.
2174 static int mpage_da_writepages(struct address_space *mapping,
2175 struct writeback_control *wbc,
2176 struct mpage_da_data *mpd)
2181 if (!mpd->get_block)
2182 return generic_writepages(mapping, wbc);
2184 mpd->lbh.b_size = 0;
2185 mpd->lbh.b_state = 0;
2186 mpd->lbh.b_blocknr = 0;
2187 mpd->first_page = 0;
2190 mpd->pages_written = 0;
2193 to_write = wbc->nr_to_write;
2195 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2198 * Handle last extent of pages
2200 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2201 if (mpage_da_map_blocks(mpd) == 0)
2202 mpage_da_submit_io(mpd);
2205 wbc->nr_to_write = to_write - mpd->pages_written;
2210 * this is a special callback for ->write_begin() only
2211 * it's intention is to return mapped block or reserve space
2213 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2214 struct buffer_head *bh_result, int create)
2217 sector_t invalid_block = ~((sector_t) 0xffff);
2219 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2222 BUG_ON(create == 0);
2223 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2226 * first, we need to know whether the block is allocated already
2227 * preallocated blocks are unmapped but should treated
2228 * the same as allocated blocks.
2230 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2231 if ((ret == 0) && !buffer_delay(bh_result)) {
2232 /* the block isn't (pre)allocated yet, let's reserve space */
2234 * XXX: __block_prepare_write() unmaps passed block,
2237 ret = ext4_da_reserve_space(inode, 1);
2239 /* not enough space to reserve */
2242 map_bh(bh_result, inode->i_sb, invalid_block);
2243 set_buffer_new(bh_result);
2244 set_buffer_delay(bh_result);
2245 } else if (ret > 0) {
2246 bh_result->b_size = (ret << inode->i_blkbits);
2248 * With sub-block writes into unwritten extents
2249 * we also need to mark the buffer as new so that
2250 * the unwritten parts of the buffer gets correctly zeroed.
2252 if (buffer_unwritten(bh_result))
2253 set_buffer_new(bh_result);
2259 #define EXT4_DELALLOC_RSVED 1
2260 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2261 struct buffer_head *bh_result, int create)
2264 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2265 loff_t disksize = EXT4_I(inode)->i_disksize;
2266 handle_t *handle = NULL;
2268 handle = ext4_journal_current_handle();
2270 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2271 bh_result, 0, 0, 0);
2274 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2275 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2279 bh_result->b_size = (ret << inode->i_blkbits);
2282 * Update on-disk size along with block allocation
2283 * we don't use 'extend_disksize' as size may change
2284 * within already allocated block -bzzz
2286 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2287 if (disksize > i_size_read(inode))
2288 disksize = i_size_read(inode);
2289 if (disksize > EXT4_I(inode)->i_disksize) {
2291 * XXX: replace with spinlock if seen contended -bzzz
2293 down_write(&EXT4_I(inode)->i_data_sem);
2294 if (disksize > EXT4_I(inode)->i_disksize)
2295 EXT4_I(inode)->i_disksize = disksize;
2296 up_write(&EXT4_I(inode)->i_data_sem);
2298 if (EXT4_I(inode)->i_disksize == disksize) {
2299 ret = ext4_mark_inode_dirty(handle, inode);
2308 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2311 * unmapped buffer is possible for holes.
2312 * delay buffer is possible with delayed allocation
2314 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2317 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2318 struct buffer_head *bh_result, int create)
2321 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2324 * we don't want to do block allocation in writepage
2325 * so call get_block_wrap with create = 0
2327 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2328 bh_result, 0, 0, 0);
2330 bh_result->b_size = (ret << inode->i_blkbits);
2337 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2338 * get called via journal_submit_inode_data_buffers (no journal handle)
2339 * get called via shrink_page_list via pdflush (no journal handle)
2340 * or grab_page_cache when doing write_begin (have journal handle)
2342 static int ext4_da_writepage(struct page *page,
2343 struct writeback_control *wbc)
2348 struct buffer_head *page_bufs;
2349 struct inode *inode = page->mapping->host;
2351 size = i_size_read(inode);
2352 if (page->index == size >> PAGE_CACHE_SHIFT)
2353 len = size & ~PAGE_CACHE_MASK;
2355 len = PAGE_CACHE_SIZE;
2357 if (page_has_buffers(page)) {
2358 page_bufs = page_buffers(page);
2359 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2360 ext4_bh_unmapped_or_delay)) {
2362 * We don't want to do block allocation
2363 * So redirty the page and return
2364 * We may reach here when we do a journal commit
2365 * via journal_submit_inode_data_buffers.
2366 * If we don't have mapping block we just ignore
2367 * them. We can also reach here via shrink_page_list
2369 redirty_page_for_writepage(wbc, page);
2375 * The test for page_has_buffers() is subtle:
2376 * We know the page is dirty but it lost buffers. That means
2377 * that at some moment in time after write_begin()/write_end()
2378 * has been called all buffers have been clean and thus they
2379 * must have been written at least once. So they are all
2380 * mapped and we can happily proceed with mapping them
2381 * and writing the page.
2383 * Try to initialize the buffer_heads and check whether
2384 * all are mapped and non delay. We don't want to
2385 * do block allocation here.
2387 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2388 ext4_normal_get_block_write);
2390 page_bufs = page_buffers(page);
2391 /* check whether all are mapped and non delay */
2392 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2393 ext4_bh_unmapped_or_delay)) {
2394 redirty_page_for_writepage(wbc, page);
2400 * We can't do block allocation here
2401 * so just redity the page and unlock
2404 redirty_page_for_writepage(wbc, page);
2408 /* now mark the buffer_heads as dirty and uptodate */
2409 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2412 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2413 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2415 ret = block_write_full_page(page,
2416 ext4_normal_get_block_write,
2423 * This is called via ext4_da_writepages() to
2424 * calulate the total number of credits to reserve to fit
2425 * a single extent allocation into a single transaction,
2426 * ext4_da_writpeages() will loop calling this before
2427 * the block allocation.
2430 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2432 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2435 * With non-extent format the journal credit needed to
2436 * insert nrblocks contiguous block is dependent on
2437 * number of contiguous block. So we will limit
2438 * number of contiguous block to a sane value
2440 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2441 (max_blocks > EXT4_MAX_TRANS_DATA))
2442 max_blocks = EXT4_MAX_TRANS_DATA;
2444 return ext4_chunk_trans_blocks(inode, max_blocks);
2447 static int ext4_da_writepages(struct address_space *mapping,
2448 struct writeback_control *wbc)
2450 handle_t *handle = NULL;
2451 struct mpage_da_data mpd;
2452 struct inode *inode = mapping->host;
2453 int needed_blocks, ret = 0, nr_to_writebump = 0;
2454 long to_write, pages_skipped = 0;
2455 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2458 * No pages to write? This is mainly a kludge to avoid starting
2459 * a transaction for special inodes like journal inode on last iput()
2460 * because that could violate lock ordering on umount
2462 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2466 * If the filesystem has aborted, it is read-only, so return
2467 * right away instead of dumping stack traces later on that
2468 * will obscure the real source of the problem. We test
2469 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2470 * the latter could be true if the filesystem is mounted
2471 * read-only, and in that case, ext4_da_writepages should
2472 * *never* be called, so if that ever happens, we would want
2475 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2479 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2480 * This make sure small files blocks are allocated in
2481 * single attempt. This ensure that small files
2482 * get less fragmented.
2484 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2485 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2486 wbc->nr_to_write = sbi->s_mb_stream_request;
2490 pages_skipped = wbc->pages_skipped;
2493 mpd.inode = mapping->host;
2496 to_write = wbc->nr_to_write;
2497 while (!ret && to_write > 0) {
2500 * we insert one extent at a time. So we need
2501 * credit needed for single extent allocation.
2502 * journalled mode is currently not supported
2505 BUG_ON(ext4_should_journal_data(inode));
2506 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2508 /* start a new transaction*/
2509 handle = ext4_journal_start(inode, needed_blocks);
2510 if (IS_ERR(handle)) {
2511 ret = PTR_ERR(handle);
2512 printk(KERN_CRIT "%s: jbd2_start: "
2513 "%ld pages, ino %lu; err %d\n", __func__,
2514 wbc->nr_to_write, inode->i_ino, ret);
2516 goto out_writepages;
2518 if (ext4_should_order_data(inode)) {
2520 * With ordered mode we need to add
2521 * the inode to the journal handl
2522 * when we do block allocation.
2524 ret = ext4_jbd2_file_inode(handle, inode);
2526 ext4_journal_stop(handle);
2527 goto out_writepages;
2530 to_write -= wbc->nr_to_write;
2532 mpd.get_block = ext4_da_get_block_write;
2533 ret = mpage_da_writepages(mapping, wbc, &mpd);
2535 ext4_journal_stop(handle);
2537 if (mpd.retval == -ENOSPC)
2538 jbd2_journal_force_commit_nested(sbi->s_journal);
2540 /* reset the retry count */
2541 if (ret == MPAGE_DA_EXTENT_TAIL) {
2543 * got one extent now try with
2546 to_write += wbc->nr_to_write;
2548 } else if (wbc->nr_to_write) {
2550 * There is no more writeout needed
2551 * or we requested for a noblocking writeout
2552 * and we found the device congested
2554 to_write += wbc->nr_to_write;
2557 wbc->nr_to_write = to_write;
2560 if (!wbc->range_cyclic && (pages_skipped != wbc->pages_skipped)) {
2561 /* We skipped pages in this loop */
2562 wbc->nr_to_write = to_write +
2563 wbc->pages_skipped - pages_skipped;
2564 wbc->pages_skipped = pages_skipped;
2569 wbc->nr_to_write = to_write - nr_to_writebump;
2573 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2574 loff_t pos, unsigned len, unsigned flags,
2575 struct page **pagep, void **fsdata)
2577 int ret, retries = 0;
2581 struct inode *inode = mapping->host;
2584 index = pos >> PAGE_CACHE_SHIFT;
2585 from = pos & (PAGE_CACHE_SIZE - 1);
2589 * With delayed allocation, we don't log the i_disksize update
2590 * if there is delayed block allocation. But we still need
2591 * to journalling the i_disksize update if writes to the end
2592 * of file which has an already mapped buffer.
2594 handle = ext4_journal_start(inode, 1);
2595 if (IS_ERR(handle)) {
2596 ret = PTR_ERR(handle);
2599 /* We cannot recurse into the filesystem as the transaction is already
2601 flags |= AOP_FLAG_NOFS;
2603 page = grab_cache_page_write_begin(mapping, index, flags);
2605 ext4_journal_stop(handle);
2611 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2612 ext4_da_get_block_prep);
2615 ext4_journal_stop(handle);
2616 page_cache_release(page);
2619 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2626 * Check if we should update i_disksize
2627 * when write to the end of file but not require block allocation
2629 static int ext4_da_should_update_i_disksize(struct page *page,
2630 unsigned long offset)
2632 struct buffer_head *bh;
2633 struct inode *inode = page->mapping->host;
2637 bh = page_buffers(page);
2638 idx = offset >> inode->i_blkbits;
2640 for (i=0; i < idx; i++)
2641 bh = bh->b_this_page;
2643 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2648 static int ext4_da_write_end(struct file *file,
2649 struct address_space *mapping,
2650 loff_t pos, unsigned len, unsigned copied,
2651 struct page *page, void *fsdata)
2653 struct inode *inode = mapping->host;
2655 handle_t *handle = ext4_journal_current_handle();
2657 unsigned long start, end;
2659 start = pos & (PAGE_CACHE_SIZE - 1);
2660 end = start + copied -1;
2663 * generic_write_end() will run mark_inode_dirty() if i_size
2664 * changes. So let's piggyback the i_disksize mark_inode_dirty
2668 new_i_size = pos + copied;
2669 if (new_i_size > EXT4_I(inode)->i_disksize) {
2670 if (ext4_da_should_update_i_disksize(page, end)) {
2671 down_write(&EXT4_I(inode)->i_data_sem);
2672 if (new_i_size > EXT4_I(inode)->i_disksize) {
2674 * Updating i_disksize when extending file
2675 * without needing block allocation
2677 if (ext4_should_order_data(inode))
2678 ret = ext4_jbd2_file_inode(handle,
2681 EXT4_I(inode)->i_disksize = new_i_size;
2683 up_write(&EXT4_I(inode)->i_data_sem);
2686 ret2 = generic_write_end(file, mapping, pos, len, copied,
2691 ret2 = ext4_journal_stop(handle);
2695 return ret ? ret : copied;
2698 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2701 * Drop reserved blocks
2703 BUG_ON(!PageLocked(page));
2704 if (!page_has_buffers(page))
2707 ext4_da_page_release_reservation(page, offset);
2710 ext4_invalidatepage(page, offset);
2716 * Force all delayed allocation blocks to be allocated for a given inode.
2718 int ext4_alloc_da_blocks(struct inode *inode)
2720 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2721 !EXT4_I(inode)->i_reserved_meta_blocks)
2725 * We do something simple for now. The filemap_flush() will
2726 * also start triggering a write of the data blocks, which is
2727 * not strictly speaking necessary (and for users of
2728 * laptop_mode, not even desirable). However, to do otherwise
2729 * would require replicating code paths in:
2731 * ext4_da_writepages() ->
2732 * write_cache_pages() ---> (via passed in callback function)
2733 * __mpage_da_writepage() -->
2734 * mpage_add_bh_to_extent()
2735 * mpage_da_map_blocks()
2737 * The problem is that write_cache_pages(), located in
2738 * mm/page-writeback.c, marks pages clean in preparation for
2739 * doing I/O, which is not desirable if we're not planning on
2742 * We could call write_cache_pages(), and then redirty all of
2743 * the pages by calling redirty_page_for_writeback() but that
2744 * would be ugly in the extreme. So instead we would need to
2745 * replicate parts of the code in the above functions,
2746 * simplifying them becuase we wouldn't actually intend to
2747 * write out the pages, but rather only collect contiguous
2748 * logical block extents, call the multi-block allocator, and
2749 * then update the buffer heads with the block allocations.
2751 * For now, though, we'll cheat by calling filemap_flush(),
2752 * which will map the blocks, and start the I/O, but not
2753 * actually wait for the I/O to complete.
2755 return filemap_flush(inode->i_mapping);
2759 * bmap() is special. It gets used by applications such as lilo and by
2760 * the swapper to find the on-disk block of a specific piece of data.
2762 * Naturally, this is dangerous if the block concerned is still in the
2763 * journal. If somebody makes a swapfile on an ext4 data-journaling
2764 * filesystem and enables swap, then they may get a nasty shock when the
2765 * data getting swapped to that swapfile suddenly gets overwritten by
2766 * the original zero's written out previously to the journal and
2767 * awaiting writeback in the kernel's buffer cache.
2769 * So, if we see any bmap calls here on a modified, data-journaled file,
2770 * take extra steps to flush any blocks which might be in the cache.
2772 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2774 struct inode *inode = mapping->host;
2778 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2779 test_opt(inode->i_sb, DELALLOC)) {
2781 * With delalloc we want to sync the file
2782 * so that we can make sure we allocate
2785 filemap_write_and_wait(mapping);
2788 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2790 * This is a REALLY heavyweight approach, but the use of
2791 * bmap on dirty files is expected to be extremely rare:
2792 * only if we run lilo or swapon on a freshly made file
2793 * do we expect this to happen.
2795 * (bmap requires CAP_SYS_RAWIO so this does not
2796 * represent an unprivileged user DOS attack --- we'd be
2797 * in trouble if mortal users could trigger this path at
2800 * NB. EXT4_STATE_JDATA is not set on files other than
2801 * regular files. If somebody wants to bmap a directory
2802 * or symlink and gets confused because the buffer
2803 * hasn't yet been flushed to disk, they deserve
2804 * everything they get.
2807 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2808 journal = EXT4_JOURNAL(inode);
2809 jbd2_journal_lock_updates(journal);
2810 err = jbd2_journal_flush(journal);
2811 jbd2_journal_unlock_updates(journal);
2817 return generic_block_bmap(mapping,block,ext4_get_block);
2820 static int bget_one(handle_t *handle, struct buffer_head *bh)
2826 static int bput_one(handle_t *handle, struct buffer_head *bh)
2833 * Note that we don't need to start a transaction unless we're journaling data
2834 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2835 * need to file the inode to the transaction's list in ordered mode because if
2836 * we are writing back data added by write(), the inode is already there and if
2837 * we are writing back data modified via mmap(), noone guarantees in which
2838 * transaction the data will hit the disk. In case we are journaling data, we
2839 * cannot start transaction directly because transaction start ranks above page
2840 * lock so we have to do some magic.
2842 * In all journaling modes block_write_full_page() will start the I/O.
2846 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2851 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2853 * Same applies to ext4_get_block(). We will deadlock on various things like
2854 * lock_journal and i_data_sem
2856 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2859 * 16May01: If we're reentered then journal_current_handle() will be
2860 * non-zero. We simply *return*.
2862 * 1 July 2001: @@@ FIXME:
2863 * In journalled data mode, a data buffer may be metadata against the
2864 * current transaction. But the same file is part of a shared mapping
2865 * and someone does a writepage() on it.
2867 * We will move the buffer onto the async_data list, but *after* it has
2868 * been dirtied. So there's a small window where we have dirty data on
2871 * Note that this only applies to the last partial page in the file. The
2872 * bit which block_write_full_page() uses prepare/commit for. (That's
2873 * broken code anyway: it's wrong for msync()).
2875 * It's a rare case: affects the final partial page, for journalled data
2876 * where the file is subject to bith write() and writepage() in the same
2877 * transction. To fix it we'll need a custom block_write_full_page().
2878 * We'll probably need that anyway for journalling writepage() output.
2880 * We don't honour synchronous mounts for writepage(). That would be
2881 * disastrous. Any write() or metadata operation will sync the fs for
2885 static int __ext4_normal_writepage(struct page *page,
2886 struct writeback_control *wbc)
2888 struct inode *inode = page->mapping->host;
2890 if (test_opt(inode->i_sb, NOBH))
2891 return nobh_writepage(page,
2892 ext4_normal_get_block_write, wbc);
2894 return block_write_full_page(page,
2895 ext4_normal_get_block_write,
2899 static int ext4_normal_writepage(struct page *page,
2900 struct writeback_control *wbc)
2902 struct inode *inode = page->mapping->host;
2903 loff_t size = i_size_read(inode);
2906 J_ASSERT(PageLocked(page));
2907 if (page->index == size >> PAGE_CACHE_SHIFT)
2908 len = size & ~PAGE_CACHE_MASK;
2910 len = PAGE_CACHE_SIZE;
2912 if (page_has_buffers(page)) {
2913 /* if page has buffers it should all be mapped
2914 * and allocated. If there are not buffers attached
2915 * to the page we know the page is dirty but it lost
2916 * buffers. That means that at some moment in time
2917 * after write_begin() / write_end() has been called
2918 * all buffers have been clean and thus they must have been
2919 * written at least once. So they are all mapped and we can
2920 * happily proceed with mapping them and writing the page.
2922 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2923 ext4_bh_unmapped_or_delay));
2926 if (!ext4_journal_current_handle())
2927 return __ext4_normal_writepage(page, wbc);
2929 redirty_page_for_writepage(wbc, page);
2934 static int __ext4_journalled_writepage(struct page *page,
2935 struct writeback_control *wbc)
2937 struct address_space *mapping = page->mapping;
2938 struct inode *inode = mapping->host;
2939 struct buffer_head *page_bufs;
2940 handle_t *handle = NULL;
2944 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2945 ext4_normal_get_block_write);
2949 page_bufs = page_buffers(page);
2950 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2952 /* As soon as we unlock the page, it can go away, but we have
2953 * references to buffers so we are safe */
2956 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2957 if (IS_ERR(handle)) {
2958 ret = PTR_ERR(handle);
2962 ret = walk_page_buffers(handle, page_bufs, 0,
2963 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2965 err = walk_page_buffers(handle, page_bufs, 0,
2966 PAGE_CACHE_SIZE, NULL, write_end_fn);
2969 err = ext4_journal_stop(handle);
2973 walk_page_buffers(handle, page_bufs, 0,
2974 PAGE_CACHE_SIZE, NULL, bput_one);
2975 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2984 static int ext4_journalled_writepage(struct page *page,
2985 struct writeback_control *wbc)
2987 struct inode *inode = page->mapping->host;
2988 loff_t size = i_size_read(inode);
2991 J_ASSERT(PageLocked(page));
2992 if (page->index == size >> PAGE_CACHE_SHIFT)
2993 len = size & ~PAGE_CACHE_MASK;
2995 len = PAGE_CACHE_SIZE;
2997 if (page_has_buffers(page)) {
2998 /* if page has buffers it should all be mapped
2999 * and allocated. If there are not buffers attached
3000 * to the page we know the page is dirty but it lost
3001 * buffers. That means that at some moment in time
3002 * after write_begin() / write_end() has been called
3003 * all buffers have been clean and thus they must have been
3004 * written at least once. So they are all mapped and we can
3005 * happily proceed with mapping them and writing the page.
3007 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3008 ext4_bh_unmapped_or_delay));
3011 if (ext4_journal_current_handle())
3014 if (PageChecked(page)) {
3016 * It's mmapped pagecache. Add buffers and journal it. There
3017 * doesn't seem much point in redirtying the page here.
3019 ClearPageChecked(page);
3020 return __ext4_journalled_writepage(page, wbc);
3023 * It may be a page full of checkpoint-mode buffers. We don't
3024 * really know unless we go poke around in the buffer_heads.
3025 * But block_write_full_page will do the right thing.
3027 return block_write_full_page(page,
3028 ext4_normal_get_block_write,
3032 redirty_page_for_writepage(wbc, page);
3037 static int ext4_readpage(struct file *file, struct page *page)
3039 return mpage_readpage(page, ext4_get_block);
3043 ext4_readpages(struct file *file, struct address_space *mapping,
3044 struct list_head *pages, unsigned nr_pages)
3046 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3049 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3051 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3054 * If it's a full truncate we just forget about the pending dirtying
3057 ClearPageChecked(page);
3059 jbd2_journal_invalidatepage(journal, page, offset);
3062 static int ext4_releasepage(struct page *page, gfp_t wait)
3064 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3066 WARN_ON(PageChecked(page));
3067 if (!page_has_buffers(page))
3069 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3073 * If the O_DIRECT write will extend the file then add this inode to the
3074 * orphan list. So recovery will truncate it back to the original size
3075 * if the machine crashes during the write.
3077 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3078 * crashes then stale disk data _may_ be exposed inside the file. But current
3079 * VFS code falls back into buffered path in that case so we are safe.
3081 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3082 const struct iovec *iov, loff_t offset,
3083 unsigned long nr_segs)
3085 struct file *file = iocb->ki_filp;
3086 struct inode *inode = file->f_mapping->host;
3087 struct ext4_inode_info *ei = EXT4_I(inode);
3091 size_t count = iov_length(iov, nr_segs);
3094 loff_t final_size = offset + count;
3096 if (final_size > inode->i_size) {
3097 /* Credits for sb + inode write */
3098 handle = ext4_journal_start(inode, 2);
3099 if (IS_ERR(handle)) {
3100 ret = PTR_ERR(handle);
3103 ret = ext4_orphan_add(handle, inode);
3105 ext4_journal_stop(handle);
3109 ei->i_disksize = inode->i_size;
3110 ext4_journal_stop(handle);
3114 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3116 ext4_get_block, NULL);
3121 /* Credits for sb + inode write */
3122 handle = ext4_journal_start(inode, 2);
3123 if (IS_ERR(handle)) {
3124 /* This is really bad luck. We've written the data
3125 * but cannot extend i_size. Bail out and pretend
3126 * the write failed... */
3127 ret = PTR_ERR(handle);
3131 ext4_orphan_del(handle, inode);
3133 loff_t end = offset + ret;
3134 if (end > inode->i_size) {
3135 ei->i_disksize = end;
3136 i_size_write(inode, end);
3138 * We're going to return a positive `ret'
3139 * here due to non-zero-length I/O, so there's
3140 * no way of reporting error returns from
3141 * ext4_mark_inode_dirty() to userspace. So
3144 ext4_mark_inode_dirty(handle, inode);
3147 err = ext4_journal_stop(handle);
3156 * Pages can be marked dirty completely asynchronously from ext4's journalling
3157 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3158 * much here because ->set_page_dirty is called under VFS locks. The page is
3159 * not necessarily locked.
3161 * We cannot just dirty the page and leave attached buffers clean, because the
3162 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3163 * or jbddirty because all the journalling code will explode.
3165 * So what we do is to mark the page "pending dirty" and next time writepage
3166 * is called, propagate that into the buffers appropriately.
3168 static int ext4_journalled_set_page_dirty(struct page *page)
3170 SetPageChecked(page);
3171 return __set_page_dirty_nobuffers(page);
3174 static const struct address_space_operations ext4_ordered_aops = {
3175 .readpage = ext4_readpage,
3176 .readpages = ext4_readpages,
3177 .writepage = ext4_normal_writepage,
3178 .sync_page = block_sync_page,
3179 .write_begin = ext4_write_begin,
3180 .write_end = ext4_ordered_write_end,
3182 .invalidatepage = ext4_invalidatepage,
3183 .releasepage = ext4_releasepage,
3184 .direct_IO = ext4_direct_IO,
3185 .migratepage = buffer_migrate_page,
3186 .is_partially_uptodate = block_is_partially_uptodate,
3189 static const struct address_space_operations ext4_writeback_aops = {
3190 .readpage = ext4_readpage,
3191 .readpages = ext4_readpages,
3192 .writepage = ext4_normal_writepage,
3193 .sync_page = block_sync_page,
3194 .write_begin = ext4_write_begin,
3195 .write_end = ext4_writeback_write_end,
3197 .invalidatepage = ext4_invalidatepage,
3198 .releasepage = ext4_releasepage,
3199 .direct_IO = ext4_direct_IO,
3200 .migratepage = buffer_migrate_page,
3201 .is_partially_uptodate = block_is_partially_uptodate,
3204 static const struct address_space_operations ext4_journalled_aops = {
3205 .readpage = ext4_readpage,
3206 .readpages = ext4_readpages,
3207 .writepage = ext4_journalled_writepage,
3208 .sync_page = block_sync_page,
3209 .write_begin = ext4_write_begin,
3210 .write_end = ext4_journalled_write_end,
3211 .set_page_dirty = ext4_journalled_set_page_dirty,
3213 .invalidatepage = ext4_invalidatepage,
3214 .releasepage = ext4_releasepage,
3215 .is_partially_uptodate = block_is_partially_uptodate,
3218 static const struct address_space_operations ext4_da_aops = {
3219 .readpage = ext4_readpage,
3220 .readpages = ext4_readpages,
3221 .writepage = ext4_da_writepage,
3222 .writepages = ext4_da_writepages,
3223 .sync_page = block_sync_page,
3224 .write_begin = ext4_da_write_begin,
3225 .write_end = ext4_da_write_end,
3227 .invalidatepage = ext4_da_invalidatepage,
3228 .releasepage = ext4_releasepage,
3229 .direct_IO = ext4_direct_IO,
3230 .migratepage = buffer_migrate_page,
3231 .is_partially_uptodate = block_is_partially_uptodate,
3234 void ext4_set_aops(struct inode *inode)
3236 if (ext4_should_order_data(inode) &&
3237 test_opt(inode->i_sb, DELALLOC))
3238 inode->i_mapping->a_ops = &ext4_da_aops;
3239 else if (ext4_should_order_data(inode))
3240 inode->i_mapping->a_ops = &ext4_ordered_aops;
3241 else if (ext4_should_writeback_data(inode) &&
3242 test_opt(inode->i_sb, DELALLOC))
3243 inode->i_mapping->a_ops = &ext4_da_aops;
3244 else if (ext4_should_writeback_data(inode))
3245 inode->i_mapping->a_ops = &ext4_writeback_aops;
3247 inode->i_mapping->a_ops = &ext4_journalled_aops;
3251 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3252 * up to the end of the block which corresponds to `from'.
3253 * This required during truncate. We need to physically zero the tail end
3254 * of that block so it doesn't yield old data if the file is later grown.
3256 int ext4_block_truncate_page(handle_t *handle,
3257 struct address_space *mapping, loff_t from)
3259 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3260 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3261 unsigned blocksize, length, pos;
3263 struct inode *inode = mapping->host;
3264 struct buffer_head *bh;
3268 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3272 blocksize = inode->i_sb->s_blocksize;
3273 length = blocksize - (offset & (blocksize - 1));
3274 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3277 * For "nobh" option, we can only work if we don't need to
3278 * read-in the page - otherwise we create buffers to do the IO.
3280 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3281 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3282 zero_user(page, offset, length);
3283 set_page_dirty(page);
3287 if (!page_has_buffers(page))
3288 create_empty_buffers(page, blocksize, 0);
3290 /* Find the buffer that contains "offset" */
3291 bh = page_buffers(page);
3293 while (offset >= pos) {
3294 bh = bh->b_this_page;
3300 if (buffer_freed(bh)) {
3301 BUFFER_TRACE(bh, "freed: skip");
3305 if (!buffer_mapped(bh)) {
3306 BUFFER_TRACE(bh, "unmapped");
3307 ext4_get_block(inode, iblock, bh, 0);
3308 /* unmapped? It's a hole - nothing to do */
3309 if (!buffer_mapped(bh)) {
3310 BUFFER_TRACE(bh, "still unmapped");
3315 /* Ok, it's mapped. Make sure it's up-to-date */
3316 if (PageUptodate(page))
3317 set_buffer_uptodate(bh);
3319 if (!buffer_uptodate(bh)) {
3321 ll_rw_block(READ, 1, &bh);
3323 /* Uhhuh. Read error. Complain and punt. */
3324 if (!buffer_uptodate(bh))
3328 if (ext4_should_journal_data(inode)) {
3329 BUFFER_TRACE(bh, "get write access");
3330 err = ext4_journal_get_write_access(handle, bh);
3335 zero_user(page, offset, length);
3337 BUFFER_TRACE(bh, "zeroed end of block");
3340 if (ext4_should_journal_data(inode)) {
3341 err = ext4_journal_dirty_metadata(handle, bh);
3343 if (ext4_should_order_data(inode))
3344 err = ext4_jbd2_file_inode(handle, inode);
3345 mark_buffer_dirty(bh);
3350 page_cache_release(page);
3355 * Probably it should be a library function... search for first non-zero word
3356 * or memcmp with zero_page, whatever is better for particular architecture.
3359 static inline int all_zeroes(__le32 *p, __le32 *q)
3368 * ext4_find_shared - find the indirect blocks for partial truncation.
3369 * @inode: inode in question
3370 * @depth: depth of the affected branch
3371 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3372 * @chain: place to store the pointers to partial indirect blocks
3373 * @top: place to the (detached) top of branch
3375 * This is a helper function used by ext4_truncate().
3377 * When we do truncate() we may have to clean the ends of several
3378 * indirect blocks but leave the blocks themselves alive. Block is
3379 * partially truncated if some data below the new i_size is refered
3380 * from it (and it is on the path to the first completely truncated
3381 * data block, indeed). We have to free the top of that path along
3382 * with everything to the right of the path. Since no allocation
3383 * past the truncation point is possible until ext4_truncate()
3384 * finishes, we may safely do the latter, but top of branch may
3385 * require special attention - pageout below the truncation point
3386 * might try to populate it.
3388 * We atomically detach the top of branch from the tree, store the
3389 * block number of its root in *@top, pointers to buffer_heads of
3390 * partially truncated blocks - in @chain[].bh and pointers to
3391 * their last elements that should not be removed - in
3392 * @chain[].p. Return value is the pointer to last filled element
3395 * The work left to caller to do the actual freeing of subtrees:
3396 * a) free the subtree starting from *@top
3397 * b) free the subtrees whose roots are stored in
3398 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3399 * c) free the subtrees growing from the inode past the @chain[0].
3400 * (no partially truncated stuff there). */
3402 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3403 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3405 Indirect *partial, *p;
3409 /* Make k index the deepest non-null offest + 1 */
3410 for (k = depth; k > 1 && !offsets[k-1]; k--)
3412 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3413 /* Writer: pointers */
3415 partial = chain + k-1;
3417 * If the branch acquired continuation since we've looked at it -
3418 * fine, it should all survive and (new) top doesn't belong to us.
3420 if (!partial->key && *partial->p)
3423 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
3426 * OK, we've found the last block that must survive. The rest of our
3427 * branch should be detached before unlocking. However, if that rest
3428 * of branch is all ours and does not grow immediately from the inode
3429 * it's easier to cheat and just decrement partial->p.
3431 if (p == chain + k - 1 && p > chain) {
3435 /* Nope, don't do this in ext4. Must leave the tree intact */
3442 while(partial > p) {
3443 brelse(partial->bh);
3451 * Zero a number of block pointers in either an inode or an indirect block.
3452 * If we restart the transaction we must again get write access to the
3453 * indirect block for further modification.
3455 * We release `count' blocks on disk, but (last - first) may be greater
3456 * than `count' because there can be holes in there.
3458 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3459 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3460 unsigned long count, __le32 *first, __le32 *last)
3463 if (try_to_extend_transaction(handle, inode)) {
3465 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3466 ext4_journal_dirty_metadata(handle, bh);
3468 ext4_mark_inode_dirty(handle, inode);
3469 ext4_journal_test_restart(handle, inode);
3471 BUFFER_TRACE(bh, "retaking write access");
3472 ext4_journal_get_write_access(handle, bh);
3477 * Any buffers which are on the journal will be in memory. We find
3478 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3479 * on them. We've already detached each block from the file, so
3480 * bforget() in jbd2_journal_forget() should be safe.
3482 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3484 for (p = first; p < last; p++) {
3485 u32 nr = le32_to_cpu(*p);
3487 struct buffer_head *tbh;
3490 tbh = sb_find_get_block(inode->i_sb, nr);
3491 ext4_forget(handle, 0, inode, tbh, nr);
3495 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3499 * ext4_free_data - free a list of data blocks
3500 * @handle: handle for this transaction
3501 * @inode: inode we are dealing with
3502 * @this_bh: indirect buffer_head which contains *@first and *@last
3503 * @first: array of block numbers
3504 * @last: points immediately past the end of array
3506 * We are freeing all blocks refered from that array (numbers are stored as
3507 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3509 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3510 * blocks are contiguous then releasing them at one time will only affect one
3511 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3512 * actually use a lot of journal space.
3514 * @this_bh will be %NULL if @first and @last point into the inode's direct
3517 static void ext4_free_data(handle_t *handle, struct inode *inode,
3518 struct buffer_head *this_bh,
3519 __le32 *first, __le32 *last)
3521 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3522 unsigned long count = 0; /* Number of blocks in the run */
3523 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3526 ext4_fsblk_t nr; /* Current block # */
3527 __le32 *p; /* Pointer into inode/ind
3528 for current block */
3531 if (this_bh) { /* For indirect block */
3532 BUFFER_TRACE(this_bh, "get_write_access");
3533 err = ext4_journal_get_write_access(handle, this_bh);
3534 /* Important: if we can't update the indirect pointers
3535 * to the blocks, we can't free them. */
3540 for (p = first; p < last; p++) {
3541 nr = le32_to_cpu(*p);
3543 /* accumulate blocks to free if they're contiguous */
3546 block_to_free_p = p;
3548 } else if (nr == block_to_free + count) {
3551 ext4_clear_blocks(handle, inode, this_bh,
3553 count, block_to_free_p, p);
3555 block_to_free_p = p;
3562 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3563 count, block_to_free_p, p);
3566 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3569 * The buffer head should have an attached journal head at this
3570 * point. However, if the data is corrupted and an indirect
3571 * block pointed to itself, it would have been detached when
3572 * the block was cleared. Check for this instead of OOPSing.
3575 ext4_journal_dirty_metadata(handle, this_bh);
3577 ext4_error(inode->i_sb, __func__,
3578 "circular indirect block detected, "
3579 "inode=%lu, block=%llu",
3581 (unsigned long long) this_bh->b_blocknr);
3586 * ext4_free_branches - free an array of branches
3587 * @handle: JBD handle for this transaction
3588 * @inode: inode we are dealing with
3589 * @parent_bh: the buffer_head which contains *@first and *@last
3590 * @first: array of block numbers
3591 * @last: pointer immediately past the end of array
3592 * @depth: depth of the branches to free
3594 * We are freeing all blocks refered from these branches (numbers are
3595 * stored as little-endian 32-bit) and updating @inode->i_blocks
3598 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3599 struct buffer_head *parent_bh,
3600 __le32 *first, __le32 *last, int depth)
3605 if (is_handle_aborted(handle))
3609 struct buffer_head *bh;
3610 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3612 while (--p >= first) {
3613 nr = le32_to_cpu(*p);
3615 continue; /* A hole */
3617 /* Go read the buffer for the next level down */
3618 bh = sb_bread(inode->i_sb, nr);
3621 * A read failure? Report error and clear slot
3625 ext4_error(inode->i_sb, "ext4_free_branches",
3626 "Read failure, inode=%lu, block=%llu",
3631 /* This zaps the entire block. Bottom up. */
3632 BUFFER_TRACE(bh, "free child branches");
3633 ext4_free_branches(handle, inode, bh,
3634 (__le32*)bh->b_data,
3635 (__le32*)bh->b_data + addr_per_block,
3639 * We've probably journalled the indirect block several
3640 * times during the truncate. But it's no longer
3641 * needed and we now drop it from the transaction via
3642 * jbd2_journal_revoke().
3644 * That's easy if it's exclusively part of this
3645 * transaction. But if it's part of the committing
3646 * transaction then jbd2_journal_forget() will simply
3647 * brelse() it. That means that if the underlying
3648 * block is reallocated in ext4_get_block(),
3649 * unmap_underlying_metadata() will find this block
3650 * and will try to get rid of it. damn, damn.
3652 * If this block has already been committed to the
3653 * journal, a revoke record will be written. And
3654 * revoke records must be emitted *before* clearing
3655 * this block's bit in the bitmaps.
3657 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3660 * Everything below this this pointer has been
3661 * released. Now let this top-of-subtree go.
3663 * We want the freeing of this indirect block to be
3664 * atomic in the journal with the updating of the
3665 * bitmap block which owns it. So make some room in
3668 * We zero the parent pointer *after* freeing its
3669 * pointee in the bitmaps, so if extend_transaction()
3670 * for some reason fails to put the bitmap changes and
3671 * the release into the same transaction, recovery
3672 * will merely complain about releasing a free block,
3673 * rather than leaking blocks.
3675 if (is_handle_aborted(handle))
3677 if (try_to_extend_transaction(handle, inode)) {
3678 ext4_mark_inode_dirty(handle, inode);
3679 ext4_journal_test_restart(handle, inode);
3682 ext4_free_blocks(handle, inode, nr, 1, 1);
3686 * The block which we have just freed is
3687 * pointed to by an indirect block: journal it
3689 BUFFER_TRACE(parent_bh, "get_write_access");
3690 if (!ext4_journal_get_write_access(handle,
3693 BUFFER_TRACE(parent_bh,
3694 "call ext4_journal_dirty_metadata");
3695 ext4_journal_dirty_metadata(handle,
3701 /* We have reached the bottom of the tree. */
3702 BUFFER_TRACE(parent_bh, "free data blocks");
3703 ext4_free_data(handle, inode, parent_bh, first, last);
3707 int ext4_can_truncate(struct inode *inode)
3709 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3711 if (S_ISREG(inode->i_mode))
3713 if (S_ISDIR(inode->i_mode))
3715 if (S_ISLNK(inode->i_mode))
3716 return !ext4_inode_is_fast_symlink(inode);
3723 * We block out ext4_get_block() block instantiations across the entire
3724 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3725 * simultaneously on behalf of the same inode.
3727 * As we work through the truncate and commmit bits of it to the journal there
3728 * is one core, guiding principle: the file's tree must always be consistent on
3729 * disk. We must be able to restart the truncate after a crash.
3731 * The file's tree may be transiently inconsistent in memory (although it
3732 * probably isn't), but whenever we close off and commit a journal transaction,
3733 * the contents of (the filesystem + the journal) must be consistent and
3734 * restartable. It's pretty simple, really: bottom up, right to left (although
3735 * left-to-right works OK too).
3737 * Note that at recovery time, journal replay occurs *before* the restart of
3738 * truncate against the orphan inode list.
3740 * The committed inode has the new, desired i_size (which is the same as
3741 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3742 * that this inode's truncate did not complete and it will again call
3743 * ext4_truncate() to have another go. So there will be instantiated blocks
3744 * to the right of the truncation point in a crashed ext4 filesystem. But
3745 * that's fine - as long as they are linked from the inode, the post-crash
3746 * ext4_truncate() run will find them and release them.
3748 void ext4_truncate(struct inode *inode)
3751 struct ext4_inode_info *ei = EXT4_I(inode);
3752 __le32 *i_data = ei->i_data;
3753 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3754 struct address_space *mapping = inode->i_mapping;
3755 ext4_lblk_t offsets[4];
3760 ext4_lblk_t last_block;
3761 unsigned blocksize = inode->i_sb->s_blocksize;
3763 if (!ext4_can_truncate(inode))
3766 if (inode->i_size == 0)
3767 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3769 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3770 ext4_ext_truncate(inode);
3774 handle = start_transaction(inode);
3776 return; /* AKPM: return what? */
3778 last_block = (inode->i_size + blocksize-1)
3779 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3781 if (inode->i_size & (blocksize - 1))
3782 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3785 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3787 goto out_stop; /* error */
3790 * OK. This truncate is going to happen. We add the inode to the
3791 * orphan list, so that if this truncate spans multiple transactions,
3792 * and we crash, we will resume the truncate when the filesystem
3793 * recovers. It also marks the inode dirty, to catch the new size.
3795 * Implication: the file must always be in a sane, consistent
3796 * truncatable state while each transaction commits.
3798 if (ext4_orphan_add(handle, inode))
3802 * From here we block out all ext4_get_block() callers who want to
3803 * modify the block allocation tree.
3805 down_write(&ei->i_data_sem);
3807 ext4_discard_reservation(inode);
3810 * The orphan list entry will now protect us from any crash which
3811 * occurs before the truncate completes, so it is now safe to propagate
3812 * the new, shorter inode size (held for now in i_size) into the
3813 * on-disk inode. We do this via i_disksize, which is the value which
3814 * ext4 *really* writes onto the disk inode.
3816 ei->i_disksize = inode->i_size;
3818 if (n == 1) { /* direct blocks */
3819 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3820 i_data + EXT4_NDIR_BLOCKS);
3824 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3825 /* Kill the top of shared branch (not detached) */
3827 if (partial == chain) {
3828 /* Shared branch grows from the inode */
3829 ext4_free_branches(handle, inode, NULL,
3830 &nr, &nr+1, (chain+n-1) - partial);
3833 * We mark the inode dirty prior to restart,
3834 * and prior to stop. No need for it here.
3837 /* Shared branch grows from an indirect block */
3838 BUFFER_TRACE(partial->bh, "get_write_access");
3839 ext4_free_branches(handle, inode, partial->bh,
3841 partial->p+1, (chain+n-1) - partial);
3844 /* Clear the ends of indirect blocks on the shared branch */
3845 while (partial > chain) {
3846 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3847 (__le32*)partial->bh->b_data+addr_per_block,
3848 (chain+n-1) - partial);
3849 BUFFER_TRACE(partial->bh, "call brelse");
3850 brelse (partial->bh);
3854 /* Kill the remaining (whole) subtrees */
3855 switch (offsets[0]) {
3857 nr = i_data[EXT4_IND_BLOCK];
3859 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3860 i_data[EXT4_IND_BLOCK] = 0;
3862 case EXT4_IND_BLOCK:
3863 nr = i_data[EXT4_DIND_BLOCK];
3865 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3866 i_data[EXT4_DIND_BLOCK] = 0;
3868 case EXT4_DIND_BLOCK:
3869 nr = i_data[EXT4_TIND_BLOCK];
3871 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3872 i_data[EXT4_TIND_BLOCK] = 0;
3874 case EXT4_TIND_BLOCK:
3878 up_write(&ei->i_data_sem);
3879 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3880 ext4_mark_inode_dirty(handle, inode);
3883 * In a multi-transaction truncate, we only make the final transaction
3890 * If this was a simple ftruncate(), and the file will remain alive
3891 * then we need to clear up the orphan record which we created above.
3892 * However, if this was a real unlink then we were called by
3893 * ext4_delete_inode(), and we allow that function to clean up the
3894 * orphan info for us.
3897 ext4_orphan_del(handle, inode);
3899 ext4_journal_stop(handle);
3902 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3903 unsigned long ino, struct ext4_iloc *iloc)
3905 ext4_group_t block_group;
3906 unsigned long offset;
3908 struct ext4_group_desc *gdp;
3910 if (!ext4_valid_inum(sb, ino)) {
3912 * This error is already checked for in namei.c unless we are
3913 * looking at an NFS filehandle, in which case no error
3919 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3920 gdp = ext4_get_group_desc(sb, block_group, NULL);
3925 * Figure out the offset within the block group inode table
3927 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3928 EXT4_INODE_SIZE(sb);
3929 block = ext4_inode_table(sb, gdp) +
3930 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3932 iloc->block_group = block_group;
3933 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3938 * ext4_get_inode_loc returns with an extra refcount against the inode's
3939 * underlying buffer_head on success. If 'in_mem' is true, we have all
3940 * data in memory that is needed to recreate the on-disk version of this
3943 static int __ext4_get_inode_loc(struct inode *inode,
3944 struct ext4_iloc *iloc, int in_mem)
3947 struct buffer_head *bh;
3949 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3953 bh = sb_getblk(inode->i_sb, block);
3955 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3956 "unable to read inode block - "
3957 "inode=%lu, block=%llu",
3958 inode->i_ino, block);
3961 if (!buffer_uptodate(bh)) {
3965 * If the buffer has the write error flag, we have failed
3966 * to write out another inode in the same block. In this
3967 * case, we don't have to read the block because we may
3968 * read the old inode data successfully.
3970 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3971 set_buffer_uptodate(bh);
3973 if (buffer_uptodate(bh)) {
3974 /* someone brought it uptodate while we waited */
3980 * If we have all information of the inode in memory and this
3981 * is the only valid inode in the block, we need not read the
3985 struct buffer_head *bitmap_bh;
3986 struct ext4_group_desc *desc;
3987 int inodes_per_buffer;
3988 int inode_offset, i;
3989 ext4_group_t block_group;
3992 block_group = (inode->i_ino - 1) /
3993 EXT4_INODES_PER_GROUP(inode->i_sb);
3994 inodes_per_buffer = bh->b_size /
3995 EXT4_INODE_SIZE(inode->i_sb);
3996 inode_offset = ((inode->i_ino - 1) %
3997 EXT4_INODES_PER_GROUP(inode->i_sb));
3998 start = inode_offset & ~(inodes_per_buffer - 1);
4000 /* Is the inode bitmap in cache? */
4001 desc = ext4_get_group_desc(inode->i_sb,
4006 bitmap_bh = sb_getblk(inode->i_sb,
4007 ext4_inode_bitmap(inode->i_sb, desc));
4012 * If the inode bitmap isn't in cache then the
4013 * optimisation may end up performing two reads instead
4014 * of one, so skip it.
4016 if (!buffer_uptodate(bitmap_bh)) {
4020 for (i = start; i < start + inodes_per_buffer; i++) {
4021 if (i == inode_offset)
4023 if (ext4_test_bit(i, bitmap_bh->b_data))
4027 if (i == start + inodes_per_buffer) {
4028 /* all other inodes are free, so skip I/O */
4029 memset(bh->b_data, 0, bh->b_size);
4030 set_buffer_uptodate(bh);
4038 * There are other valid inodes in the buffer, this inode
4039 * has in-inode xattrs, or we don't have this inode in memory.
4040 * Read the block from disk.
4043 bh->b_end_io = end_buffer_read_sync;
4044 submit_bh(READ_META, bh);
4046 if (!buffer_uptodate(bh)) {
4047 ext4_error(inode->i_sb, "ext4_get_inode_loc",
4048 "unable to read inode block - "
4049 "inode=%lu, block=%llu",
4050 inode->i_ino, block);
4060 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4062 /* We have all inode data except xattrs in memory here. */
4063 return __ext4_get_inode_loc(inode, iloc,
4064 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4067 void ext4_set_inode_flags(struct inode *inode)
4069 unsigned int flags = EXT4_I(inode)->i_flags;
4071 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4072 if (flags & EXT4_SYNC_FL)
4073 inode->i_flags |= S_SYNC;
4074 if (flags & EXT4_APPEND_FL)
4075 inode->i_flags |= S_APPEND;
4076 if (flags & EXT4_IMMUTABLE_FL)
4077 inode->i_flags |= S_IMMUTABLE;
4078 if (flags & EXT4_NOATIME_FL)
4079 inode->i_flags |= S_NOATIME;
4080 if (flags & EXT4_DIRSYNC_FL)
4081 inode->i_flags |= S_DIRSYNC;
4084 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4085 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4087 unsigned int flags = ei->vfs_inode.i_flags;
4089 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4090 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4092 ei->i_flags |= EXT4_SYNC_FL;
4093 if (flags & S_APPEND)
4094 ei->i_flags |= EXT4_APPEND_FL;
4095 if (flags & S_IMMUTABLE)
4096 ei->i_flags |= EXT4_IMMUTABLE_FL;
4097 if (flags & S_NOATIME)
4098 ei->i_flags |= EXT4_NOATIME_FL;
4099 if (flags & S_DIRSYNC)
4100 ei->i_flags |= EXT4_DIRSYNC_FL;
4102 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4103 struct ext4_inode_info *ei)
4106 struct inode *inode = &(ei->vfs_inode);
4107 struct super_block *sb = inode->i_sb;
4109 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4110 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4111 /* we are using combined 48 bit field */
4112 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4113 le32_to_cpu(raw_inode->i_blocks_lo);
4114 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4115 /* i_blocks represent file system block size */
4116 return i_blocks << (inode->i_blkbits - 9);
4121 return le32_to_cpu(raw_inode->i_blocks_lo);
4125 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4127 struct ext4_iloc iloc;
4128 struct ext4_inode *raw_inode;
4129 struct ext4_inode_info *ei;
4130 struct buffer_head *bh;
4131 struct inode *inode;
4135 inode = iget_locked(sb, ino);
4137 return ERR_PTR(-ENOMEM);
4138 if (!(inode->i_state & I_NEW))
4142 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4143 ei->i_acl = EXT4_ACL_NOT_CACHED;
4144 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4146 ei->i_block_alloc_info = NULL;
4148 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4152 raw_inode = ext4_raw_inode(&iloc);
4153 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4154 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4155 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4156 if(!(test_opt (inode->i_sb, NO_UID32))) {
4157 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4158 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4160 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4163 ei->i_dir_start_lookup = 0;
4164 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4165 /* We now have enough fields to check if the inode was active or not.
4166 * This is needed because nfsd might try to access dead inodes
4167 * the test is that same one that e2fsck uses
4168 * NeilBrown 1999oct15
4170 if (inode->i_nlink == 0) {
4171 if (inode->i_mode == 0 ||
4172 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4173 /* this inode is deleted */
4178 /* The only unlinked inodes we let through here have
4179 * valid i_mode and are being read by the orphan
4180 * recovery code: that's fine, we're about to complete
4181 * the process of deleting those. */
4183 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4184 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4185 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4186 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4188 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4189 inode->i_size = ext4_isize(raw_inode);
4190 ei->i_disksize = inode->i_size;
4191 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4192 ei->i_block_group = iloc.block_group;
4194 * NOTE! The in-memory inode i_data array is in little-endian order
4195 * even on big-endian machines: we do NOT byteswap the block numbers!
4197 for (block = 0; block < EXT4_N_BLOCKS; block++)
4198 ei->i_data[block] = raw_inode->i_block[block];
4199 INIT_LIST_HEAD(&ei->i_orphan);
4201 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4202 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4203 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4204 EXT4_INODE_SIZE(inode->i_sb)) {
4209 if (ei->i_extra_isize == 0) {
4210 /* The extra space is currently unused. Use it. */
4211 ei->i_extra_isize = sizeof(struct ext4_inode) -
4212 EXT4_GOOD_OLD_INODE_SIZE;
4214 __le32 *magic = (void *)raw_inode +
4215 EXT4_GOOD_OLD_INODE_SIZE +
4217 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4218 ei->i_state |= EXT4_STATE_XATTR;
4221 ei->i_extra_isize = 0;
4223 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4224 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4225 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4226 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4228 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4229 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4230 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4232 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4235 if (ei->i_file_acl &&
4237 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4238 EXT4_SB(sb)->s_gdb_count)) ||
4239 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4240 ext4_error(sb, __func__,
4241 "bad extended attribute block %llu in inode #%lu",
4242 ei->i_file_acl, inode->i_ino);
4247 if (S_ISREG(inode->i_mode)) {
4248 inode->i_op = &ext4_file_inode_operations;
4249 inode->i_fop = &ext4_file_operations;
4250 ext4_set_aops(inode);
4251 } else if (S_ISDIR(inode->i_mode)) {
4252 inode->i_op = &ext4_dir_inode_operations;
4253 inode->i_fop = &ext4_dir_operations;
4254 } else if (S_ISLNK(inode->i_mode)) {
4255 if (ext4_inode_is_fast_symlink(inode))
4256 inode->i_op = &ext4_fast_symlink_inode_operations;
4258 inode->i_op = &ext4_symlink_inode_operations;
4259 ext4_set_aops(inode);
4261 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4262 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4263 inode->i_op = &ext4_special_inode_operations;
4264 if (raw_inode->i_block[0])
4265 init_special_inode(inode, inode->i_mode,
4266 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4268 init_special_inode(inode, inode->i_mode,
4269 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4273 ext4_error(inode->i_sb, __func__,
4274 "bogus i_mode (%o) for inode=%lu",
4275 inode->i_mode, inode->i_ino);
4279 ext4_set_inode_flags(inode);
4280 unlock_new_inode(inode);
4285 return ERR_PTR(ret);
4288 static int ext4_inode_blocks_set(handle_t *handle,
4289 struct ext4_inode *raw_inode,
4290 struct ext4_inode_info *ei)
4292 struct inode *inode = &(ei->vfs_inode);
4293 u64 i_blocks = inode->i_blocks;
4294 struct super_block *sb = inode->i_sb;
4297 if (i_blocks <= ~0U) {
4299 * i_blocks can be represnted in a 32 bit variable
4300 * as multiple of 512 bytes
4302 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4303 raw_inode->i_blocks_high = 0;
4304 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4305 } else if (i_blocks <= 0xffffffffffffULL) {
4307 * i_blocks can be represented in a 48 bit variable
4308 * as multiple of 512 bytes
4310 err = ext4_update_rocompat_feature(handle, sb,
4311 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4314 /* i_block is stored in the split 48 bit fields */
4315 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4316 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4317 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4320 * i_blocks should be represented in a 48 bit variable
4321 * as multiple of file system block size
4323 err = ext4_update_rocompat_feature(handle, sb,
4324 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4327 ei->i_flags |= EXT4_HUGE_FILE_FL;
4328 /* i_block is stored in file system block size */
4329 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4330 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4331 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4338 * Post the struct inode info into an on-disk inode location in the
4339 * buffer-cache. This gobbles the caller's reference to the
4340 * buffer_head in the inode location struct.
4342 * The caller must have write access to iloc->bh.
4344 static int ext4_do_update_inode(handle_t *handle,
4345 struct inode *inode,
4346 struct ext4_iloc *iloc)
4348 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4349 struct ext4_inode_info *ei = EXT4_I(inode);
4350 struct buffer_head *bh = iloc->bh;
4351 int err = 0, rc, block;
4353 /* For fields not not tracking in the in-memory inode,
4354 * initialise them to zero for new inodes. */
4355 if (ei->i_state & EXT4_STATE_NEW)
4356 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4358 ext4_get_inode_flags(ei);
4359 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4360 if(!(test_opt(inode->i_sb, NO_UID32))) {
4361 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4362 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4364 * Fix up interoperability with old kernels. Otherwise, old inodes get
4365 * re-used with the upper 16 bits of the uid/gid intact
4368 raw_inode->i_uid_high =
4369 cpu_to_le16(high_16_bits(inode->i_uid));
4370 raw_inode->i_gid_high =
4371 cpu_to_le16(high_16_bits(inode->i_gid));
4373 raw_inode->i_uid_high = 0;
4374 raw_inode->i_gid_high = 0;
4377 raw_inode->i_uid_low =
4378 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4379 raw_inode->i_gid_low =
4380 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4381 raw_inode->i_uid_high = 0;
4382 raw_inode->i_gid_high = 0;
4384 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4386 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4387 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4388 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4389 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4391 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4393 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4394 /* clear the migrate flag in the raw_inode */
4395 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4396 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4397 cpu_to_le32(EXT4_OS_HURD))
4398 raw_inode->i_file_acl_high =
4399 cpu_to_le16(ei->i_file_acl >> 32);
4400 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4401 ext4_isize_set(raw_inode, ei->i_disksize);
4402 if (ei->i_disksize > 0x7fffffffULL) {
4403 struct super_block *sb = inode->i_sb;
4404 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4405 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4406 EXT4_SB(sb)->s_es->s_rev_level ==
4407 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4408 /* If this is the first large file
4409 * created, add a flag to the superblock.
4411 err = ext4_journal_get_write_access(handle,
4412 EXT4_SB(sb)->s_sbh);
4415 ext4_update_dynamic_rev(sb);
4416 EXT4_SET_RO_COMPAT_FEATURE(sb,
4417 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4420 err = ext4_journal_dirty_metadata(handle,
4421 EXT4_SB(sb)->s_sbh);
4424 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4425 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4426 if (old_valid_dev(inode->i_rdev)) {
4427 raw_inode->i_block[0] =
4428 cpu_to_le32(old_encode_dev(inode->i_rdev));
4429 raw_inode->i_block[1] = 0;
4431 raw_inode->i_block[0] = 0;
4432 raw_inode->i_block[1] =
4433 cpu_to_le32(new_encode_dev(inode->i_rdev));
4434 raw_inode->i_block[2] = 0;
4436 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4437 raw_inode->i_block[block] = ei->i_data[block];
4439 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4440 if (ei->i_extra_isize) {
4441 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4442 raw_inode->i_version_hi =
4443 cpu_to_le32(inode->i_version >> 32);
4444 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4448 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4449 rc = ext4_journal_dirty_metadata(handle, bh);
4452 ei->i_state &= ~EXT4_STATE_NEW;
4456 ext4_std_error(inode->i_sb, err);
4461 * ext4_write_inode()
4463 * We are called from a few places:
4465 * - Within generic_file_write() for O_SYNC files.
4466 * Here, there will be no transaction running. We wait for any running
4467 * trasnaction to commit.
4469 * - Within sys_sync(), kupdate and such.
4470 * We wait on commit, if tol to.
4472 * - Within prune_icache() (PF_MEMALLOC == true)
4473 * Here we simply return. We can't afford to block kswapd on the
4476 * In all cases it is actually safe for us to return without doing anything,
4477 * because the inode has been copied into a raw inode buffer in
4478 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4481 * Note that we are absolutely dependent upon all inode dirtiers doing the
4482 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4483 * which we are interested.
4485 * It would be a bug for them to not do this. The code:
4487 * mark_inode_dirty(inode)
4489 * inode->i_size = expr;
4491 * is in error because a kswapd-driven write_inode() could occur while
4492 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4493 * will no longer be on the superblock's dirty inode list.
4495 int ext4_write_inode(struct inode *inode, int wait)
4497 if (current->flags & PF_MEMALLOC)
4500 if (ext4_journal_current_handle()) {
4501 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4509 return ext4_force_commit(inode->i_sb);
4515 * Called from notify_change.
4517 * We want to trap VFS attempts to truncate the file as soon as
4518 * possible. In particular, we want to make sure that when the VFS
4519 * shrinks i_size, we put the inode on the orphan list and modify
4520 * i_disksize immediately, so that during the subsequent flushing of
4521 * dirty pages and freeing of disk blocks, we can guarantee that any
4522 * commit will leave the blocks being flushed in an unused state on
4523 * disk. (On recovery, the inode will get truncated and the blocks will
4524 * be freed, so we have a strong guarantee that no future commit will
4525 * leave these blocks visible to the user.)
4527 * Another thing we have to assure is that if we are in ordered mode
4528 * and inode is still attached to the committing transaction, we must
4529 * we start writeout of all the dirty pages which are being truncated.
4530 * This way we are sure that all the data written in the previous
4531 * transaction are already on disk (truncate waits for pages under
4534 * Called with inode->i_mutex down.
4536 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4538 struct inode *inode = dentry->d_inode;
4540 const unsigned int ia_valid = attr->ia_valid;
4542 error = inode_change_ok(inode, attr);
4546 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4547 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4550 /* (user+group)*(old+new) structure, inode write (sb,
4551 * inode block, ? - but truncate inode update has it) */
4552 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4553 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4554 if (IS_ERR(handle)) {
4555 error = PTR_ERR(handle);
4558 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4560 ext4_journal_stop(handle);
4563 /* Update corresponding info in inode so that everything is in
4564 * one transaction */
4565 if (attr->ia_valid & ATTR_UID)
4566 inode->i_uid = attr->ia_uid;
4567 if (attr->ia_valid & ATTR_GID)
4568 inode->i_gid = attr->ia_gid;
4569 error = ext4_mark_inode_dirty(handle, inode);
4570 ext4_journal_stop(handle);
4573 if (attr->ia_valid & ATTR_SIZE) {
4574 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4575 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4577 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4584 if (S_ISREG(inode->i_mode) &&
4585 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4588 handle = ext4_journal_start(inode, 3);
4589 if (IS_ERR(handle)) {
4590 error = PTR_ERR(handle);
4594 error = ext4_orphan_add(handle, inode);
4595 EXT4_I(inode)->i_disksize = attr->ia_size;
4596 rc = ext4_mark_inode_dirty(handle, inode);
4599 ext4_journal_stop(handle);
4601 if (ext4_should_order_data(inode)) {
4602 error = ext4_begin_ordered_truncate(inode,
4605 /* Do as much error cleanup as possible */
4606 handle = ext4_journal_start(inode, 3);
4607 if (IS_ERR(handle)) {
4608 ext4_orphan_del(NULL, inode);
4611 ext4_orphan_del(handle, inode);
4612 ext4_journal_stop(handle);
4618 rc = inode_setattr(inode, attr);
4620 /* If inode_setattr's call to ext4_truncate failed to get a
4621 * transaction handle at all, we need to clean up the in-core
4622 * orphan list manually. */
4624 ext4_orphan_del(NULL, inode);
4626 if (!rc && (ia_valid & ATTR_MODE))
4627 rc = ext4_acl_chmod(inode);
4630 ext4_std_error(inode->i_sb, error);
4636 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4639 struct inode *inode;
4640 unsigned long delalloc_blocks;
4642 inode = dentry->d_inode;
4643 generic_fillattr(inode, stat);
4646 * We can't update i_blocks if the block allocation is delayed
4647 * otherwise in the case of system crash before the real block
4648 * allocation is done, we will have i_blocks inconsistent with
4649 * on-disk file blocks.
4650 * We always keep i_blocks updated together with real
4651 * allocation. But to not confuse with user, stat
4652 * will return the blocks that include the delayed allocation
4653 * blocks for this file.
4655 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4656 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4657 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4659 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4663 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4668 /* if nrblocks are contiguous */
4671 * With N contiguous data blocks, it need at most
4672 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4673 * 2 dindirect blocks
4676 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4677 return indirects + 3;
4680 * if nrblocks are not contiguous, worse case, each block touch
4681 * a indirect block, and each indirect block touch a double indirect
4682 * block, plus a triple indirect block
4684 indirects = nrblocks * 2 + 1;
4688 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4690 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4691 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4692 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4696 * Account for index blocks, block groups bitmaps and block group
4697 * descriptor blocks if modify datablocks and index blocks
4698 * worse case, the indexs blocks spread over different block groups
4700 * If datablocks are discontiguous, they are possible to spread over
4701 * different block groups too. If they are contiugous, with flexbg,
4702 * they could still across block group boundary.
4704 * Also account for superblock, inode, quota and xattr blocks
4706 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4708 int groups, gdpblocks;
4713 * How many index blocks need to touch to modify nrblocks?
4714 * The "Chunk" flag indicating whether the nrblocks is
4715 * physically contiguous on disk
4717 * For Direct IO and fallocate, they calls get_block to allocate
4718 * one single extent at a time, so they could set the "Chunk" flag
4720 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4725 * Now let's see how many group bitmaps and group descriptors need
4735 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4736 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4737 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4738 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4740 /* bitmaps and block group descriptor blocks */
4741 ret += groups + gdpblocks;
4743 /* Blocks for super block, inode, quota and xattr blocks */
4744 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4750 * Calulate the total number of credits to reserve to fit
4751 * the modification of a single pages into a single transaction,
4752 * which may include multiple chunks of block allocations.
4754 * This could be called via ext4_write_begin()
4756 * We need to consider the worse case, when
4757 * one new block per extent.
4759 int ext4_writepage_trans_blocks(struct inode *inode)
4761 int bpp = ext4_journal_blocks_per_page(inode);
4764 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4766 /* Account for data blocks for journalled mode */
4767 if (ext4_should_journal_data(inode))
4773 * Calculate the journal credits for a chunk of data modification.
4775 * This is called from DIO, fallocate or whoever calling
4776 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4778 * journal buffers for data blocks are not included here, as DIO
4779 * and fallocate do no need to journal data buffers.
4781 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4783 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4787 * The caller must have previously called ext4_reserve_inode_write().
4788 * Give this, we know that the caller already has write access to iloc->bh.
4790 int ext4_mark_iloc_dirty(handle_t *handle,
4791 struct inode *inode, struct ext4_iloc *iloc)
4795 if (test_opt(inode->i_sb, I_VERSION))
4796 inode_inc_iversion(inode);
4798 /* the do_update_inode consumes one bh->b_count */
4801 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4802 err = ext4_do_update_inode(handle, inode, iloc);
4808 * On success, We end up with an outstanding reference count against
4809 * iloc->bh. This _must_ be cleaned up later.
4813 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4814 struct ext4_iloc *iloc)
4818 err = ext4_get_inode_loc(inode, iloc);
4820 BUFFER_TRACE(iloc->bh, "get_write_access");
4821 err = ext4_journal_get_write_access(handle, iloc->bh);
4828 ext4_std_error(inode->i_sb, err);
4833 * Expand an inode by new_extra_isize bytes.
4834 * Returns 0 on success or negative error number on failure.
4836 static int ext4_expand_extra_isize(struct inode *inode,
4837 unsigned int new_extra_isize,
4838 struct ext4_iloc iloc,
4841 struct ext4_inode *raw_inode;
4842 struct ext4_xattr_ibody_header *header;
4843 struct ext4_xattr_entry *entry;
4845 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4848 raw_inode = ext4_raw_inode(&iloc);
4850 header = IHDR(inode, raw_inode);
4851 entry = IFIRST(header);
4853 /* No extended attributes present */
4854 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4855 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4856 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4858 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4862 /* try to expand with EAs present */
4863 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4868 * What we do here is to mark the in-core inode as clean with respect to inode
4869 * dirtiness (it may still be data-dirty).
4870 * This means that the in-core inode may be reaped by prune_icache
4871 * without having to perform any I/O. This is a very good thing,
4872 * because *any* task may call prune_icache - even ones which
4873 * have a transaction open against a different journal.
4875 * Is this cheating? Not really. Sure, we haven't written the
4876 * inode out, but prune_icache isn't a user-visible syncing function.
4877 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4878 * we start and wait on commits.
4880 * Is this efficient/effective? Well, we're being nice to the system
4881 * by cleaning up our inodes proactively so they can be reaped
4882 * without I/O. But we are potentially leaving up to five seconds'
4883 * worth of inodes floating about which prune_icache wants us to
4884 * write out. One way to fix that would be to get prune_icache()
4885 * to do a write_super() to free up some memory. It has the desired
4888 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4890 struct ext4_iloc iloc;
4891 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4892 static unsigned int mnt_count;
4896 err = ext4_reserve_inode_write(handle, inode, &iloc);
4897 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4898 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4900 * We need extra buffer credits since we may write into EA block
4901 * with this same handle. If journal_extend fails, then it will
4902 * only result in a minor loss of functionality for that inode.
4903 * If this is felt to be critical, then e2fsck should be run to
4904 * force a large enough s_min_extra_isize.
4906 if ((jbd2_journal_extend(handle,
4907 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4908 ret = ext4_expand_extra_isize(inode,
4909 sbi->s_want_extra_isize,
4912 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4914 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4915 ext4_warning(inode->i_sb, __func__,
4916 "Unable to expand inode %lu. Delete"
4917 " some EAs or run e2fsck.",
4920 le16_to_cpu(sbi->s_es->s_mnt_count);
4926 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4931 * ext4_dirty_inode() is called from __mark_inode_dirty()
4933 * We're really interested in the case where a file is being extended.
4934 * i_size has been changed by generic_commit_write() and we thus need
4935 * to include the updated inode in the current transaction.
4937 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4938 * are allocated to the file.
4940 * If the inode is marked synchronous, we don't honour that here - doing
4941 * so would cause a commit on atime updates, which we don't bother doing.
4942 * We handle synchronous inodes at the highest possible level.
4944 void ext4_dirty_inode(struct inode *inode)
4946 handle_t *current_handle = ext4_journal_current_handle();
4949 handle = ext4_journal_start(inode, 2);
4952 if (current_handle &&
4953 current_handle->h_transaction != handle->h_transaction) {
4954 /* This task has a transaction open against a different fs */
4955 printk(KERN_EMERG "%s: transactions do not match!\n",
4958 jbd_debug(5, "marking dirty. outer handle=%p\n",
4960 ext4_mark_inode_dirty(handle, inode);
4962 ext4_journal_stop(handle);
4969 * Bind an inode's backing buffer_head into this transaction, to prevent
4970 * it from being flushed to disk early. Unlike
4971 * ext4_reserve_inode_write, this leaves behind no bh reference and
4972 * returns no iloc structure, so the caller needs to repeat the iloc
4973 * lookup to mark the inode dirty later.
4975 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4977 struct ext4_iloc iloc;
4981 err = ext4_get_inode_loc(inode, &iloc);
4983 BUFFER_TRACE(iloc.bh, "get_write_access");
4984 err = jbd2_journal_get_write_access(handle, iloc.bh);
4986 err = ext4_journal_dirty_metadata(handle,
4991 ext4_std_error(inode->i_sb, err);
4996 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5003 * We have to be very careful here: changing a data block's
5004 * journaling status dynamically is dangerous. If we write a
5005 * data block to the journal, change the status and then delete
5006 * that block, we risk forgetting to revoke the old log record
5007 * from the journal and so a subsequent replay can corrupt data.
5008 * So, first we make sure that the journal is empty and that
5009 * nobody is changing anything.
5012 journal = EXT4_JOURNAL(inode);
5013 if (is_journal_aborted(journal))
5016 jbd2_journal_lock_updates(journal);
5017 jbd2_journal_flush(journal);
5020 * OK, there are no updates running now, and all cached data is
5021 * synced to disk. We are now in a completely consistent state
5022 * which doesn't have anything in the journal, and we know that
5023 * no filesystem updates are running, so it is safe to modify
5024 * the inode's in-core data-journaling state flag now.
5028 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5030 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5031 ext4_set_aops(inode);
5033 jbd2_journal_unlock_updates(journal);
5035 /* Finally we can mark the inode as dirty. */
5037 handle = ext4_journal_start(inode, 1);
5039 return PTR_ERR(handle);
5041 err = ext4_mark_inode_dirty(handle, inode);
5043 ext4_journal_stop(handle);
5044 ext4_std_error(inode->i_sb, err);
5049 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5051 return !buffer_mapped(bh);
5054 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5056 struct page *page = vmf->page;
5060 struct file *file = vma->vm_file;
5061 struct inode *inode = file->f_path.dentry->d_inode;
5062 struct address_space *mapping = inode->i_mapping;
5065 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5066 * get i_mutex because we are already holding mmap_sem.
5068 down_read(&inode->i_alloc_sem);
5069 size = i_size_read(inode);
5070 if (page->mapping != mapping || size <= page_offset(page)
5071 || !PageUptodate(page)) {
5072 /* page got truncated from under us? */
5076 if (PageMappedToDisk(page))
5079 if (page->index == size >> PAGE_CACHE_SHIFT)
5080 len = size & ~PAGE_CACHE_MASK;
5082 len = PAGE_CACHE_SIZE;
5084 if (page_has_buffers(page)) {
5085 /* return if we have all the buffers mapped */
5086 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5091 * OK, we need to fill the hole... Do write_begin write_end
5092 * to do block allocation/reservation.We are not holding
5093 * inode.i__mutex here. That allow * parallel write_begin,
5094 * write_end call. lock_page prevent this from happening
5095 * on the same page though
5097 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5098 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
5101 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5102 len, len, page, NULL);
5108 ret = VM_FAULT_SIGBUS;
5109 up_read(&inode->i_alloc_sem);