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/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode *inode,
50 return jbd2_journal_begin_ordered_truncate(
51 EXT4_SB(inode->i_sb)->s_journal,
52 &EXT4_I(inode)->jinode,
56 static void ext4_invalidatepage(struct page *page, unsigned long offset);
59 * Test whether an inode is a fast symlink.
61 static int ext4_inode_is_fast_symlink(struct inode *inode)
63 int ea_blocks = EXT4_I(inode)->i_file_acl ?
64 (inode->i_sb->s_blocksize >> 9) : 0;
66 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
70 * The ext4 forget function must perform a revoke if we are freeing data
71 * which has been journaled. Metadata (eg. indirect blocks) must be
72 * revoked in all cases.
74 * "bh" may be NULL: a metadata block may have been freed from memory
75 * but there may still be a record of it in the journal, and that record
76 * still needs to be revoked.
78 * If the handle isn't valid we're not journaling so there's nothing to do.
80 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
81 struct buffer_head *bh, ext4_fsblk_t blocknr)
85 if (!ext4_handle_valid(handle))
90 BUFFER_TRACE(bh, "enter");
92 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 bh, is_metadata, inode->i_mode,
95 test_opt(inode->i_sb, DATA_FLAGS));
97 /* Never use the revoke function if we are doing full data
98 * journaling: there is no need to, and a V1 superblock won't
99 * support it. Otherwise, only skip the revoke on un-journaled
102 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
103 (!is_metadata && !ext4_should_journal_data(inode))) {
105 BUFFER_TRACE(bh, "call jbd2_journal_forget");
106 return ext4_journal_forget(handle, bh);
112 * data!=journal && (is_metadata || should_journal_data(inode))
114 BUFFER_TRACE(bh, "call ext4_journal_revoke");
115 err = ext4_journal_revoke(handle, blocknr, bh);
117 ext4_abort(inode->i_sb, __func__,
118 "error %d when attempting revoke", err);
119 BUFFER_TRACE(bh, "exit");
124 * Work out how many blocks we need to proceed with the next chunk of a
125 * truncate transaction.
127 static unsigned long blocks_for_truncate(struct inode *inode)
131 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
133 /* Give ourselves just enough room to cope with inodes in which
134 * i_blocks is corrupt: we've seen disk corruptions in the past
135 * which resulted in random data in an inode which looked enough
136 * like a regular file for ext4 to try to delete it. Things
137 * will go a bit crazy if that happens, but at least we should
138 * try not to panic the whole kernel. */
142 /* But we need to bound the transaction so we don't overflow the
144 if (needed > EXT4_MAX_TRANS_DATA)
145 needed = EXT4_MAX_TRANS_DATA;
147 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
151 * Truncate transactions can be complex and absolutely huge. So we need to
152 * be able to restart the transaction at a conventient checkpoint to make
153 * sure we don't overflow the journal.
155 * start_transaction gets us a new handle for a truncate transaction,
156 * and extend_transaction tries to extend the existing one a bit. If
157 * extend fails, we need to propagate the failure up and restart the
158 * transaction in the top-level truncate loop. --sct
160 static handle_t *start_transaction(struct inode *inode)
164 result = ext4_journal_start(inode, blocks_for_truncate(inode));
168 ext4_std_error(inode->i_sb, PTR_ERR(result));
173 * Try to extend this transaction for the purposes of truncation.
175 * Returns 0 if we managed to create more room. If we can't create more
176 * room, and the transaction must be restarted we return 1.
178 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
180 if (!ext4_handle_valid(handle))
182 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
184 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
190 * Restart the transaction associated with *handle. This does a commit,
191 * so before we call here everything must be consistently dirtied against
194 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
196 BUG_ON(EXT4_JOURNAL(inode) == NULL);
197 jbd_debug(2, "restarting handle %p\n", handle);
198 return ext4_journal_restart(handle, blocks_for_truncate(inode));
202 * Called at the last iput() if i_nlink is zero.
204 void ext4_delete_inode(struct inode *inode)
209 if (ext4_should_order_data(inode))
210 ext4_begin_ordered_truncate(inode, 0);
211 truncate_inode_pages(&inode->i_data, 0);
213 if (is_bad_inode(inode))
216 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
217 if (IS_ERR(handle)) {
218 ext4_std_error(inode->i_sb, PTR_ERR(handle));
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext4_orphan_del(NULL, inode);
229 ext4_handle_sync(handle);
231 err = ext4_mark_inode_dirty(handle, inode);
233 ext4_warning(inode->i_sb, __func__,
234 "couldn't mark inode dirty (err %d)", err);
238 ext4_truncate(inode);
241 * ext4_ext_truncate() doesn't reserve any slop when it
242 * restarts journal transactions; therefore there may not be
243 * enough credits left in the handle to remove the inode from
244 * the orphan list and set the dtime field.
246 if (!ext4_handle_has_enough_credits(handle, 3)) {
247 err = ext4_journal_extend(handle, 3);
249 err = ext4_journal_restart(handle, 3);
251 ext4_warning(inode->i_sb, __func__,
252 "couldn't extend journal (err %d)", err);
254 ext4_journal_stop(handle);
260 * Kill off the orphan record which ext4_truncate created.
261 * AKPM: I think this can be inside the above `if'.
262 * Note that ext4_orphan_del() has to be able to cope with the
263 * deletion of a non-existent orphan - this is because we don't
264 * know if ext4_truncate() actually created an orphan record.
265 * (Well, we could do this if we need to, but heck - it works)
267 ext4_orphan_del(handle, inode);
268 EXT4_I(inode)->i_dtime = get_seconds();
271 * One subtle ordering requirement: if anything has gone wrong
272 * (transaction abort, IO errors, whatever), then we can still
273 * do these next steps (the fs will already have been marked as
274 * having errors), but we can't free the inode if the mark_dirty
277 if (ext4_mark_inode_dirty(handle, inode))
278 /* If that failed, just do the required in-core inode clear. */
281 ext4_free_inode(handle, inode);
282 ext4_journal_stop(handle);
285 clear_inode(inode); /* We must guarantee clearing of inode... */
291 struct buffer_head *bh;
294 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
296 p->key = *(p->p = v);
301 * ext4_block_to_path - parse the block number into array of offsets
302 * @inode: inode in question (we are only interested in its superblock)
303 * @i_block: block number to be parsed
304 * @offsets: array to store the offsets in
305 * @boundary: set this non-zero if the referred-to block is likely to be
306 * followed (on disk) by an indirect block.
308 * To store the locations of file's data ext4 uses a data structure common
309 * for UNIX filesystems - tree of pointers anchored in the inode, with
310 * data blocks at leaves and indirect blocks in intermediate nodes.
311 * This function translates the block number into path in that tree -
312 * return value is the path length and @offsets[n] is the offset of
313 * pointer to (n+1)th node in the nth one. If @block is out of range
314 * (negative or too large) warning is printed and zero returned.
316 * Note: function doesn't find node addresses, so no IO is needed. All
317 * we need to know is the capacity of indirect blocks (taken from the
322 * Portability note: the last comparison (check that we fit into triple
323 * indirect block) is spelled differently, because otherwise on an
324 * architecture with 32-bit longs and 8Kb pages we might get into trouble
325 * if our filesystem had 8Kb blocks. We might use long long, but that would
326 * kill us on x86. Oh, well, at least the sign propagation does not matter -
327 * i_block would have to be negative in the very beginning, so we would not
331 static int ext4_block_to_path(struct inode *inode,
333 ext4_lblk_t offsets[4], int *boundary)
335 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
336 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
337 const long direct_blocks = EXT4_NDIR_BLOCKS,
338 indirect_blocks = ptrs,
339 double_blocks = (1 << (ptrs_bits * 2));
344 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
345 } else if (i_block < direct_blocks) {
346 offsets[n++] = i_block;
347 final = direct_blocks;
348 } else if ((i_block -= direct_blocks) < indirect_blocks) {
349 offsets[n++] = EXT4_IND_BLOCK;
350 offsets[n++] = i_block;
352 } else if ((i_block -= indirect_blocks) < double_blocks) {
353 offsets[n++] = EXT4_DIND_BLOCK;
354 offsets[n++] = i_block >> ptrs_bits;
355 offsets[n++] = i_block & (ptrs - 1);
357 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
358 offsets[n++] = EXT4_TIND_BLOCK;
359 offsets[n++] = i_block >> (ptrs_bits * 2);
360 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
361 offsets[n++] = i_block & (ptrs - 1);
364 ext4_warning(inode->i_sb, "ext4_block_to_path",
365 "block %lu > max in inode %lu",
366 i_block + direct_blocks +
367 indirect_blocks + double_blocks, inode->i_ino);
370 *boundary = final - 1 - (i_block & (ptrs - 1));
374 static int __ext4_check_blockref(const char *function, struct inode *inode,
375 __le32 *p, unsigned int max) {
377 unsigned int maxblocks = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es);
379 while (bref < p+max) {
380 if (unlikely(le32_to_cpu(*bref) >= maxblocks)) {
381 ext4_error(inode->i_sb, function,
382 "block reference %u >= max (%u) "
383 "in inode #%lu, offset=%d",
384 le32_to_cpu(*bref), maxblocks,
385 inode->i_ino, (int)(bref-p));
394 #define ext4_check_indirect_blockref(inode, bh) \
395 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
396 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
398 #define ext4_check_inode_blockref(inode) \
399 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
403 * ext4_get_branch - read the chain of indirect blocks leading to data
404 * @inode: inode in question
405 * @depth: depth of the chain (1 - direct pointer, etc.)
406 * @offsets: offsets of pointers in inode/indirect blocks
407 * @chain: place to store the result
408 * @err: here we store the error value
410 * Function fills the array of triples <key, p, bh> and returns %NULL
411 * if everything went OK or the pointer to the last filled triple
412 * (incomplete one) otherwise. Upon the return chain[i].key contains
413 * the number of (i+1)-th block in the chain (as it is stored in memory,
414 * i.e. little-endian 32-bit), chain[i].p contains the address of that
415 * number (it points into struct inode for i==0 and into the bh->b_data
416 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
417 * block for i>0 and NULL for i==0. In other words, it holds the block
418 * numbers of the chain, addresses they were taken from (and where we can
419 * verify that chain did not change) and buffer_heads hosting these
422 * Function stops when it stumbles upon zero pointer (absent block)
423 * (pointer to last triple returned, *@err == 0)
424 * or when it gets an IO error reading an indirect block
425 * (ditto, *@err == -EIO)
426 * or when it reads all @depth-1 indirect blocks successfully and finds
427 * the whole chain, all way to the data (returns %NULL, *err == 0).
429 * Need to be called with
430 * down_read(&EXT4_I(inode)->i_data_sem)
432 static Indirect *ext4_get_branch(struct inode *inode, int depth,
433 ext4_lblk_t *offsets,
434 Indirect chain[4], int *err)
436 struct super_block *sb = inode->i_sb;
438 struct buffer_head *bh;
441 /* i_data is not going away, no lock needed */
442 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
446 bh = sb_getblk(sb, le32_to_cpu(p->key));
450 if (!bh_uptodate_or_lock(bh)) {
451 if (bh_submit_read(bh) < 0) {
455 /* validate block references */
456 if (ext4_check_indirect_blockref(inode, bh)) {
462 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
476 * ext4_find_near - find a place for allocation with sufficient locality
478 * @ind: descriptor of indirect block.
480 * This function returns the preferred place for block allocation.
481 * It is used when heuristic for sequential allocation fails.
483 * + if there is a block to the left of our position - allocate near it.
484 * + if pointer will live in indirect block - allocate near that block.
485 * + if pointer will live in inode - allocate in the same
488 * In the latter case we colour the starting block by the callers PID to
489 * prevent it from clashing with concurrent allocations for a different inode
490 * in the same block group. The PID is used here so that functionally related
491 * files will be close-by on-disk.
493 * Caller must make sure that @ind is valid and will stay that way.
495 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
497 struct ext4_inode_info *ei = EXT4_I(inode);
498 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
500 ext4_fsblk_t bg_start;
501 ext4_fsblk_t last_block;
502 ext4_grpblk_t colour;
503 ext4_group_t block_group;
504 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
506 /* Try to find previous block */
507 for (p = ind->p - 1; p >= start; p--) {
509 return le32_to_cpu(*p);
512 /* No such thing, so let's try location of indirect block */
514 return ind->bh->b_blocknr;
517 * It is going to be referred to from the inode itself? OK, just put it
518 * into the same cylinder group then.
520 block_group = ei->i_block_group;
521 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
522 block_group &= ~(flex_size-1);
523 if (S_ISREG(inode->i_mode))
526 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
527 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
530 * If we are doing delayed allocation, we don't need take
531 * colour into account.
533 if (test_opt(inode->i_sb, DELALLOC))
536 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
537 colour = (current->pid % 16) *
538 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
540 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
541 return bg_start + colour;
545 * ext4_find_goal - find a preferred place for allocation.
547 * @block: block we want
548 * @partial: pointer to the last triple within a chain
550 * Normally this function find the preferred place for block allocation,
553 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
557 * XXX need to get goal block from mballoc's data structures
560 return ext4_find_near(inode, partial);
564 * ext4_blks_to_allocate: Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
575 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
578 unsigned int count = 0;
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
589 count += blocks_to_boundary + 1;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @indirect_blks: the number of blocks need to allocate for indirect
606 * @new_blocks: on return it will store the new block numbers for
607 * the indirect blocks(if needed) and the first direct block,
608 * @blks: on return it will store the total number of allocated
611 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
612 ext4_lblk_t iblock, ext4_fsblk_t goal,
613 int indirect_blks, int blks,
614 ext4_fsblk_t new_blocks[4], int *err)
616 struct ext4_allocation_request ar;
618 unsigned long count = 0, blk_allocated = 0;
620 ext4_fsblk_t current_block = 0;
624 * Here we try to allocate the requested multiple blocks at once,
625 * on a best-effort basis.
626 * To build a branch, we should allocate blocks for
627 * the indirect blocks(if not allocated yet), and at least
628 * the first direct block of this branch. That's the
629 * minimum number of blocks need to allocate(required)
631 /* first we try to allocate the indirect blocks */
632 target = indirect_blks;
635 /* allocating blocks for indirect blocks and direct blocks */
636 current_block = ext4_new_meta_blocks(handle, inode,
642 /* allocate blocks for indirect blocks */
643 while (index < indirect_blks && count) {
644 new_blocks[index++] = current_block++;
649 * save the new block number
650 * for the first direct block
652 new_blocks[index] = current_block;
653 printk(KERN_INFO "%s returned more blocks than "
654 "requested\n", __func__);
660 target = blks - count ;
661 blk_allocated = count;
664 /* Now allocate data blocks */
665 memset(&ar, 0, sizeof(ar));
670 if (S_ISREG(inode->i_mode))
671 /* enable in-core preallocation only for regular files */
672 ar.flags = EXT4_MB_HINT_DATA;
674 current_block = ext4_mb_new_blocks(handle, &ar, err);
676 if (*err && (target == blks)) {
678 * if the allocation failed and we didn't allocate
684 if (target == blks) {
686 * save the new block number
687 * for the first direct block
689 new_blocks[index] = current_block;
691 blk_allocated += ar.len;
694 /* total number of blocks allocated for direct blocks */
699 for (i = 0; i < index; i++)
700 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
705 * ext4_alloc_branch - allocate and set up a chain of blocks.
707 * @indirect_blks: number of allocated indirect blocks
708 * @blks: number of allocated direct blocks
709 * @offsets: offsets (in the blocks) to store the pointers to next.
710 * @branch: place to store the chain in.
712 * This function allocates blocks, zeroes out all but the last one,
713 * links them into chain and (if we are synchronous) writes them to disk.
714 * In other words, it prepares a branch that can be spliced onto the
715 * inode. It stores the information about that chain in the branch[], in
716 * the same format as ext4_get_branch() would do. We are calling it after
717 * we had read the existing part of chain and partial points to the last
718 * triple of that (one with zero ->key). Upon the exit we have the same
719 * picture as after the successful ext4_get_block(), except that in one
720 * place chain is disconnected - *branch->p is still zero (we did not
721 * set the last link), but branch->key contains the number that should
722 * be placed into *branch->p to fill that gap.
724 * If allocation fails we free all blocks we've allocated (and forget
725 * their buffer_heads) and return the error value the from failed
726 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
727 * as described above and return 0.
729 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
730 ext4_lblk_t iblock, int indirect_blks,
731 int *blks, ext4_fsblk_t goal,
732 ext4_lblk_t *offsets, Indirect *branch)
734 int blocksize = inode->i_sb->s_blocksize;
737 struct buffer_head *bh;
739 ext4_fsblk_t new_blocks[4];
740 ext4_fsblk_t current_block;
742 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
743 *blks, new_blocks, &err);
747 branch[0].key = cpu_to_le32(new_blocks[0]);
749 * metadata blocks and data blocks are allocated.
751 for (n = 1; n <= indirect_blks; n++) {
753 * Get buffer_head for parent block, zero it out
754 * and set the pointer to new one, then send
757 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
760 BUFFER_TRACE(bh, "call get_create_access");
761 err = ext4_journal_get_create_access(handle, bh);
768 memset(bh->b_data, 0, blocksize);
769 branch[n].p = (__le32 *) bh->b_data + offsets[n];
770 branch[n].key = cpu_to_le32(new_blocks[n]);
771 *branch[n].p = branch[n].key;
772 if (n == indirect_blks) {
773 current_block = new_blocks[n];
775 * End of chain, update the last new metablock of
776 * the chain to point to the new allocated
777 * data blocks numbers
779 for (i=1; i < num; i++)
780 *(branch[n].p + i) = cpu_to_le32(++current_block);
782 BUFFER_TRACE(bh, "marking uptodate");
783 set_buffer_uptodate(bh);
786 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
787 err = ext4_handle_dirty_metadata(handle, inode, bh);
794 /* Allocation failed, free what we already allocated */
795 for (i = 1; i <= n ; i++) {
796 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
797 ext4_journal_forget(handle, branch[i].bh);
799 for (i = 0; i < indirect_blks; i++)
800 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
802 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
808 * ext4_splice_branch - splice the allocated branch onto inode.
810 * @block: (logical) number of block we are adding
811 * @chain: chain of indirect blocks (with a missing link - see
813 * @where: location of missing link
814 * @num: number of indirect blocks we are adding
815 * @blks: number of direct blocks we are adding
817 * This function fills the missing link and does all housekeeping needed in
818 * inode (->i_blocks, etc.). In case of success we end up with the full
819 * chain to new block and return 0.
821 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
822 ext4_lblk_t block, Indirect *where, int num, int blks)
826 ext4_fsblk_t current_block;
829 * If we're splicing into a [td]indirect block (as opposed to the
830 * inode) then we need to get write access to the [td]indirect block
834 BUFFER_TRACE(where->bh, "get_write_access");
835 err = ext4_journal_get_write_access(handle, where->bh);
841 *where->p = where->key;
844 * Update the host buffer_head or inode to point to more just allocated
845 * direct blocks blocks
847 if (num == 0 && blks > 1) {
848 current_block = le32_to_cpu(where->key) + 1;
849 for (i = 1; i < blks; i++)
850 *(where->p + i) = cpu_to_le32(current_block++);
853 /* We are done with atomic stuff, now do the rest of housekeeping */
855 inode->i_ctime = ext4_current_time(inode);
856 ext4_mark_inode_dirty(handle, inode);
858 /* had we spliced it onto indirect block? */
861 * If we spliced it onto an indirect block, we haven't
862 * altered the inode. Note however that if it is being spliced
863 * onto an indirect block at the very end of the file (the
864 * file is growing) then we *will* alter the inode to reflect
865 * the new i_size. But that is not done here - it is done in
866 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
868 jbd_debug(5, "splicing indirect only\n");
869 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
870 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
875 * OK, we spliced it into the inode itself on a direct block.
876 * Inode was dirtied above.
878 jbd_debug(5, "splicing direct\n");
883 for (i = 1; i <= num; i++) {
884 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
885 ext4_journal_forget(handle, where[i].bh);
886 ext4_free_blocks(handle, inode,
887 le32_to_cpu(where[i-1].key), 1, 0);
889 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
895 * Allocation strategy is simple: if we have to allocate something, we will
896 * have to go the whole way to leaf. So let's do it before attaching anything
897 * to tree, set linkage between the newborn blocks, write them if sync is
898 * required, recheck the path, free and repeat if check fails, otherwise
899 * set the last missing link (that will protect us from any truncate-generated
900 * removals - all blocks on the path are immune now) and possibly force the
901 * write on the parent block.
902 * That has a nice additional property: no special recovery from the failed
903 * allocations is needed - we simply release blocks and do not touch anything
904 * reachable from inode.
906 * `handle' can be NULL if create == 0.
908 * return > 0, # of blocks mapped or allocated.
909 * return = 0, if plain lookup failed.
910 * return < 0, error case.
913 * Need to be called with
914 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
915 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
917 static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
918 ext4_lblk_t iblock, unsigned int maxblocks,
919 struct buffer_head *bh_result,
920 int create, int extend_disksize)
923 ext4_lblk_t offsets[4];
928 int blocks_to_boundary = 0;
930 struct ext4_inode_info *ei = EXT4_I(inode);
932 ext4_fsblk_t first_block = 0;
936 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
937 J_ASSERT(handle != NULL || create == 0);
938 depth = ext4_block_to_path(inode, iblock, offsets,
939 &blocks_to_boundary);
944 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
946 /* Simplest case - block found, no allocation needed */
948 first_block = le32_to_cpu(chain[depth - 1].key);
949 clear_buffer_new(bh_result);
952 while (count < maxblocks && count <= blocks_to_boundary) {
955 blk = le32_to_cpu(*(chain[depth-1].p + count));
957 if (blk == first_block + count)
965 /* Next simple case - plain lookup or failed read of indirect block */
966 if (!create || err == -EIO)
970 * Okay, we need to do block allocation.
972 goal = ext4_find_goal(inode, iblock, partial);
974 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks = (chain + depth) - partial - 1;
978 * Next look up the indirect map to count the totoal number of
979 * direct blocks to allocate for this branch.
981 count = ext4_blks_to_allocate(partial, indirect_blks,
982 maxblocks, blocks_to_boundary);
984 * Block out ext4_truncate while we alter the tree
986 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
988 offsets + (partial - chain), partial);
991 * The ext4_splice_branch call will free and forget any buffers
992 * on the new chain if there is a failure, but that risks using
993 * up transaction credits, especially for bitmaps where the
994 * credits cannot be returned. Can we handle this somehow? We
995 * may need to return -EAGAIN upwards in the worst case. --sct
998 err = ext4_splice_branch(handle, inode, iblock,
999 partial, indirect_blks, count);
1001 * i_disksize growing is protected by i_data_sem. Don't forget to
1002 * protect it if you're about to implement concurrent
1003 * ext4_get_block() -bzzz
1005 if (!err && extend_disksize) {
1006 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
1007 if (disksize > i_size_read(inode))
1008 disksize = i_size_read(inode);
1009 if (disksize > ei->i_disksize)
1010 ei->i_disksize = disksize;
1015 set_buffer_new(bh_result);
1017 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1018 if (count > blocks_to_boundary)
1019 set_buffer_boundary(bh_result);
1021 /* Clean up and exit */
1022 partial = chain + depth - 1; /* the whole chain */
1024 while (partial > chain) {
1025 BUFFER_TRACE(partial->bh, "call brelse");
1026 brelse(partial->bh);
1029 BUFFER_TRACE(bh_result, "returned");
1034 qsize_t ext4_get_reserved_space(struct inode *inode)
1036 unsigned long long total;
1038 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1039 total = EXT4_I(inode)->i_reserved_data_blocks +
1040 EXT4_I(inode)->i_reserved_meta_blocks;
1041 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1046 * Calculate the number of metadata blocks need to reserve
1047 * to allocate @blocks for non extent file based file
1049 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1051 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1052 int ind_blks, dind_blks, tind_blks;
1054 /* number of new indirect blocks needed */
1055 ind_blks = (blocks + icap - 1) / icap;
1057 dind_blks = (ind_blks + icap - 1) / icap;
1061 return ind_blks + dind_blks + tind_blks;
1065 * Calculate the number of metadata blocks need to reserve
1066 * to allocate given number of blocks
1068 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1073 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1074 return ext4_ext_calc_metadata_amount(inode, blocks);
1076 return ext4_indirect_calc_metadata_amount(inode, blocks);
1079 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1081 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1082 int total, mdb, mdb_free;
1084 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1085 /* recalculate the number of metablocks still need to be reserved */
1086 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1087 mdb = ext4_calc_metadata_amount(inode, total);
1089 /* figure out how many metablocks to release */
1090 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1091 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1094 /* Account for allocated meta_blocks */
1095 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1097 /* update fs dirty blocks counter */
1098 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1099 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1100 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1103 /* update per-inode reservations */
1104 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1105 EXT4_I(inode)->i_reserved_data_blocks -= used;
1106 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1109 * free those over-booking quota for metadata blocks
1112 vfs_dq_release_reservation_block(inode, mdb_free);
1115 * If we have done all the pending block allocations and if
1116 * there aren't any writers on the inode, we can discard the
1117 * inode's preallocations.
1119 if (!total && (atomic_read(&inode->i_writecount) == 0))
1120 ext4_discard_preallocations(inode);
1124 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1125 * and returns if the blocks are already mapped.
1127 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1128 * and store the allocated blocks in the result buffer head and mark it
1131 * If file type is extents based, it will call ext4_ext_get_blocks(),
1132 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1135 * On success, it returns the number of blocks being mapped or allocate.
1136 * if create==0 and the blocks are pre-allocated and uninitialized block,
1137 * the result buffer head is unmapped. If the create ==1, it will make sure
1138 * the buffer head is mapped.
1140 * It returns 0 if plain look up failed (blocks have not been allocated), in
1141 * that casem, buffer head is unmapped
1143 * It returns the error in case of allocation failure.
1145 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1146 unsigned int max_blocks, struct buffer_head *bh,
1147 int create, int extend_disksize, int flag)
1151 clear_buffer_mapped(bh);
1154 * Try to see if we can get the block without requesting
1155 * for new file system block.
1157 down_read((&EXT4_I(inode)->i_data_sem));
1158 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1159 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1162 retval = ext4_get_blocks_handle(handle,
1163 inode, block, max_blocks, bh, 0, 0);
1165 up_read((&EXT4_I(inode)->i_data_sem));
1167 /* If it is only a block(s) look up */
1172 * Returns if the blocks have already allocated
1174 * Note that if blocks have been preallocated
1175 * ext4_ext_get_block() returns th create = 0
1176 * with buffer head unmapped.
1178 if (retval > 0 && buffer_mapped(bh))
1182 * New blocks allocate and/or writing to uninitialized extent
1183 * will possibly result in updating i_data, so we take
1184 * the write lock of i_data_sem, and call get_blocks()
1185 * with create == 1 flag.
1187 down_write((&EXT4_I(inode)->i_data_sem));
1190 * if the caller is from delayed allocation writeout path
1191 * we have already reserved fs blocks for allocation
1192 * let the underlying get_block() function know to
1193 * avoid double accounting
1196 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1198 * We need to check for EXT4 here because migrate
1199 * could have changed the inode type in between
1201 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1202 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1203 bh, create, extend_disksize);
1205 retval = ext4_get_blocks_handle(handle, inode, block,
1206 max_blocks, bh, create, extend_disksize);
1208 if (retval > 0 && buffer_new(bh)) {
1210 * We allocated new blocks which will result in
1211 * i_data's format changing. Force the migrate
1212 * to fail by clearing migrate flags
1214 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1220 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1222 * Update reserved blocks/metadata blocks
1223 * after successful block allocation
1224 * which were deferred till now
1226 if ((retval > 0) && buffer_delay(bh))
1227 ext4_da_update_reserve_space(inode, retval);
1230 up_write((&EXT4_I(inode)->i_data_sem));
1234 /* Maximum number of blocks we map for direct IO at once. */
1235 #define DIO_MAX_BLOCKS 4096
1237 int ext4_get_block(struct inode *inode, sector_t iblock,
1238 struct buffer_head *bh_result, int create)
1240 handle_t *handle = ext4_journal_current_handle();
1241 int ret = 0, started = 0;
1242 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1245 if (create && !handle) {
1246 /* Direct IO write... */
1247 if (max_blocks > DIO_MAX_BLOCKS)
1248 max_blocks = DIO_MAX_BLOCKS;
1249 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1250 handle = ext4_journal_start(inode, dio_credits);
1251 if (IS_ERR(handle)) {
1252 ret = PTR_ERR(handle);
1258 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1259 max_blocks, bh_result, create, 0, 0);
1261 bh_result->b_size = (ret << inode->i_blkbits);
1265 ext4_journal_stop(handle);
1271 * `handle' can be NULL if create is zero
1273 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1274 ext4_lblk_t block, int create, int *errp)
1276 struct buffer_head dummy;
1279 J_ASSERT(handle != NULL || create == 0);
1282 dummy.b_blocknr = -1000;
1283 buffer_trace_init(&dummy.b_history);
1284 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1285 &dummy, create, 1, 0);
1287 * ext4_get_blocks_handle() returns number of blocks
1288 * mapped. 0 in case of a HOLE.
1296 if (!err && buffer_mapped(&dummy)) {
1297 struct buffer_head *bh;
1298 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1303 if (buffer_new(&dummy)) {
1304 J_ASSERT(create != 0);
1305 J_ASSERT(handle != NULL);
1308 * Now that we do not always journal data, we should
1309 * keep in mind whether this should always journal the
1310 * new buffer as metadata. For now, regular file
1311 * writes use ext4_get_block instead, so it's not a
1315 BUFFER_TRACE(bh, "call get_create_access");
1316 fatal = ext4_journal_get_create_access(handle, bh);
1317 if (!fatal && !buffer_uptodate(bh)) {
1318 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1319 set_buffer_uptodate(bh);
1322 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1323 err = ext4_handle_dirty_metadata(handle, inode, bh);
1327 BUFFER_TRACE(bh, "not a new buffer");
1340 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1341 ext4_lblk_t block, int create, int *err)
1343 struct buffer_head *bh;
1345 bh = ext4_getblk(handle, inode, block, create, err);
1348 if (buffer_uptodate(bh))
1350 ll_rw_block(READ_META, 1, &bh);
1352 if (buffer_uptodate(bh))
1359 static int walk_page_buffers(handle_t *handle,
1360 struct buffer_head *head,
1364 int (*fn)(handle_t *handle,
1365 struct buffer_head *bh))
1367 struct buffer_head *bh;
1368 unsigned block_start, block_end;
1369 unsigned blocksize = head->b_size;
1371 struct buffer_head *next;
1373 for (bh = head, block_start = 0;
1374 ret == 0 && (bh != head || !block_start);
1375 block_start = block_end, bh = next)
1377 next = bh->b_this_page;
1378 block_end = block_start + blocksize;
1379 if (block_end <= from || block_start >= to) {
1380 if (partial && !buffer_uptodate(bh))
1384 err = (*fn)(handle, bh);
1392 * To preserve ordering, it is essential that the hole instantiation and
1393 * the data write be encapsulated in a single transaction. We cannot
1394 * close off a transaction and start a new one between the ext4_get_block()
1395 * and the commit_write(). So doing the jbd2_journal_start at the start of
1396 * prepare_write() is the right place.
1398 * Also, this function can nest inside ext4_writepage() ->
1399 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1400 * has generated enough buffer credits to do the whole page. So we won't
1401 * block on the journal in that case, which is good, because the caller may
1404 * By accident, ext4 can be reentered when a transaction is open via
1405 * quota file writes. If we were to commit the transaction while thus
1406 * reentered, there can be a deadlock - we would be holding a quota
1407 * lock, and the commit would never complete if another thread had a
1408 * transaction open and was blocking on the quota lock - a ranking
1411 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1412 * will _not_ run commit under these circumstances because handle->h_ref
1413 * is elevated. We'll still have enough credits for the tiny quotafile
1416 static int do_journal_get_write_access(handle_t *handle,
1417 struct buffer_head *bh)
1419 if (!buffer_mapped(bh) || buffer_freed(bh))
1421 return ext4_journal_get_write_access(handle, bh);
1424 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1425 loff_t pos, unsigned len, unsigned flags,
1426 struct page **pagep, void **fsdata)
1428 struct inode *inode = mapping->host;
1429 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1436 trace_mark(ext4_write_begin,
1437 "dev %s ino %lu pos %llu len %u flags %u",
1438 inode->i_sb->s_id, inode->i_ino,
1439 (unsigned long long) pos, len, flags);
1440 index = pos >> PAGE_CACHE_SHIFT;
1441 from = pos & (PAGE_CACHE_SIZE - 1);
1445 handle = ext4_journal_start(inode, needed_blocks);
1446 if (IS_ERR(handle)) {
1447 ret = PTR_ERR(handle);
1451 /* We cannot recurse into the filesystem as the transaction is already
1453 flags |= AOP_FLAG_NOFS;
1455 page = grab_cache_page_write_begin(mapping, index, flags);
1457 ext4_journal_stop(handle);
1463 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1466 if (!ret && ext4_should_journal_data(inode)) {
1467 ret = walk_page_buffers(handle, page_buffers(page),
1468 from, to, NULL, do_journal_get_write_access);
1473 ext4_journal_stop(handle);
1474 page_cache_release(page);
1476 * block_write_begin may have instantiated a few blocks
1477 * outside i_size. Trim these off again. Don't need
1478 * i_size_read because we hold i_mutex.
1480 if (pos + len > inode->i_size)
1481 vmtruncate(inode, inode->i_size);
1484 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1490 /* For write_end() in data=journal mode */
1491 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1493 if (!buffer_mapped(bh) || buffer_freed(bh))
1495 set_buffer_uptodate(bh);
1496 return ext4_handle_dirty_metadata(handle, NULL, bh);
1500 * We need to pick up the new inode size which generic_commit_write gave us
1501 * `file' can be NULL - eg, when called from page_symlink().
1503 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1504 * buffers are managed internally.
1506 static int ext4_ordered_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;
1515 trace_mark(ext4_ordered_write_end,
1516 "dev %s ino %lu pos %llu len %u copied %u",
1517 inode->i_sb->s_id, inode->i_ino,
1518 (unsigned long long) pos, len, copied);
1519 ret = ext4_jbd2_file_inode(handle, inode);
1524 new_i_size = pos + copied;
1525 if (new_i_size > EXT4_I(inode)->i_disksize) {
1526 ext4_update_i_disksize(inode, new_i_size);
1527 /* We need to mark inode dirty even if
1528 * new_i_size is less that inode->i_size
1529 * bu greater than i_disksize.(hint delalloc)
1531 ext4_mark_inode_dirty(handle, inode);
1534 ret2 = generic_write_end(file, mapping, pos, len, copied,
1540 ret2 = ext4_journal_stop(handle);
1544 return ret ? ret : copied;
1547 static int ext4_writeback_write_end(struct file *file,
1548 struct address_space *mapping,
1549 loff_t pos, unsigned len, unsigned copied,
1550 struct page *page, void *fsdata)
1552 handle_t *handle = ext4_journal_current_handle();
1553 struct inode *inode = mapping->host;
1557 trace_mark(ext4_writeback_write_end,
1558 "dev %s ino %lu pos %llu len %u copied %u",
1559 inode->i_sb->s_id, inode->i_ino,
1560 (unsigned long long) pos, len, copied);
1561 new_i_size = pos + copied;
1562 if (new_i_size > EXT4_I(inode)->i_disksize) {
1563 ext4_update_i_disksize(inode, new_i_size);
1564 /* We need to mark inode dirty even if
1565 * new_i_size is less that inode->i_size
1566 * bu greater than i_disksize.(hint delalloc)
1568 ext4_mark_inode_dirty(handle, inode);
1571 ret2 = generic_write_end(file, mapping, pos, len, copied,
1577 ret2 = ext4_journal_stop(handle);
1581 return ret ? ret : copied;
1584 static int ext4_journalled_write_end(struct file *file,
1585 struct address_space *mapping,
1586 loff_t pos, unsigned len, unsigned copied,
1587 struct page *page, void *fsdata)
1589 handle_t *handle = ext4_journal_current_handle();
1590 struct inode *inode = mapping->host;
1596 trace_mark(ext4_journalled_write_end,
1597 "dev %s ino %lu pos %llu len %u copied %u",
1598 inode->i_sb->s_id, inode->i_ino,
1599 (unsigned long long) pos, len, copied);
1600 from = pos & (PAGE_CACHE_SIZE - 1);
1604 if (!PageUptodate(page))
1606 page_zero_new_buffers(page, from+copied, to);
1609 ret = walk_page_buffers(handle, page_buffers(page), from,
1610 to, &partial, write_end_fn);
1612 SetPageUptodate(page);
1613 new_i_size = pos + copied;
1614 if (new_i_size > inode->i_size)
1615 i_size_write(inode, pos+copied);
1616 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1617 if (new_i_size > EXT4_I(inode)->i_disksize) {
1618 ext4_update_i_disksize(inode, new_i_size);
1619 ret2 = ext4_mark_inode_dirty(handle, inode);
1625 ret2 = ext4_journal_stop(handle);
1628 page_cache_release(page);
1630 return ret ? ret : copied;
1633 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1636 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1637 unsigned long md_needed, mdblocks, total = 0;
1640 * recalculate the amount of metadata blocks to reserve
1641 * in order to allocate nrblocks
1642 * worse case is one extent per block
1645 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1646 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1647 mdblocks = ext4_calc_metadata_amount(inode, total);
1648 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1650 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1651 total = md_needed + nrblocks;
1654 * Make quota reservation here to prevent quota overflow
1655 * later. Real quota accounting is done at pages writeout
1658 if (vfs_dq_reserve_block(inode, total)) {
1659 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1663 if (ext4_claim_free_blocks(sbi, total)) {
1664 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1665 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1669 vfs_dq_release_reservation_block(inode, total);
1672 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1673 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1675 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1676 return 0; /* success */
1679 static void ext4_da_release_space(struct inode *inode, int to_free)
1681 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1682 int total, mdb, mdb_free, release;
1685 return; /* Nothing to release, exit */
1687 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1689 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1691 * if there is no reserved blocks, but we try to free some
1692 * then the counter is messed up somewhere.
1693 * but since this function is called from invalidate
1694 * page, it's harmless to return without any action
1696 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1697 "blocks for inode %lu, but there is no reserved "
1698 "data blocks\n", to_free, inode->i_ino);
1699 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1703 /* recalculate the number of metablocks still need to be reserved */
1704 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1705 mdb = ext4_calc_metadata_amount(inode, total);
1707 /* figure out how many metablocks to release */
1708 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1709 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1711 release = to_free + mdb_free;
1713 /* update fs dirty blocks counter for truncate case */
1714 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1716 /* update per-inode reservations */
1717 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1718 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1720 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1721 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1722 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1724 vfs_dq_release_reservation_block(inode, release);
1727 static void ext4_da_page_release_reservation(struct page *page,
1728 unsigned long offset)
1731 struct buffer_head *head, *bh;
1732 unsigned int curr_off = 0;
1734 head = page_buffers(page);
1737 unsigned int next_off = curr_off + bh->b_size;
1739 if ((offset <= curr_off) && (buffer_delay(bh))) {
1741 clear_buffer_delay(bh);
1743 curr_off = next_off;
1744 } while ((bh = bh->b_this_page) != head);
1745 ext4_da_release_space(page->mapping->host, to_release);
1749 * Delayed allocation stuff
1752 struct mpage_da_data {
1753 struct inode *inode;
1754 sector_t b_blocknr; /* start block number of extent */
1755 size_t b_size; /* size of extent */
1756 unsigned long b_state; /* state of the extent */
1757 unsigned long first_page, next_page; /* extent of pages */
1758 struct writeback_control *wbc;
1765 * mpage_da_submit_io - walks through extent of pages and try to write
1766 * them with writepage() call back
1768 * @mpd->inode: inode
1769 * @mpd->first_page: first page of the extent
1770 * @mpd->next_page: page after the last page of the extent
1772 * By the time mpage_da_submit_io() is called we expect all blocks
1773 * to be allocated. this may be wrong if allocation failed.
1775 * As pages are already locked by write_cache_pages(), we can't use it
1777 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1780 struct pagevec pvec;
1781 unsigned long index, end;
1782 int ret = 0, err, nr_pages, i;
1783 struct inode *inode = mpd->inode;
1784 struct address_space *mapping = inode->i_mapping;
1786 BUG_ON(mpd->next_page <= mpd->first_page);
1788 * We need to start from the first_page to the next_page - 1
1789 * to make sure we also write the mapped dirty buffer_heads.
1790 * If we look at mpd->b_blocknr we would only be looking
1791 * at the currently mapped buffer_heads.
1793 index = mpd->first_page;
1794 end = mpd->next_page - 1;
1796 pagevec_init(&pvec, 0);
1797 while (index <= end) {
1798 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1801 for (i = 0; i < nr_pages; i++) {
1802 struct page *page = pvec.pages[i];
1804 index = page->index;
1809 BUG_ON(!PageLocked(page));
1810 BUG_ON(PageWriteback(page));
1812 pages_skipped = mpd->wbc->pages_skipped;
1813 err = mapping->a_ops->writepage(page, mpd->wbc);
1814 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1816 * have successfully written the page
1817 * without skipping the same
1819 mpd->pages_written++;
1821 * In error case, we have to continue because
1822 * remaining pages are still locked
1823 * XXX: unlock and re-dirty them?
1828 pagevec_release(&pvec);
1834 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1836 * @mpd->inode - inode to walk through
1837 * @exbh->b_blocknr - first block on a disk
1838 * @exbh->b_size - amount of space in bytes
1839 * @logical - first logical block to start assignment with
1841 * the function goes through all passed space and put actual disk
1842 * block numbers into buffer heads, dropping BH_Delay
1844 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1845 struct buffer_head *exbh)
1847 struct inode *inode = mpd->inode;
1848 struct address_space *mapping = inode->i_mapping;
1849 int blocks = exbh->b_size >> inode->i_blkbits;
1850 sector_t pblock = exbh->b_blocknr, cur_logical;
1851 struct buffer_head *head, *bh;
1853 struct pagevec pvec;
1856 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1857 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1858 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1860 pagevec_init(&pvec, 0);
1862 while (index <= end) {
1863 /* XXX: optimize tail */
1864 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1867 for (i = 0; i < nr_pages; i++) {
1868 struct page *page = pvec.pages[i];
1870 index = page->index;
1875 BUG_ON(!PageLocked(page));
1876 BUG_ON(PageWriteback(page));
1877 BUG_ON(!page_has_buffers(page));
1879 bh = page_buffers(page);
1882 /* skip blocks out of the range */
1884 if (cur_logical >= logical)
1887 } while ((bh = bh->b_this_page) != head);
1890 if (cur_logical >= logical + blocks)
1892 if (buffer_delay(bh)) {
1893 bh->b_blocknr = pblock;
1894 clear_buffer_delay(bh);
1895 bh->b_bdev = inode->i_sb->s_bdev;
1896 } else if (buffer_unwritten(bh)) {
1897 bh->b_blocknr = pblock;
1898 clear_buffer_unwritten(bh);
1899 set_buffer_mapped(bh);
1901 bh->b_bdev = inode->i_sb->s_bdev;
1902 } else if (buffer_mapped(bh))
1903 BUG_ON(bh->b_blocknr != pblock);
1907 } while ((bh = bh->b_this_page) != head);
1909 pagevec_release(&pvec);
1915 * __unmap_underlying_blocks - just a helper function to unmap
1916 * set of blocks described by @bh
1918 static inline void __unmap_underlying_blocks(struct inode *inode,
1919 struct buffer_head *bh)
1921 struct block_device *bdev = inode->i_sb->s_bdev;
1924 blocks = bh->b_size >> inode->i_blkbits;
1925 for (i = 0; i < blocks; i++)
1926 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1929 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1930 sector_t logical, long blk_cnt)
1934 struct pagevec pvec;
1935 struct inode *inode = mpd->inode;
1936 struct address_space *mapping = inode->i_mapping;
1938 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1939 end = (logical + blk_cnt - 1) >>
1940 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1941 while (index <= end) {
1942 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1945 for (i = 0; i < nr_pages; i++) {
1946 struct page *page = pvec.pages[i];
1947 index = page->index;
1952 BUG_ON(!PageLocked(page));
1953 BUG_ON(PageWriteback(page));
1954 block_invalidatepage(page, 0);
1955 ClearPageUptodate(page);
1962 static void ext4_print_free_blocks(struct inode *inode)
1964 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1965 printk(KERN_EMERG "Total free blocks count %lld\n",
1966 ext4_count_free_blocks(inode->i_sb));
1967 printk(KERN_EMERG "Free/Dirty block details\n");
1968 printk(KERN_EMERG "free_blocks=%lld\n",
1969 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1970 printk(KERN_EMERG "dirty_blocks=%lld\n",
1971 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1972 printk(KERN_EMERG "Block reservation details\n");
1973 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1974 EXT4_I(inode)->i_reserved_data_blocks);
1975 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1976 EXT4_I(inode)->i_reserved_meta_blocks);
1980 #define EXT4_DELALLOC_RSVED 1
1981 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
1982 struct buffer_head *bh_result, int create)
1985 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1986 loff_t disksize = EXT4_I(inode)->i_disksize;
1987 handle_t *handle = NULL;
1989 handle = ext4_journal_current_handle();
1991 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
1992 bh_result, create, 0, EXT4_DELALLOC_RSVED);
1996 bh_result->b_size = (ret << inode->i_blkbits);
1998 if (ext4_should_order_data(inode)) {
2000 retval = ext4_jbd2_file_inode(handle, inode);
2003 * Failed to add inode for ordered mode. Don't
2010 * Update on-disk size along with block allocation we don't
2011 * use 'extend_disksize' as size may change within already
2012 * allocated block -bzzz
2014 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2015 if (disksize > i_size_read(inode))
2016 disksize = i_size_read(inode);
2017 if (disksize > EXT4_I(inode)->i_disksize) {
2018 ext4_update_i_disksize(inode, disksize);
2019 ret = ext4_mark_inode_dirty(handle, inode);
2026 * mpage_da_map_blocks - go through given space
2028 * @mpd - bh describing space
2030 * The function skips space we know is already mapped to disk blocks.
2033 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2036 struct buffer_head new;
2040 * We consider only non-mapped and non-allocated blocks
2042 if ((mpd->b_state & (1 << BH_Mapped)) &&
2043 !(mpd->b_state & (1 << BH_Delay)))
2045 new.b_state = mpd->b_state;
2047 new.b_size = mpd->b_size;
2048 next = mpd->b_blocknr;
2050 * If we didn't accumulate anything
2051 * to write simply return
2056 err = ext4_da_get_block_write(mpd->inode, next, &new, 1);
2059 * If get block returns with error we simply
2060 * return. Later writepage will redirty the page and
2061 * writepages will find the dirty page again
2066 if (err == -ENOSPC &&
2067 ext4_count_free_blocks(mpd->inode->i_sb)) {
2073 * get block failure will cause us to loop in
2074 * writepages, because a_ops->writepage won't be able
2075 * to make progress. The page will be redirtied by
2076 * writepage and writepages will again try to write
2079 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2080 "at logical offset %llu with max blocks "
2081 "%zd with error %d\n",
2082 __func__, mpd->inode->i_ino,
2083 (unsigned long long)next,
2084 mpd->b_size >> mpd->inode->i_blkbits, err);
2085 printk(KERN_EMERG "This should not happen.!! "
2086 "Data will be lost\n");
2087 if (err == -ENOSPC) {
2088 ext4_print_free_blocks(mpd->inode);
2090 /* invlaidate all the pages */
2091 ext4_da_block_invalidatepages(mpd, next,
2092 mpd->b_size >> mpd->inode->i_blkbits);
2095 BUG_ON(new.b_size == 0);
2097 if (buffer_new(&new))
2098 __unmap_underlying_blocks(mpd->inode, &new);
2101 * If blocks are delayed marked, we need to
2102 * put actual blocknr and drop delayed bit
2104 if ((mpd->b_state & (1 << BH_Delay)) ||
2105 (mpd->b_state & (1 << BH_Unwritten)))
2106 mpage_put_bnr_to_bhs(mpd, next, &new);
2111 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2112 (1 << BH_Delay) | (1 << BH_Unwritten))
2115 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2117 * @mpd->lbh - extent of blocks
2118 * @logical - logical number of the block in the file
2119 * @bh - bh of the block (used to access block's state)
2121 * the function is used to collect contig. blocks in same state
2123 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2124 sector_t logical, size_t b_size,
2125 unsigned long b_state)
2128 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2130 /* check if thereserved journal credits might overflow */
2131 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2132 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2134 * With non-extent format we are limited by the journal
2135 * credit available. Total credit needed to insert
2136 * nrblocks contiguous blocks is dependent on the
2137 * nrblocks. So limit nrblocks.
2140 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2141 EXT4_MAX_TRANS_DATA) {
2143 * Adding the new buffer_head would make it cross the
2144 * allowed limit for which we have journal credit
2145 * reserved. So limit the new bh->b_size
2147 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2148 mpd->inode->i_blkbits;
2149 /* we will do mpage_da_submit_io in the next loop */
2153 * First block in the extent
2155 if (mpd->b_size == 0) {
2156 mpd->b_blocknr = logical;
2157 mpd->b_size = b_size;
2158 mpd->b_state = b_state & BH_FLAGS;
2162 next = mpd->b_blocknr + nrblocks;
2164 * Can we merge the block to our big extent?
2166 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2167 mpd->b_size += b_size;
2173 * We couldn't merge the block to our extent, so we
2174 * need to flush current extent and start new one
2176 if (mpage_da_map_blocks(mpd) == 0)
2177 mpage_da_submit_io(mpd);
2183 * __mpage_da_writepage - finds extent of pages and blocks
2185 * @page: page to consider
2186 * @wbc: not used, we just follow rules
2189 * The function finds extents of pages and scan them for all blocks.
2191 static int __mpage_da_writepage(struct page *page,
2192 struct writeback_control *wbc, void *data)
2194 struct mpage_da_data *mpd = data;
2195 struct inode *inode = mpd->inode;
2196 struct buffer_head *bh, *head;
2201 * Rest of the page in the page_vec
2202 * redirty then and skip then. We will
2203 * try to to write them again after
2204 * starting a new transaction
2206 redirty_page_for_writepage(wbc, page);
2208 return MPAGE_DA_EXTENT_TAIL;
2211 * Can we merge this page to current extent?
2213 if (mpd->next_page != page->index) {
2215 * Nope, we can't. So, we map non-allocated blocks
2216 * and start IO on them using writepage()
2218 if (mpd->next_page != mpd->first_page) {
2219 if (mpage_da_map_blocks(mpd) == 0)
2220 mpage_da_submit_io(mpd);
2222 * skip rest of the page in the page_vec
2225 redirty_page_for_writepage(wbc, page);
2227 return MPAGE_DA_EXTENT_TAIL;
2231 * Start next extent of pages ...
2233 mpd->first_page = page->index;
2243 mpd->next_page = page->index + 1;
2244 logical = (sector_t) page->index <<
2245 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2247 if (!page_has_buffers(page)) {
2248 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2249 (1 << BH_Dirty) | (1 << BH_Uptodate));
2251 return MPAGE_DA_EXTENT_TAIL;
2254 * Page with regular buffer heads, just add all dirty ones
2256 head = page_buffers(page);
2259 BUG_ON(buffer_locked(bh));
2261 * We need to try to allocate
2262 * unmapped blocks in the same page.
2263 * Otherwise we won't make progress
2264 * with the page in ext4_da_writepage
2266 if (buffer_dirty(bh) &&
2267 (!buffer_mapped(bh) || buffer_delay(bh))) {
2268 mpage_add_bh_to_extent(mpd, logical,
2272 return MPAGE_DA_EXTENT_TAIL;
2273 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2275 * mapped dirty buffer. We need to update
2276 * the b_state because we look at
2277 * b_state in mpage_da_map_blocks. We don't
2278 * update b_size because if we find an
2279 * unmapped buffer_head later we need to
2280 * use the b_state flag of that buffer_head.
2282 if (mpd->b_size == 0)
2283 mpd->b_state = bh->b_state & BH_FLAGS;
2286 } while ((bh = bh->b_this_page) != head);
2293 * this is a special callback for ->write_begin() only
2294 * it's intention is to return mapped block or reserve space
2296 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2297 struct buffer_head *bh_result, int create)
2301 BUG_ON(create == 0);
2302 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2305 * first, we need to know whether the block is allocated already
2306 * preallocated blocks are unmapped but should treated
2307 * the same as allocated blocks.
2309 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2310 if ((ret == 0) && !buffer_delay(bh_result)) {
2311 /* the block isn't (pre)allocated yet, let's reserve space */
2313 * XXX: __block_prepare_write() unmaps passed block,
2316 ret = ext4_da_reserve_space(inode, 1);
2318 /* not enough space to reserve */
2321 map_bh(bh_result, inode->i_sb, 0);
2322 set_buffer_new(bh_result);
2323 set_buffer_delay(bh_result);
2324 } else if (ret > 0) {
2325 bh_result->b_size = (ret << inode->i_blkbits);
2327 * With sub-block writes into unwritten extents
2328 * we also need to mark the buffer as new so that
2329 * the unwritten parts of the buffer gets correctly zeroed.
2331 if (buffer_unwritten(bh_result))
2332 set_buffer_new(bh_result);
2339 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2342 * unmapped buffer is possible for holes.
2343 * delay buffer is possible with delayed allocation
2345 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2348 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2349 struct buffer_head *bh_result, int create)
2352 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2355 * we don't want to do block allocation in writepage
2356 * so call get_block_wrap with create = 0
2358 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2359 bh_result, 0, 0, 0);
2361 bh_result->b_size = (ret << inode->i_blkbits);
2368 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2369 * get called via journal_submit_inode_data_buffers (no journal handle)
2370 * get called via shrink_page_list via pdflush (no journal handle)
2371 * or grab_page_cache when doing write_begin (have journal handle)
2373 static int ext4_da_writepage(struct page *page,
2374 struct writeback_control *wbc)
2379 struct buffer_head *page_bufs;
2380 struct inode *inode = page->mapping->host;
2382 trace_mark(ext4_da_writepage,
2383 "dev %s ino %lu page_index %lu",
2384 inode->i_sb->s_id, inode->i_ino, page->index);
2385 size = i_size_read(inode);
2386 if (page->index == size >> PAGE_CACHE_SHIFT)
2387 len = size & ~PAGE_CACHE_MASK;
2389 len = PAGE_CACHE_SIZE;
2391 if (page_has_buffers(page)) {
2392 page_bufs = page_buffers(page);
2393 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2394 ext4_bh_unmapped_or_delay)) {
2396 * We don't want to do block allocation
2397 * So redirty the page and return
2398 * We may reach here when we do a journal commit
2399 * via journal_submit_inode_data_buffers.
2400 * If we don't have mapping block we just ignore
2401 * them. We can also reach here via shrink_page_list
2403 redirty_page_for_writepage(wbc, page);
2409 * The test for page_has_buffers() is subtle:
2410 * We know the page is dirty but it lost buffers. That means
2411 * that at some moment in time after write_begin()/write_end()
2412 * has been called all buffers have been clean and thus they
2413 * must have been written at least once. So they are all
2414 * mapped and we can happily proceed with mapping them
2415 * and writing the page.
2417 * Try to initialize the buffer_heads and check whether
2418 * all are mapped and non delay. We don't want to
2419 * do block allocation here.
2421 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2422 ext4_normal_get_block_write);
2424 page_bufs = page_buffers(page);
2425 /* check whether all are mapped and non delay */
2426 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2427 ext4_bh_unmapped_or_delay)) {
2428 redirty_page_for_writepage(wbc, page);
2434 * We can't do block allocation here
2435 * so just redity the page and unlock
2438 redirty_page_for_writepage(wbc, page);
2442 /* now mark the buffer_heads as dirty and uptodate */
2443 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2446 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2447 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2449 ret = block_write_full_page(page,
2450 ext4_normal_get_block_write,
2457 * This is called via ext4_da_writepages() to
2458 * calulate the total number of credits to reserve to fit
2459 * a single extent allocation into a single transaction,
2460 * ext4_da_writpeages() will loop calling this before
2461 * the block allocation.
2464 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2466 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2469 * With non-extent format the journal credit needed to
2470 * insert nrblocks contiguous block is dependent on
2471 * number of contiguous block. So we will limit
2472 * number of contiguous block to a sane value
2474 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2475 (max_blocks > EXT4_MAX_TRANS_DATA))
2476 max_blocks = EXT4_MAX_TRANS_DATA;
2478 return ext4_chunk_trans_blocks(inode, max_blocks);
2481 static int ext4_da_writepages(struct address_space *mapping,
2482 struct writeback_control *wbc)
2485 int range_whole = 0;
2486 handle_t *handle = NULL;
2487 struct mpage_da_data mpd;
2488 struct inode *inode = mapping->host;
2489 int no_nrwrite_index_update;
2490 int pages_written = 0;
2492 int range_cyclic, cycled = 1, io_done = 0;
2493 int needed_blocks, ret = 0, nr_to_writebump = 0;
2494 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2496 trace_mark(ext4_da_writepages,
2497 "dev %s ino %lu nr_t_write %ld "
2498 "pages_skipped %ld range_start %llu "
2499 "range_end %llu nonblocking %d "
2500 "for_kupdate %d for_reclaim %d "
2501 "for_writepages %d range_cyclic %d",
2502 inode->i_sb->s_id, inode->i_ino,
2503 wbc->nr_to_write, wbc->pages_skipped,
2504 (unsigned long long) wbc->range_start,
2505 (unsigned long long) wbc->range_end,
2506 wbc->nonblocking, wbc->for_kupdate,
2507 wbc->for_reclaim, wbc->for_writepages,
2511 * No pages to write? This is mainly a kludge to avoid starting
2512 * a transaction for special inodes like journal inode on last iput()
2513 * because that could violate lock ordering on umount
2515 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2519 * If the filesystem has aborted, it is read-only, so return
2520 * right away instead of dumping stack traces later on that
2521 * will obscure the real source of the problem. We test
2522 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2523 * the latter could be true if the filesystem is mounted
2524 * read-only, and in that case, ext4_da_writepages should
2525 * *never* be called, so if that ever happens, we would want
2528 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2532 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2533 * This make sure small files blocks are allocated in
2534 * single attempt. This ensure that small files
2535 * get less fragmented.
2537 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2538 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2539 wbc->nr_to_write = sbi->s_mb_stream_request;
2541 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2544 range_cyclic = wbc->range_cyclic;
2545 if (wbc->range_cyclic) {
2546 index = mapping->writeback_index;
2549 wbc->range_start = index << PAGE_CACHE_SHIFT;
2550 wbc->range_end = LLONG_MAX;
2551 wbc->range_cyclic = 0;
2553 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2556 mpd.inode = mapping->host;
2559 * we don't want write_cache_pages to update
2560 * nr_to_write and writeback_index
2562 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2563 wbc->no_nrwrite_index_update = 1;
2564 pages_skipped = wbc->pages_skipped;
2567 while (!ret && wbc->nr_to_write > 0) {
2570 * we insert one extent at a time. So we need
2571 * credit needed for single extent allocation.
2572 * journalled mode is currently not supported
2575 BUG_ON(ext4_should_journal_data(inode));
2576 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2578 /* start a new transaction*/
2579 handle = ext4_journal_start(inode, needed_blocks);
2580 if (IS_ERR(handle)) {
2581 ret = PTR_ERR(handle);
2582 printk(KERN_CRIT "%s: jbd2_start: "
2583 "%ld pages, ino %lu; err %d\n", __func__,
2584 wbc->nr_to_write, inode->i_ino, ret);
2586 goto out_writepages;
2590 * Now call __mpage_da_writepage to find the next
2591 * contiguous region of logical blocks that need
2592 * blocks to be allocated by ext4. We don't actually
2593 * submit the blocks for I/O here, even though
2594 * write_cache_pages thinks it will, and will set the
2595 * pages as clean for write before calling
2596 * __mpage_da_writepage().
2604 mpd.pages_written = 0;
2606 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2609 * If we have a contigous extent of pages and we
2610 * haven't done the I/O yet, map the blocks and submit
2613 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2614 if (mpage_da_map_blocks(&mpd) == 0)
2615 mpage_da_submit_io(&mpd);
2617 ret = MPAGE_DA_EXTENT_TAIL;
2619 wbc->nr_to_write -= mpd.pages_written;
2621 ext4_journal_stop(handle);
2623 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2624 /* commit the transaction which would
2625 * free blocks released in the transaction
2628 jbd2_journal_force_commit_nested(sbi->s_journal);
2629 wbc->pages_skipped = pages_skipped;
2631 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2633 * got one extent now try with
2636 pages_written += mpd.pages_written;
2637 wbc->pages_skipped = pages_skipped;
2640 } else if (wbc->nr_to_write)
2642 * There is no more writeout needed
2643 * or we requested for a noblocking writeout
2644 * and we found the device congested
2648 if (!io_done && !cycled) {
2651 wbc->range_start = index << PAGE_CACHE_SHIFT;
2652 wbc->range_end = mapping->writeback_index - 1;
2655 if (pages_skipped != wbc->pages_skipped)
2656 printk(KERN_EMERG "This should not happen leaving %s "
2657 "with nr_to_write = %ld ret = %d\n",
2658 __func__, wbc->nr_to_write, ret);
2661 index += pages_written;
2662 wbc->range_cyclic = range_cyclic;
2663 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2665 * set the writeback_index so that range_cyclic
2666 * mode will write it back later
2668 mapping->writeback_index = index;
2671 if (!no_nrwrite_index_update)
2672 wbc->no_nrwrite_index_update = 0;
2673 wbc->nr_to_write -= nr_to_writebump;
2674 trace_mark(ext4_da_writepage_result,
2675 "dev %s ino %lu ret %d pages_written %d "
2676 "pages_skipped %ld congestion %d "
2677 "more_io %d no_nrwrite_index_update %d",
2678 inode->i_sb->s_id, inode->i_ino, ret,
2679 pages_written, wbc->pages_skipped,
2680 wbc->encountered_congestion, wbc->more_io,
2681 wbc->no_nrwrite_index_update);
2685 #define FALL_BACK_TO_NONDELALLOC 1
2686 static int ext4_nonda_switch(struct super_block *sb)
2688 s64 free_blocks, dirty_blocks;
2689 struct ext4_sb_info *sbi = EXT4_SB(sb);
2692 * switch to non delalloc mode if we are running low
2693 * on free block. The free block accounting via percpu
2694 * counters can get slightly wrong with percpu_counter_batch getting
2695 * accumulated on each CPU without updating global counters
2696 * Delalloc need an accurate free block accounting. So switch
2697 * to non delalloc when we are near to error range.
2699 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2700 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2701 if (2 * free_blocks < 3 * dirty_blocks ||
2702 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2704 * free block count is less that 150% of dirty blocks
2705 * or free blocks is less that watermark
2712 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2713 loff_t pos, unsigned len, unsigned flags,
2714 struct page **pagep, void **fsdata)
2716 int ret, retries = 0;
2720 struct inode *inode = mapping->host;
2723 index = pos >> PAGE_CACHE_SHIFT;
2724 from = pos & (PAGE_CACHE_SIZE - 1);
2727 if (ext4_nonda_switch(inode->i_sb)) {
2728 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2729 return ext4_write_begin(file, mapping, pos,
2730 len, flags, pagep, fsdata);
2732 *fsdata = (void *)0;
2734 trace_mark(ext4_da_write_begin,
2735 "dev %s ino %lu pos %llu len %u flags %u",
2736 inode->i_sb->s_id, inode->i_ino,
2737 (unsigned long long) pos, len, flags);
2740 * With delayed allocation, we don't log the i_disksize update
2741 * if there is delayed block allocation. But we still need
2742 * to journalling the i_disksize update if writes to the end
2743 * of file which has an already mapped buffer.
2745 handle = ext4_journal_start(inode, 1);
2746 if (IS_ERR(handle)) {
2747 ret = PTR_ERR(handle);
2750 /* We cannot recurse into the filesystem as the transaction is already
2752 flags |= AOP_FLAG_NOFS;
2754 page = grab_cache_page_write_begin(mapping, index, flags);
2756 ext4_journal_stop(handle);
2762 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2763 ext4_da_get_block_prep);
2766 ext4_journal_stop(handle);
2767 page_cache_release(page);
2769 * block_write_begin may have instantiated a few blocks
2770 * outside i_size. Trim these off again. Don't need
2771 * i_size_read because we hold i_mutex.
2773 if (pos + len > inode->i_size)
2774 vmtruncate(inode, inode->i_size);
2777 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2784 * Check if we should update i_disksize
2785 * when write to the end of file but not require block allocation
2787 static int ext4_da_should_update_i_disksize(struct page *page,
2788 unsigned long offset)
2790 struct buffer_head *bh;
2791 struct inode *inode = page->mapping->host;
2795 bh = page_buffers(page);
2796 idx = offset >> inode->i_blkbits;
2798 for (i = 0; i < idx; i++)
2799 bh = bh->b_this_page;
2801 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2806 static int ext4_da_write_end(struct file *file,
2807 struct address_space *mapping,
2808 loff_t pos, unsigned len, unsigned copied,
2809 struct page *page, void *fsdata)
2811 struct inode *inode = mapping->host;
2813 handle_t *handle = ext4_journal_current_handle();
2815 unsigned long start, end;
2816 int write_mode = (int)(unsigned long)fsdata;
2818 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2819 if (ext4_should_order_data(inode)) {
2820 return ext4_ordered_write_end(file, mapping, pos,
2821 len, copied, page, fsdata);
2822 } else if (ext4_should_writeback_data(inode)) {
2823 return ext4_writeback_write_end(file, mapping, pos,
2824 len, copied, page, fsdata);
2830 trace_mark(ext4_da_write_end,
2831 "dev %s ino %lu pos %llu len %u copied %u",
2832 inode->i_sb->s_id, inode->i_ino,
2833 (unsigned long long) pos, len, copied);
2834 start = pos & (PAGE_CACHE_SIZE - 1);
2835 end = start + copied - 1;
2838 * generic_write_end() will run mark_inode_dirty() if i_size
2839 * changes. So let's piggyback the i_disksize mark_inode_dirty
2843 new_i_size = pos + copied;
2844 if (new_i_size > EXT4_I(inode)->i_disksize) {
2845 if (ext4_da_should_update_i_disksize(page, end)) {
2846 down_write(&EXT4_I(inode)->i_data_sem);
2847 if (new_i_size > EXT4_I(inode)->i_disksize) {
2849 * Updating i_disksize when extending file
2850 * without needing block allocation
2852 if (ext4_should_order_data(inode))
2853 ret = ext4_jbd2_file_inode(handle,
2856 EXT4_I(inode)->i_disksize = new_i_size;
2858 up_write(&EXT4_I(inode)->i_data_sem);
2859 /* We need to mark inode dirty even if
2860 * new_i_size is less that inode->i_size
2861 * bu greater than i_disksize.(hint delalloc)
2863 ext4_mark_inode_dirty(handle, inode);
2866 ret2 = generic_write_end(file, mapping, pos, len, copied,
2871 ret2 = ext4_journal_stop(handle);
2875 return ret ? ret : copied;
2878 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2881 * Drop reserved blocks
2883 BUG_ON(!PageLocked(page));
2884 if (!page_has_buffers(page))
2887 ext4_da_page_release_reservation(page, offset);
2890 ext4_invalidatepage(page, offset);
2896 * Force all delayed allocation blocks to be allocated for a given inode.
2898 int ext4_alloc_da_blocks(struct inode *inode)
2900 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2901 !EXT4_I(inode)->i_reserved_meta_blocks)
2905 * We do something simple for now. The filemap_flush() will
2906 * also start triggering a write of the data blocks, which is
2907 * not strictly speaking necessary (and for users of
2908 * laptop_mode, not even desirable). However, to do otherwise
2909 * would require replicating code paths in:
2911 * ext4_da_writepages() ->
2912 * write_cache_pages() ---> (via passed in callback function)
2913 * __mpage_da_writepage() -->
2914 * mpage_add_bh_to_extent()
2915 * mpage_da_map_blocks()
2917 * The problem is that write_cache_pages(), located in
2918 * mm/page-writeback.c, marks pages clean in preparation for
2919 * doing I/O, which is not desirable if we're not planning on
2922 * We could call write_cache_pages(), and then redirty all of
2923 * the pages by calling redirty_page_for_writeback() but that
2924 * would be ugly in the extreme. So instead we would need to
2925 * replicate parts of the code in the above functions,
2926 * simplifying them becuase we wouldn't actually intend to
2927 * write out the pages, but rather only collect contiguous
2928 * logical block extents, call the multi-block allocator, and
2929 * then update the buffer heads with the block allocations.
2931 * For now, though, we'll cheat by calling filemap_flush(),
2932 * which will map the blocks, and start the I/O, but not
2933 * actually wait for the I/O to complete.
2935 return filemap_flush(inode->i_mapping);
2939 * bmap() is special. It gets used by applications such as lilo and by
2940 * the swapper to find the on-disk block of a specific piece of data.
2942 * Naturally, this is dangerous if the block concerned is still in the
2943 * journal. If somebody makes a swapfile on an ext4 data-journaling
2944 * filesystem and enables swap, then they may get a nasty shock when the
2945 * data getting swapped to that swapfile suddenly gets overwritten by
2946 * the original zero's written out previously to the journal and
2947 * awaiting writeback in the kernel's buffer cache.
2949 * So, if we see any bmap calls here on a modified, data-journaled file,
2950 * take extra steps to flush any blocks which might be in the cache.
2952 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2954 struct inode *inode = mapping->host;
2958 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2959 test_opt(inode->i_sb, DELALLOC)) {
2961 * With delalloc we want to sync the file
2962 * so that we can make sure we allocate
2965 filemap_write_and_wait(mapping);
2968 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2970 * This is a REALLY heavyweight approach, but the use of
2971 * bmap on dirty files is expected to be extremely rare:
2972 * only if we run lilo or swapon on a freshly made file
2973 * do we expect this to happen.
2975 * (bmap requires CAP_SYS_RAWIO so this does not
2976 * represent an unprivileged user DOS attack --- we'd be
2977 * in trouble if mortal users could trigger this path at
2980 * NB. EXT4_STATE_JDATA is not set on files other than
2981 * regular files. If somebody wants to bmap a directory
2982 * or symlink and gets confused because the buffer
2983 * hasn't yet been flushed to disk, they deserve
2984 * everything they get.
2987 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2988 journal = EXT4_JOURNAL(inode);
2989 jbd2_journal_lock_updates(journal);
2990 err = jbd2_journal_flush(journal);
2991 jbd2_journal_unlock_updates(journal);
2997 return generic_block_bmap(mapping, block, ext4_get_block);
3000 static int bget_one(handle_t *handle, struct buffer_head *bh)
3006 static int bput_one(handle_t *handle, struct buffer_head *bh)
3013 * Note that we don't need to start a transaction unless we're journaling data
3014 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3015 * need to file the inode to the transaction's list in ordered mode because if
3016 * we are writing back data added by write(), the inode is already there and if
3017 * we are writing back data modified via mmap(), noone guarantees in which
3018 * transaction the data will hit the disk. In case we are journaling data, we
3019 * cannot start transaction directly because transaction start ranks above page
3020 * lock so we have to do some magic.
3022 * In all journaling modes block_write_full_page() will start the I/O.
3026 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3031 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3033 * Same applies to ext4_get_block(). We will deadlock on various things like
3034 * lock_journal and i_data_sem
3036 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3039 * 16May01: If we're reentered then journal_current_handle() will be
3040 * non-zero. We simply *return*.
3042 * 1 July 2001: @@@ FIXME:
3043 * In journalled data mode, a data buffer may be metadata against the
3044 * current transaction. But the same file is part of a shared mapping
3045 * and someone does a writepage() on it.
3047 * We will move the buffer onto the async_data list, but *after* it has
3048 * been dirtied. So there's a small window where we have dirty data on
3051 * Note that this only applies to the last partial page in the file. The
3052 * bit which block_write_full_page() uses prepare/commit for. (That's
3053 * broken code anyway: it's wrong for msync()).
3055 * It's a rare case: affects the final partial page, for journalled data
3056 * where the file is subject to bith write() and writepage() in the same
3057 * transction. To fix it we'll need a custom block_write_full_page().
3058 * We'll probably need that anyway for journalling writepage() output.
3060 * We don't honour synchronous mounts for writepage(). That would be
3061 * disastrous. Any write() or metadata operation will sync the fs for
3065 static int __ext4_normal_writepage(struct page *page,
3066 struct writeback_control *wbc)
3068 struct inode *inode = page->mapping->host;
3070 if (test_opt(inode->i_sb, NOBH))
3071 return nobh_writepage(page,
3072 ext4_normal_get_block_write, wbc);
3074 return block_write_full_page(page,
3075 ext4_normal_get_block_write,
3079 static int ext4_normal_writepage(struct page *page,
3080 struct writeback_control *wbc)
3082 struct inode *inode = page->mapping->host;
3083 loff_t size = i_size_read(inode);
3086 trace_mark(ext4_normal_writepage,
3087 "dev %s ino %lu page_index %lu",
3088 inode->i_sb->s_id, inode->i_ino, page->index);
3089 J_ASSERT(PageLocked(page));
3090 if (page->index == size >> PAGE_CACHE_SHIFT)
3091 len = size & ~PAGE_CACHE_MASK;
3093 len = PAGE_CACHE_SIZE;
3095 if (page_has_buffers(page)) {
3096 /* if page has buffers it should all be mapped
3097 * and allocated. If there are not buffers attached
3098 * to the page we know the page is dirty but it lost
3099 * buffers. That means that at some moment in time
3100 * after write_begin() / write_end() has been called
3101 * all buffers have been clean and thus they must have been
3102 * written at least once. So they are all mapped and we can
3103 * happily proceed with mapping them and writing the page.
3105 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3106 ext4_bh_unmapped_or_delay));
3109 if (!ext4_journal_current_handle())
3110 return __ext4_normal_writepage(page, wbc);
3112 redirty_page_for_writepage(wbc, page);
3117 static int __ext4_journalled_writepage(struct page *page,
3118 struct writeback_control *wbc)
3120 struct address_space *mapping = page->mapping;
3121 struct inode *inode = mapping->host;
3122 struct buffer_head *page_bufs;
3123 handle_t *handle = NULL;
3127 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3128 ext4_normal_get_block_write);
3132 page_bufs = page_buffers(page);
3133 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3135 /* As soon as we unlock the page, it can go away, but we have
3136 * references to buffers so we are safe */
3139 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3140 if (IS_ERR(handle)) {
3141 ret = PTR_ERR(handle);
3145 ret = walk_page_buffers(handle, page_bufs, 0,
3146 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3148 err = walk_page_buffers(handle, page_bufs, 0,
3149 PAGE_CACHE_SIZE, NULL, write_end_fn);
3152 err = ext4_journal_stop(handle);
3156 walk_page_buffers(handle, page_bufs, 0,
3157 PAGE_CACHE_SIZE, NULL, bput_one);
3158 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3167 static int ext4_journalled_writepage(struct page *page,
3168 struct writeback_control *wbc)
3170 struct inode *inode = page->mapping->host;
3171 loff_t size = i_size_read(inode);
3174 trace_mark(ext4_journalled_writepage,
3175 "dev %s ino %lu page_index %lu",
3176 inode->i_sb->s_id, inode->i_ino, page->index);
3177 J_ASSERT(PageLocked(page));
3178 if (page->index == size >> PAGE_CACHE_SHIFT)
3179 len = size & ~PAGE_CACHE_MASK;
3181 len = PAGE_CACHE_SIZE;
3183 if (page_has_buffers(page)) {
3184 /* if page has buffers it should all be mapped
3185 * and allocated. If there are not buffers attached
3186 * to the page we know the page is dirty but it lost
3187 * buffers. That means that at some moment in time
3188 * after write_begin() / write_end() has been called
3189 * all buffers have been clean and thus they must have been
3190 * written at least once. So they are all mapped and we can
3191 * happily proceed with mapping them and writing the page.
3193 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3194 ext4_bh_unmapped_or_delay));
3197 if (ext4_journal_current_handle())
3200 if (PageChecked(page)) {
3202 * It's mmapped pagecache. Add buffers and journal it. There
3203 * doesn't seem much point in redirtying the page here.
3205 ClearPageChecked(page);
3206 return __ext4_journalled_writepage(page, wbc);
3209 * It may be a page full of checkpoint-mode buffers. We don't
3210 * really know unless we go poke around in the buffer_heads.
3211 * But block_write_full_page will do the right thing.
3213 return block_write_full_page(page,
3214 ext4_normal_get_block_write,
3218 redirty_page_for_writepage(wbc, page);
3223 static int ext4_readpage(struct file *file, struct page *page)
3225 return mpage_readpage(page, ext4_get_block);
3229 ext4_readpages(struct file *file, struct address_space *mapping,
3230 struct list_head *pages, unsigned nr_pages)
3232 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3235 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3237 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3240 * If it's a full truncate we just forget about the pending dirtying
3243 ClearPageChecked(page);
3246 jbd2_journal_invalidatepage(journal, page, offset);
3248 block_invalidatepage(page, offset);
3251 static int ext4_releasepage(struct page *page, gfp_t wait)
3253 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3255 WARN_ON(PageChecked(page));
3256 if (!page_has_buffers(page))
3259 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3261 return try_to_free_buffers(page);
3265 * If the O_DIRECT write will extend the file then add this inode to the
3266 * orphan list. So recovery will truncate it back to the original size
3267 * if the machine crashes during the write.
3269 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3270 * crashes then stale disk data _may_ be exposed inside the file. But current
3271 * VFS code falls back into buffered path in that case so we are safe.
3273 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3274 const struct iovec *iov, loff_t offset,
3275 unsigned long nr_segs)
3277 struct file *file = iocb->ki_filp;
3278 struct inode *inode = file->f_mapping->host;
3279 struct ext4_inode_info *ei = EXT4_I(inode);
3283 size_t count = iov_length(iov, nr_segs);
3286 loff_t final_size = offset + count;
3288 if (final_size > inode->i_size) {
3289 /* Credits for sb + inode write */
3290 handle = ext4_journal_start(inode, 2);
3291 if (IS_ERR(handle)) {
3292 ret = PTR_ERR(handle);
3295 ret = ext4_orphan_add(handle, inode);
3297 ext4_journal_stop(handle);
3301 ei->i_disksize = inode->i_size;
3302 ext4_journal_stop(handle);
3306 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3308 ext4_get_block, NULL);
3313 /* Credits for sb + inode write */
3314 handle = ext4_journal_start(inode, 2);
3315 if (IS_ERR(handle)) {
3316 /* This is really bad luck. We've written the data
3317 * but cannot extend i_size. Bail out and pretend
3318 * the write failed... */
3319 ret = PTR_ERR(handle);
3323 ext4_orphan_del(handle, inode);
3325 loff_t end = offset + ret;
3326 if (end > inode->i_size) {
3327 ei->i_disksize = end;
3328 i_size_write(inode, end);
3330 * We're going to return a positive `ret'
3331 * here due to non-zero-length I/O, so there's
3332 * no way of reporting error returns from
3333 * ext4_mark_inode_dirty() to userspace. So
3336 ext4_mark_inode_dirty(handle, inode);
3339 err = ext4_journal_stop(handle);
3348 * Pages can be marked dirty completely asynchronously from ext4's journalling
3349 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3350 * much here because ->set_page_dirty is called under VFS locks. The page is
3351 * not necessarily locked.
3353 * We cannot just dirty the page and leave attached buffers clean, because the
3354 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3355 * or jbddirty because all the journalling code will explode.
3357 * So what we do is to mark the page "pending dirty" and next time writepage
3358 * is called, propagate that into the buffers appropriately.
3360 static int ext4_journalled_set_page_dirty(struct page *page)
3362 SetPageChecked(page);
3363 return __set_page_dirty_nobuffers(page);
3366 static const struct address_space_operations ext4_ordered_aops = {
3367 .readpage = ext4_readpage,
3368 .readpages = ext4_readpages,
3369 .writepage = ext4_normal_writepage,
3370 .sync_page = block_sync_page,
3371 .write_begin = ext4_write_begin,
3372 .write_end = ext4_ordered_write_end,
3374 .invalidatepage = ext4_invalidatepage,
3375 .releasepage = ext4_releasepage,
3376 .direct_IO = ext4_direct_IO,
3377 .migratepage = buffer_migrate_page,
3378 .is_partially_uptodate = block_is_partially_uptodate,
3381 static const struct address_space_operations ext4_writeback_aops = {
3382 .readpage = ext4_readpage,
3383 .readpages = ext4_readpages,
3384 .writepage = ext4_normal_writepage,
3385 .sync_page = block_sync_page,
3386 .write_begin = ext4_write_begin,
3387 .write_end = ext4_writeback_write_end,
3389 .invalidatepage = ext4_invalidatepage,
3390 .releasepage = ext4_releasepage,
3391 .direct_IO = ext4_direct_IO,
3392 .migratepage = buffer_migrate_page,
3393 .is_partially_uptodate = block_is_partially_uptodate,
3396 static const struct address_space_operations ext4_journalled_aops = {
3397 .readpage = ext4_readpage,
3398 .readpages = ext4_readpages,
3399 .writepage = ext4_journalled_writepage,
3400 .sync_page = block_sync_page,
3401 .write_begin = ext4_write_begin,
3402 .write_end = ext4_journalled_write_end,
3403 .set_page_dirty = ext4_journalled_set_page_dirty,
3405 .invalidatepage = ext4_invalidatepage,
3406 .releasepage = ext4_releasepage,
3407 .is_partially_uptodate = block_is_partially_uptodate,
3410 static const struct address_space_operations ext4_da_aops = {
3411 .readpage = ext4_readpage,
3412 .readpages = ext4_readpages,
3413 .writepage = ext4_da_writepage,
3414 .writepages = ext4_da_writepages,
3415 .sync_page = block_sync_page,
3416 .write_begin = ext4_da_write_begin,
3417 .write_end = ext4_da_write_end,
3419 .invalidatepage = ext4_da_invalidatepage,
3420 .releasepage = ext4_releasepage,
3421 .direct_IO = ext4_direct_IO,
3422 .migratepage = buffer_migrate_page,
3423 .is_partially_uptodate = block_is_partially_uptodate,
3426 void ext4_set_aops(struct inode *inode)
3428 if (ext4_should_order_data(inode) &&
3429 test_opt(inode->i_sb, DELALLOC))
3430 inode->i_mapping->a_ops = &ext4_da_aops;
3431 else if (ext4_should_order_data(inode))
3432 inode->i_mapping->a_ops = &ext4_ordered_aops;
3433 else if (ext4_should_writeback_data(inode) &&
3434 test_opt(inode->i_sb, DELALLOC))
3435 inode->i_mapping->a_ops = &ext4_da_aops;
3436 else if (ext4_should_writeback_data(inode))
3437 inode->i_mapping->a_ops = &ext4_writeback_aops;
3439 inode->i_mapping->a_ops = &ext4_journalled_aops;
3443 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3444 * up to the end of the block which corresponds to `from'.
3445 * This required during truncate. We need to physically zero the tail end
3446 * of that block so it doesn't yield old data if the file is later grown.
3448 int ext4_block_truncate_page(handle_t *handle,
3449 struct address_space *mapping, loff_t from)
3451 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3452 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3453 unsigned blocksize, length, pos;
3455 struct inode *inode = mapping->host;
3456 struct buffer_head *bh;
3460 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3464 blocksize = inode->i_sb->s_blocksize;
3465 length = blocksize - (offset & (blocksize - 1));
3466 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3469 * For "nobh" option, we can only work if we don't need to
3470 * read-in the page - otherwise we create buffers to do the IO.
3472 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3473 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3474 zero_user(page, offset, length);
3475 set_page_dirty(page);
3479 if (!page_has_buffers(page))
3480 create_empty_buffers(page, blocksize, 0);
3482 /* Find the buffer that contains "offset" */
3483 bh = page_buffers(page);
3485 while (offset >= pos) {
3486 bh = bh->b_this_page;
3492 if (buffer_freed(bh)) {
3493 BUFFER_TRACE(bh, "freed: skip");
3497 if (!buffer_mapped(bh)) {
3498 BUFFER_TRACE(bh, "unmapped");
3499 ext4_get_block(inode, iblock, bh, 0);
3500 /* unmapped? It's a hole - nothing to do */
3501 if (!buffer_mapped(bh)) {
3502 BUFFER_TRACE(bh, "still unmapped");
3507 /* Ok, it's mapped. Make sure it's up-to-date */
3508 if (PageUptodate(page))
3509 set_buffer_uptodate(bh);
3511 if (!buffer_uptodate(bh)) {
3513 ll_rw_block(READ, 1, &bh);
3515 /* Uhhuh. Read error. Complain and punt. */
3516 if (!buffer_uptodate(bh))
3520 if (ext4_should_journal_data(inode)) {
3521 BUFFER_TRACE(bh, "get write access");
3522 err = ext4_journal_get_write_access(handle, bh);
3527 zero_user(page, offset, length);
3529 BUFFER_TRACE(bh, "zeroed end of block");
3532 if (ext4_should_journal_data(inode)) {
3533 err = ext4_handle_dirty_metadata(handle, inode, bh);
3535 if (ext4_should_order_data(inode))
3536 err = ext4_jbd2_file_inode(handle, inode);
3537 mark_buffer_dirty(bh);
3542 page_cache_release(page);
3547 * Probably it should be a library function... search for first non-zero word
3548 * or memcmp with zero_page, whatever is better for particular architecture.
3551 static inline int all_zeroes(__le32 *p, __le32 *q)
3560 * ext4_find_shared - find the indirect blocks for partial truncation.
3561 * @inode: inode in question
3562 * @depth: depth of the affected branch
3563 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3564 * @chain: place to store the pointers to partial indirect blocks
3565 * @top: place to the (detached) top of branch
3567 * This is a helper function used by ext4_truncate().
3569 * When we do truncate() we may have to clean the ends of several
3570 * indirect blocks but leave the blocks themselves alive. Block is
3571 * partially truncated if some data below the new i_size is refered
3572 * from it (and it is on the path to the first completely truncated
3573 * data block, indeed). We have to free the top of that path along
3574 * with everything to the right of the path. Since no allocation
3575 * past the truncation point is possible until ext4_truncate()
3576 * finishes, we may safely do the latter, but top of branch may
3577 * require special attention - pageout below the truncation point
3578 * might try to populate it.
3580 * We atomically detach the top of branch from the tree, store the
3581 * block number of its root in *@top, pointers to buffer_heads of
3582 * partially truncated blocks - in @chain[].bh and pointers to
3583 * their last elements that should not be removed - in
3584 * @chain[].p. Return value is the pointer to last filled element
3587 * The work left to caller to do the actual freeing of subtrees:
3588 * a) free the subtree starting from *@top
3589 * b) free the subtrees whose roots are stored in
3590 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3591 * c) free the subtrees growing from the inode past the @chain[0].
3592 * (no partially truncated stuff there). */
3594 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3595 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3597 Indirect *partial, *p;
3601 /* Make k index the deepest non-null offest + 1 */
3602 for (k = depth; k > 1 && !offsets[k-1]; k--)
3604 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3605 /* Writer: pointers */
3607 partial = chain + k-1;
3609 * If the branch acquired continuation since we've looked at it -
3610 * fine, it should all survive and (new) top doesn't belong to us.
3612 if (!partial->key && *partial->p)
3615 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3618 * OK, we've found the last block that must survive. The rest of our
3619 * branch should be detached before unlocking. However, if that rest
3620 * of branch is all ours and does not grow immediately from the inode
3621 * it's easier to cheat and just decrement partial->p.
3623 if (p == chain + k - 1 && p > chain) {
3627 /* Nope, don't do this in ext4. Must leave the tree intact */
3634 while (partial > p) {
3635 brelse(partial->bh);
3643 * Zero a number of block pointers in either an inode or an indirect block.
3644 * If we restart the transaction we must again get write access to the
3645 * indirect block for further modification.
3647 * We release `count' blocks on disk, but (last - first) may be greater
3648 * than `count' because there can be holes in there.
3650 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3651 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3652 unsigned long count, __le32 *first, __le32 *last)
3655 if (try_to_extend_transaction(handle, inode)) {
3657 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3658 ext4_handle_dirty_metadata(handle, inode, bh);
3660 ext4_mark_inode_dirty(handle, inode);
3661 ext4_journal_test_restart(handle, inode);
3663 BUFFER_TRACE(bh, "retaking write access");
3664 ext4_journal_get_write_access(handle, bh);
3669 * Any buffers which are on the journal will be in memory. We find
3670 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3671 * on them. We've already detached each block from the file, so
3672 * bforget() in jbd2_journal_forget() should be safe.
3674 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3676 for (p = first; p < last; p++) {
3677 u32 nr = le32_to_cpu(*p);
3679 struct buffer_head *tbh;
3682 tbh = sb_find_get_block(inode->i_sb, nr);
3683 ext4_forget(handle, 0, inode, tbh, nr);
3687 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3691 * ext4_free_data - free a list of data blocks
3692 * @handle: handle for this transaction
3693 * @inode: inode we are dealing with
3694 * @this_bh: indirect buffer_head which contains *@first and *@last
3695 * @first: array of block numbers
3696 * @last: points immediately past the end of array
3698 * We are freeing all blocks refered from that array (numbers are stored as
3699 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3701 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3702 * blocks are contiguous then releasing them at one time will only affect one
3703 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3704 * actually use a lot of journal space.
3706 * @this_bh will be %NULL if @first and @last point into the inode's direct
3709 static void ext4_free_data(handle_t *handle, struct inode *inode,
3710 struct buffer_head *this_bh,
3711 __le32 *first, __le32 *last)
3713 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3714 unsigned long count = 0; /* Number of blocks in the run */
3715 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3718 ext4_fsblk_t nr; /* Current block # */
3719 __le32 *p; /* Pointer into inode/ind
3720 for current block */
3723 if (this_bh) { /* For indirect block */
3724 BUFFER_TRACE(this_bh, "get_write_access");
3725 err = ext4_journal_get_write_access(handle, this_bh);
3726 /* Important: if we can't update the indirect pointers
3727 * to the blocks, we can't free them. */
3732 for (p = first; p < last; p++) {
3733 nr = le32_to_cpu(*p);
3735 /* accumulate blocks to free if they're contiguous */
3738 block_to_free_p = p;
3740 } else if (nr == block_to_free + count) {
3743 ext4_clear_blocks(handle, inode, this_bh,
3745 count, block_to_free_p, p);
3747 block_to_free_p = p;
3754 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3755 count, block_to_free_p, p);
3758 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3761 * The buffer head should have an attached journal head at this
3762 * point. However, if the data is corrupted and an indirect
3763 * block pointed to itself, it would have been detached when
3764 * the block was cleared. Check for this instead of OOPSing.
3766 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3767 ext4_handle_dirty_metadata(handle, inode, this_bh);
3769 ext4_error(inode->i_sb, __func__,
3770 "circular indirect block detected, "
3771 "inode=%lu, block=%llu",
3773 (unsigned long long) this_bh->b_blocknr);
3778 * ext4_free_branches - free an array of branches
3779 * @handle: JBD handle for this transaction
3780 * @inode: inode we are dealing with
3781 * @parent_bh: the buffer_head which contains *@first and *@last
3782 * @first: array of block numbers
3783 * @last: pointer immediately past the end of array
3784 * @depth: depth of the branches to free
3786 * We are freeing all blocks refered from these branches (numbers are
3787 * stored as little-endian 32-bit) and updating @inode->i_blocks
3790 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3791 struct buffer_head *parent_bh,
3792 __le32 *first, __le32 *last, int depth)
3797 if (ext4_handle_is_aborted(handle))
3801 struct buffer_head *bh;
3802 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3804 while (--p >= first) {
3805 nr = le32_to_cpu(*p);
3807 continue; /* A hole */
3809 /* Go read the buffer for the next level down */
3810 bh = sb_bread(inode->i_sb, nr);
3813 * A read failure? Report error and clear slot
3817 ext4_error(inode->i_sb, "ext4_free_branches",
3818 "Read failure, inode=%lu, block=%llu",
3823 /* This zaps the entire block. Bottom up. */
3824 BUFFER_TRACE(bh, "free child branches");
3825 ext4_free_branches(handle, inode, bh,
3826 (__le32 *) bh->b_data,
3827 (__le32 *) bh->b_data + addr_per_block,
3831 * We've probably journalled the indirect block several
3832 * times during the truncate. But it's no longer
3833 * needed and we now drop it from the transaction via
3834 * jbd2_journal_revoke().
3836 * That's easy if it's exclusively part of this
3837 * transaction. But if it's part of the committing
3838 * transaction then jbd2_journal_forget() will simply
3839 * brelse() it. That means that if the underlying
3840 * block is reallocated in ext4_get_block(),
3841 * unmap_underlying_metadata() will find this block
3842 * and will try to get rid of it. damn, damn.
3844 * If this block has already been committed to the
3845 * journal, a revoke record will be written. And
3846 * revoke records must be emitted *before* clearing
3847 * this block's bit in the bitmaps.
3849 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3852 * Everything below this this pointer has been
3853 * released. Now let this top-of-subtree go.
3855 * We want the freeing of this indirect block to be
3856 * atomic in the journal with the updating of the
3857 * bitmap block which owns it. So make some room in
3860 * We zero the parent pointer *after* freeing its
3861 * pointee in the bitmaps, so if extend_transaction()
3862 * for some reason fails to put the bitmap changes and
3863 * the release into the same transaction, recovery
3864 * will merely complain about releasing a free block,
3865 * rather than leaking blocks.
3867 if (ext4_handle_is_aborted(handle))
3869 if (try_to_extend_transaction(handle, inode)) {
3870 ext4_mark_inode_dirty(handle, inode);
3871 ext4_journal_test_restart(handle, inode);
3874 ext4_free_blocks(handle, inode, nr, 1, 1);
3878 * The block which we have just freed is
3879 * pointed to by an indirect block: journal it
3881 BUFFER_TRACE(parent_bh, "get_write_access");
3882 if (!ext4_journal_get_write_access(handle,
3885 BUFFER_TRACE(parent_bh,
3886 "call ext4_handle_dirty_metadata");
3887 ext4_handle_dirty_metadata(handle,
3894 /* We have reached the bottom of the tree. */
3895 BUFFER_TRACE(parent_bh, "free data blocks");
3896 ext4_free_data(handle, inode, parent_bh, first, last);
3900 int ext4_can_truncate(struct inode *inode)
3902 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3904 if (S_ISREG(inode->i_mode))
3906 if (S_ISDIR(inode->i_mode))
3908 if (S_ISLNK(inode->i_mode))
3909 return !ext4_inode_is_fast_symlink(inode);
3916 * We block out ext4_get_block() block instantiations across the entire
3917 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3918 * simultaneously on behalf of the same inode.
3920 * As we work through the truncate and commmit bits of it to the journal there
3921 * is one core, guiding principle: the file's tree must always be consistent on
3922 * disk. We must be able to restart the truncate after a crash.
3924 * The file's tree may be transiently inconsistent in memory (although it
3925 * probably isn't), but whenever we close off and commit a journal transaction,
3926 * the contents of (the filesystem + the journal) must be consistent and
3927 * restartable. It's pretty simple, really: bottom up, right to left (although
3928 * left-to-right works OK too).
3930 * Note that at recovery time, journal replay occurs *before* the restart of
3931 * truncate against the orphan inode list.
3933 * The committed inode has the new, desired i_size (which is the same as
3934 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3935 * that this inode's truncate did not complete and it will again call
3936 * ext4_truncate() to have another go. So there will be instantiated blocks
3937 * to the right of the truncation point in a crashed ext4 filesystem. But
3938 * that's fine - as long as they are linked from the inode, the post-crash
3939 * ext4_truncate() run will find them and release them.
3941 void ext4_truncate(struct inode *inode)
3944 struct ext4_inode_info *ei = EXT4_I(inode);
3945 __le32 *i_data = ei->i_data;
3946 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3947 struct address_space *mapping = inode->i_mapping;
3948 ext4_lblk_t offsets[4];
3953 ext4_lblk_t last_block;
3954 unsigned blocksize = inode->i_sb->s_blocksize;
3956 if (!ext4_can_truncate(inode))
3959 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
3960 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
3962 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3963 ext4_ext_truncate(inode);
3967 handle = start_transaction(inode);
3969 return; /* AKPM: return what? */
3971 last_block = (inode->i_size + blocksize-1)
3972 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3974 if (inode->i_size & (blocksize - 1))
3975 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3978 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3980 goto out_stop; /* error */
3983 * OK. This truncate is going to happen. We add the inode to the
3984 * orphan list, so that if this truncate spans multiple transactions,
3985 * and we crash, we will resume the truncate when the filesystem
3986 * recovers. It also marks the inode dirty, to catch the new size.
3988 * Implication: the file must always be in a sane, consistent
3989 * truncatable state while each transaction commits.
3991 if (ext4_orphan_add(handle, inode))
3995 * From here we block out all ext4_get_block() callers who want to
3996 * modify the block allocation tree.
3998 down_write(&ei->i_data_sem);
4000 ext4_discard_preallocations(inode);
4003 * The orphan list entry will now protect us from any crash which
4004 * occurs before the truncate completes, so it is now safe to propagate
4005 * the new, shorter inode size (held for now in i_size) into the
4006 * on-disk inode. We do this via i_disksize, which is the value which
4007 * ext4 *really* writes onto the disk inode.
4009 ei->i_disksize = inode->i_size;
4011 if (n == 1) { /* direct blocks */
4012 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4013 i_data + EXT4_NDIR_BLOCKS);
4017 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4018 /* Kill the top of shared branch (not detached) */
4020 if (partial == chain) {
4021 /* Shared branch grows from the inode */
4022 ext4_free_branches(handle, inode, NULL,
4023 &nr, &nr+1, (chain+n-1) - partial);
4026 * We mark the inode dirty prior to restart,
4027 * and prior to stop. No need for it here.
4030 /* Shared branch grows from an indirect block */
4031 BUFFER_TRACE(partial->bh, "get_write_access");
4032 ext4_free_branches(handle, inode, partial->bh,
4034 partial->p+1, (chain+n-1) - partial);
4037 /* Clear the ends of indirect blocks on the shared branch */
4038 while (partial > chain) {
4039 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4040 (__le32*)partial->bh->b_data+addr_per_block,
4041 (chain+n-1) - partial);
4042 BUFFER_TRACE(partial->bh, "call brelse");
4043 brelse (partial->bh);
4047 /* Kill the remaining (whole) subtrees */
4048 switch (offsets[0]) {
4050 nr = i_data[EXT4_IND_BLOCK];
4052 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4053 i_data[EXT4_IND_BLOCK] = 0;
4055 case EXT4_IND_BLOCK:
4056 nr = i_data[EXT4_DIND_BLOCK];
4058 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4059 i_data[EXT4_DIND_BLOCK] = 0;
4061 case EXT4_DIND_BLOCK:
4062 nr = i_data[EXT4_TIND_BLOCK];
4064 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4065 i_data[EXT4_TIND_BLOCK] = 0;
4067 case EXT4_TIND_BLOCK:
4071 up_write(&ei->i_data_sem);
4072 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4073 ext4_mark_inode_dirty(handle, inode);
4076 * In a multi-transaction truncate, we only make the final transaction
4080 ext4_handle_sync(handle);
4083 * If this was a simple ftruncate(), and the file will remain alive
4084 * then we need to clear up the orphan record which we created above.
4085 * However, if this was a real unlink then we were called by
4086 * ext4_delete_inode(), and we allow that function to clean up the
4087 * orphan info for us.
4090 ext4_orphan_del(handle, inode);
4092 ext4_journal_stop(handle);
4096 * ext4_get_inode_loc returns with an extra refcount against the inode's
4097 * underlying buffer_head on success. If 'in_mem' is true, we have all
4098 * data in memory that is needed to recreate the on-disk version of this
4101 static int __ext4_get_inode_loc(struct inode *inode,
4102 struct ext4_iloc *iloc, int in_mem)
4104 struct ext4_group_desc *gdp;
4105 struct buffer_head *bh;
4106 struct super_block *sb = inode->i_sb;
4108 int inodes_per_block, inode_offset;
4111 if (!ext4_valid_inum(sb, inode->i_ino))
4114 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4115 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4120 * Figure out the offset within the block group inode table
4122 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4123 inode_offset = ((inode->i_ino - 1) %
4124 EXT4_INODES_PER_GROUP(sb));
4125 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4126 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4128 bh = sb_getblk(sb, block);
4130 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4131 "inode block - inode=%lu, block=%llu",
4132 inode->i_ino, block);
4135 if (!buffer_uptodate(bh)) {
4139 * If the buffer has the write error flag, we have failed
4140 * to write out another inode in the same block. In this
4141 * case, we don't have to read the block because we may
4142 * read the old inode data successfully.
4144 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4145 set_buffer_uptodate(bh);
4147 if (buffer_uptodate(bh)) {
4148 /* someone brought it uptodate while we waited */
4154 * If we have all information of the inode in memory and this
4155 * is the only valid inode in the block, we need not read the
4159 struct buffer_head *bitmap_bh;
4162 start = inode_offset & ~(inodes_per_block - 1);
4164 /* Is the inode bitmap in cache? */
4165 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4170 * If the inode bitmap isn't in cache then the
4171 * optimisation may end up performing two reads instead
4172 * of one, so skip it.
4174 if (!buffer_uptodate(bitmap_bh)) {
4178 for (i = start; i < start + inodes_per_block; i++) {
4179 if (i == inode_offset)
4181 if (ext4_test_bit(i, bitmap_bh->b_data))
4185 if (i == start + inodes_per_block) {
4186 /* all other inodes are free, so skip I/O */
4187 memset(bh->b_data, 0, bh->b_size);
4188 set_buffer_uptodate(bh);
4196 * If we need to do any I/O, try to pre-readahead extra
4197 * blocks from the inode table.
4199 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4200 ext4_fsblk_t b, end, table;
4203 table = ext4_inode_table(sb, gdp);
4204 /* s_inode_readahead_blks is always a power of 2 */
4205 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4208 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4209 num = EXT4_INODES_PER_GROUP(sb);
4210 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4211 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4212 num -= ext4_itable_unused_count(sb, gdp);
4213 table += num / inodes_per_block;
4217 sb_breadahead(sb, b++);
4221 * There are other valid inodes in the buffer, this inode
4222 * has in-inode xattrs, or we don't have this inode in memory.
4223 * Read the block from disk.
4226 bh->b_end_io = end_buffer_read_sync;
4227 submit_bh(READ_META, bh);
4229 if (!buffer_uptodate(bh)) {
4230 ext4_error(sb, __func__,
4231 "unable to read inode block - inode=%lu, "
4232 "block=%llu", inode->i_ino, block);
4242 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4244 /* We have all inode data except xattrs in memory here. */
4245 return __ext4_get_inode_loc(inode, iloc,
4246 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4249 void ext4_set_inode_flags(struct inode *inode)
4251 unsigned int flags = EXT4_I(inode)->i_flags;
4253 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4254 if (flags & EXT4_SYNC_FL)
4255 inode->i_flags |= S_SYNC;
4256 if (flags & EXT4_APPEND_FL)
4257 inode->i_flags |= S_APPEND;
4258 if (flags & EXT4_IMMUTABLE_FL)
4259 inode->i_flags |= S_IMMUTABLE;
4260 if (flags & EXT4_NOATIME_FL)
4261 inode->i_flags |= S_NOATIME;
4262 if (flags & EXT4_DIRSYNC_FL)
4263 inode->i_flags |= S_DIRSYNC;
4266 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4267 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4269 unsigned int flags = ei->vfs_inode.i_flags;
4271 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4272 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4274 ei->i_flags |= EXT4_SYNC_FL;
4275 if (flags & S_APPEND)
4276 ei->i_flags |= EXT4_APPEND_FL;
4277 if (flags & S_IMMUTABLE)
4278 ei->i_flags |= EXT4_IMMUTABLE_FL;
4279 if (flags & S_NOATIME)
4280 ei->i_flags |= EXT4_NOATIME_FL;
4281 if (flags & S_DIRSYNC)
4282 ei->i_flags |= EXT4_DIRSYNC_FL;
4284 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4285 struct ext4_inode_info *ei)
4288 struct inode *inode = &(ei->vfs_inode);
4289 struct super_block *sb = inode->i_sb;
4291 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4292 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4293 /* we are using combined 48 bit field */
4294 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4295 le32_to_cpu(raw_inode->i_blocks_lo);
4296 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4297 /* i_blocks represent file system block size */
4298 return i_blocks << (inode->i_blkbits - 9);
4303 return le32_to_cpu(raw_inode->i_blocks_lo);
4307 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4309 struct ext4_iloc iloc;
4310 struct ext4_inode *raw_inode;
4311 struct ext4_inode_info *ei;
4312 struct buffer_head *bh;
4313 struct inode *inode;
4317 inode = iget_locked(sb, ino);
4319 return ERR_PTR(-ENOMEM);
4320 if (!(inode->i_state & I_NEW))
4324 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4325 ei->i_acl = EXT4_ACL_NOT_CACHED;
4326 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4329 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4333 raw_inode = ext4_raw_inode(&iloc);
4334 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4335 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4336 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4337 if (!(test_opt(inode->i_sb, NO_UID32))) {
4338 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4339 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4341 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4344 ei->i_dir_start_lookup = 0;
4345 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4346 /* We now have enough fields to check if the inode was active or not.
4347 * This is needed because nfsd might try to access dead inodes
4348 * the test is that same one that e2fsck uses
4349 * NeilBrown 1999oct15
4351 if (inode->i_nlink == 0) {
4352 if (inode->i_mode == 0 ||
4353 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4354 /* this inode is deleted */
4359 /* The only unlinked inodes we let through here have
4360 * valid i_mode and are being read by the orphan
4361 * recovery code: that's fine, we're about to complete
4362 * the process of deleting those. */
4364 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4365 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4366 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4367 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4369 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4370 inode->i_size = ext4_isize(raw_inode);
4371 ei->i_disksize = inode->i_size;
4372 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4373 ei->i_block_group = iloc.block_group;
4374 ei->i_last_alloc_group = ~0;
4376 * NOTE! The in-memory inode i_data array is in little-endian order
4377 * even on big-endian machines: we do NOT byteswap the block numbers!
4379 for (block = 0; block < EXT4_N_BLOCKS; block++)
4380 ei->i_data[block] = raw_inode->i_block[block];
4381 INIT_LIST_HEAD(&ei->i_orphan);
4383 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4384 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4385 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4386 EXT4_INODE_SIZE(inode->i_sb)) {
4391 if (ei->i_extra_isize == 0) {
4392 /* The extra space is currently unused. Use it. */
4393 ei->i_extra_isize = sizeof(struct ext4_inode) -
4394 EXT4_GOOD_OLD_INODE_SIZE;
4396 __le32 *magic = (void *)raw_inode +
4397 EXT4_GOOD_OLD_INODE_SIZE +
4399 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4400 ei->i_state |= EXT4_STATE_XATTR;
4403 ei->i_extra_isize = 0;
4405 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4406 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4407 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4408 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4410 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4411 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4412 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4414 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4418 if (ei->i_file_acl &&
4420 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4421 EXT4_SB(sb)->s_gdb_count)) ||
4422 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4423 ext4_error(sb, __func__,
4424 "bad extended attribute block %llu in inode #%lu",
4425 ei->i_file_acl, inode->i_ino);
4428 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4429 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4430 (S_ISLNK(inode->i_mode) &&
4431 !ext4_inode_is_fast_symlink(inode)))
4432 /* Validate extent which is part of inode */
4433 ret = ext4_ext_check_inode(inode);
4434 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4435 (S_ISLNK(inode->i_mode) &&
4436 !ext4_inode_is_fast_symlink(inode))) {
4437 /* Validate block references which are part of inode */
4438 ret = ext4_check_inode_blockref(inode);
4445 if (S_ISREG(inode->i_mode)) {
4446 inode->i_op = &ext4_file_inode_operations;
4447 inode->i_fop = &ext4_file_operations;
4448 ext4_set_aops(inode);
4449 } else if (S_ISDIR(inode->i_mode)) {
4450 inode->i_op = &ext4_dir_inode_operations;
4451 inode->i_fop = &ext4_dir_operations;
4452 } else if (S_ISLNK(inode->i_mode)) {
4453 if (ext4_inode_is_fast_symlink(inode)) {
4454 inode->i_op = &ext4_fast_symlink_inode_operations;
4455 nd_terminate_link(ei->i_data, inode->i_size,
4456 sizeof(ei->i_data) - 1);
4458 inode->i_op = &ext4_symlink_inode_operations;
4459 ext4_set_aops(inode);
4461 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4462 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4463 inode->i_op = &ext4_special_inode_operations;
4464 if (raw_inode->i_block[0])
4465 init_special_inode(inode, inode->i_mode,
4466 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4468 init_special_inode(inode, inode->i_mode,
4469 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4473 ext4_error(inode->i_sb, __func__,
4474 "bogus i_mode (%o) for inode=%lu",
4475 inode->i_mode, inode->i_ino);
4479 ext4_set_inode_flags(inode);
4480 unlock_new_inode(inode);
4485 return ERR_PTR(ret);
4488 static int ext4_inode_blocks_set(handle_t *handle,
4489 struct ext4_inode *raw_inode,
4490 struct ext4_inode_info *ei)
4492 struct inode *inode = &(ei->vfs_inode);
4493 u64 i_blocks = inode->i_blocks;
4494 struct super_block *sb = inode->i_sb;
4496 if (i_blocks <= ~0U) {
4498 * i_blocks can be represnted in a 32 bit variable
4499 * as multiple of 512 bytes
4501 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4502 raw_inode->i_blocks_high = 0;
4503 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4506 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4509 if (i_blocks <= 0xffffffffffffULL) {
4511 * i_blocks can be represented in a 48 bit variable
4512 * as multiple of 512 bytes
4514 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4515 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4516 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4518 ei->i_flags |= EXT4_HUGE_FILE_FL;
4519 /* i_block is stored in file system block size */
4520 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4521 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4522 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4528 * Post the struct inode info into an on-disk inode location in the
4529 * buffer-cache. This gobbles the caller's reference to the
4530 * buffer_head in the inode location struct.
4532 * The caller must have write access to iloc->bh.
4534 static int ext4_do_update_inode(handle_t *handle,
4535 struct inode *inode,
4536 struct ext4_iloc *iloc)
4538 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4539 struct ext4_inode_info *ei = EXT4_I(inode);
4540 struct buffer_head *bh = iloc->bh;
4541 int err = 0, rc, block;
4543 /* For fields not not tracking in the in-memory inode,
4544 * initialise them to zero for new inodes. */
4545 if (ei->i_state & EXT4_STATE_NEW)
4546 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4548 ext4_get_inode_flags(ei);
4549 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4550 if (!(test_opt(inode->i_sb, NO_UID32))) {
4551 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4552 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4554 * Fix up interoperability with old kernels. Otherwise, old inodes get
4555 * re-used with the upper 16 bits of the uid/gid intact
4558 raw_inode->i_uid_high =
4559 cpu_to_le16(high_16_bits(inode->i_uid));
4560 raw_inode->i_gid_high =
4561 cpu_to_le16(high_16_bits(inode->i_gid));
4563 raw_inode->i_uid_high = 0;
4564 raw_inode->i_gid_high = 0;
4567 raw_inode->i_uid_low =
4568 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4569 raw_inode->i_gid_low =
4570 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4571 raw_inode->i_uid_high = 0;
4572 raw_inode->i_gid_high = 0;
4574 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4576 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4577 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4578 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4579 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4581 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4583 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4584 /* clear the migrate flag in the raw_inode */
4585 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4586 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4587 cpu_to_le32(EXT4_OS_HURD))
4588 raw_inode->i_file_acl_high =
4589 cpu_to_le16(ei->i_file_acl >> 32);
4590 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4591 ext4_isize_set(raw_inode, ei->i_disksize);
4592 if (ei->i_disksize > 0x7fffffffULL) {
4593 struct super_block *sb = inode->i_sb;
4594 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4595 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4596 EXT4_SB(sb)->s_es->s_rev_level ==
4597 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4598 /* If this is the first large file
4599 * created, add a flag to the superblock.
4601 err = ext4_journal_get_write_access(handle,
4602 EXT4_SB(sb)->s_sbh);
4605 ext4_update_dynamic_rev(sb);
4606 EXT4_SET_RO_COMPAT_FEATURE(sb,
4607 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4609 ext4_handle_sync(handle);
4610 err = ext4_handle_dirty_metadata(handle, inode,
4611 EXT4_SB(sb)->s_sbh);
4614 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4615 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4616 if (old_valid_dev(inode->i_rdev)) {
4617 raw_inode->i_block[0] =
4618 cpu_to_le32(old_encode_dev(inode->i_rdev));
4619 raw_inode->i_block[1] = 0;
4621 raw_inode->i_block[0] = 0;
4622 raw_inode->i_block[1] =
4623 cpu_to_le32(new_encode_dev(inode->i_rdev));
4624 raw_inode->i_block[2] = 0;
4626 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4627 raw_inode->i_block[block] = ei->i_data[block];
4629 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4630 if (ei->i_extra_isize) {
4631 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4632 raw_inode->i_version_hi =
4633 cpu_to_le32(inode->i_version >> 32);
4634 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4637 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4638 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4641 ei->i_state &= ~EXT4_STATE_NEW;
4645 ext4_std_error(inode->i_sb, err);
4650 * ext4_write_inode()
4652 * We are called from a few places:
4654 * - Within generic_file_write() for O_SYNC files.
4655 * Here, there will be no transaction running. We wait for any running
4656 * trasnaction to commit.
4658 * - Within sys_sync(), kupdate and such.
4659 * We wait on commit, if tol to.
4661 * - Within prune_icache() (PF_MEMALLOC == true)
4662 * Here we simply return. We can't afford to block kswapd on the
4665 * In all cases it is actually safe for us to return without doing anything,
4666 * because the inode has been copied into a raw inode buffer in
4667 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4670 * Note that we are absolutely dependent upon all inode dirtiers doing the
4671 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4672 * which we are interested.
4674 * It would be a bug for them to not do this. The code:
4676 * mark_inode_dirty(inode)
4678 * inode->i_size = expr;
4680 * is in error because a kswapd-driven write_inode() could occur while
4681 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4682 * will no longer be on the superblock's dirty inode list.
4684 int ext4_write_inode(struct inode *inode, int wait)
4686 if (current->flags & PF_MEMALLOC)
4689 if (ext4_journal_current_handle()) {
4690 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4698 return ext4_force_commit(inode->i_sb);
4701 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4705 mark_buffer_dirty(bh);
4706 if (inode && inode_needs_sync(inode)) {
4707 sync_dirty_buffer(bh);
4708 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4709 ext4_error(inode->i_sb, __func__,
4710 "IO error syncing inode, "
4711 "inode=%lu, block=%llu",
4713 (unsigned long long)bh->b_blocknr);
4723 * Called from notify_change.
4725 * We want to trap VFS attempts to truncate the file as soon as
4726 * possible. In particular, we want to make sure that when the VFS
4727 * shrinks i_size, we put the inode on the orphan list and modify
4728 * i_disksize immediately, so that during the subsequent flushing of
4729 * dirty pages and freeing of disk blocks, we can guarantee that any
4730 * commit will leave the blocks being flushed in an unused state on
4731 * disk. (On recovery, the inode will get truncated and the blocks will
4732 * be freed, so we have a strong guarantee that no future commit will
4733 * leave these blocks visible to the user.)
4735 * Another thing we have to assure is that if we are in ordered mode
4736 * and inode is still attached to the committing transaction, we must
4737 * we start writeout of all the dirty pages which are being truncated.
4738 * This way we are sure that all the data written in the previous
4739 * transaction are already on disk (truncate waits for pages under
4742 * Called with inode->i_mutex down.
4744 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4746 struct inode *inode = dentry->d_inode;
4748 const unsigned int ia_valid = attr->ia_valid;
4750 error = inode_change_ok(inode, attr);
4754 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4755 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4758 /* (user+group)*(old+new) structure, inode write (sb,
4759 * inode block, ? - but truncate inode update has it) */
4760 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4761 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4762 if (IS_ERR(handle)) {
4763 error = PTR_ERR(handle);
4766 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4768 ext4_journal_stop(handle);
4771 /* Update corresponding info in inode so that everything is in
4772 * one transaction */
4773 if (attr->ia_valid & ATTR_UID)
4774 inode->i_uid = attr->ia_uid;
4775 if (attr->ia_valid & ATTR_GID)
4776 inode->i_gid = attr->ia_gid;
4777 error = ext4_mark_inode_dirty(handle, inode);
4778 ext4_journal_stop(handle);
4781 if (attr->ia_valid & ATTR_SIZE) {
4782 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4783 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4785 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4792 if (S_ISREG(inode->i_mode) &&
4793 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4796 handle = ext4_journal_start(inode, 3);
4797 if (IS_ERR(handle)) {
4798 error = PTR_ERR(handle);
4802 error = ext4_orphan_add(handle, inode);
4803 EXT4_I(inode)->i_disksize = attr->ia_size;
4804 rc = ext4_mark_inode_dirty(handle, inode);
4807 ext4_journal_stop(handle);
4809 if (ext4_should_order_data(inode)) {
4810 error = ext4_begin_ordered_truncate(inode,
4813 /* Do as much error cleanup as possible */
4814 handle = ext4_journal_start(inode, 3);
4815 if (IS_ERR(handle)) {
4816 ext4_orphan_del(NULL, inode);
4819 ext4_orphan_del(handle, inode);
4820 ext4_journal_stop(handle);
4826 rc = inode_setattr(inode, attr);
4828 /* If inode_setattr's call to ext4_truncate failed to get a
4829 * transaction handle at all, we need to clean up the in-core
4830 * orphan list manually. */
4832 ext4_orphan_del(NULL, inode);
4834 if (!rc && (ia_valid & ATTR_MODE))
4835 rc = ext4_acl_chmod(inode);
4838 ext4_std_error(inode->i_sb, error);
4844 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4847 struct inode *inode;
4848 unsigned long delalloc_blocks;
4850 inode = dentry->d_inode;
4851 generic_fillattr(inode, stat);
4854 * We can't update i_blocks if the block allocation is delayed
4855 * otherwise in the case of system crash before the real block
4856 * allocation is done, we will have i_blocks inconsistent with
4857 * on-disk file blocks.
4858 * We always keep i_blocks updated together with real
4859 * allocation. But to not confuse with user, stat
4860 * will return the blocks that include the delayed allocation
4861 * blocks for this file.
4863 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4864 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4865 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4867 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4871 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4876 /* if nrblocks are contiguous */
4879 * With N contiguous data blocks, it need at most
4880 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4881 * 2 dindirect blocks
4884 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4885 return indirects + 3;
4888 * if nrblocks are not contiguous, worse case, each block touch
4889 * a indirect block, and each indirect block touch a double indirect
4890 * block, plus a triple indirect block
4892 indirects = nrblocks * 2 + 1;
4896 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4898 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4899 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4900 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4904 * Account for index blocks, block groups bitmaps and block group
4905 * descriptor blocks if modify datablocks and index blocks
4906 * worse case, the indexs blocks spread over different block groups
4908 * If datablocks are discontiguous, they are possible to spread over
4909 * different block groups too. If they are contiugous, with flexbg,
4910 * they could still across block group boundary.
4912 * Also account for superblock, inode, quota and xattr blocks
4914 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4916 int groups, gdpblocks;
4921 * How many index blocks need to touch to modify nrblocks?
4922 * The "Chunk" flag indicating whether the nrblocks is
4923 * physically contiguous on disk
4925 * For Direct IO and fallocate, they calls get_block to allocate
4926 * one single extent at a time, so they could set the "Chunk" flag
4928 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4933 * Now let's see how many group bitmaps and group descriptors need
4943 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4944 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4945 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4946 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4948 /* bitmaps and block group descriptor blocks */
4949 ret += groups + gdpblocks;
4951 /* Blocks for super block, inode, quota and xattr blocks */
4952 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4958 * Calulate the total number of credits to reserve to fit
4959 * the modification of a single pages into a single transaction,
4960 * which may include multiple chunks of block allocations.
4962 * This could be called via ext4_write_begin()
4964 * We need to consider the worse case, when
4965 * one new block per extent.
4967 int ext4_writepage_trans_blocks(struct inode *inode)
4969 int bpp = ext4_journal_blocks_per_page(inode);
4972 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4974 /* Account for data blocks for journalled mode */
4975 if (ext4_should_journal_data(inode))
4981 * Calculate the journal credits for a chunk of data modification.
4983 * This is called from DIO, fallocate or whoever calling
4984 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4986 * journal buffers for data blocks are not included here, as DIO
4987 * and fallocate do no need to journal data buffers.
4989 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4991 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4995 * The caller must have previously called ext4_reserve_inode_write().
4996 * Give this, we know that the caller already has write access to iloc->bh.
4998 int ext4_mark_iloc_dirty(handle_t *handle,
4999 struct inode *inode, struct ext4_iloc *iloc)
5003 if (test_opt(inode->i_sb, I_VERSION))
5004 inode_inc_iversion(inode);
5006 /* the do_update_inode consumes one bh->b_count */
5009 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5010 err = ext4_do_update_inode(handle, inode, iloc);
5016 * On success, We end up with an outstanding reference count against
5017 * iloc->bh. This _must_ be cleaned up later.
5021 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5022 struct ext4_iloc *iloc)
5026 err = ext4_get_inode_loc(inode, iloc);
5028 BUFFER_TRACE(iloc->bh, "get_write_access");
5029 err = ext4_journal_get_write_access(handle, iloc->bh);
5035 ext4_std_error(inode->i_sb, err);
5040 * Expand an inode by new_extra_isize bytes.
5041 * Returns 0 on success or negative error number on failure.
5043 static int ext4_expand_extra_isize(struct inode *inode,
5044 unsigned int new_extra_isize,
5045 struct ext4_iloc iloc,
5048 struct ext4_inode *raw_inode;
5049 struct ext4_xattr_ibody_header *header;
5050 struct ext4_xattr_entry *entry;
5052 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5055 raw_inode = ext4_raw_inode(&iloc);
5057 header = IHDR(inode, raw_inode);
5058 entry = IFIRST(header);
5060 /* No extended attributes present */
5061 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5062 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5063 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5065 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5069 /* try to expand with EAs present */
5070 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5075 * What we do here is to mark the in-core inode as clean with respect to inode
5076 * dirtiness (it may still be data-dirty).
5077 * This means that the in-core inode may be reaped by prune_icache
5078 * without having to perform any I/O. This is a very good thing,
5079 * because *any* task may call prune_icache - even ones which
5080 * have a transaction open against a different journal.
5082 * Is this cheating? Not really. Sure, we haven't written the
5083 * inode out, but prune_icache isn't a user-visible syncing function.
5084 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5085 * we start and wait on commits.
5087 * Is this efficient/effective? Well, we're being nice to the system
5088 * by cleaning up our inodes proactively so they can be reaped
5089 * without I/O. But we are potentially leaving up to five seconds'
5090 * worth of inodes floating about which prune_icache wants us to
5091 * write out. One way to fix that would be to get prune_icache()
5092 * to do a write_super() to free up some memory. It has the desired
5095 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5097 struct ext4_iloc iloc;
5098 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5099 static unsigned int mnt_count;
5103 err = ext4_reserve_inode_write(handle, inode, &iloc);
5104 if (ext4_handle_valid(handle) &&
5105 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5106 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5108 * We need extra buffer credits since we may write into EA block
5109 * with this same handle. If journal_extend fails, then it will
5110 * only result in a minor loss of functionality for that inode.
5111 * If this is felt to be critical, then e2fsck should be run to
5112 * force a large enough s_min_extra_isize.
5114 if ((jbd2_journal_extend(handle,
5115 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5116 ret = ext4_expand_extra_isize(inode,
5117 sbi->s_want_extra_isize,
5120 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5122 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5123 ext4_warning(inode->i_sb, __func__,
5124 "Unable to expand inode %lu. Delete"
5125 " some EAs or run e2fsck.",
5128 le16_to_cpu(sbi->s_es->s_mnt_count);
5134 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5139 * ext4_dirty_inode() is called from __mark_inode_dirty()
5141 * We're really interested in the case where a file is being extended.
5142 * i_size has been changed by generic_commit_write() and we thus need
5143 * to include the updated inode in the current transaction.
5145 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5146 * are allocated to the file.
5148 * If the inode is marked synchronous, we don't honour that here - doing
5149 * so would cause a commit on atime updates, which we don't bother doing.
5150 * We handle synchronous inodes at the highest possible level.
5152 void ext4_dirty_inode(struct inode *inode)
5154 handle_t *current_handle = ext4_journal_current_handle();
5157 if (!ext4_handle_valid(current_handle)) {
5158 ext4_mark_inode_dirty(current_handle, inode);
5162 handle = ext4_journal_start(inode, 2);
5165 if (current_handle &&
5166 current_handle->h_transaction != handle->h_transaction) {
5167 /* This task has a transaction open against a different fs */
5168 printk(KERN_EMERG "%s: transactions do not match!\n",
5171 jbd_debug(5, "marking dirty. outer handle=%p\n",
5173 ext4_mark_inode_dirty(handle, inode);
5175 ext4_journal_stop(handle);
5182 * Bind an inode's backing buffer_head into this transaction, to prevent
5183 * it from being flushed to disk early. Unlike
5184 * ext4_reserve_inode_write, this leaves behind no bh reference and
5185 * returns no iloc structure, so the caller needs to repeat the iloc
5186 * lookup to mark the inode dirty later.
5188 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5190 struct ext4_iloc iloc;
5194 err = ext4_get_inode_loc(inode, &iloc);
5196 BUFFER_TRACE(iloc.bh, "get_write_access");
5197 err = jbd2_journal_get_write_access(handle, iloc.bh);
5199 err = ext4_handle_dirty_metadata(handle,
5205 ext4_std_error(inode->i_sb, err);
5210 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5217 * We have to be very careful here: changing a data block's
5218 * journaling status dynamically is dangerous. If we write a
5219 * data block to the journal, change the status and then delete
5220 * that block, we risk forgetting to revoke the old log record
5221 * from the journal and so a subsequent replay can corrupt data.
5222 * So, first we make sure that the journal is empty and that
5223 * nobody is changing anything.
5226 journal = EXT4_JOURNAL(inode);
5229 if (is_journal_aborted(journal))
5232 jbd2_journal_lock_updates(journal);
5233 jbd2_journal_flush(journal);
5236 * OK, there are no updates running now, and all cached data is
5237 * synced to disk. We are now in a completely consistent state
5238 * which doesn't have anything in the journal, and we know that
5239 * no filesystem updates are running, so it is safe to modify
5240 * the inode's in-core data-journaling state flag now.
5244 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5246 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5247 ext4_set_aops(inode);
5249 jbd2_journal_unlock_updates(journal);
5251 /* Finally we can mark the inode as dirty. */
5253 handle = ext4_journal_start(inode, 1);
5255 return PTR_ERR(handle);
5257 err = ext4_mark_inode_dirty(handle, inode);
5258 ext4_handle_sync(handle);
5259 ext4_journal_stop(handle);
5260 ext4_std_error(inode->i_sb, err);
5265 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5267 return !buffer_mapped(bh);
5270 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5272 struct page *page = vmf->page;
5277 struct file *file = vma->vm_file;
5278 struct inode *inode = file->f_path.dentry->d_inode;
5279 struct address_space *mapping = inode->i_mapping;
5282 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5283 * get i_mutex because we are already holding mmap_sem.
5285 down_read(&inode->i_alloc_sem);
5286 size = i_size_read(inode);
5287 if (page->mapping != mapping || size <= page_offset(page)
5288 || !PageUptodate(page)) {
5289 /* page got truncated from under us? */
5293 if (PageMappedToDisk(page))
5296 if (page->index == size >> PAGE_CACHE_SHIFT)
5297 len = size & ~PAGE_CACHE_MASK;
5299 len = PAGE_CACHE_SIZE;
5301 if (page_has_buffers(page)) {
5302 /* return if we have all the buffers mapped */
5303 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5308 * OK, we need to fill the hole... Do write_begin write_end
5309 * to do block allocation/reservation.We are not holding
5310 * inode.i__mutex here. That allow * parallel write_begin,
5311 * write_end call. lock_page prevent this from happening
5312 * on the same page though
5314 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5315 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5318 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5319 len, len, page, fsdata);
5325 ret = VM_FAULT_SIGBUS;
5326 up_read(&inode->i_alloc_sem);