2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode *inode,
49 return jbd2_journal_begin_ordered_truncate(
50 EXT4_SB(inode->i_sb)->s_journal,
51 &EXT4_I(inode)->jinode,
55 static void ext4_invalidatepage(struct page *page, unsigned long offset);
58 * Test whether an inode is a fast symlink.
60 static int ext4_inode_is_fast_symlink(struct inode *inode)
62 int ea_blocks = EXT4_I(inode)->i_file_acl ?
63 (inode->i_sb->s_blocksize >> 9) : 0;
65 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
69 * The ext4 forget function must perform a revoke if we are freeing data
70 * which has been journaled. Metadata (eg. indirect blocks) must be
71 * revoked in all cases.
73 * "bh" may be NULL: a metadata block may have been freed from memory
74 * but there may still be a record of it in the journal, and that record
75 * still needs to be revoked.
77 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
78 struct buffer_head *bh, ext4_fsblk_t blocknr)
84 BUFFER_TRACE(bh, "enter");
86 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
88 bh, is_metadata, inode->i_mode,
89 test_opt(inode->i_sb, DATA_FLAGS));
91 /* Never use the revoke function if we are doing full data
92 * journaling: there is no need to, and a V1 superblock won't
93 * support it. Otherwise, only skip the revoke on un-journaled
96 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
97 (!is_metadata && !ext4_should_journal_data(inode))) {
99 BUFFER_TRACE(bh, "call jbd2_journal_forget");
100 return ext4_journal_forget(handle, bh);
106 * data!=journal && (is_metadata || should_journal_data(inode))
108 BUFFER_TRACE(bh, "call ext4_journal_revoke");
109 err = ext4_journal_revoke(handle, blocknr, bh);
111 ext4_abort(inode->i_sb, __func__,
112 "error %d when attempting revoke", err);
113 BUFFER_TRACE(bh, "exit");
118 * Work out how many blocks we need to proceed with the next chunk of a
119 * truncate transaction.
121 static unsigned long blocks_for_truncate(struct inode *inode)
125 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
127 /* Give ourselves just enough room to cope with inodes in which
128 * i_blocks is corrupt: we've seen disk corruptions in the past
129 * which resulted in random data in an inode which looked enough
130 * like a regular file for ext4 to try to delete it. Things
131 * will go a bit crazy if that happens, but at least we should
132 * try not to panic the whole kernel. */
136 /* But we need to bound the transaction so we don't overflow the
138 if (needed > EXT4_MAX_TRANS_DATA)
139 needed = EXT4_MAX_TRANS_DATA;
141 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
145 * Truncate transactions can be complex and absolutely huge. So we need to
146 * be able to restart the transaction at a conventient checkpoint to make
147 * sure we don't overflow the journal.
149 * start_transaction gets us a new handle for a truncate transaction,
150 * and extend_transaction tries to extend the existing one a bit. If
151 * extend fails, we need to propagate the failure up and restart the
152 * transaction in the top-level truncate loop. --sct
154 static handle_t *start_transaction(struct inode *inode)
158 result = ext4_journal_start(inode, blocks_for_truncate(inode));
162 ext4_std_error(inode->i_sb, PTR_ERR(result));
167 * Try to extend this transaction for the purposes of truncation.
169 * Returns 0 if we managed to create more room. If we can't create more
170 * room, and the transaction must be restarted we return 1.
172 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
174 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
176 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
182 * Restart the transaction associated with *handle. This does a commit,
183 * so before we call here everything must be consistently dirtied against
186 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
188 jbd_debug(2, "restarting handle %p\n", handle);
189 return ext4_journal_restart(handle, blocks_for_truncate(inode));
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_delete_inode (struct inode * inode)
200 if (ext4_should_order_data(inode))
201 ext4_begin_ordered_truncate(inode, 0);
202 truncate_inode_pages(&inode->i_data, 0);
204 if (is_bad_inode(inode))
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL, inode);
222 err = ext4_mark_inode_dirty(handle, inode);
224 ext4_warning(inode->i_sb, __func__,
225 "couldn't mark inode dirty (err %d)", err);
229 ext4_truncate(inode);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (handle->h_buffer_credits < 3) {
238 err = ext4_journal_extend(handle, 3);
240 err = ext4_journal_restart(handle, 3);
242 ext4_warning(inode->i_sb, __func__,
243 "couldn't extend journal (err %d)", err);
245 ext4_journal_stop(handle);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle, inode);
259 EXT4_I(inode)->i_dtime = get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle, inode))
269 /* If that failed, just do the required in-core inode clear. */
272 ext4_free_inode(handle, inode);
273 ext4_journal_stop(handle);
276 clear_inode(inode); /* We must guarantee clearing of inode... */
282 struct buffer_head *bh;
285 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287 p->key = *(p->p = v);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t offsets[4], int *boundary)
326 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
327 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
328 const long direct_blocks = EXT4_NDIR_BLOCKS,
329 indirect_blocks = ptrs,
330 double_blocks = (1 << (ptrs_bits * 2));
335 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
336 } else if (i_block < direct_blocks) {
337 offsets[n++] = i_block;
338 final = direct_blocks;
339 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
340 offsets[n++] = EXT4_IND_BLOCK;
341 offsets[n++] = i_block;
343 } else if ((i_block -= indirect_blocks) < double_blocks) {
344 offsets[n++] = EXT4_DIND_BLOCK;
345 offsets[n++] = i_block >> ptrs_bits;
346 offsets[n++] = i_block & (ptrs - 1);
348 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
349 offsets[n++] = EXT4_TIND_BLOCK;
350 offsets[n++] = i_block >> (ptrs_bits * 2);
351 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
352 offsets[n++] = i_block & (ptrs - 1);
355 ext4_warning(inode->i_sb, "ext4_block_to_path",
356 "block %lu > max in inode %lu",
357 i_block + direct_blocks +
358 indirect_blocks + double_blocks, inode->i_ino);
361 *boundary = final - 1 - (i_block & (ptrs - 1));
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect *ext4_get_branch(struct inode *inode, int depth,
396 ext4_lblk_t *offsets,
397 Indirect chain[4], int *err)
399 struct super_block *sb = inode->i_sb;
401 struct buffer_head *bh;
404 /* i_data is not going away, no lock needed */
405 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
409 bh = sb_bread(sb, le32_to_cpu(p->key));
412 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
426 * ext4_find_near - find a place for allocation with sufficient locality
428 * @ind: descriptor of indirect block.
430 * This function returns the preferred place for block allocation.
431 * It is used when heuristic for sequential allocation fails.
433 * + if there is a block to the left of our position - allocate near it.
434 * + if pointer will live in indirect block - allocate near that block.
435 * + if pointer will live in inode - allocate in the same
438 * In the latter case we colour the starting block by the callers PID to
439 * prevent it from clashing with concurrent allocations for a different inode
440 * in the same block group. The PID is used here so that functionally related
441 * files will be close-by on-disk.
443 * Caller must make sure that @ind is valid and will stay that way.
445 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
447 struct ext4_inode_info *ei = EXT4_I(inode);
448 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
450 ext4_fsblk_t bg_start;
451 ext4_fsblk_t last_block;
452 ext4_grpblk_t colour;
454 /* Try to find previous block */
455 for (p = ind->p - 1; p >= start; p--) {
457 return le32_to_cpu(*p);
460 /* No such thing, so let's try location of indirect block */
462 return ind->bh->b_blocknr;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
469 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
471 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
472 colour = (current->pid % 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
475 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
476 return bg_start + colour;
480 * ext4_find_goal - find a preferred place for allocation.
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
485 * Normally this function find the preferred place for block allocation,
488 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
491 struct ext4_block_alloc_info *block_i;
493 block_i = EXT4_I(inode)->i_block_alloc_info;
496 * try the heuristic for sequential allocation,
497 * failing that at least try to get decent locality.
499 if (block_i && (block == block_i->last_alloc_logical_block + 1)
500 && (block_i->last_alloc_physical_block != 0)) {
501 return block_i->last_alloc_physical_block + 1;
504 return ext4_find_near(inode, partial);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
520 int blocks_to_boundary)
522 unsigned long count = 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks < blocks_to_boundary + 1)
533 count += blocks_to_boundary + 1;
538 while (count < blks && count <= blocks_to_boundary &&
539 le32_to_cpu(*(branch[0].p + count)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
556 ext4_lblk_t iblock, ext4_fsblk_t goal,
557 int indirect_blks, int blks,
558 ext4_fsblk_t new_blocks[4], int *err)
561 unsigned long count = 0, blk_allocated = 0;
563 ext4_fsblk_t current_block = 0;
567 * Here we try to allocate the requested multiple blocks at once,
568 * on a best-effort basis.
569 * To build a branch, we should allocate blocks for
570 * the indirect blocks(if not allocated yet), and at least
571 * the first direct block of this branch. That's the
572 * minimum number of blocks need to allocate(required)
574 /* first we try to allocate the indirect blocks */
575 target = indirect_blks;
578 /* allocating blocks for indirect blocks and direct blocks */
579 current_block = ext4_new_meta_blocks(handle, inode,
585 /* allocate blocks for indirect blocks */
586 while (index < indirect_blks && count) {
587 new_blocks[index++] = current_block++;
592 * save the new block number
593 * for the first direct block
595 new_blocks[index] = current_block;
596 printk(KERN_INFO "%s returned more blocks than "
597 "requested\n", __func__);
603 target = blks - count ;
604 blk_allocated = count;
607 /* Now allocate data blocks */
609 /* allocating blocks for data blocks */
610 current_block = ext4_new_blocks(handle, inode, iblock,
612 if (*err && (target == blks)) {
614 * if the allocation failed and we didn't allocate
620 if (target == blks) {
622 * save the new block number
623 * for the first direct block
625 new_blocks[index] = current_block;
627 blk_allocated += count;
630 /* total number of blocks allocated for direct blocks */
635 for (i = 0; i <index; i++)
636 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
641 * ext4_alloc_branch - allocate and set up a chain of blocks.
643 * @indirect_blks: number of allocated indirect blocks
644 * @blks: number of allocated direct blocks
645 * @offsets: offsets (in the blocks) to store the pointers to next.
646 * @branch: place to store the chain in.
648 * This function allocates blocks, zeroes out all but the last one,
649 * links them into chain and (if we are synchronous) writes them to disk.
650 * In other words, it prepares a branch that can be spliced onto the
651 * inode. It stores the information about that chain in the branch[], in
652 * the same format as ext4_get_branch() would do. We are calling it after
653 * we had read the existing part of chain and partial points to the last
654 * triple of that (one with zero ->key). Upon the exit we have the same
655 * picture as after the successful ext4_get_block(), except that in one
656 * place chain is disconnected - *branch->p is still zero (we did not
657 * set the last link), but branch->key contains the number that should
658 * be placed into *branch->p to fill that gap.
660 * If allocation fails we free all blocks we've allocated (and forget
661 * their buffer_heads) and return the error value the from failed
662 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
663 * as described above and return 0.
665 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
666 ext4_lblk_t iblock, int indirect_blks,
667 int *blks, ext4_fsblk_t goal,
668 ext4_lblk_t *offsets, Indirect *branch)
670 int blocksize = inode->i_sb->s_blocksize;
673 struct buffer_head *bh;
675 ext4_fsblk_t new_blocks[4];
676 ext4_fsblk_t current_block;
678 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
679 *blks, new_blocks, &err);
683 branch[0].key = cpu_to_le32(new_blocks[0]);
685 * metadata blocks and data blocks are allocated.
687 for (n = 1; n <= indirect_blks; n++) {
689 * Get buffer_head for parent block, zero it out
690 * and set the pointer to new one, then send
693 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
696 BUFFER_TRACE(bh, "call get_create_access");
697 err = ext4_journal_get_create_access(handle, bh);
704 memset(bh->b_data, 0, blocksize);
705 branch[n].p = (__le32 *) bh->b_data + offsets[n];
706 branch[n].key = cpu_to_le32(new_blocks[n]);
707 *branch[n].p = branch[n].key;
708 if ( n == indirect_blks) {
709 current_block = new_blocks[n];
711 * End of chain, update the last new metablock of
712 * the chain to point to the new allocated
713 * data blocks numbers
715 for (i=1; i < num; i++)
716 *(branch[n].p + i) = cpu_to_le32(++current_block);
718 BUFFER_TRACE(bh, "marking uptodate");
719 set_buffer_uptodate(bh);
722 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
723 err = ext4_journal_dirty_metadata(handle, bh);
730 /* Allocation failed, free what we already allocated */
731 for (i = 1; i <= n ; i++) {
732 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
733 ext4_journal_forget(handle, branch[i].bh);
735 for (i = 0; i <indirect_blks; i++)
736 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
738 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
744 * ext4_splice_branch - splice the allocated branch onto inode.
746 * @block: (logical) number of block we are adding
747 * @chain: chain of indirect blocks (with a missing link - see
749 * @where: location of missing link
750 * @num: number of indirect blocks we are adding
751 * @blks: number of direct blocks we are adding
753 * This function fills the missing link and does all housekeeping needed in
754 * inode (->i_blocks, etc.). In case of success we end up with the full
755 * chain to new block and return 0.
757 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
758 ext4_lblk_t block, Indirect *where, int num, int blks)
762 struct ext4_block_alloc_info *block_i;
763 ext4_fsblk_t current_block;
765 block_i = EXT4_I(inode)->i_block_alloc_info;
767 * If we're splicing into a [td]indirect block (as opposed to the
768 * inode) then we need to get write access to the [td]indirect block
772 BUFFER_TRACE(where->bh, "get_write_access");
773 err = ext4_journal_get_write_access(handle, where->bh);
779 *where->p = where->key;
782 * Update the host buffer_head or inode to point to more just allocated
783 * direct blocks blocks
785 if (num == 0 && blks > 1) {
786 current_block = le32_to_cpu(where->key) + 1;
787 for (i = 1; i < blks; i++)
788 *(where->p + i ) = cpu_to_le32(current_block++);
792 * update the most recently allocated logical & physical block
793 * in i_block_alloc_info, to assist find the proper goal block for next
797 block_i->last_alloc_logical_block = block + blks - 1;
798 block_i->last_alloc_physical_block =
799 le32_to_cpu(where[num].key) + blks - 1;
802 /* We are done with atomic stuff, now do the rest of housekeeping */
804 inode->i_ctime = ext4_current_time(inode);
805 ext4_mark_inode_dirty(handle, inode);
807 /* had we spliced it onto indirect block? */
810 * If we spliced it onto an indirect block, we haven't
811 * altered the inode. Note however that if it is being spliced
812 * onto an indirect block at the very end of the file (the
813 * file is growing) then we *will* alter the inode to reflect
814 * the new i_size. But that is not done here - it is done in
815 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
817 jbd_debug(5, "splicing indirect only\n");
818 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
819 err = ext4_journal_dirty_metadata(handle, where->bh);
824 * OK, we spliced it into the inode itself on a direct block.
825 * Inode was dirtied above.
827 jbd_debug(5, "splicing direct\n");
832 for (i = 1; i <= num; i++) {
833 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
834 ext4_journal_forget(handle, where[i].bh);
835 ext4_free_blocks(handle, inode,
836 le32_to_cpu(where[i-1].key), 1, 0);
838 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
844 * Allocation strategy is simple: if we have to allocate something, we will
845 * have to go the whole way to leaf. So let's do it before attaching anything
846 * to tree, set linkage between the newborn blocks, write them if sync is
847 * required, recheck the path, free and repeat if check fails, otherwise
848 * set the last missing link (that will protect us from any truncate-generated
849 * removals - all blocks on the path are immune now) and possibly force the
850 * write on the parent block.
851 * That has a nice additional property: no special recovery from the failed
852 * allocations is needed - we simply release blocks and do not touch anything
853 * reachable from inode.
855 * `handle' can be NULL if create == 0.
857 * return > 0, # of blocks mapped or allocated.
858 * return = 0, if plain lookup failed.
859 * return < 0, error case.
862 * Need to be called with
863 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
864 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
866 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
867 ext4_lblk_t iblock, unsigned long maxblocks,
868 struct buffer_head *bh_result,
869 int create, int extend_disksize)
872 ext4_lblk_t offsets[4];
877 int blocks_to_boundary = 0;
879 struct ext4_inode_info *ei = EXT4_I(inode);
881 ext4_fsblk_t first_block = 0;
885 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
886 J_ASSERT(handle != NULL || create == 0);
887 depth = ext4_block_to_path(inode, iblock, offsets,
888 &blocks_to_boundary);
893 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
895 /* Simplest case - block found, no allocation needed */
897 first_block = le32_to_cpu(chain[depth - 1].key);
898 clear_buffer_new(bh_result);
901 while (count < maxblocks && count <= blocks_to_boundary) {
904 blk = le32_to_cpu(*(chain[depth-1].p + count));
906 if (blk == first_block + count)
914 /* Next simple case - plain lookup or failed read of indirect block */
915 if (!create || err == -EIO)
919 * Okay, we need to do block allocation. Lazily initialize the block
920 * allocation info here if necessary
922 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
923 ext4_init_block_alloc_info(inode);
925 goal = ext4_find_goal(inode, iblock, partial);
927 /* the number of blocks need to allocate for [d,t]indirect blocks */
928 indirect_blks = (chain + depth) - partial - 1;
931 * Next look up the indirect map to count the totoal number of
932 * direct blocks to allocate for this branch.
934 count = ext4_blks_to_allocate(partial, indirect_blks,
935 maxblocks, blocks_to_boundary);
937 * Block out ext4_truncate while we alter the tree
939 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
941 offsets + (partial - chain), partial);
944 * The ext4_splice_branch call will free and forget any buffers
945 * on the new chain if there is a failure, but that risks using
946 * up transaction credits, especially for bitmaps where the
947 * credits cannot be returned. Can we handle this somehow? We
948 * may need to return -EAGAIN upwards in the worst case. --sct
951 err = ext4_splice_branch(handle, inode, iblock,
952 partial, indirect_blks, count);
954 * i_disksize growing is protected by i_data_sem. Don't forget to
955 * protect it if you're about to implement concurrent
956 * ext4_get_block() -bzzz
958 if (!err && extend_disksize) {
959 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
960 if (disksize > i_size_read(inode))
961 disksize = i_size_read(inode);
962 if (disksize > ei->i_disksize)
963 ei->i_disksize = disksize;
968 set_buffer_new(bh_result);
970 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
971 if (count > blocks_to_boundary)
972 set_buffer_boundary(bh_result);
974 /* Clean up and exit */
975 partial = chain + depth - 1; /* the whole chain */
977 while (partial > chain) {
978 BUFFER_TRACE(partial->bh, "call brelse");
982 BUFFER_TRACE(bh_result, "returned");
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate @blocks for non extent file based file
991 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
993 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
994 int ind_blks, dind_blks, tind_blks;
996 /* number of new indirect blocks needed */
997 ind_blks = (blocks + icap - 1) / icap;
999 dind_blks = (ind_blks + icap - 1) / icap;
1003 return ind_blks + dind_blks + tind_blks;
1007 * Calculate the number of metadata blocks need to reserve
1008 * to allocate given number of blocks
1010 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1015 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1016 return ext4_ext_calc_metadata_amount(inode, blocks);
1018 return ext4_indirect_calc_metadata_amount(inode, blocks);
1021 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1023 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1024 int total, mdb, mdb_free;
1026 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1027 /* recalculate the number of metablocks still need to be reserved */
1028 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1029 mdb = ext4_calc_metadata_amount(inode, total);
1031 /* figure out how many metablocks to release */
1032 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1033 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1035 /* Account for allocated meta_blocks */
1036 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1038 /* update fs free blocks counter for truncate case */
1039 percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
1041 /* update per-inode reservations */
1042 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1043 EXT4_I(inode)->i_reserved_data_blocks -= used;
1045 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1046 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1047 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1048 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1051 * If we have done all the pending block allocations and if
1052 * there aren't any writers on the inode, we can discard the
1053 * inode's preallocations.
1055 if (!total && (atomic_read(&inode->i_writecount) == 0))
1056 ext4_discard_reservation(inode);
1060 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1061 * and returns if the blocks are already mapped.
1063 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1064 * and store the allocated blocks in the result buffer head and mark it
1067 * If file type is extents based, it will call ext4_ext_get_blocks(),
1068 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1071 * On success, it returns the number of blocks being mapped or allocate.
1072 * if create==0 and the blocks are pre-allocated and uninitialized block,
1073 * the result buffer head is unmapped. If the create ==1, it will make sure
1074 * the buffer head is mapped.
1076 * It returns 0 if plain look up failed (blocks have not been allocated), in
1077 * that casem, buffer head is unmapped
1079 * It returns the error in case of allocation failure.
1081 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1082 unsigned long max_blocks, struct buffer_head *bh,
1083 int create, int extend_disksize, int flag)
1087 clear_buffer_mapped(bh);
1088 clear_buffer_unwritten(bh);
1091 * Try to see if we can get the block without requesting
1092 * for new file system block.
1094 down_read((&EXT4_I(inode)->i_data_sem));
1095 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1096 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1099 retval = ext4_get_blocks_handle(handle,
1100 inode, block, max_blocks, bh, 0, 0);
1102 up_read((&EXT4_I(inode)->i_data_sem));
1104 /* If it is only a block(s) look up */
1109 * Returns if the blocks have already allocated
1111 * Note that if blocks have been preallocated
1112 * ext4_ext_get_block() returns th create = 0
1113 * with buffer head unmapped.
1115 if (retval > 0 && buffer_mapped(bh))
1119 * When we call get_blocks without the create flag, the
1120 * BH_Unwritten flag could have gotten set if the blocks
1121 * requested were part of a uninitialized extent. We need to
1122 * clear this flag now that we are committed to convert all or
1123 * part of the uninitialized extent to be an initialized
1124 * extent. This is because we need to avoid the combination
1125 * of BH_Unwritten and BH_Mapped flags being simultaneously
1126 * set on the buffer_head.
1128 clear_buffer_unwritten(bh);
1131 * New blocks allocate and/or writing to uninitialized extent
1132 * will possibly result in updating i_data, so we take
1133 * the write lock of i_data_sem, and call get_blocks()
1134 * with create == 1 flag.
1136 down_write((&EXT4_I(inode)->i_data_sem));
1139 * if the caller is from delayed allocation writeout path
1140 * we have already reserved fs blocks for allocation
1141 * let the underlying get_block() function know to
1142 * avoid double accounting
1145 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1147 * We need to check for EXT4 here because migrate
1148 * could have changed the inode type in between
1150 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1151 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1152 bh, create, extend_disksize);
1154 retval = ext4_get_blocks_handle(handle, inode, block,
1155 max_blocks, bh, create, extend_disksize);
1157 if (retval > 0 && buffer_new(bh)) {
1159 * We allocated new blocks which will result in
1160 * i_data's format changing. Force the migrate
1161 * to fail by clearing migrate flags
1163 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1169 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1171 * Update reserved blocks/metadata blocks
1172 * after successful block allocation
1173 * which were deferred till now
1175 if ((retval > 0) && buffer_delay(bh))
1176 ext4_da_update_reserve_space(inode, retval);
1179 up_write((&EXT4_I(inode)->i_data_sem));
1183 /* Maximum number of blocks we map for direct IO at once. */
1184 #define DIO_MAX_BLOCKS 4096
1186 static int ext4_get_block(struct inode *inode, sector_t iblock,
1187 struct buffer_head *bh_result, int create)
1189 handle_t *handle = ext4_journal_current_handle();
1190 int ret = 0, started = 0;
1191 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1194 if (create && !handle) {
1195 /* Direct IO write... */
1196 if (max_blocks > DIO_MAX_BLOCKS)
1197 max_blocks = DIO_MAX_BLOCKS;
1198 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1199 handle = ext4_journal_start(inode, dio_credits);
1200 if (IS_ERR(handle)) {
1201 ret = PTR_ERR(handle);
1207 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1208 max_blocks, bh_result, create, 0, 0);
1210 bh_result->b_size = (ret << inode->i_blkbits);
1214 ext4_journal_stop(handle);
1220 * `handle' can be NULL if create is zero
1222 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1223 ext4_lblk_t block, int create, int *errp)
1225 struct buffer_head dummy;
1228 J_ASSERT(handle != NULL || create == 0);
1231 dummy.b_blocknr = -1000;
1232 buffer_trace_init(&dummy.b_history);
1233 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1234 &dummy, create, 1, 0);
1236 * ext4_get_blocks_handle() returns number of blocks
1237 * mapped. 0 in case of a HOLE.
1245 if (!err && buffer_mapped(&dummy)) {
1246 struct buffer_head *bh;
1247 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1252 if (buffer_new(&dummy)) {
1253 J_ASSERT(create != 0);
1254 J_ASSERT(handle != NULL);
1257 * Now that we do not always journal data, we should
1258 * keep in mind whether this should always journal the
1259 * new buffer as metadata. For now, regular file
1260 * writes use ext4_get_block instead, so it's not a
1264 BUFFER_TRACE(bh, "call get_create_access");
1265 fatal = ext4_journal_get_create_access(handle, bh);
1266 if (!fatal && !buffer_uptodate(bh)) {
1267 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1268 set_buffer_uptodate(bh);
1271 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1272 err = ext4_journal_dirty_metadata(handle, bh);
1276 BUFFER_TRACE(bh, "not a new buffer");
1289 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1290 ext4_lblk_t block, int create, int *err)
1292 struct buffer_head * bh;
1294 bh = ext4_getblk(handle, inode, block, create, err);
1297 if (buffer_uptodate(bh))
1299 ll_rw_block(READ_META, 1, &bh);
1301 if (buffer_uptodate(bh))
1308 static int walk_page_buffers( handle_t *handle,
1309 struct buffer_head *head,
1313 int (*fn)( handle_t *handle,
1314 struct buffer_head *bh))
1316 struct buffer_head *bh;
1317 unsigned block_start, block_end;
1318 unsigned blocksize = head->b_size;
1320 struct buffer_head *next;
1322 for ( bh = head, block_start = 0;
1323 ret == 0 && (bh != head || !block_start);
1324 block_start = block_end, bh = next)
1326 next = bh->b_this_page;
1327 block_end = block_start + blocksize;
1328 if (block_end <= from || block_start >= to) {
1329 if (partial && !buffer_uptodate(bh))
1333 err = (*fn)(handle, bh);
1341 * To preserve ordering, it is essential that the hole instantiation and
1342 * the data write be encapsulated in a single transaction. We cannot
1343 * close off a transaction and start a new one between the ext4_get_block()
1344 * and the commit_write(). So doing the jbd2_journal_start at the start of
1345 * prepare_write() is the right place.
1347 * Also, this function can nest inside ext4_writepage() ->
1348 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1349 * has generated enough buffer credits to do the whole page. So we won't
1350 * block on the journal in that case, which is good, because the caller may
1353 * By accident, ext4 can be reentered when a transaction is open via
1354 * quota file writes. If we were to commit the transaction while thus
1355 * reentered, there can be a deadlock - we would be holding a quota
1356 * lock, and the commit would never complete if another thread had a
1357 * transaction open and was blocking on the quota lock - a ranking
1360 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1361 * will _not_ run commit under these circumstances because handle->h_ref
1362 * is elevated. We'll still have enough credits for the tiny quotafile
1365 static int do_journal_get_write_access(handle_t *handle,
1366 struct buffer_head *bh)
1368 if (!buffer_mapped(bh) || buffer_freed(bh))
1370 return ext4_journal_get_write_access(handle, bh);
1373 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1374 loff_t pos, unsigned len, unsigned flags,
1375 struct page **pagep, void **fsdata)
1377 struct inode *inode = mapping->host;
1378 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1385 index = pos >> PAGE_CACHE_SHIFT;
1386 from = pos & (PAGE_CACHE_SIZE - 1);
1390 handle = ext4_journal_start(inode, needed_blocks);
1391 if (IS_ERR(handle)) {
1392 ret = PTR_ERR(handle);
1396 /* We cannot recurse into the filesystem as the transaction is already
1398 flags |= AOP_FLAG_NOFS;
1400 page = grab_cache_page_write_begin(mapping, index, flags);
1402 ext4_journal_stop(handle);
1408 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1411 if (!ret && ext4_should_journal_data(inode)) {
1412 ret = walk_page_buffers(handle, page_buffers(page),
1413 from, to, NULL, do_journal_get_write_access);
1418 ext4_journal_stop(handle);
1419 page_cache_release(page);
1422 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1428 /* For write_end() in data=journal mode */
1429 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1431 if (!buffer_mapped(bh) || buffer_freed(bh))
1433 set_buffer_uptodate(bh);
1434 return ext4_journal_dirty_metadata(handle, bh);
1438 * We need to pick up the new inode size which generic_commit_write gave us
1439 * `file' can be NULL - eg, when called from page_symlink().
1441 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1442 * buffers are managed internally.
1444 static int ext4_ordered_write_end(struct file *file,
1445 struct address_space *mapping,
1446 loff_t pos, unsigned len, unsigned copied,
1447 struct page *page, void *fsdata)
1449 handle_t *handle = ext4_journal_current_handle();
1450 struct inode *inode = mapping->host;
1453 ret = ext4_jbd2_file_inode(handle, inode);
1457 * generic_write_end() will run mark_inode_dirty() if i_size
1458 * changes. So let's piggyback the i_disksize mark_inode_dirty
1463 new_i_size = pos + copied;
1464 if (new_i_size > EXT4_I(inode)->i_disksize)
1465 EXT4_I(inode)->i_disksize = new_i_size;
1466 ret2 = generic_write_end(file, mapping, pos, len, copied,
1472 ret2 = ext4_journal_stop(handle);
1476 return ret ? ret : copied;
1479 static int ext4_writeback_write_end(struct file *file,
1480 struct address_space *mapping,
1481 loff_t pos, unsigned len, unsigned copied,
1482 struct page *page, void *fsdata)
1484 handle_t *handle = ext4_journal_current_handle();
1485 struct inode *inode = mapping->host;
1489 new_i_size = pos + copied;
1490 if (new_i_size > EXT4_I(inode)->i_disksize)
1491 EXT4_I(inode)->i_disksize = new_i_size;
1493 ret2 = generic_write_end(file, mapping, pos, len, copied,
1499 ret2 = ext4_journal_stop(handle);
1503 return ret ? ret : copied;
1506 static int ext4_journalled_write_end(struct file *file,
1507 struct address_space *mapping,
1508 loff_t pos, unsigned len, unsigned copied,
1509 struct page *page, void *fsdata)
1511 handle_t *handle = ext4_journal_current_handle();
1512 struct inode *inode = mapping->host;
1517 from = pos & (PAGE_CACHE_SIZE - 1);
1521 if (!PageUptodate(page))
1523 page_zero_new_buffers(page, from+copied, to);
1526 ret = walk_page_buffers(handle, page_buffers(page), from,
1527 to, &partial, write_end_fn);
1529 SetPageUptodate(page);
1530 if (pos+copied > inode->i_size)
1531 i_size_write(inode, pos+copied);
1532 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1533 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1534 EXT4_I(inode)->i_disksize = inode->i_size;
1535 ret2 = ext4_mark_inode_dirty(handle, inode);
1541 ret2 = ext4_journal_stop(handle);
1544 page_cache_release(page);
1546 return ret ? ret : copied;
1549 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1551 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1552 unsigned long md_needed, mdblocks, total = 0;
1555 * recalculate the amount of metadata blocks to reserve
1556 * in order to allocate nrblocks
1557 * worse case is one extent per block
1559 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1560 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1561 mdblocks = ext4_calc_metadata_amount(inode, total);
1562 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1564 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1565 total = md_needed + nrblocks;
1567 if (ext4_claim_free_blocks(sbi, total)) {
1568 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1571 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1572 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1574 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1575 return 0; /* success */
1578 static void ext4_da_release_space(struct inode *inode, int to_free)
1580 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1581 int total, mdb, mdb_free, release;
1584 return; /* Nothing to release, exit */
1586 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1588 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1590 * if there is no reserved blocks, but we try to free some
1591 * then the counter is messed up somewhere.
1592 * but since this function is called from invalidate
1593 * page, it's harmless to return without any action
1595 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1596 "blocks for inode %lu, but there is no reserved "
1597 "data blocks\n", to_free, inode->i_ino);
1598 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1602 /* recalculate the number of metablocks still need to be reserved */
1603 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1604 mdb = ext4_calc_metadata_amount(inode, total);
1606 /* figure out how many metablocks to release */
1607 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1608 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1610 release = to_free + mdb_free;
1612 /* update fs free blocks counter for truncate case */
1613 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1615 /* update per-inode reservations */
1616 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1617 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1619 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1620 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1621 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1624 static void ext4_da_page_release_reservation(struct page *page,
1625 unsigned long offset)
1628 struct buffer_head *head, *bh;
1629 unsigned int curr_off = 0;
1631 head = page_buffers(page);
1634 unsigned int next_off = curr_off + bh->b_size;
1636 if ((offset <= curr_off) && (buffer_delay(bh))) {
1638 clear_buffer_delay(bh);
1640 curr_off = next_off;
1641 } while ((bh = bh->b_this_page) != head);
1642 ext4_da_release_space(page->mapping->host, to_release);
1646 * Delayed allocation stuff
1649 struct mpage_da_data {
1650 struct inode *inode;
1651 struct buffer_head lbh; /* extent of blocks */
1652 unsigned long first_page, next_page; /* extent of pages */
1653 get_block_t *get_block;
1654 struct writeback_control *wbc;
1660 * mpage_da_submit_io - walks through extent of pages and try to write
1661 * them with writepage() call back
1663 * @mpd->inode: inode
1664 * @mpd->first_page: first page of the extent
1665 * @mpd->next_page: page after the last page of the extent
1666 * @mpd->get_block: the filesystem's block mapper function
1668 * By the time mpage_da_submit_io() is called we expect all blocks
1669 * to be allocated. this may be wrong if allocation failed.
1671 * As pages are already locked by write_cache_pages(), we can't use it
1673 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1676 struct pagevec pvec;
1677 unsigned long index, end;
1678 int ret = 0, err, nr_pages, i;
1679 struct inode *inode = mpd->inode;
1680 struct address_space *mapping = inode->i_mapping;
1682 BUG_ON(mpd->next_page <= mpd->first_page);
1684 * We need to start from the first_page to the next_page - 1
1685 * to make sure we also write the mapped dirty buffer_heads.
1686 * If we look at mpd->lbh.b_blocknr we would only be looking
1687 * at the currently mapped buffer_heads.
1689 index = mpd->first_page;
1690 end = mpd->next_page - 1;
1692 pagevec_init(&pvec, 0);
1693 while (index <= end) {
1694 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1697 for (i = 0; i < nr_pages; i++) {
1698 struct page *page = pvec.pages[i];
1700 index = page->index;
1705 BUG_ON(!PageLocked(page));
1706 BUG_ON(PageWriteback(page));
1708 pages_skipped = mpd->wbc->pages_skipped;
1709 err = mapping->a_ops->writepage(page, mpd->wbc);
1711 mpd->pages_written++;
1713 * In error case, we have to continue because
1714 * remaining pages are still locked
1715 * XXX: unlock and re-dirty them?
1720 pagevec_release(&pvec);
1726 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1728 * @mpd->inode - inode to walk through
1729 * @exbh->b_blocknr - first block on a disk
1730 * @exbh->b_size - amount of space in bytes
1731 * @logical - first logical block to start assignment with
1733 * the function goes through all passed space and put actual disk
1734 * block numbers into buffer heads, dropping BH_Delay
1736 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1737 struct buffer_head *exbh)
1739 struct inode *inode = mpd->inode;
1740 struct address_space *mapping = inode->i_mapping;
1741 int blocks = exbh->b_size >> inode->i_blkbits;
1742 sector_t pblock = exbh->b_blocknr, cur_logical;
1743 struct buffer_head *head, *bh;
1745 struct pagevec pvec;
1748 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1749 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1750 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1752 pagevec_init(&pvec, 0);
1754 while (index <= end) {
1755 /* XXX: optimize tail */
1756 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1759 for (i = 0; i < nr_pages; i++) {
1760 struct page *page = pvec.pages[i];
1762 index = page->index;
1767 BUG_ON(!PageLocked(page));
1768 BUG_ON(PageWriteback(page));
1769 BUG_ON(!page_has_buffers(page));
1771 bh = page_buffers(page);
1774 /* skip blocks out of the range */
1776 if (cur_logical >= logical)
1779 } while ((bh = bh->b_this_page) != head);
1782 if (cur_logical >= logical + blocks)
1784 if (buffer_delay(bh)) {
1785 bh->b_blocknr = pblock;
1786 clear_buffer_delay(bh);
1787 bh->b_bdev = inode->i_sb->s_bdev;
1788 } else if (buffer_unwritten(bh)) {
1789 bh->b_blocknr = pblock;
1790 clear_buffer_unwritten(bh);
1791 set_buffer_mapped(bh);
1793 bh->b_bdev = inode->i_sb->s_bdev;
1794 } else if (buffer_mapped(bh))
1795 BUG_ON(bh->b_blocknr != pblock);
1799 } while ((bh = bh->b_this_page) != head);
1801 pagevec_release(&pvec);
1807 * __unmap_underlying_blocks - just a helper function to unmap
1808 * set of blocks described by @bh
1810 static inline void __unmap_underlying_blocks(struct inode *inode,
1811 struct buffer_head *bh)
1813 struct block_device *bdev = inode->i_sb->s_bdev;
1816 blocks = bh->b_size >> inode->i_blkbits;
1817 for (i = 0; i < blocks; i++)
1818 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1821 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1822 sector_t logical, long blk_cnt)
1826 struct pagevec pvec;
1827 struct inode *inode = mpd->inode;
1828 struct address_space *mapping = inode->i_mapping;
1830 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1831 end = (logical + blk_cnt - 1) >>
1832 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1833 while (index <= end) {
1834 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1837 for (i = 0; i < nr_pages; i++) {
1838 struct page *page = pvec.pages[i];
1839 index = page->index;
1844 BUG_ON(!PageLocked(page));
1845 BUG_ON(PageWriteback(page));
1846 block_invalidatepage(page, 0);
1847 ClearPageUptodate(page);
1855 * mpage_da_map_blocks - go through given space
1857 * @mpd->lbh - bh describing space
1858 * @mpd->get_block - the filesystem's block mapper function
1860 * The function skips space we know is already mapped to disk blocks.
1863 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1866 struct buffer_head *lbh = &mpd->lbh;
1867 sector_t next = lbh->b_blocknr;
1868 struct buffer_head new;
1871 * We consider only non-mapped and non-allocated blocks
1873 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1876 new.b_state = lbh->b_state;
1878 new.b_size = lbh->b_size;
1881 * If we didn't accumulate anything
1882 * to write simply return
1886 err = mpd->get_block(mpd->inode, next, &new, 1);
1889 /* If get block returns with error
1890 * we simply return. Later writepage
1891 * will redirty the page and writepages
1892 * will find the dirty page again
1897 * get block failure will cause us
1898 * to loop in writepages. Because
1899 * a_ops->writepage won't be able to
1900 * make progress. The page will be redirtied
1901 * by writepage and writepages will again
1902 * try to write the same.
1904 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1905 "at logical offset %llu with max blocks "
1906 "%zd with error %d\n",
1907 __func__, mpd->inode->i_ino,
1908 (unsigned long long)next,
1909 lbh->b_size >> mpd->inode->i_blkbits, err);
1910 printk(KERN_EMERG "This should not happen.!! "
1911 "Data will be lost\n");
1912 /* invlaidate all the pages */
1913 ext4_da_block_invalidatepages(mpd, next,
1914 lbh->b_size >> mpd->inode->i_blkbits);
1917 BUG_ON(new.b_size == 0);
1919 if (buffer_new(&new))
1920 __unmap_underlying_blocks(mpd->inode, &new);
1923 * If blocks are delayed marked, we need to
1924 * put actual blocknr and drop delayed bit
1926 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1927 mpage_put_bnr_to_bhs(mpd, next, &new);
1932 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1933 (1 << BH_Delay) | (1 << BH_Unwritten))
1936 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1938 * @mpd->lbh - extent of blocks
1939 * @logical - logical number of the block in the file
1940 * @bh - bh of the block (used to access block's state)
1942 * the function is used to collect contig. blocks in same state
1944 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1945 sector_t logical, struct buffer_head *bh)
1948 size_t b_size = bh->b_size;
1949 struct buffer_head *lbh = &mpd->lbh;
1950 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1952 /* check if thereserved journal credits might overflow */
1953 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1954 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1956 * With non-extent format we are limited by the journal
1957 * credit available. Total credit needed to insert
1958 * nrblocks contiguous blocks is dependent on the
1959 * nrblocks. So limit nrblocks.
1962 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1963 EXT4_MAX_TRANS_DATA) {
1965 * Adding the new buffer_head would make it cross the
1966 * allowed limit for which we have journal credit
1967 * reserved. So limit the new bh->b_size
1969 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1970 mpd->inode->i_blkbits;
1971 /* we will do mpage_da_submit_io in the next loop */
1975 * First block in the extent
1977 if (lbh->b_size == 0) {
1978 lbh->b_blocknr = logical;
1979 lbh->b_size = b_size;
1980 lbh->b_state = bh->b_state & BH_FLAGS;
1984 next = lbh->b_blocknr + nrblocks;
1986 * Can we merge the block to our big extent?
1988 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1989 lbh->b_size += b_size;
1995 * We couldn't merge the block to our extent, so we
1996 * need to flush current extent and start new one
1998 if (mpage_da_map_blocks(mpd) == 0)
1999 mpage_da_submit_io(mpd);
2005 * __mpage_da_writepage - finds extent of pages and blocks
2007 * @page: page to consider
2008 * @wbc: not used, we just follow rules
2011 * The function finds extents of pages and scan them for all blocks.
2013 static int __mpage_da_writepage(struct page *page,
2014 struct writeback_control *wbc, void *data)
2016 struct mpage_da_data *mpd = data;
2017 struct inode *inode = mpd->inode;
2018 struct buffer_head *bh, *head, fake;
2023 * Rest of the page in the page_vec
2024 * redirty then and skip then. We will
2025 * try to to write them again after
2026 * starting a new transaction
2028 redirty_page_for_writepage(wbc, page);
2030 return MPAGE_DA_EXTENT_TAIL;
2033 * Can we merge this page to current extent?
2035 if (mpd->next_page != page->index) {
2037 * Nope, we can't. So, we map non-allocated blocks
2038 * and start IO on them using writepage()
2040 if (mpd->next_page != mpd->first_page) {
2041 if (mpage_da_map_blocks(mpd) == 0)
2042 mpage_da_submit_io(mpd);
2044 * skip rest of the page in the page_vec
2047 redirty_page_for_writepage(wbc, page);
2049 return MPAGE_DA_EXTENT_TAIL;
2053 * Start next extent of pages ...
2055 mpd->first_page = page->index;
2060 mpd->lbh.b_size = 0;
2061 mpd->lbh.b_state = 0;
2062 mpd->lbh.b_blocknr = 0;
2065 mpd->next_page = page->index + 1;
2066 logical = (sector_t) page->index <<
2067 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2069 if (!page_has_buffers(page)) {
2071 * There is no attached buffer heads yet (mmap?)
2072 * we treat the page asfull of dirty blocks
2075 bh->b_size = PAGE_CACHE_SIZE;
2077 set_buffer_dirty(bh);
2078 set_buffer_uptodate(bh);
2079 mpage_add_bh_to_extent(mpd, logical, bh);
2081 return MPAGE_DA_EXTENT_TAIL;
2084 * Page with regular buffer heads, just add all dirty ones
2086 head = page_buffers(page);
2089 BUG_ON(buffer_locked(bh));
2091 * We need to try to allocate
2092 * unmapped blocks in the same page.
2093 * Otherwise we won't make progress
2094 * with the page in ext4_da_writepage
2096 if (buffer_dirty(bh) &&
2097 (!buffer_mapped(bh) || buffer_delay(bh))) {
2098 mpage_add_bh_to_extent(mpd, logical, bh);
2100 return MPAGE_DA_EXTENT_TAIL;
2101 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2103 * mapped dirty buffer. We need to update
2104 * the b_state because we look at
2105 * b_state in mpage_da_map_blocks. We don't
2106 * update b_size because if we find an
2107 * unmapped buffer_head later we need to
2108 * use the b_state flag of that buffer_head.
2110 if (mpd->lbh.b_size == 0)
2112 bh->b_state & BH_FLAGS;
2115 } while ((bh = bh->b_this_page) != head);
2122 * mpage_da_writepages - walk the list of dirty pages of the given
2123 * address space, allocates non-allocated blocks, maps newly-allocated
2124 * blocks to existing bhs and issue IO them
2126 * @mapping: address space structure to write
2127 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2128 * @get_block: the filesystem's block mapper function.
2130 * This is a library function, which implements the writepages()
2131 * address_space_operation.
2133 static int mpage_da_writepages(struct address_space *mapping,
2134 struct writeback_control *wbc,
2135 get_block_t get_block)
2137 struct mpage_da_data mpd;
2142 return generic_writepages(mapping, wbc);
2145 mpd.inode = mapping->host;
2147 mpd.lbh.b_state = 0;
2148 mpd.lbh.b_blocknr = 0;
2151 mpd.get_block = get_block;
2153 mpd.pages_written = 0;
2155 to_write = wbc->nr_to_write;
2157 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
2160 * Handle last extent of pages
2162 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2163 if (mpage_da_map_blocks(&mpd) == 0)
2164 mpage_da_submit_io(&mpd);
2167 wbc->nr_to_write = to_write - mpd.pages_written;
2172 * this is a special callback for ->write_begin() only
2173 * it's intention is to return mapped block or reserve space
2175 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2176 struct buffer_head *bh_result, int create)
2179 sector_t invalid_block = ~((sector_t) 0xffff);
2181 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2184 BUG_ON(create == 0);
2185 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2188 * first, we need to know whether the block is allocated already
2189 * preallocated blocks are unmapped but should treated
2190 * the same as allocated blocks.
2192 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2193 if ((ret == 0) && !buffer_delay(bh_result)) {
2194 /* the block isn't (pre)allocated yet, let's reserve space */
2196 * XXX: __block_prepare_write() unmaps passed block,
2199 ret = ext4_da_reserve_space(inode, 1);
2201 /* not enough space to reserve */
2204 map_bh(bh_result, inode->i_sb, invalid_block);
2205 set_buffer_new(bh_result);
2206 set_buffer_delay(bh_result);
2207 } else if (ret > 0) {
2208 bh_result->b_size = (ret << inode->i_blkbits);
2210 * With sub-block writes into unwritten extents
2211 * we also need to mark the buffer as new so that
2212 * the unwritten parts of the buffer gets correctly zeroed.
2214 if (buffer_unwritten(bh_result))
2215 set_buffer_new(bh_result);
2221 #define EXT4_DELALLOC_RSVED 1
2222 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2223 struct buffer_head *bh_result, int create)
2226 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2227 loff_t disksize = EXT4_I(inode)->i_disksize;
2228 handle_t *handle = NULL;
2230 handle = ext4_journal_current_handle();
2232 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2233 bh_result, 0, 0, 0);
2236 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2237 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2241 bh_result->b_size = (ret << inode->i_blkbits);
2244 * Update on-disk size along with block allocation
2245 * we don't use 'extend_disksize' as size may change
2246 * within already allocated block -bzzz
2248 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2249 if (disksize > i_size_read(inode))
2250 disksize = i_size_read(inode);
2251 if (disksize > EXT4_I(inode)->i_disksize) {
2253 * XXX: replace with spinlock if seen contended -bzzz
2255 down_write(&EXT4_I(inode)->i_data_sem);
2256 if (disksize > EXT4_I(inode)->i_disksize)
2257 EXT4_I(inode)->i_disksize = disksize;
2258 up_write(&EXT4_I(inode)->i_data_sem);
2260 if (EXT4_I(inode)->i_disksize == disksize) {
2261 ret = ext4_mark_inode_dirty(handle, inode);
2270 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2273 * unmapped buffer is possible for holes.
2274 * delay buffer is possible with delayed allocation
2276 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2279 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2280 struct buffer_head *bh_result, int create)
2283 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2286 * we don't want to do block allocation in writepage
2287 * so call get_block_wrap with create = 0
2289 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2290 bh_result, 0, 0, 0);
2292 bh_result->b_size = (ret << inode->i_blkbits);
2299 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2300 * get called via journal_submit_inode_data_buffers (no journal handle)
2301 * get called via shrink_page_list via pdflush (no journal handle)
2302 * or grab_page_cache when doing write_begin (have journal handle)
2304 static int ext4_da_writepage(struct page *page,
2305 struct writeback_control *wbc)
2310 struct buffer_head *page_bufs;
2311 struct inode *inode = page->mapping->host;
2313 size = i_size_read(inode);
2314 if (page->index == size >> PAGE_CACHE_SHIFT)
2315 len = size & ~PAGE_CACHE_MASK;
2317 len = PAGE_CACHE_SIZE;
2319 if (page_has_buffers(page)) {
2320 page_bufs = page_buffers(page);
2321 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2322 ext4_bh_unmapped_or_delay)) {
2324 * We don't want to do block allocation
2325 * So redirty the page and return
2326 * We may reach here when we do a journal commit
2327 * via journal_submit_inode_data_buffers.
2328 * If we don't have mapping block we just ignore
2329 * them. We can also reach here via shrink_page_list
2331 redirty_page_for_writepage(wbc, page);
2337 * The test for page_has_buffers() is subtle:
2338 * We know the page is dirty but it lost buffers. That means
2339 * that at some moment in time after write_begin()/write_end()
2340 * has been called all buffers have been clean and thus they
2341 * must have been written at least once. So they are all
2342 * mapped and we can happily proceed with mapping them
2343 * and writing the page.
2345 * Try to initialize the buffer_heads and check whether
2346 * all are mapped and non delay. We don't want to
2347 * do block allocation here.
2349 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2350 ext4_normal_get_block_write);
2352 page_bufs = page_buffers(page);
2353 /* check whether all are mapped and non delay */
2354 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2355 ext4_bh_unmapped_or_delay)) {
2356 redirty_page_for_writepage(wbc, page);
2362 * We can't do block allocation here
2363 * so just redity the page and unlock
2366 redirty_page_for_writepage(wbc, page);
2370 /* now mark the buffer_heads as dirty and uptodate */
2371 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2374 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2375 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2377 ret = block_write_full_page(page,
2378 ext4_normal_get_block_write,
2385 * This is called via ext4_da_writepages() to
2386 * calulate the total number of credits to reserve to fit
2387 * a single extent allocation into a single transaction,
2388 * ext4_da_writpeages() will loop calling this before
2389 * the block allocation.
2392 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2394 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2397 * With non-extent format the journal credit needed to
2398 * insert nrblocks contiguous block is dependent on
2399 * number of contiguous block. So we will limit
2400 * number of contiguous block to a sane value
2402 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2403 (max_blocks > EXT4_MAX_TRANS_DATA))
2404 max_blocks = EXT4_MAX_TRANS_DATA;
2406 return ext4_chunk_trans_blocks(inode, max_blocks);
2409 static int ext4_da_writepages(struct address_space *mapping,
2410 struct writeback_control *wbc)
2412 handle_t *handle = NULL;
2413 loff_t range_start = 0;
2414 struct inode *inode = mapping->host;
2415 int needed_blocks, ret = 0, nr_to_writebump = 0;
2416 long to_write, pages_skipped = 0;
2417 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2420 * No pages to write? This is mainly a kludge to avoid starting
2421 * a transaction for special inodes like journal inode on last iput()
2422 * because that could violate lock ordering on umount
2424 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2428 * If the filesystem has aborted, it is read-only, so return
2429 * right away instead of dumping stack traces later on that
2430 * will obscure the real source of the problem. We test
2431 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2432 * the latter could be true if the filesystem is mounted
2433 * read-only, and in that case, ext4_da_writepages should
2434 * *never* be called, so if that ever happens, we would want
2437 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2441 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2442 * This make sure small files blocks are allocated in
2443 * single attempt. This ensure that small files
2444 * get less fragmented.
2446 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2447 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2448 wbc->nr_to_write = sbi->s_mb_stream_request;
2451 if (!wbc->range_cyclic)
2453 * If range_cyclic is not set force range_cont
2454 * and save the old writeback_index
2456 wbc->range_cont = 1;
2458 range_start = wbc->range_start;
2459 pages_skipped = wbc->pages_skipped;
2462 to_write = wbc->nr_to_write;
2463 while (!ret && to_write > 0) {
2466 * we insert one extent at a time. So we need
2467 * credit needed for single extent allocation.
2468 * journalled mode is currently not supported
2471 BUG_ON(ext4_should_journal_data(inode));
2472 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2474 /* start a new transaction*/
2475 handle = ext4_journal_start(inode, needed_blocks);
2476 if (IS_ERR(handle)) {
2477 ret = PTR_ERR(handle);
2478 printk(KERN_CRIT "%s: jbd2_start: "
2479 "%ld pages, ino %lu; err %d\n", __func__,
2480 wbc->nr_to_write, inode->i_ino, ret);
2482 goto out_writepages;
2484 if (ext4_should_order_data(inode)) {
2486 * With ordered mode we need to add
2487 * the inode to the journal handl
2488 * when we do block allocation.
2490 ret = ext4_jbd2_file_inode(handle, inode);
2492 ext4_journal_stop(handle);
2493 goto out_writepages;
2497 to_write -= wbc->nr_to_write;
2498 ret = mpage_da_writepages(mapping, wbc,
2499 ext4_da_get_block_write);
2500 ext4_journal_stop(handle);
2501 if (ret == MPAGE_DA_EXTENT_TAIL) {
2503 * got one extent now try with
2506 to_write += wbc->nr_to_write;
2508 } else if (wbc->nr_to_write) {
2510 * There is no more writeout needed
2511 * or we requested for a noblocking writeout
2512 * and we found the device congested
2514 to_write += wbc->nr_to_write;
2517 wbc->nr_to_write = to_write;
2520 if (wbc->range_cont && (pages_skipped != wbc->pages_skipped)) {
2521 /* We skipped pages in this loop */
2522 wbc->range_start = range_start;
2523 wbc->nr_to_write = to_write +
2524 wbc->pages_skipped - pages_skipped;
2525 wbc->pages_skipped = pages_skipped;
2530 wbc->nr_to_write = to_write - nr_to_writebump;
2531 wbc->range_start = range_start;
2535 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2536 loff_t pos, unsigned len, unsigned flags,
2537 struct page **pagep, void **fsdata)
2539 int ret, retries = 0;
2543 struct inode *inode = mapping->host;
2546 index = pos >> PAGE_CACHE_SHIFT;
2547 from = pos & (PAGE_CACHE_SIZE - 1);
2552 * With delayed allocation, we don't log the i_disksize update
2553 * if there is delayed block allocation. But we still need
2554 * to journalling the i_disksize update if writes to the end
2555 * of file which has an already mapped buffer.
2557 handle = ext4_journal_start(inode, 1);
2558 if (IS_ERR(handle)) {
2559 ret = PTR_ERR(handle);
2562 /* We cannot recurse into the filesystem as the transaction is already
2564 flags |= AOP_FLAG_NOFS;
2566 page = grab_cache_page_write_begin(mapping, index, flags);
2568 ext4_journal_stop(handle);
2574 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2575 ext4_da_get_block_prep);
2578 ext4_journal_stop(handle);
2579 page_cache_release(page);
2582 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2589 * Check if we should update i_disksize
2590 * when write to the end of file but not require block allocation
2592 static int ext4_da_should_update_i_disksize(struct page *page,
2593 unsigned long offset)
2595 struct buffer_head *bh;
2596 struct inode *inode = page->mapping->host;
2600 bh = page_buffers(page);
2601 idx = offset >> inode->i_blkbits;
2603 for (i=0; i < idx; i++)
2604 bh = bh->b_this_page;
2606 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2611 static int ext4_da_write_end(struct file *file,
2612 struct address_space *mapping,
2613 loff_t pos, unsigned len, unsigned copied,
2614 struct page *page, void *fsdata)
2616 struct inode *inode = mapping->host;
2618 handle_t *handle = ext4_journal_current_handle();
2620 unsigned long start, end;
2622 start = pos & (PAGE_CACHE_SIZE - 1);
2623 end = start + copied -1;
2626 * generic_write_end() will run mark_inode_dirty() if i_size
2627 * changes. So let's piggyback the i_disksize mark_inode_dirty
2631 new_i_size = pos + copied;
2632 if (new_i_size > EXT4_I(inode)->i_disksize) {
2633 if (ext4_da_should_update_i_disksize(page, end)) {
2634 down_write(&EXT4_I(inode)->i_data_sem);
2635 if (new_i_size > EXT4_I(inode)->i_disksize) {
2637 * Updating i_disksize when extending file
2638 * without needing block allocation
2640 if (ext4_should_order_data(inode))
2641 ret = ext4_jbd2_file_inode(handle,
2644 EXT4_I(inode)->i_disksize = new_i_size;
2646 up_write(&EXT4_I(inode)->i_data_sem);
2649 ret2 = generic_write_end(file, mapping, pos, len, copied,
2654 ret2 = ext4_journal_stop(handle);
2658 return ret ? ret : copied;
2661 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2664 * Drop reserved blocks
2666 BUG_ON(!PageLocked(page));
2667 if (!page_has_buffers(page))
2670 ext4_da_page_release_reservation(page, offset);
2673 ext4_invalidatepage(page, offset);
2679 * Force all delayed allocation blocks to be allocated for a given inode.
2681 int ext4_alloc_da_blocks(struct inode *inode)
2683 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2684 !EXT4_I(inode)->i_reserved_meta_blocks)
2688 * We do something simple for now. The filemap_flush() will
2689 * also start triggering a write of the data blocks, which is
2690 * not strictly speaking necessary (and for users of
2691 * laptop_mode, not even desirable). However, to do otherwise
2692 * would require replicating code paths in:
2694 * ext4_da_writepages() ->
2695 * write_cache_pages() ---> (via passed in callback function)
2696 * __mpage_da_writepage() -->
2697 * mpage_add_bh_to_extent()
2698 * mpage_da_map_blocks()
2700 * The problem is that write_cache_pages(), located in
2701 * mm/page-writeback.c, marks pages clean in preparation for
2702 * doing I/O, which is not desirable if we're not planning on
2705 * We could call write_cache_pages(), and then redirty all of
2706 * the pages by calling redirty_page_for_writeback() but that
2707 * would be ugly in the extreme. So instead we would need to
2708 * replicate parts of the code in the above functions,
2709 * simplifying them becuase we wouldn't actually intend to
2710 * write out the pages, but rather only collect contiguous
2711 * logical block extents, call the multi-block allocator, and
2712 * then update the buffer heads with the block allocations.
2714 * For now, though, we'll cheat by calling filemap_flush(),
2715 * which will map the blocks, and start the I/O, but not
2716 * actually wait for the I/O to complete.
2718 return filemap_flush(inode->i_mapping);
2722 * bmap() is special. It gets used by applications such as lilo and by
2723 * the swapper to find the on-disk block of a specific piece of data.
2725 * Naturally, this is dangerous if the block concerned is still in the
2726 * journal. If somebody makes a swapfile on an ext4 data-journaling
2727 * filesystem and enables swap, then they may get a nasty shock when the
2728 * data getting swapped to that swapfile suddenly gets overwritten by
2729 * the original zero's written out previously to the journal and
2730 * awaiting writeback in the kernel's buffer cache.
2732 * So, if we see any bmap calls here on a modified, data-journaled file,
2733 * take extra steps to flush any blocks which might be in the cache.
2735 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2737 struct inode *inode = mapping->host;
2741 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2742 test_opt(inode->i_sb, DELALLOC)) {
2744 * With delalloc we want to sync the file
2745 * so that we can make sure we allocate
2748 filemap_write_and_wait(mapping);
2751 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2753 * This is a REALLY heavyweight approach, but the use of
2754 * bmap on dirty files is expected to be extremely rare:
2755 * only if we run lilo or swapon on a freshly made file
2756 * do we expect this to happen.
2758 * (bmap requires CAP_SYS_RAWIO so this does not
2759 * represent an unprivileged user DOS attack --- we'd be
2760 * in trouble if mortal users could trigger this path at
2763 * NB. EXT4_STATE_JDATA is not set on files other than
2764 * regular files. If somebody wants to bmap a directory
2765 * or symlink and gets confused because the buffer
2766 * hasn't yet been flushed to disk, they deserve
2767 * everything they get.
2770 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2771 journal = EXT4_JOURNAL(inode);
2772 jbd2_journal_lock_updates(journal);
2773 err = jbd2_journal_flush(journal);
2774 jbd2_journal_unlock_updates(journal);
2780 return generic_block_bmap(mapping,block,ext4_get_block);
2783 static int bget_one(handle_t *handle, struct buffer_head *bh)
2789 static int bput_one(handle_t *handle, struct buffer_head *bh)
2796 * Note that we don't need to start a transaction unless we're journaling data
2797 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2798 * need to file the inode to the transaction's list in ordered mode because if
2799 * we are writing back data added by write(), the inode is already there and if
2800 * we are writing back data modified via mmap(), noone guarantees in which
2801 * transaction the data will hit the disk. In case we are journaling data, we
2802 * cannot start transaction directly because transaction start ranks above page
2803 * lock so we have to do some magic.
2805 * In all journaling modes block_write_full_page() will start the I/O.
2809 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2814 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2816 * Same applies to ext4_get_block(). We will deadlock on various things like
2817 * lock_journal and i_data_sem
2819 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2822 * 16May01: If we're reentered then journal_current_handle() will be
2823 * non-zero. We simply *return*.
2825 * 1 July 2001: @@@ FIXME:
2826 * In journalled data mode, a data buffer may be metadata against the
2827 * current transaction. But the same file is part of a shared mapping
2828 * and someone does a writepage() on it.
2830 * We will move the buffer onto the async_data list, but *after* it has
2831 * been dirtied. So there's a small window where we have dirty data on
2834 * Note that this only applies to the last partial page in the file. The
2835 * bit which block_write_full_page() uses prepare/commit for. (That's
2836 * broken code anyway: it's wrong for msync()).
2838 * It's a rare case: affects the final partial page, for journalled data
2839 * where the file is subject to bith write() and writepage() in the same
2840 * transction. To fix it we'll need a custom block_write_full_page().
2841 * We'll probably need that anyway for journalling writepage() output.
2843 * We don't honour synchronous mounts for writepage(). That would be
2844 * disastrous. Any write() or metadata operation will sync the fs for
2848 static int __ext4_normal_writepage(struct page *page,
2849 struct writeback_control *wbc)
2851 struct inode *inode = page->mapping->host;
2853 if (test_opt(inode->i_sb, NOBH))
2854 return nobh_writepage(page,
2855 ext4_normal_get_block_write, wbc);
2857 return block_write_full_page(page,
2858 ext4_normal_get_block_write,
2862 static int ext4_normal_writepage(struct page *page,
2863 struct writeback_control *wbc)
2865 struct inode *inode = page->mapping->host;
2866 loff_t size = i_size_read(inode);
2869 J_ASSERT(PageLocked(page));
2870 if (page->index == size >> PAGE_CACHE_SHIFT)
2871 len = size & ~PAGE_CACHE_MASK;
2873 len = PAGE_CACHE_SIZE;
2875 if (page_has_buffers(page)) {
2876 /* if page has buffers it should all be mapped
2877 * and allocated. If there are not buffers attached
2878 * to the page we know the page is dirty but it lost
2879 * buffers. That means that at some moment in time
2880 * after write_begin() / write_end() has been called
2881 * all buffers have been clean and thus they must have been
2882 * written at least once. So they are all mapped and we can
2883 * happily proceed with mapping them and writing the page.
2885 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2886 ext4_bh_unmapped_or_delay));
2889 if (!ext4_journal_current_handle())
2890 return __ext4_normal_writepage(page, wbc);
2892 redirty_page_for_writepage(wbc, page);
2897 static int __ext4_journalled_writepage(struct page *page,
2898 struct writeback_control *wbc)
2900 struct address_space *mapping = page->mapping;
2901 struct inode *inode = mapping->host;
2902 struct buffer_head *page_bufs;
2903 handle_t *handle = NULL;
2907 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2908 ext4_normal_get_block_write);
2912 page_bufs = page_buffers(page);
2913 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2915 /* As soon as we unlock the page, it can go away, but we have
2916 * references to buffers so we are safe */
2919 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2920 if (IS_ERR(handle)) {
2921 ret = PTR_ERR(handle);
2925 ret = walk_page_buffers(handle, page_bufs, 0,
2926 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2928 err = walk_page_buffers(handle, page_bufs, 0,
2929 PAGE_CACHE_SIZE, NULL, write_end_fn);
2932 err = ext4_journal_stop(handle);
2936 walk_page_buffers(handle, page_bufs, 0,
2937 PAGE_CACHE_SIZE, NULL, bput_one);
2938 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2947 static int ext4_journalled_writepage(struct page *page,
2948 struct writeback_control *wbc)
2950 struct inode *inode = page->mapping->host;
2951 loff_t size = i_size_read(inode);
2954 J_ASSERT(PageLocked(page));
2955 if (page->index == size >> PAGE_CACHE_SHIFT)
2956 len = size & ~PAGE_CACHE_MASK;
2958 len = PAGE_CACHE_SIZE;
2960 if (page_has_buffers(page)) {
2961 /* if page has buffers it should all be mapped
2962 * and allocated. If there are not buffers attached
2963 * to the page we know the page is dirty but it lost
2964 * buffers. That means that at some moment in time
2965 * after write_begin() / write_end() has been called
2966 * all buffers have been clean and thus they must have been
2967 * written at least once. So they are all mapped and we can
2968 * happily proceed with mapping them and writing the page.
2970 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2971 ext4_bh_unmapped_or_delay));
2974 if (ext4_journal_current_handle())
2977 if (PageChecked(page)) {
2979 * It's mmapped pagecache. Add buffers and journal it. There
2980 * doesn't seem much point in redirtying the page here.
2982 ClearPageChecked(page);
2983 return __ext4_journalled_writepage(page, wbc);
2986 * It may be a page full of checkpoint-mode buffers. We don't
2987 * really know unless we go poke around in the buffer_heads.
2988 * But block_write_full_page will do the right thing.
2990 return block_write_full_page(page,
2991 ext4_normal_get_block_write,
2995 redirty_page_for_writepage(wbc, page);
3000 static int ext4_readpage(struct file *file, struct page *page)
3002 return mpage_readpage(page, ext4_get_block);
3006 ext4_readpages(struct file *file, struct address_space *mapping,
3007 struct list_head *pages, unsigned nr_pages)
3009 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3012 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3014 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3017 * If it's a full truncate we just forget about the pending dirtying
3020 ClearPageChecked(page);
3022 jbd2_journal_invalidatepage(journal, page, offset);
3025 static int ext4_releasepage(struct page *page, gfp_t wait)
3027 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3029 WARN_ON(PageChecked(page));
3030 if (!page_has_buffers(page))
3032 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3036 * If the O_DIRECT write will extend the file then add this inode to the
3037 * orphan list. So recovery will truncate it back to the original size
3038 * if the machine crashes during the write.
3040 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3041 * crashes then stale disk data _may_ be exposed inside the file. But current
3042 * VFS code falls back into buffered path in that case so we are safe.
3044 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3045 const struct iovec *iov, loff_t offset,
3046 unsigned long nr_segs)
3048 struct file *file = iocb->ki_filp;
3049 struct inode *inode = file->f_mapping->host;
3050 struct ext4_inode_info *ei = EXT4_I(inode);
3054 size_t count = iov_length(iov, nr_segs);
3057 loff_t final_size = offset + count;
3059 if (final_size > inode->i_size) {
3060 /* Credits for sb + inode write */
3061 handle = ext4_journal_start(inode, 2);
3062 if (IS_ERR(handle)) {
3063 ret = PTR_ERR(handle);
3066 ret = ext4_orphan_add(handle, inode);
3068 ext4_journal_stop(handle);
3072 ei->i_disksize = inode->i_size;
3073 ext4_journal_stop(handle);
3077 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3079 ext4_get_block, NULL);
3084 /* Credits for sb + inode write */
3085 handle = ext4_journal_start(inode, 2);
3086 if (IS_ERR(handle)) {
3087 /* This is really bad luck. We've written the data
3088 * but cannot extend i_size. Bail out and pretend
3089 * the write failed... */
3090 ret = PTR_ERR(handle);
3094 ext4_orphan_del(handle, inode);
3096 loff_t end = offset + ret;
3097 if (end > inode->i_size) {
3098 ei->i_disksize = end;
3099 i_size_write(inode, end);
3101 * We're going to return a positive `ret'
3102 * here due to non-zero-length I/O, so there's
3103 * no way of reporting error returns from
3104 * ext4_mark_inode_dirty() to userspace. So
3107 ext4_mark_inode_dirty(handle, inode);
3110 err = ext4_journal_stop(handle);
3119 * Pages can be marked dirty completely asynchronously from ext4's journalling
3120 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3121 * much here because ->set_page_dirty is called under VFS locks. The page is
3122 * not necessarily locked.
3124 * We cannot just dirty the page and leave attached buffers clean, because the
3125 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3126 * or jbddirty because all the journalling code will explode.
3128 * So what we do is to mark the page "pending dirty" and next time writepage
3129 * is called, propagate that into the buffers appropriately.
3131 static int ext4_journalled_set_page_dirty(struct page *page)
3133 SetPageChecked(page);
3134 return __set_page_dirty_nobuffers(page);
3137 static const struct address_space_operations ext4_ordered_aops = {
3138 .readpage = ext4_readpage,
3139 .readpages = ext4_readpages,
3140 .writepage = ext4_normal_writepage,
3141 .sync_page = block_sync_page,
3142 .write_begin = ext4_write_begin,
3143 .write_end = ext4_ordered_write_end,
3145 .invalidatepage = ext4_invalidatepage,
3146 .releasepage = ext4_releasepage,
3147 .direct_IO = ext4_direct_IO,
3148 .migratepage = buffer_migrate_page,
3149 .is_partially_uptodate = block_is_partially_uptodate,
3152 static const struct address_space_operations ext4_writeback_aops = {
3153 .readpage = ext4_readpage,
3154 .readpages = ext4_readpages,
3155 .writepage = ext4_normal_writepage,
3156 .sync_page = block_sync_page,
3157 .write_begin = ext4_write_begin,
3158 .write_end = ext4_writeback_write_end,
3160 .invalidatepage = ext4_invalidatepage,
3161 .releasepage = ext4_releasepage,
3162 .direct_IO = ext4_direct_IO,
3163 .migratepage = buffer_migrate_page,
3164 .is_partially_uptodate = block_is_partially_uptodate,
3167 static const struct address_space_operations ext4_journalled_aops = {
3168 .readpage = ext4_readpage,
3169 .readpages = ext4_readpages,
3170 .writepage = ext4_journalled_writepage,
3171 .sync_page = block_sync_page,
3172 .write_begin = ext4_write_begin,
3173 .write_end = ext4_journalled_write_end,
3174 .set_page_dirty = ext4_journalled_set_page_dirty,
3176 .invalidatepage = ext4_invalidatepage,
3177 .releasepage = ext4_releasepage,
3178 .is_partially_uptodate = block_is_partially_uptodate,
3181 static const struct address_space_operations ext4_da_aops = {
3182 .readpage = ext4_readpage,
3183 .readpages = ext4_readpages,
3184 .writepage = ext4_da_writepage,
3185 .writepages = ext4_da_writepages,
3186 .sync_page = block_sync_page,
3187 .write_begin = ext4_da_write_begin,
3188 .write_end = ext4_da_write_end,
3190 .invalidatepage = ext4_da_invalidatepage,
3191 .releasepage = ext4_releasepage,
3192 .direct_IO = ext4_direct_IO,
3193 .migratepage = buffer_migrate_page,
3194 .is_partially_uptodate = block_is_partially_uptodate,
3197 void ext4_set_aops(struct inode *inode)
3199 if (ext4_should_order_data(inode) &&
3200 test_opt(inode->i_sb, DELALLOC))
3201 inode->i_mapping->a_ops = &ext4_da_aops;
3202 else if (ext4_should_order_data(inode))
3203 inode->i_mapping->a_ops = &ext4_ordered_aops;
3204 else if (ext4_should_writeback_data(inode) &&
3205 test_opt(inode->i_sb, DELALLOC))
3206 inode->i_mapping->a_ops = &ext4_da_aops;
3207 else if (ext4_should_writeback_data(inode))
3208 inode->i_mapping->a_ops = &ext4_writeback_aops;
3210 inode->i_mapping->a_ops = &ext4_journalled_aops;
3214 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3215 * up to the end of the block which corresponds to `from'.
3216 * This required during truncate. We need to physically zero the tail end
3217 * of that block so it doesn't yield old data if the file is later grown.
3219 int ext4_block_truncate_page(handle_t *handle,
3220 struct address_space *mapping, loff_t from)
3222 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3223 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3224 unsigned blocksize, length, pos;
3226 struct inode *inode = mapping->host;
3227 struct buffer_head *bh;
3231 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3235 blocksize = inode->i_sb->s_blocksize;
3236 length = blocksize - (offset & (blocksize - 1));
3237 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3240 * For "nobh" option, we can only work if we don't need to
3241 * read-in the page - otherwise we create buffers to do the IO.
3243 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3244 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3245 zero_user(page, offset, length);
3246 set_page_dirty(page);
3250 if (!page_has_buffers(page))
3251 create_empty_buffers(page, blocksize, 0);
3253 /* Find the buffer that contains "offset" */
3254 bh = page_buffers(page);
3256 while (offset >= pos) {
3257 bh = bh->b_this_page;
3263 if (buffer_freed(bh)) {
3264 BUFFER_TRACE(bh, "freed: skip");
3268 if (!buffer_mapped(bh)) {
3269 BUFFER_TRACE(bh, "unmapped");
3270 ext4_get_block(inode, iblock, bh, 0);
3271 /* unmapped? It's a hole - nothing to do */
3272 if (!buffer_mapped(bh)) {
3273 BUFFER_TRACE(bh, "still unmapped");
3278 /* Ok, it's mapped. Make sure it's up-to-date */
3279 if (PageUptodate(page))
3280 set_buffer_uptodate(bh);
3282 if (!buffer_uptodate(bh)) {
3284 ll_rw_block(READ, 1, &bh);
3286 /* Uhhuh. Read error. Complain and punt. */
3287 if (!buffer_uptodate(bh))
3291 if (ext4_should_journal_data(inode)) {
3292 BUFFER_TRACE(bh, "get write access");
3293 err = ext4_journal_get_write_access(handle, bh);
3298 zero_user(page, offset, length);
3300 BUFFER_TRACE(bh, "zeroed end of block");
3303 if (ext4_should_journal_data(inode)) {
3304 err = ext4_journal_dirty_metadata(handle, bh);
3306 if (ext4_should_order_data(inode))
3307 err = ext4_jbd2_file_inode(handle, inode);
3308 mark_buffer_dirty(bh);
3313 page_cache_release(page);
3318 * Probably it should be a library function... search for first non-zero word
3319 * or memcmp with zero_page, whatever is better for particular architecture.
3322 static inline int all_zeroes(__le32 *p, __le32 *q)
3331 * ext4_find_shared - find the indirect blocks for partial truncation.
3332 * @inode: inode in question
3333 * @depth: depth of the affected branch
3334 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3335 * @chain: place to store the pointers to partial indirect blocks
3336 * @top: place to the (detached) top of branch
3338 * This is a helper function used by ext4_truncate().
3340 * When we do truncate() we may have to clean the ends of several
3341 * indirect blocks but leave the blocks themselves alive. Block is
3342 * partially truncated if some data below the new i_size is refered
3343 * from it (and it is on the path to the first completely truncated
3344 * data block, indeed). We have to free the top of that path along
3345 * with everything to the right of the path. Since no allocation
3346 * past the truncation point is possible until ext4_truncate()
3347 * finishes, we may safely do the latter, but top of branch may
3348 * require special attention - pageout below the truncation point
3349 * might try to populate it.
3351 * We atomically detach the top of branch from the tree, store the
3352 * block number of its root in *@top, pointers to buffer_heads of
3353 * partially truncated blocks - in @chain[].bh and pointers to
3354 * their last elements that should not be removed - in
3355 * @chain[].p. Return value is the pointer to last filled element
3358 * The work left to caller to do the actual freeing of subtrees:
3359 * a) free the subtree starting from *@top
3360 * b) free the subtrees whose roots are stored in
3361 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3362 * c) free the subtrees growing from the inode past the @chain[0].
3363 * (no partially truncated stuff there). */
3365 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3366 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3368 Indirect *partial, *p;
3372 /* Make k index the deepest non-null offest + 1 */
3373 for (k = depth; k > 1 && !offsets[k-1]; k--)
3375 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3376 /* Writer: pointers */
3378 partial = chain + k-1;
3380 * If the branch acquired continuation since we've looked at it -
3381 * fine, it should all survive and (new) top doesn't belong to us.
3383 if (!partial->key && *partial->p)
3386 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
3389 * OK, we've found the last block that must survive. The rest of our
3390 * branch should be detached before unlocking. However, if that rest
3391 * of branch is all ours and does not grow immediately from the inode
3392 * it's easier to cheat and just decrement partial->p.
3394 if (p == chain + k - 1 && p > chain) {
3398 /* Nope, don't do this in ext4. Must leave the tree intact */
3405 while(partial > p) {
3406 brelse(partial->bh);
3414 * Zero a number of block pointers in either an inode or an indirect block.
3415 * If we restart the transaction we must again get write access to the
3416 * indirect block for further modification.
3418 * We release `count' blocks on disk, but (last - first) may be greater
3419 * than `count' because there can be holes in there.
3421 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3422 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3423 unsigned long count, __le32 *first, __le32 *last)
3426 if (try_to_extend_transaction(handle, inode)) {
3428 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3429 ext4_journal_dirty_metadata(handle, bh);
3431 ext4_mark_inode_dirty(handle, inode);
3432 ext4_journal_test_restart(handle, inode);
3434 BUFFER_TRACE(bh, "retaking write access");
3435 ext4_journal_get_write_access(handle, bh);
3440 * Any buffers which are on the journal will be in memory. We find
3441 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3442 * on them. We've already detached each block from the file, so
3443 * bforget() in jbd2_journal_forget() should be safe.
3445 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3447 for (p = first; p < last; p++) {
3448 u32 nr = le32_to_cpu(*p);
3450 struct buffer_head *tbh;
3453 tbh = sb_find_get_block(inode->i_sb, nr);
3454 ext4_forget(handle, 0, inode, tbh, nr);
3458 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3462 * ext4_free_data - free a list of data blocks
3463 * @handle: handle for this transaction
3464 * @inode: inode we are dealing with
3465 * @this_bh: indirect buffer_head which contains *@first and *@last
3466 * @first: array of block numbers
3467 * @last: points immediately past the end of array
3469 * We are freeing all blocks refered from that array (numbers are stored as
3470 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3472 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3473 * blocks are contiguous then releasing them at one time will only affect one
3474 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3475 * actually use a lot of journal space.
3477 * @this_bh will be %NULL if @first and @last point into the inode's direct
3480 static void ext4_free_data(handle_t *handle, struct inode *inode,
3481 struct buffer_head *this_bh,
3482 __le32 *first, __le32 *last)
3484 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3485 unsigned long count = 0; /* Number of blocks in the run */
3486 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3489 ext4_fsblk_t nr; /* Current block # */
3490 __le32 *p; /* Pointer into inode/ind
3491 for current block */
3494 if (this_bh) { /* For indirect block */
3495 BUFFER_TRACE(this_bh, "get_write_access");
3496 err = ext4_journal_get_write_access(handle, this_bh);
3497 /* Important: if we can't update the indirect pointers
3498 * to the blocks, we can't free them. */
3503 for (p = first; p < last; p++) {
3504 nr = le32_to_cpu(*p);
3506 /* accumulate blocks to free if they're contiguous */
3509 block_to_free_p = p;
3511 } else if (nr == block_to_free + count) {
3514 ext4_clear_blocks(handle, inode, this_bh,
3516 count, block_to_free_p, p);
3518 block_to_free_p = p;
3525 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3526 count, block_to_free_p, p);
3529 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3532 * The buffer head should have an attached journal head at this
3533 * point. However, if the data is corrupted and an indirect
3534 * block pointed to itself, it would have been detached when
3535 * the block was cleared. Check for this instead of OOPSing.
3538 ext4_journal_dirty_metadata(handle, this_bh);
3540 ext4_error(inode->i_sb, __func__,
3541 "circular indirect block detected, "
3542 "inode=%lu, block=%llu",
3544 (unsigned long long) this_bh->b_blocknr);
3549 * ext4_free_branches - free an array of branches
3550 * @handle: JBD handle for this transaction
3551 * @inode: inode we are dealing with
3552 * @parent_bh: the buffer_head which contains *@first and *@last
3553 * @first: array of block numbers
3554 * @last: pointer immediately past the end of array
3555 * @depth: depth of the branches to free
3557 * We are freeing all blocks refered from these branches (numbers are
3558 * stored as little-endian 32-bit) and updating @inode->i_blocks
3561 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3562 struct buffer_head *parent_bh,
3563 __le32 *first, __le32 *last, int depth)
3568 if (is_handle_aborted(handle))
3572 struct buffer_head *bh;
3573 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3575 while (--p >= first) {
3576 nr = le32_to_cpu(*p);
3578 continue; /* A hole */
3580 /* Go read the buffer for the next level down */
3581 bh = sb_bread(inode->i_sb, nr);
3584 * A read failure? Report error and clear slot
3588 ext4_error(inode->i_sb, "ext4_free_branches",
3589 "Read failure, inode=%lu, block=%llu",
3594 /* This zaps the entire block. Bottom up. */
3595 BUFFER_TRACE(bh, "free child branches");
3596 ext4_free_branches(handle, inode, bh,
3597 (__le32*)bh->b_data,
3598 (__le32*)bh->b_data + addr_per_block,
3602 * We've probably journalled the indirect block several
3603 * times during the truncate. But it's no longer
3604 * needed and we now drop it from the transaction via
3605 * jbd2_journal_revoke().
3607 * That's easy if it's exclusively part of this
3608 * transaction. But if it's part of the committing
3609 * transaction then jbd2_journal_forget() will simply
3610 * brelse() it. That means that if the underlying
3611 * block is reallocated in ext4_get_block(),
3612 * unmap_underlying_metadata() will find this block
3613 * and will try to get rid of it. damn, damn.
3615 * If this block has already been committed to the
3616 * journal, a revoke record will be written. And
3617 * revoke records must be emitted *before* clearing
3618 * this block's bit in the bitmaps.
3620 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3623 * Everything below this this pointer has been
3624 * released. Now let this top-of-subtree go.
3626 * We want the freeing of this indirect block to be
3627 * atomic in the journal with the updating of the
3628 * bitmap block which owns it. So make some room in
3631 * We zero the parent pointer *after* freeing its
3632 * pointee in the bitmaps, so if extend_transaction()
3633 * for some reason fails to put the bitmap changes and
3634 * the release into the same transaction, recovery
3635 * will merely complain about releasing a free block,
3636 * rather than leaking blocks.
3638 if (is_handle_aborted(handle))
3640 if (try_to_extend_transaction(handle, inode)) {
3641 ext4_mark_inode_dirty(handle, inode);
3642 ext4_journal_test_restart(handle, inode);
3645 ext4_free_blocks(handle, inode, nr, 1, 1);
3649 * The block which we have just freed is
3650 * pointed to by an indirect block: journal it
3652 BUFFER_TRACE(parent_bh, "get_write_access");
3653 if (!ext4_journal_get_write_access(handle,
3656 BUFFER_TRACE(parent_bh,
3657 "call ext4_journal_dirty_metadata");
3658 ext4_journal_dirty_metadata(handle,
3664 /* We have reached the bottom of the tree. */
3665 BUFFER_TRACE(parent_bh, "free data blocks");
3666 ext4_free_data(handle, inode, parent_bh, first, last);
3670 int ext4_can_truncate(struct inode *inode)
3672 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3674 if (S_ISREG(inode->i_mode))
3676 if (S_ISDIR(inode->i_mode))
3678 if (S_ISLNK(inode->i_mode))
3679 return !ext4_inode_is_fast_symlink(inode);
3686 * We block out ext4_get_block() block instantiations across the entire
3687 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3688 * simultaneously on behalf of the same inode.
3690 * As we work through the truncate and commmit bits of it to the journal there
3691 * is one core, guiding principle: the file's tree must always be consistent on
3692 * disk. We must be able to restart the truncate after a crash.
3694 * The file's tree may be transiently inconsistent in memory (although it
3695 * probably isn't), but whenever we close off and commit a journal transaction,
3696 * the contents of (the filesystem + the journal) must be consistent and
3697 * restartable. It's pretty simple, really: bottom up, right to left (although
3698 * left-to-right works OK too).
3700 * Note that at recovery time, journal replay occurs *before* the restart of
3701 * truncate against the orphan inode list.
3703 * The committed inode has the new, desired i_size (which is the same as
3704 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3705 * that this inode's truncate did not complete and it will again call
3706 * ext4_truncate() to have another go. So there will be instantiated blocks
3707 * to the right of the truncation point in a crashed ext4 filesystem. But
3708 * that's fine - as long as they are linked from the inode, the post-crash
3709 * ext4_truncate() run will find them and release them.
3711 void ext4_truncate(struct inode *inode)
3714 struct ext4_inode_info *ei = EXT4_I(inode);
3715 __le32 *i_data = ei->i_data;
3716 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3717 struct address_space *mapping = inode->i_mapping;
3718 ext4_lblk_t offsets[4];
3723 ext4_lblk_t last_block;
3724 unsigned blocksize = inode->i_sb->s_blocksize;
3726 if (!ext4_can_truncate(inode))
3729 if (inode->i_size == 0)
3730 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3732 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3733 ext4_ext_truncate(inode);
3737 handle = start_transaction(inode);
3739 return; /* AKPM: return what? */
3741 last_block = (inode->i_size + blocksize-1)
3742 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3744 if (inode->i_size & (blocksize - 1))
3745 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3748 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3750 goto out_stop; /* error */
3753 * OK. This truncate is going to happen. We add the inode to the
3754 * orphan list, so that if this truncate spans multiple transactions,
3755 * and we crash, we will resume the truncate when the filesystem
3756 * recovers. It also marks the inode dirty, to catch the new size.
3758 * Implication: the file must always be in a sane, consistent
3759 * truncatable state while each transaction commits.
3761 if (ext4_orphan_add(handle, inode))
3765 * From here we block out all ext4_get_block() callers who want to
3766 * modify the block allocation tree.
3768 down_write(&ei->i_data_sem);
3770 ext4_discard_reservation(inode);
3773 * The orphan list entry will now protect us from any crash which
3774 * occurs before the truncate completes, so it is now safe to propagate
3775 * the new, shorter inode size (held for now in i_size) into the
3776 * on-disk inode. We do this via i_disksize, which is the value which
3777 * ext4 *really* writes onto the disk inode.
3779 ei->i_disksize = inode->i_size;
3781 if (n == 1) { /* direct blocks */
3782 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3783 i_data + EXT4_NDIR_BLOCKS);
3787 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3788 /* Kill the top of shared branch (not detached) */
3790 if (partial == chain) {
3791 /* Shared branch grows from the inode */
3792 ext4_free_branches(handle, inode, NULL,
3793 &nr, &nr+1, (chain+n-1) - partial);
3796 * We mark the inode dirty prior to restart,
3797 * and prior to stop. No need for it here.
3800 /* Shared branch grows from an indirect block */
3801 BUFFER_TRACE(partial->bh, "get_write_access");
3802 ext4_free_branches(handle, inode, partial->bh,
3804 partial->p+1, (chain+n-1) - partial);
3807 /* Clear the ends of indirect blocks on the shared branch */
3808 while (partial > chain) {
3809 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3810 (__le32*)partial->bh->b_data+addr_per_block,
3811 (chain+n-1) - partial);
3812 BUFFER_TRACE(partial->bh, "call brelse");
3813 brelse (partial->bh);
3817 /* Kill the remaining (whole) subtrees */
3818 switch (offsets[0]) {
3820 nr = i_data[EXT4_IND_BLOCK];
3822 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3823 i_data[EXT4_IND_BLOCK] = 0;
3825 case EXT4_IND_BLOCK:
3826 nr = i_data[EXT4_DIND_BLOCK];
3828 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3829 i_data[EXT4_DIND_BLOCK] = 0;
3831 case EXT4_DIND_BLOCK:
3832 nr = i_data[EXT4_TIND_BLOCK];
3834 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3835 i_data[EXT4_TIND_BLOCK] = 0;
3837 case EXT4_TIND_BLOCK:
3841 up_write(&ei->i_data_sem);
3842 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3843 ext4_mark_inode_dirty(handle, inode);
3846 * In a multi-transaction truncate, we only make the final transaction
3853 * If this was a simple ftruncate(), and the file will remain alive
3854 * then we need to clear up the orphan record which we created above.
3855 * However, if this was a real unlink then we were called by
3856 * ext4_delete_inode(), and we allow that function to clean up the
3857 * orphan info for us.
3860 ext4_orphan_del(handle, inode);
3862 ext4_journal_stop(handle);
3865 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3866 unsigned long ino, struct ext4_iloc *iloc)
3868 ext4_group_t block_group;
3869 unsigned long offset;
3871 struct ext4_group_desc *gdp;
3873 if (!ext4_valid_inum(sb, ino)) {
3875 * This error is already checked for in namei.c unless we are
3876 * looking at an NFS filehandle, in which case no error
3882 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3883 gdp = ext4_get_group_desc(sb, block_group, NULL);
3888 * Figure out the offset within the block group inode table
3890 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3891 EXT4_INODE_SIZE(sb);
3892 block = ext4_inode_table(sb, gdp) +
3893 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3895 iloc->block_group = block_group;
3896 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3901 * ext4_get_inode_loc returns with an extra refcount against the inode's
3902 * underlying buffer_head on success. If 'in_mem' is true, we have all
3903 * data in memory that is needed to recreate the on-disk version of this
3906 static int __ext4_get_inode_loc(struct inode *inode,
3907 struct ext4_iloc *iloc, int in_mem)
3910 struct buffer_head *bh;
3912 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3916 bh = sb_getblk(inode->i_sb, block);
3918 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3919 "unable to read inode block - "
3920 "inode=%lu, block=%llu",
3921 inode->i_ino, block);
3924 if (!buffer_uptodate(bh)) {
3928 * If the buffer has the write error flag, we have failed
3929 * to write out another inode in the same block. In this
3930 * case, we don't have to read the block because we may
3931 * read the old inode data successfully.
3933 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3934 set_buffer_uptodate(bh);
3936 if (buffer_uptodate(bh)) {
3937 /* someone brought it uptodate while we waited */
3943 * If we have all information of the inode in memory and this
3944 * is the only valid inode in the block, we need not read the
3948 struct buffer_head *bitmap_bh;
3949 struct ext4_group_desc *desc;
3950 int inodes_per_buffer;
3951 int inode_offset, i;
3952 ext4_group_t block_group;
3955 block_group = (inode->i_ino - 1) /
3956 EXT4_INODES_PER_GROUP(inode->i_sb);
3957 inodes_per_buffer = bh->b_size /
3958 EXT4_INODE_SIZE(inode->i_sb);
3959 inode_offset = ((inode->i_ino - 1) %
3960 EXT4_INODES_PER_GROUP(inode->i_sb));
3961 start = inode_offset & ~(inodes_per_buffer - 1);
3963 /* Is the inode bitmap in cache? */
3964 desc = ext4_get_group_desc(inode->i_sb,
3969 bitmap_bh = sb_getblk(inode->i_sb,
3970 ext4_inode_bitmap(inode->i_sb, desc));
3975 * If the inode bitmap isn't in cache then the
3976 * optimisation may end up performing two reads instead
3977 * of one, so skip it.
3979 if (!buffer_uptodate(bitmap_bh)) {
3983 for (i = start; i < start + inodes_per_buffer; i++) {
3984 if (i == inode_offset)
3986 if (ext4_test_bit(i, bitmap_bh->b_data))
3990 if (i == start + inodes_per_buffer) {
3991 /* all other inodes are free, so skip I/O */
3992 memset(bh->b_data, 0, bh->b_size);
3993 set_buffer_uptodate(bh);
4001 * There are other valid inodes in the buffer, this inode
4002 * has in-inode xattrs, or we don't have this inode in memory.
4003 * Read the block from disk.
4006 bh->b_end_io = end_buffer_read_sync;
4007 submit_bh(READ_META, bh);
4009 if (!buffer_uptodate(bh)) {
4010 ext4_error(inode->i_sb, "ext4_get_inode_loc",
4011 "unable to read inode block - "
4012 "inode=%lu, block=%llu",
4013 inode->i_ino, block);
4023 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4025 /* We have all inode data except xattrs in memory here. */
4026 return __ext4_get_inode_loc(inode, iloc,
4027 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4030 void ext4_set_inode_flags(struct inode *inode)
4032 unsigned int flags = EXT4_I(inode)->i_flags;
4034 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4035 if (flags & EXT4_SYNC_FL)
4036 inode->i_flags |= S_SYNC;
4037 if (flags & EXT4_APPEND_FL)
4038 inode->i_flags |= S_APPEND;
4039 if (flags & EXT4_IMMUTABLE_FL)
4040 inode->i_flags |= S_IMMUTABLE;
4041 if (flags & EXT4_NOATIME_FL)
4042 inode->i_flags |= S_NOATIME;
4043 if (flags & EXT4_DIRSYNC_FL)
4044 inode->i_flags |= S_DIRSYNC;
4047 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4048 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4050 unsigned int flags = ei->vfs_inode.i_flags;
4052 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4053 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4055 ei->i_flags |= EXT4_SYNC_FL;
4056 if (flags & S_APPEND)
4057 ei->i_flags |= EXT4_APPEND_FL;
4058 if (flags & S_IMMUTABLE)
4059 ei->i_flags |= EXT4_IMMUTABLE_FL;
4060 if (flags & S_NOATIME)
4061 ei->i_flags |= EXT4_NOATIME_FL;
4062 if (flags & S_DIRSYNC)
4063 ei->i_flags |= EXT4_DIRSYNC_FL;
4065 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4066 struct ext4_inode_info *ei)
4069 struct inode *inode = &(ei->vfs_inode);
4070 struct super_block *sb = inode->i_sb;
4072 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4073 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4074 /* we are using combined 48 bit field */
4075 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4076 le32_to_cpu(raw_inode->i_blocks_lo);
4077 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4078 /* i_blocks represent file system block size */
4079 return i_blocks << (inode->i_blkbits - 9);
4084 return le32_to_cpu(raw_inode->i_blocks_lo);
4088 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4090 struct ext4_iloc iloc;
4091 struct ext4_inode *raw_inode;
4092 struct ext4_inode_info *ei;
4093 struct buffer_head *bh;
4094 struct inode *inode;
4098 inode = iget_locked(sb, ino);
4100 return ERR_PTR(-ENOMEM);
4101 if (!(inode->i_state & I_NEW))
4105 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
4106 ei->i_acl = EXT4_ACL_NOT_CACHED;
4107 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4109 ei->i_block_alloc_info = NULL;
4111 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4115 raw_inode = ext4_raw_inode(&iloc);
4116 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4117 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4118 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4119 if(!(test_opt (inode->i_sb, NO_UID32))) {
4120 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4121 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4123 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4126 ei->i_dir_start_lookup = 0;
4127 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4128 /* We now have enough fields to check if the inode was active or not.
4129 * This is needed because nfsd might try to access dead inodes
4130 * the test is that same one that e2fsck uses
4131 * NeilBrown 1999oct15
4133 if (inode->i_nlink == 0) {
4134 if (inode->i_mode == 0 ||
4135 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4136 /* this inode is deleted */
4141 /* The only unlinked inodes we let through here have
4142 * valid i_mode and are being read by the orphan
4143 * recovery code: that's fine, we're about to complete
4144 * the process of deleting those. */
4146 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4147 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4148 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4149 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4151 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4152 inode->i_size = ext4_isize(raw_inode);
4153 ei->i_disksize = inode->i_size;
4154 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4155 ei->i_block_group = iloc.block_group;
4157 * NOTE! The in-memory inode i_data array is in little-endian order
4158 * even on big-endian machines: we do NOT byteswap the block numbers!
4160 for (block = 0; block < EXT4_N_BLOCKS; block++)
4161 ei->i_data[block] = raw_inode->i_block[block];
4162 INIT_LIST_HEAD(&ei->i_orphan);
4164 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4165 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4166 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4167 EXT4_INODE_SIZE(inode->i_sb)) {
4172 if (ei->i_extra_isize == 0) {
4173 /* The extra space is currently unused. Use it. */
4174 ei->i_extra_isize = sizeof(struct ext4_inode) -
4175 EXT4_GOOD_OLD_INODE_SIZE;
4177 __le32 *magic = (void *)raw_inode +
4178 EXT4_GOOD_OLD_INODE_SIZE +
4180 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4181 ei->i_state |= EXT4_STATE_XATTR;
4184 ei->i_extra_isize = 0;
4186 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4187 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4188 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4189 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4191 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4192 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4193 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4195 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4198 if (ei->i_file_acl &&
4200 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4201 EXT4_SB(sb)->s_gdb_count)) ||
4202 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4203 ext4_error(sb, __func__,
4204 "bad extended attribute block %llu in inode #%lu",
4205 ei->i_file_acl, inode->i_ino);
4210 if (S_ISREG(inode->i_mode)) {
4211 inode->i_op = &ext4_file_inode_operations;
4212 inode->i_fop = &ext4_file_operations;
4213 ext4_set_aops(inode);
4214 } else if (S_ISDIR(inode->i_mode)) {
4215 inode->i_op = &ext4_dir_inode_operations;
4216 inode->i_fop = &ext4_dir_operations;
4217 } else if (S_ISLNK(inode->i_mode)) {
4218 if (ext4_inode_is_fast_symlink(inode))
4219 inode->i_op = &ext4_fast_symlink_inode_operations;
4221 inode->i_op = &ext4_symlink_inode_operations;
4222 ext4_set_aops(inode);
4224 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4225 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4226 inode->i_op = &ext4_special_inode_operations;
4227 if (raw_inode->i_block[0])
4228 init_special_inode(inode, inode->i_mode,
4229 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4231 init_special_inode(inode, inode->i_mode,
4232 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4236 ext4_error(inode->i_sb, __func__,
4237 "bogus i_mode (%o) for inode=%lu",
4238 inode->i_mode, inode->i_ino);
4242 ext4_set_inode_flags(inode);
4243 unlock_new_inode(inode);
4248 return ERR_PTR(ret);
4251 static int ext4_inode_blocks_set(handle_t *handle,
4252 struct ext4_inode *raw_inode,
4253 struct ext4_inode_info *ei)
4255 struct inode *inode = &(ei->vfs_inode);
4256 u64 i_blocks = inode->i_blocks;
4257 struct super_block *sb = inode->i_sb;
4260 if (i_blocks <= ~0U) {
4262 * i_blocks can be represnted in a 32 bit variable
4263 * as multiple of 512 bytes
4265 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4266 raw_inode->i_blocks_high = 0;
4267 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4268 } else if (i_blocks <= 0xffffffffffffULL) {
4270 * i_blocks can be represented in a 48 bit variable
4271 * as multiple of 512 bytes
4273 err = ext4_update_rocompat_feature(handle, sb,
4274 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4277 /* i_block is stored in the split 48 bit fields */
4278 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4279 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4280 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4283 * i_blocks should be represented in a 48 bit variable
4284 * as multiple of file system block size
4286 err = ext4_update_rocompat_feature(handle, sb,
4287 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4290 ei->i_flags |= EXT4_HUGE_FILE_FL;
4291 /* i_block is stored in file system block size */
4292 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4293 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4294 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4301 * Post the struct inode info into an on-disk inode location in the
4302 * buffer-cache. This gobbles the caller's reference to the
4303 * buffer_head in the inode location struct.
4305 * The caller must have write access to iloc->bh.
4307 static int ext4_do_update_inode(handle_t *handle,
4308 struct inode *inode,
4309 struct ext4_iloc *iloc)
4311 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4312 struct ext4_inode_info *ei = EXT4_I(inode);
4313 struct buffer_head *bh = iloc->bh;
4314 int err = 0, rc, block;
4316 /* For fields not not tracking in the in-memory inode,
4317 * initialise them to zero for new inodes. */
4318 if (ei->i_state & EXT4_STATE_NEW)
4319 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4321 ext4_get_inode_flags(ei);
4322 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4323 if(!(test_opt(inode->i_sb, NO_UID32))) {
4324 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4325 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4327 * Fix up interoperability with old kernels. Otherwise, old inodes get
4328 * re-used with the upper 16 bits of the uid/gid intact
4331 raw_inode->i_uid_high =
4332 cpu_to_le16(high_16_bits(inode->i_uid));
4333 raw_inode->i_gid_high =
4334 cpu_to_le16(high_16_bits(inode->i_gid));
4336 raw_inode->i_uid_high = 0;
4337 raw_inode->i_gid_high = 0;
4340 raw_inode->i_uid_low =
4341 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4342 raw_inode->i_gid_low =
4343 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4344 raw_inode->i_uid_high = 0;
4345 raw_inode->i_gid_high = 0;
4347 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4349 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4350 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4351 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4352 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4354 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4356 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4357 /* clear the migrate flag in the raw_inode */
4358 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4359 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4360 cpu_to_le32(EXT4_OS_HURD))
4361 raw_inode->i_file_acl_high =
4362 cpu_to_le16(ei->i_file_acl >> 32);
4363 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4364 ext4_isize_set(raw_inode, ei->i_disksize);
4365 if (ei->i_disksize > 0x7fffffffULL) {
4366 struct super_block *sb = inode->i_sb;
4367 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4368 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4369 EXT4_SB(sb)->s_es->s_rev_level ==
4370 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4371 /* If this is the first large file
4372 * created, add a flag to the superblock.
4374 err = ext4_journal_get_write_access(handle,
4375 EXT4_SB(sb)->s_sbh);
4378 ext4_update_dynamic_rev(sb);
4379 EXT4_SET_RO_COMPAT_FEATURE(sb,
4380 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4383 err = ext4_journal_dirty_metadata(handle,
4384 EXT4_SB(sb)->s_sbh);
4387 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4388 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4389 if (old_valid_dev(inode->i_rdev)) {
4390 raw_inode->i_block[0] =
4391 cpu_to_le32(old_encode_dev(inode->i_rdev));
4392 raw_inode->i_block[1] = 0;
4394 raw_inode->i_block[0] = 0;
4395 raw_inode->i_block[1] =
4396 cpu_to_le32(new_encode_dev(inode->i_rdev));
4397 raw_inode->i_block[2] = 0;
4399 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4400 raw_inode->i_block[block] = ei->i_data[block];
4402 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4403 if (ei->i_extra_isize) {
4404 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4405 raw_inode->i_version_hi =
4406 cpu_to_le32(inode->i_version >> 32);
4407 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4411 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4412 rc = ext4_journal_dirty_metadata(handle, bh);
4415 ei->i_state &= ~EXT4_STATE_NEW;
4419 ext4_std_error(inode->i_sb, err);
4424 * ext4_write_inode()
4426 * We are called from a few places:
4428 * - Within generic_file_write() for O_SYNC files.
4429 * Here, there will be no transaction running. We wait for any running
4430 * trasnaction to commit.
4432 * - Within sys_sync(), kupdate and such.
4433 * We wait on commit, if tol to.
4435 * - Within prune_icache() (PF_MEMALLOC == true)
4436 * Here we simply return. We can't afford to block kswapd on the
4439 * In all cases it is actually safe for us to return without doing anything,
4440 * because the inode has been copied into a raw inode buffer in
4441 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4444 * Note that we are absolutely dependent upon all inode dirtiers doing the
4445 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4446 * which we are interested.
4448 * It would be a bug for them to not do this. The code:
4450 * mark_inode_dirty(inode)
4452 * inode->i_size = expr;
4454 * is in error because a kswapd-driven write_inode() could occur while
4455 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4456 * will no longer be on the superblock's dirty inode list.
4458 int ext4_write_inode(struct inode *inode, int wait)
4460 if (current->flags & PF_MEMALLOC)
4463 if (ext4_journal_current_handle()) {
4464 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4472 return ext4_force_commit(inode->i_sb);
4478 * Called from notify_change.
4480 * We want to trap VFS attempts to truncate the file as soon as
4481 * possible. In particular, we want to make sure that when the VFS
4482 * shrinks i_size, we put the inode on the orphan list and modify
4483 * i_disksize immediately, so that during the subsequent flushing of
4484 * dirty pages and freeing of disk blocks, we can guarantee that any
4485 * commit will leave the blocks being flushed in an unused state on
4486 * disk. (On recovery, the inode will get truncated and the blocks will
4487 * be freed, so we have a strong guarantee that no future commit will
4488 * leave these blocks visible to the user.)
4490 * Another thing we have to assure is that if we are in ordered mode
4491 * and inode is still attached to the committing transaction, we must
4492 * we start writeout of all the dirty pages which are being truncated.
4493 * This way we are sure that all the data written in the previous
4494 * transaction are already on disk (truncate waits for pages under
4497 * Called with inode->i_mutex down.
4499 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4501 struct inode *inode = dentry->d_inode;
4503 const unsigned int ia_valid = attr->ia_valid;
4505 error = inode_change_ok(inode, attr);
4509 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4510 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4513 /* (user+group)*(old+new) structure, inode write (sb,
4514 * inode block, ? - but truncate inode update has it) */
4515 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4516 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4517 if (IS_ERR(handle)) {
4518 error = PTR_ERR(handle);
4521 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4523 ext4_journal_stop(handle);
4526 /* Update corresponding info in inode so that everything is in
4527 * one transaction */
4528 if (attr->ia_valid & ATTR_UID)
4529 inode->i_uid = attr->ia_uid;
4530 if (attr->ia_valid & ATTR_GID)
4531 inode->i_gid = attr->ia_gid;
4532 error = ext4_mark_inode_dirty(handle, inode);
4533 ext4_journal_stop(handle);
4536 if (attr->ia_valid & ATTR_SIZE) {
4537 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4538 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4540 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4547 if (S_ISREG(inode->i_mode) &&
4548 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4551 handle = ext4_journal_start(inode, 3);
4552 if (IS_ERR(handle)) {
4553 error = PTR_ERR(handle);
4557 error = ext4_orphan_add(handle, inode);
4558 EXT4_I(inode)->i_disksize = attr->ia_size;
4559 rc = ext4_mark_inode_dirty(handle, inode);
4562 ext4_journal_stop(handle);
4564 if (ext4_should_order_data(inode)) {
4565 error = ext4_begin_ordered_truncate(inode,
4568 /* Do as much error cleanup as possible */
4569 handle = ext4_journal_start(inode, 3);
4570 if (IS_ERR(handle)) {
4571 ext4_orphan_del(NULL, inode);
4574 ext4_orphan_del(handle, inode);
4575 ext4_journal_stop(handle);
4581 rc = inode_setattr(inode, attr);
4583 /* If inode_setattr's call to ext4_truncate failed to get a
4584 * transaction handle at all, we need to clean up the in-core
4585 * orphan list manually. */
4587 ext4_orphan_del(NULL, inode);
4589 if (!rc && (ia_valid & ATTR_MODE))
4590 rc = ext4_acl_chmod(inode);
4593 ext4_std_error(inode->i_sb, error);
4599 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4602 struct inode *inode;
4603 unsigned long delalloc_blocks;
4605 inode = dentry->d_inode;
4606 generic_fillattr(inode, stat);
4609 * We can't update i_blocks if the block allocation is delayed
4610 * otherwise in the case of system crash before the real block
4611 * allocation is done, we will have i_blocks inconsistent with
4612 * on-disk file blocks.
4613 * We always keep i_blocks updated together with real
4614 * allocation. But to not confuse with user, stat
4615 * will return the blocks that include the delayed allocation
4616 * blocks for this file.
4618 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4619 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4620 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4622 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4626 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4631 /* if nrblocks are contiguous */
4634 * With N contiguous data blocks, it need at most
4635 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4636 * 2 dindirect blocks
4639 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4640 return indirects + 3;
4643 * if nrblocks are not contiguous, worse case, each block touch
4644 * a indirect block, and each indirect block touch a double indirect
4645 * block, plus a triple indirect block
4647 indirects = nrblocks * 2 + 1;
4651 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4653 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4654 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4655 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4659 * Account for index blocks, block groups bitmaps and block group
4660 * descriptor blocks if modify datablocks and index blocks
4661 * worse case, the indexs blocks spread over different block groups
4663 * If datablocks are discontiguous, they are possible to spread over
4664 * different block groups too. If they are contiugous, with flexbg,
4665 * they could still across block group boundary.
4667 * Also account for superblock, inode, quota and xattr blocks
4669 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4671 int groups, gdpblocks;
4676 * How many index blocks need to touch to modify nrblocks?
4677 * The "Chunk" flag indicating whether the nrblocks is
4678 * physically contiguous on disk
4680 * For Direct IO and fallocate, they calls get_block to allocate
4681 * one single extent at a time, so they could set the "Chunk" flag
4683 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4688 * Now let's see how many group bitmaps and group descriptors need
4698 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4699 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4700 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4701 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4703 /* bitmaps and block group descriptor blocks */
4704 ret += groups + gdpblocks;
4706 /* Blocks for super block, inode, quota and xattr blocks */
4707 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4713 * Calulate the total number of credits to reserve to fit
4714 * the modification of a single pages into a single transaction,
4715 * which may include multiple chunks of block allocations.
4717 * This could be called via ext4_write_begin()
4719 * We need to consider the worse case, when
4720 * one new block per extent.
4722 int ext4_writepage_trans_blocks(struct inode *inode)
4724 int bpp = ext4_journal_blocks_per_page(inode);
4727 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4729 /* Account for data blocks for journalled mode */
4730 if (ext4_should_journal_data(inode))
4736 * Calculate the journal credits for a chunk of data modification.
4738 * This is called from DIO, fallocate or whoever calling
4739 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4741 * journal buffers for data blocks are not included here, as DIO
4742 * and fallocate do no need to journal data buffers.
4744 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4746 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4750 * The caller must have previously called ext4_reserve_inode_write().
4751 * Give this, we know that the caller already has write access to iloc->bh.
4753 int ext4_mark_iloc_dirty(handle_t *handle,
4754 struct inode *inode, struct ext4_iloc *iloc)
4758 if (test_opt(inode->i_sb, I_VERSION))
4759 inode_inc_iversion(inode);
4761 /* the do_update_inode consumes one bh->b_count */
4764 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4765 err = ext4_do_update_inode(handle, inode, iloc);
4771 * On success, We end up with an outstanding reference count against
4772 * iloc->bh. This _must_ be cleaned up later.
4776 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4777 struct ext4_iloc *iloc)
4781 err = ext4_get_inode_loc(inode, iloc);
4783 BUFFER_TRACE(iloc->bh, "get_write_access");
4784 err = ext4_journal_get_write_access(handle, iloc->bh);
4791 ext4_std_error(inode->i_sb, err);
4796 * Expand an inode by new_extra_isize bytes.
4797 * Returns 0 on success or negative error number on failure.
4799 static int ext4_expand_extra_isize(struct inode *inode,
4800 unsigned int new_extra_isize,
4801 struct ext4_iloc iloc,
4804 struct ext4_inode *raw_inode;
4805 struct ext4_xattr_ibody_header *header;
4806 struct ext4_xattr_entry *entry;
4808 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4811 raw_inode = ext4_raw_inode(&iloc);
4813 header = IHDR(inode, raw_inode);
4814 entry = IFIRST(header);
4816 /* No extended attributes present */
4817 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4818 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4819 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4821 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4825 /* try to expand with EAs present */
4826 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4831 * What we do here is to mark the in-core inode as clean with respect to inode
4832 * dirtiness (it may still be data-dirty).
4833 * This means that the in-core inode may be reaped by prune_icache
4834 * without having to perform any I/O. This is a very good thing,
4835 * because *any* task may call prune_icache - even ones which
4836 * have a transaction open against a different journal.
4838 * Is this cheating? Not really. Sure, we haven't written the
4839 * inode out, but prune_icache isn't a user-visible syncing function.
4840 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4841 * we start and wait on commits.
4843 * Is this efficient/effective? Well, we're being nice to the system
4844 * by cleaning up our inodes proactively so they can be reaped
4845 * without I/O. But we are potentially leaving up to five seconds'
4846 * worth of inodes floating about which prune_icache wants us to
4847 * write out. One way to fix that would be to get prune_icache()
4848 * to do a write_super() to free up some memory. It has the desired
4851 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4853 struct ext4_iloc iloc;
4854 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4855 static unsigned int mnt_count;
4859 err = ext4_reserve_inode_write(handle, inode, &iloc);
4860 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4861 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4863 * We need extra buffer credits since we may write into EA block
4864 * with this same handle. If journal_extend fails, then it will
4865 * only result in a minor loss of functionality for that inode.
4866 * If this is felt to be critical, then e2fsck should be run to
4867 * force a large enough s_min_extra_isize.
4869 if ((jbd2_journal_extend(handle,
4870 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4871 ret = ext4_expand_extra_isize(inode,
4872 sbi->s_want_extra_isize,
4875 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4877 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4878 ext4_warning(inode->i_sb, __func__,
4879 "Unable to expand inode %lu. Delete"
4880 " some EAs or run e2fsck.",
4883 le16_to_cpu(sbi->s_es->s_mnt_count);
4889 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4894 * ext4_dirty_inode() is called from __mark_inode_dirty()
4896 * We're really interested in the case where a file is being extended.
4897 * i_size has been changed by generic_commit_write() and we thus need
4898 * to include the updated inode in the current transaction.
4900 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4901 * are allocated to the file.
4903 * If the inode is marked synchronous, we don't honour that here - doing
4904 * so would cause a commit on atime updates, which we don't bother doing.
4905 * We handle synchronous inodes at the highest possible level.
4907 void ext4_dirty_inode(struct inode *inode)
4909 handle_t *current_handle = ext4_journal_current_handle();
4912 handle = ext4_journal_start(inode, 2);
4915 if (current_handle &&
4916 current_handle->h_transaction != handle->h_transaction) {
4917 /* This task has a transaction open against a different fs */
4918 printk(KERN_EMERG "%s: transactions do not match!\n",
4921 jbd_debug(5, "marking dirty. outer handle=%p\n",
4923 ext4_mark_inode_dirty(handle, inode);
4925 ext4_journal_stop(handle);
4932 * Bind an inode's backing buffer_head into this transaction, to prevent
4933 * it from being flushed to disk early. Unlike
4934 * ext4_reserve_inode_write, this leaves behind no bh reference and
4935 * returns no iloc structure, so the caller needs to repeat the iloc
4936 * lookup to mark the inode dirty later.
4938 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4940 struct ext4_iloc iloc;
4944 err = ext4_get_inode_loc(inode, &iloc);
4946 BUFFER_TRACE(iloc.bh, "get_write_access");
4947 err = jbd2_journal_get_write_access(handle, iloc.bh);
4949 err = ext4_journal_dirty_metadata(handle,
4954 ext4_std_error(inode->i_sb, err);
4959 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4966 * We have to be very careful here: changing a data block's
4967 * journaling status dynamically is dangerous. If we write a
4968 * data block to the journal, change the status and then delete
4969 * that block, we risk forgetting to revoke the old log record
4970 * from the journal and so a subsequent replay can corrupt data.
4971 * So, first we make sure that the journal is empty and that
4972 * nobody is changing anything.
4975 journal = EXT4_JOURNAL(inode);
4976 if (is_journal_aborted(journal))
4979 jbd2_journal_lock_updates(journal);
4980 jbd2_journal_flush(journal);
4983 * OK, there are no updates running now, and all cached data is
4984 * synced to disk. We are now in a completely consistent state
4985 * which doesn't have anything in the journal, and we know that
4986 * no filesystem updates are running, so it is safe to modify
4987 * the inode's in-core data-journaling state flag now.
4991 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4993 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4994 ext4_set_aops(inode);
4996 jbd2_journal_unlock_updates(journal);
4998 /* Finally we can mark the inode as dirty. */
5000 handle = ext4_journal_start(inode, 1);
5002 return PTR_ERR(handle);
5004 err = ext4_mark_inode_dirty(handle, inode);
5006 ext4_journal_stop(handle);
5007 ext4_std_error(inode->i_sb, err);
5012 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5014 return !buffer_mapped(bh);
5017 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5019 struct page *page = vmf->page;
5023 struct file *file = vma->vm_file;
5024 struct inode *inode = file->f_path.dentry->d_inode;
5025 struct address_space *mapping = inode->i_mapping;
5028 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5029 * get i_mutex because we are already holding mmap_sem.
5031 down_read(&inode->i_alloc_sem);
5032 size = i_size_read(inode);
5033 if (page->mapping != mapping || size <= page_offset(page)
5034 || !PageUptodate(page)) {
5035 /* page got truncated from under us? */
5039 if (PageMappedToDisk(page))
5042 if (page->index == size >> PAGE_CACHE_SHIFT)
5043 len = size & ~PAGE_CACHE_MASK;
5045 len = PAGE_CACHE_SIZE;
5047 if (page_has_buffers(page)) {
5048 /* return if we have all the buffers mapped */
5049 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5054 * OK, we need to fill the hole... Do write_begin write_end
5055 * to do block allocation/reservation.We are not holding
5056 * inode.i__mutex here. That allow * parallel write_begin,
5057 * write_end call. lock_page prevent this from happening
5058 * on the same page though
5060 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5061 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
5064 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5065 len, len, page, NULL);
5071 ret = VM_FAULT_SIGBUS;
5072 up_read(&inode->i_alloc_sem);