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 * The ext4_ind_get_blocks() function handles non-extents inodes
896 * (i.e., using the traditional indirect/double-indirect i_blocks
897 * scheme) for ext4_get_blocks().
899 * Allocation strategy is simple: if we have to allocate something, we will
900 * have to go the whole way to leaf. So let's do it before attaching anything
901 * to tree, set linkage between the newborn blocks, write them if sync is
902 * required, recheck the path, free and repeat if check fails, otherwise
903 * set the last missing link (that will protect us from any truncate-generated
904 * removals - all blocks on the path are immune now) and possibly force the
905 * write on the parent block.
906 * That has a nice additional property: no special recovery from the failed
907 * allocations is needed - we simply release blocks and do not touch anything
908 * reachable from inode.
910 * `handle' can be NULL if create == 0.
912 * return > 0, # of blocks mapped or allocated.
913 * return = 0, if plain lookup failed.
914 * return < 0, error case.
916 * The ext4_ind_get_blocks() function should be called with
917 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
918 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
919 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
922 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
923 ext4_lblk_t iblock, unsigned int maxblocks,
924 struct buffer_head *bh_result,
928 ext4_lblk_t offsets[4];
933 int blocks_to_boundary = 0;
935 struct ext4_inode_info *ei = EXT4_I(inode);
937 ext4_fsblk_t first_block = 0;
941 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
942 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
943 depth = ext4_block_to_path(inode, iblock, offsets,
944 &blocks_to_boundary);
949 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
951 /* Simplest case - block found, no allocation needed */
953 first_block = le32_to_cpu(chain[depth - 1].key);
954 clear_buffer_new(bh_result);
957 while (count < maxblocks && count <= blocks_to_boundary) {
960 blk = le32_to_cpu(*(chain[depth-1].p + count));
962 if (blk == first_block + count)
970 /* Next simple case - plain lookup or failed read of indirect block */
971 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
975 * Okay, we need to do block allocation.
977 goal = ext4_find_goal(inode, iblock, partial);
979 /* the number of blocks need to allocate for [d,t]indirect blocks */
980 indirect_blks = (chain + depth) - partial - 1;
983 * Next look up the indirect map to count the totoal number of
984 * direct blocks to allocate for this branch.
986 count = ext4_blks_to_allocate(partial, indirect_blks,
987 maxblocks, blocks_to_boundary);
989 * Block out ext4_truncate while we alter the tree
991 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
993 offsets + (partial - chain), partial);
996 * The ext4_splice_branch call will free and forget any buffers
997 * on the new chain if there is a failure, but that risks using
998 * up transaction credits, especially for bitmaps where the
999 * credits cannot be returned. Can we handle this somehow? We
1000 * may need to return -EAGAIN upwards in the worst case. --sct
1003 err = ext4_splice_branch(handle, inode, iblock,
1004 partial, indirect_blks, count);
1006 * i_disksize growing is protected by i_data_sem. Don't forget to
1007 * protect it if you're about to implement concurrent
1008 * ext4_get_block() -bzzz
1010 if (!err && (flags & EXT4_GET_BLOCKS_EXTEND_DISKSIZE)) {
1011 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
1012 if (disksize > i_size_read(inode))
1013 disksize = i_size_read(inode);
1014 if (disksize > ei->i_disksize)
1015 ei->i_disksize = disksize;
1020 set_buffer_new(bh_result);
1022 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1023 if (count > blocks_to_boundary)
1024 set_buffer_boundary(bh_result);
1026 /* Clean up and exit */
1027 partial = chain + depth - 1; /* the whole chain */
1029 while (partial > chain) {
1030 BUFFER_TRACE(partial->bh, "call brelse");
1031 brelse(partial->bh);
1034 BUFFER_TRACE(bh_result, "returned");
1039 qsize_t ext4_get_reserved_space(struct inode *inode)
1041 unsigned long long total;
1043 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1044 total = EXT4_I(inode)->i_reserved_data_blocks +
1045 EXT4_I(inode)->i_reserved_meta_blocks;
1046 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1051 * Calculate the number of metadata blocks need to reserve
1052 * to allocate @blocks for non extent file based file
1054 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1056 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1057 int ind_blks, dind_blks, tind_blks;
1059 /* number of new indirect blocks needed */
1060 ind_blks = (blocks + icap - 1) / icap;
1062 dind_blks = (ind_blks + icap - 1) / icap;
1066 return ind_blks + dind_blks + tind_blks;
1070 * Calculate the number of metadata blocks need to reserve
1071 * to allocate given number of blocks
1073 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1078 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1079 return ext4_ext_calc_metadata_amount(inode, blocks);
1081 return ext4_indirect_calc_metadata_amount(inode, blocks);
1084 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1086 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1087 int total, mdb, mdb_free;
1089 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1090 /* recalculate the number of metablocks still need to be reserved */
1091 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1092 mdb = ext4_calc_metadata_amount(inode, total);
1094 /* figure out how many metablocks to release */
1095 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1096 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1099 /* Account for allocated meta_blocks */
1100 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1102 /* update fs dirty blocks counter */
1103 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1104 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1105 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1108 /* update per-inode reservations */
1109 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1110 EXT4_I(inode)->i_reserved_data_blocks -= used;
1111 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1114 * free those over-booking quota for metadata blocks
1117 vfs_dq_release_reservation_block(inode, mdb_free);
1120 * If we have done all the pending block allocations and if
1121 * there aren't any writers on the inode, we can discard the
1122 * inode's preallocations.
1124 if (!total && (atomic_read(&inode->i_writecount) == 0))
1125 ext4_discard_preallocations(inode);
1129 * The ext4_get_blocks() function tries to look up the requested blocks,
1130 * and returns if the blocks are already mapped.
1132 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1133 * and store the allocated blocks in the result buffer head and mark it
1136 * If file type is extents based, it will call ext4_ext_get_blocks(),
1137 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1140 * On success, it returns the number of blocks being mapped or allocate.
1141 * if create==0 and the blocks are pre-allocated and uninitialized block,
1142 * the result buffer head is unmapped. If the create ==1, it will make sure
1143 * the buffer head is mapped.
1145 * It returns 0 if plain look up failed (blocks have not been allocated), in
1146 * that casem, buffer head is unmapped
1148 * It returns the error in case of allocation failure.
1150 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1151 unsigned int max_blocks, struct buffer_head *bh,
1156 clear_buffer_mapped(bh);
1157 clear_buffer_unwritten(bh);
1160 * Try to see if we can get the block without requesting a new
1161 * file system block.
1163 down_read((&EXT4_I(inode)->i_data_sem));
1164 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1165 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1168 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1171 up_read((&EXT4_I(inode)->i_data_sem));
1173 /* If it is only a block(s) look up */
1174 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1178 * Returns if the blocks have already allocated
1180 * Note that if blocks have been preallocated
1181 * ext4_ext_get_block() returns th create = 0
1182 * with buffer head unmapped.
1184 if (retval > 0 && buffer_mapped(bh))
1188 * When we call get_blocks without the create flag, the
1189 * BH_Unwritten flag could have gotten set if the blocks
1190 * requested were part of a uninitialized extent. We need to
1191 * clear this flag now that we are committed to convert all or
1192 * part of the uninitialized extent to be an initialized
1193 * extent. This is because we need to avoid the combination
1194 * of BH_Unwritten and BH_Mapped flags being simultaneously
1195 * set on the buffer_head.
1197 clear_buffer_unwritten(bh);
1200 * New blocks allocate and/or writing to uninitialized extent
1201 * will possibly result in updating i_data, so we take
1202 * the write lock of i_data_sem, and call get_blocks()
1203 * with create == 1 flag.
1205 down_write((&EXT4_I(inode)->i_data_sem));
1208 * if the caller is from delayed allocation writeout path
1209 * we have already reserved fs blocks for allocation
1210 * let the underlying get_block() function know to
1211 * avoid double accounting
1213 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1214 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1216 * We need to check for EXT4 here because migrate
1217 * could have changed the inode type in between
1219 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1220 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1223 retval = ext4_ind_get_blocks(handle, inode, block,
1224 max_blocks, bh, flags);
1226 if (retval > 0 && buffer_new(bh)) {
1228 * We allocated new blocks which will result in
1229 * i_data's format changing. Force the migrate
1230 * to fail by clearing migrate flags
1232 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1237 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1238 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1241 * Update reserved blocks/metadata blocks after successful
1242 * block allocation which had been deferred till now.
1244 if ((retval > 0) && (flags & EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE))
1245 ext4_da_update_reserve_space(inode, retval);
1247 up_write((&EXT4_I(inode)->i_data_sem));
1251 /* Maximum number of blocks we map for direct IO at once. */
1252 #define DIO_MAX_BLOCKS 4096
1254 int ext4_get_block(struct inode *inode, sector_t iblock,
1255 struct buffer_head *bh_result, int create)
1257 handle_t *handle = ext4_journal_current_handle();
1258 int ret = 0, started = 0;
1259 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1262 if (create && !handle) {
1263 /* Direct IO write... */
1264 if (max_blocks > DIO_MAX_BLOCKS)
1265 max_blocks = DIO_MAX_BLOCKS;
1266 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1267 handle = ext4_journal_start(inode, dio_credits);
1268 if (IS_ERR(handle)) {
1269 ret = PTR_ERR(handle);
1275 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1276 create ? EXT4_GET_BLOCKS_CREATE : 0);
1278 bh_result->b_size = (ret << inode->i_blkbits);
1282 ext4_journal_stop(handle);
1288 * `handle' can be NULL if create is zero
1290 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1291 ext4_lblk_t block, int create, int *errp)
1293 struct buffer_head dummy;
1295 int flags = EXT4_GET_BLOCKS_EXTEND_DISKSIZE;
1297 J_ASSERT(handle != NULL || create == 0);
1300 dummy.b_blocknr = -1000;
1301 buffer_trace_init(&dummy.b_history);
1303 flags |= EXT4_GET_BLOCKS_CREATE;
1304 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1306 * ext4_get_blocks() returns number of blocks mapped. 0 in
1315 if (!err && buffer_mapped(&dummy)) {
1316 struct buffer_head *bh;
1317 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1322 if (buffer_new(&dummy)) {
1323 J_ASSERT(create != 0);
1324 J_ASSERT(handle != NULL);
1327 * Now that we do not always journal data, we should
1328 * keep in mind whether this should always journal the
1329 * new buffer as metadata. For now, regular file
1330 * writes use ext4_get_block instead, so it's not a
1334 BUFFER_TRACE(bh, "call get_create_access");
1335 fatal = ext4_journal_get_create_access(handle, bh);
1336 if (!fatal && !buffer_uptodate(bh)) {
1337 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1338 set_buffer_uptodate(bh);
1341 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1342 err = ext4_handle_dirty_metadata(handle, inode, bh);
1346 BUFFER_TRACE(bh, "not a new buffer");
1359 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1360 ext4_lblk_t block, int create, int *err)
1362 struct buffer_head *bh;
1364 bh = ext4_getblk(handle, inode, block, create, err);
1367 if (buffer_uptodate(bh))
1369 ll_rw_block(READ_META, 1, &bh);
1371 if (buffer_uptodate(bh))
1378 static int walk_page_buffers(handle_t *handle,
1379 struct buffer_head *head,
1383 int (*fn)(handle_t *handle,
1384 struct buffer_head *bh))
1386 struct buffer_head *bh;
1387 unsigned block_start, block_end;
1388 unsigned blocksize = head->b_size;
1390 struct buffer_head *next;
1392 for (bh = head, block_start = 0;
1393 ret == 0 && (bh != head || !block_start);
1394 block_start = block_end, bh = next)
1396 next = bh->b_this_page;
1397 block_end = block_start + blocksize;
1398 if (block_end <= from || block_start >= to) {
1399 if (partial && !buffer_uptodate(bh))
1403 err = (*fn)(handle, bh);
1411 * To preserve ordering, it is essential that the hole instantiation and
1412 * the data write be encapsulated in a single transaction. We cannot
1413 * close off a transaction and start a new one between the ext4_get_block()
1414 * and the commit_write(). So doing the jbd2_journal_start at the start of
1415 * prepare_write() is the right place.
1417 * Also, this function can nest inside ext4_writepage() ->
1418 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1419 * has generated enough buffer credits to do the whole page. So we won't
1420 * block on the journal in that case, which is good, because the caller may
1423 * By accident, ext4 can be reentered when a transaction is open via
1424 * quota file writes. If we were to commit the transaction while thus
1425 * reentered, there can be a deadlock - we would be holding a quota
1426 * lock, and the commit would never complete if another thread had a
1427 * transaction open and was blocking on the quota lock - a ranking
1430 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1431 * will _not_ run commit under these circumstances because handle->h_ref
1432 * is elevated. We'll still have enough credits for the tiny quotafile
1435 static int do_journal_get_write_access(handle_t *handle,
1436 struct buffer_head *bh)
1438 if (!buffer_mapped(bh) || buffer_freed(bh))
1440 return ext4_journal_get_write_access(handle, bh);
1443 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1444 loff_t pos, unsigned len, unsigned flags,
1445 struct page **pagep, void **fsdata)
1447 struct inode *inode = mapping->host;
1448 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1455 trace_mark(ext4_write_begin,
1456 "dev %s ino %lu pos %llu len %u flags %u",
1457 inode->i_sb->s_id, inode->i_ino,
1458 (unsigned long long) pos, len, flags);
1459 index = pos >> PAGE_CACHE_SHIFT;
1460 from = pos & (PAGE_CACHE_SIZE - 1);
1464 handle = ext4_journal_start(inode, needed_blocks);
1465 if (IS_ERR(handle)) {
1466 ret = PTR_ERR(handle);
1470 /* We cannot recurse into the filesystem as the transaction is already
1472 flags |= AOP_FLAG_NOFS;
1474 page = grab_cache_page_write_begin(mapping, index, flags);
1476 ext4_journal_stop(handle);
1482 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1485 if (!ret && ext4_should_journal_data(inode)) {
1486 ret = walk_page_buffers(handle, page_buffers(page),
1487 from, to, NULL, do_journal_get_write_access);
1492 ext4_journal_stop(handle);
1493 page_cache_release(page);
1495 * block_write_begin may have instantiated a few blocks
1496 * outside i_size. Trim these off again. Don't need
1497 * i_size_read because we hold i_mutex.
1499 if (pos + len > inode->i_size)
1500 vmtruncate(inode, inode->i_size);
1503 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1509 /* For write_end() in data=journal mode */
1510 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1512 if (!buffer_mapped(bh) || buffer_freed(bh))
1514 set_buffer_uptodate(bh);
1515 return ext4_handle_dirty_metadata(handle, NULL, bh);
1519 * We need to pick up the new inode size which generic_commit_write gave us
1520 * `file' can be NULL - eg, when called from page_symlink().
1522 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1523 * buffers are managed internally.
1525 static int ext4_ordered_write_end(struct file *file,
1526 struct address_space *mapping,
1527 loff_t pos, unsigned len, unsigned copied,
1528 struct page *page, void *fsdata)
1530 handle_t *handle = ext4_journal_current_handle();
1531 struct inode *inode = mapping->host;
1534 trace_mark(ext4_ordered_write_end,
1535 "dev %s ino %lu pos %llu len %u copied %u",
1536 inode->i_sb->s_id, inode->i_ino,
1537 (unsigned long long) pos, len, copied);
1538 ret = ext4_jbd2_file_inode(handle, inode);
1543 new_i_size = pos + copied;
1544 if (new_i_size > EXT4_I(inode)->i_disksize) {
1545 ext4_update_i_disksize(inode, new_i_size);
1546 /* We need to mark inode dirty even if
1547 * new_i_size is less that inode->i_size
1548 * bu greater than i_disksize.(hint delalloc)
1550 ext4_mark_inode_dirty(handle, inode);
1553 ret2 = generic_write_end(file, mapping, pos, len, copied,
1559 ret2 = ext4_journal_stop(handle);
1563 return ret ? ret : copied;
1566 static int ext4_writeback_write_end(struct file *file,
1567 struct address_space *mapping,
1568 loff_t pos, unsigned len, unsigned copied,
1569 struct page *page, void *fsdata)
1571 handle_t *handle = ext4_journal_current_handle();
1572 struct inode *inode = mapping->host;
1576 trace_mark(ext4_writeback_write_end,
1577 "dev %s ino %lu pos %llu len %u copied %u",
1578 inode->i_sb->s_id, inode->i_ino,
1579 (unsigned long long) pos, len, copied);
1580 new_i_size = pos + copied;
1581 if (new_i_size > EXT4_I(inode)->i_disksize) {
1582 ext4_update_i_disksize(inode, new_i_size);
1583 /* We need to mark inode dirty even if
1584 * new_i_size is less that inode->i_size
1585 * bu greater than i_disksize.(hint delalloc)
1587 ext4_mark_inode_dirty(handle, inode);
1590 ret2 = generic_write_end(file, mapping, pos, len, copied,
1596 ret2 = ext4_journal_stop(handle);
1600 return ret ? ret : copied;
1603 static int ext4_journalled_write_end(struct file *file,
1604 struct address_space *mapping,
1605 loff_t pos, unsigned len, unsigned copied,
1606 struct page *page, void *fsdata)
1608 handle_t *handle = ext4_journal_current_handle();
1609 struct inode *inode = mapping->host;
1615 trace_mark(ext4_journalled_write_end,
1616 "dev %s ino %lu pos %llu len %u copied %u",
1617 inode->i_sb->s_id, inode->i_ino,
1618 (unsigned long long) pos, len, copied);
1619 from = pos & (PAGE_CACHE_SIZE - 1);
1623 if (!PageUptodate(page))
1625 page_zero_new_buffers(page, from+copied, to);
1628 ret = walk_page_buffers(handle, page_buffers(page), from,
1629 to, &partial, write_end_fn);
1631 SetPageUptodate(page);
1632 new_i_size = pos + copied;
1633 if (new_i_size > inode->i_size)
1634 i_size_write(inode, pos+copied);
1635 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1636 if (new_i_size > EXT4_I(inode)->i_disksize) {
1637 ext4_update_i_disksize(inode, new_i_size);
1638 ret2 = ext4_mark_inode_dirty(handle, inode);
1644 ret2 = ext4_journal_stop(handle);
1647 page_cache_release(page);
1649 return ret ? ret : copied;
1652 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1655 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1656 unsigned long md_needed, mdblocks, total = 0;
1659 * recalculate the amount of metadata blocks to reserve
1660 * in order to allocate nrblocks
1661 * worse case is one extent per block
1664 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1665 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1666 mdblocks = ext4_calc_metadata_amount(inode, total);
1667 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1669 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1670 total = md_needed + nrblocks;
1673 * Make quota reservation here to prevent quota overflow
1674 * later. Real quota accounting is done at pages writeout
1677 if (vfs_dq_reserve_block(inode, total)) {
1678 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1682 if (ext4_claim_free_blocks(sbi, total)) {
1683 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1684 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1688 vfs_dq_release_reservation_block(inode, total);
1691 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1692 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1694 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1695 return 0; /* success */
1698 static void ext4_da_release_space(struct inode *inode, int to_free)
1700 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1701 int total, mdb, mdb_free, release;
1704 return; /* Nothing to release, exit */
1706 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1708 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1710 * if there is no reserved blocks, but we try to free some
1711 * then the counter is messed up somewhere.
1712 * but since this function is called from invalidate
1713 * page, it's harmless to return without any action
1715 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1716 "blocks for inode %lu, but there is no reserved "
1717 "data blocks\n", to_free, inode->i_ino);
1718 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1722 /* recalculate the number of metablocks still need to be reserved */
1723 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1724 mdb = ext4_calc_metadata_amount(inode, total);
1726 /* figure out how many metablocks to release */
1727 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1728 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1730 release = to_free + mdb_free;
1732 /* update fs dirty blocks counter for truncate case */
1733 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1735 /* update per-inode reservations */
1736 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1737 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1739 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1740 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1741 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1743 vfs_dq_release_reservation_block(inode, release);
1746 static void ext4_da_page_release_reservation(struct page *page,
1747 unsigned long offset)
1750 struct buffer_head *head, *bh;
1751 unsigned int curr_off = 0;
1753 head = page_buffers(page);
1756 unsigned int next_off = curr_off + bh->b_size;
1758 if ((offset <= curr_off) && (buffer_delay(bh))) {
1760 clear_buffer_delay(bh);
1762 curr_off = next_off;
1763 } while ((bh = bh->b_this_page) != head);
1764 ext4_da_release_space(page->mapping->host, to_release);
1768 * Delayed allocation stuff
1771 struct mpage_da_data {
1772 struct inode *inode;
1773 sector_t b_blocknr; /* start block number of extent */
1774 size_t b_size; /* size of extent */
1775 unsigned long b_state; /* state of the extent */
1776 unsigned long first_page, next_page; /* extent of pages */
1777 struct writeback_control *wbc;
1784 * mpage_da_submit_io - walks through extent of pages and try to write
1785 * them with writepage() call back
1787 * @mpd->inode: inode
1788 * @mpd->first_page: first page of the extent
1789 * @mpd->next_page: page after the last page of the extent
1791 * By the time mpage_da_submit_io() is called we expect all blocks
1792 * to be allocated. this may be wrong if allocation failed.
1794 * As pages are already locked by write_cache_pages(), we can't use it
1796 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1799 struct pagevec pvec;
1800 unsigned long index, end;
1801 int ret = 0, err, nr_pages, i;
1802 struct inode *inode = mpd->inode;
1803 struct address_space *mapping = inode->i_mapping;
1805 BUG_ON(mpd->next_page <= mpd->first_page);
1807 * We need to start from the first_page to the next_page - 1
1808 * to make sure we also write the mapped dirty buffer_heads.
1809 * If we look at mpd->b_blocknr we would only be looking
1810 * at the currently mapped buffer_heads.
1812 index = mpd->first_page;
1813 end = mpd->next_page - 1;
1815 pagevec_init(&pvec, 0);
1816 while (index <= end) {
1817 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1820 for (i = 0; i < nr_pages; i++) {
1821 struct page *page = pvec.pages[i];
1823 index = page->index;
1828 BUG_ON(!PageLocked(page));
1829 BUG_ON(PageWriteback(page));
1831 pages_skipped = mpd->wbc->pages_skipped;
1832 err = mapping->a_ops->writepage(page, mpd->wbc);
1833 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1835 * have successfully written the page
1836 * without skipping the same
1838 mpd->pages_written++;
1840 * In error case, we have to continue because
1841 * remaining pages are still locked
1842 * XXX: unlock and re-dirty them?
1847 pagevec_release(&pvec);
1853 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1855 * @mpd->inode - inode to walk through
1856 * @exbh->b_blocknr - first block on a disk
1857 * @exbh->b_size - amount of space in bytes
1858 * @logical - first logical block to start assignment with
1860 * the function goes through all passed space and put actual disk
1861 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1863 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1864 struct buffer_head *exbh)
1866 struct inode *inode = mpd->inode;
1867 struct address_space *mapping = inode->i_mapping;
1868 int blocks = exbh->b_size >> inode->i_blkbits;
1869 sector_t pblock = exbh->b_blocknr, cur_logical;
1870 struct buffer_head *head, *bh;
1872 struct pagevec pvec;
1875 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1876 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1877 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1879 pagevec_init(&pvec, 0);
1881 while (index <= end) {
1882 /* XXX: optimize tail */
1883 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1886 for (i = 0; i < nr_pages; i++) {
1887 struct page *page = pvec.pages[i];
1889 index = page->index;
1894 BUG_ON(!PageLocked(page));
1895 BUG_ON(PageWriteback(page));
1896 BUG_ON(!page_has_buffers(page));
1898 bh = page_buffers(page);
1901 /* skip blocks out of the range */
1903 if (cur_logical >= logical)
1906 } while ((bh = bh->b_this_page) != head);
1909 if (cur_logical >= logical + blocks)
1912 if (buffer_delay(bh) ||
1913 buffer_unwritten(bh)) {
1915 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
1917 if (buffer_delay(bh)) {
1918 clear_buffer_delay(bh);
1919 bh->b_blocknr = pblock;
1922 * unwritten already should have
1923 * blocknr assigned. Verify that
1925 clear_buffer_unwritten(bh);
1926 BUG_ON(bh->b_blocknr != pblock);
1929 } else if (buffer_mapped(bh))
1930 BUG_ON(bh->b_blocknr != pblock);
1934 } while ((bh = bh->b_this_page) != head);
1936 pagevec_release(&pvec);
1942 * __unmap_underlying_blocks - just a helper function to unmap
1943 * set of blocks described by @bh
1945 static inline void __unmap_underlying_blocks(struct inode *inode,
1946 struct buffer_head *bh)
1948 struct block_device *bdev = inode->i_sb->s_bdev;
1951 blocks = bh->b_size >> inode->i_blkbits;
1952 for (i = 0; i < blocks; i++)
1953 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1956 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1957 sector_t logical, long blk_cnt)
1961 struct pagevec pvec;
1962 struct inode *inode = mpd->inode;
1963 struct address_space *mapping = inode->i_mapping;
1965 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1966 end = (logical + blk_cnt - 1) >>
1967 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1968 while (index <= end) {
1969 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1972 for (i = 0; i < nr_pages; i++) {
1973 struct page *page = pvec.pages[i];
1974 index = page->index;
1979 BUG_ON(!PageLocked(page));
1980 BUG_ON(PageWriteback(page));
1981 block_invalidatepage(page, 0);
1982 ClearPageUptodate(page);
1989 static void ext4_print_free_blocks(struct inode *inode)
1991 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1992 printk(KERN_EMERG "Total free blocks count %lld\n",
1993 ext4_count_free_blocks(inode->i_sb));
1994 printk(KERN_EMERG "Free/Dirty block details\n");
1995 printk(KERN_EMERG "free_blocks=%lld\n",
1996 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1997 printk(KERN_EMERG "dirty_blocks=%lld\n",
1998 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1999 printk(KERN_EMERG "Block reservation details\n");
2000 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
2001 EXT4_I(inode)->i_reserved_data_blocks);
2002 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
2003 EXT4_I(inode)->i_reserved_meta_blocks);
2008 * mpage_da_map_blocks - go through given space
2010 * @mpd - bh describing space
2012 * The function skips space we know is already mapped to disk blocks.
2015 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2017 int err, blks, get_blocks_flags;
2018 struct buffer_head new;
2019 sector_t next = mpd->b_blocknr;
2020 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2021 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2022 handle_t *handle = NULL;
2025 * We consider only non-mapped and non-allocated blocks
2027 if ((mpd->b_state & (1 << BH_Mapped)) &&
2028 !(mpd->b_state & (1 << BH_Delay)) &&
2029 !(mpd->b_state & (1 << BH_Unwritten)))
2033 * If we didn't accumulate anything to write simply return
2038 handle = ext4_journal_current_handle();
2042 * Call ext4_get_blocks() to allocate any delayed allocation
2043 * blocks, or to convert an uninitialized extent to be
2044 * initialized (in the case where we have written into
2045 * one or more preallocated blocks).
2047 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2048 * indicate that we are on the delayed allocation path. This
2049 * affects functions in many different parts of the allocation
2050 * call path. This flag exists primarily because we don't
2051 * want to change *many* call functions, so ext4_get_blocks()
2052 * will set the magic i_delalloc_reserved_flag once the
2053 * inode's allocation semaphore is taken.
2055 * If the blocks in questions were delalloc blocks, set
2056 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2057 * variables are updated after the blocks have been allocated.
2060 get_blocks_flags = (EXT4_GET_BLOCKS_CREATE |
2061 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2062 if (mpd->b_state & (1 << BH_Delay))
2063 get_blocks_flags |= EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE;
2064 blks = ext4_get_blocks(handle, mpd->inode, next, max_blocks,
2065 &new, get_blocks_flags);
2069 * If get block returns with error we simply
2070 * return. Later writepage will redirty the page and
2071 * writepages will find the dirty page again
2076 if (err == -ENOSPC &&
2077 ext4_count_free_blocks(mpd->inode->i_sb)) {
2083 * get block failure will cause us to loop in
2084 * writepages, because a_ops->writepage won't be able
2085 * to make progress. The page will be redirtied by
2086 * writepage and writepages will again try to write
2089 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2090 "at logical offset %llu with max blocks "
2091 "%zd with error %d\n",
2092 __func__, mpd->inode->i_ino,
2093 (unsigned long long)next,
2094 mpd->b_size >> mpd->inode->i_blkbits, err);
2095 printk(KERN_EMERG "This should not happen.!! "
2096 "Data will be lost\n");
2097 if (err == -ENOSPC) {
2098 ext4_print_free_blocks(mpd->inode);
2100 /* invalidate all the pages */
2101 ext4_da_block_invalidatepages(mpd, next,
2102 mpd->b_size >> mpd->inode->i_blkbits);
2107 new.b_size = (blks << mpd->inode->i_blkbits);
2109 if (buffer_new(&new))
2110 __unmap_underlying_blocks(mpd->inode, &new);
2113 * If blocks are delayed marked, we need to
2114 * put actual blocknr and drop delayed bit
2116 if ((mpd->b_state & (1 << BH_Delay)) ||
2117 (mpd->b_state & (1 << BH_Unwritten)))
2118 mpage_put_bnr_to_bhs(mpd, next, &new);
2120 if (ext4_should_order_data(mpd->inode)) {
2121 err = ext4_jbd2_file_inode(handle, mpd->inode);
2127 * Update on-disk size along with block allocation we don't
2128 * use EXT4_GET_BLOCKS_EXTEND_DISKSIZE as size may change
2129 * within already allocated block -bzzz
2131 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2132 if (disksize > i_size_read(mpd->inode))
2133 disksize = i_size_read(mpd->inode);
2134 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2135 ext4_update_i_disksize(mpd->inode, disksize);
2136 return ext4_mark_inode_dirty(handle, mpd->inode);
2142 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2143 (1 << BH_Delay) | (1 << BH_Unwritten))
2146 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2148 * @mpd->lbh - extent of blocks
2149 * @logical - logical number of the block in the file
2150 * @bh - bh of the block (used to access block's state)
2152 * the function is used to collect contig. blocks in same state
2154 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2155 sector_t logical, size_t b_size,
2156 unsigned long b_state)
2159 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2161 /* check if thereserved journal credits might overflow */
2162 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2163 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2165 * With non-extent format we are limited by the journal
2166 * credit available. Total credit needed to insert
2167 * nrblocks contiguous blocks is dependent on the
2168 * nrblocks. So limit nrblocks.
2171 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2172 EXT4_MAX_TRANS_DATA) {
2174 * Adding the new buffer_head would make it cross the
2175 * allowed limit for which we have journal credit
2176 * reserved. So limit the new bh->b_size
2178 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2179 mpd->inode->i_blkbits;
2180 /* we will do mpage_da_submit_io in the next loop */
2184 * First block in the extent
2186 if (mpd->b_size == 0) {
2187 mpd->b_blocknr = logical;
2188 mpd->b_size = b_size;
2189 mpd->b_state = b_state & BH_FLAGS;
2193 next = mpd->b_blocknr + nrblocks;
2195 * Can we merge the block to our big extent?
2197 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2198 mpd->b_size += b_size;
2204 * We couldn't merge the block to our extent, so we
2205 * need to flush current extent and start new one
2207 if (mpage_da_map_blocks(mpd) == 0)
2208 mpage_da_submit_io(mpd);
2213 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2216 * unmapped buffer is possible for holes.
2217 * delay buffer is possible with delayed allocation.
2218 * We also need to consider unwritten buffer as unmapped.
2220 return (!buffer_mapped(bh) || buffer_delay(bh) ||
2221 buffer_unwritten(bh)) && buffer_dirty(bh);
2225 * __mpage_da_writepage - finds extent of pages and blocks
2227 * @page: page to consider
2228 * @wbc: not used, we just follow rules
2231 * The function finds extents of pages and scan them for all blocks.
2233 static int __mpage_da_writepage(struct page *page,
2234 struct writeback_control *wbc, void *data)
2236 struct mpage_da_data *mpd = data;
2237 struct inode *inode = mpd->inode;
2238 struct buffer_head *bh, *head;
2243 * Rest of the page in the page_vec
2244 * redirty then and skip then. We will
2245 * try to to write them again after
2246 * starting a new transaction
2248 redirty_page_for_writepage(wbc, page);
2250 return MPAGE_DA_EXTENT_TAIL;
2253 * Can we merge this page to current extent?
2255 if (mpd->next_page != page->index) {
2257 * Nope, we can't. So, we map non-allocated blocks
2258 * and start IO on them using writepage()
2260 if (mpd->next_page != mpd->first_page) {
2261 if (mpage_da_map_blocks(mpd) == 0)
2262 mpage_da_submit_io(mpd);
2264 * skip rest of the page in the page_vec
2267 redirty_page_for_writepage(wbc, page);
2269 return MPAGE_DA_EXTENT_TAIL;
2273 * Start next extent of pages ...
2275 mpd->first_page = page->index;
2285 mpd->next_page = page->index + 1;
2286 logical = (sector_t) page->index <<
2287 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2289 if (!page_has_buffers(page)) {
2290 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2291 (1 << BH_Dirty) | (1 << BH_Uptodate));
2293 return MPAGE_DA_EXTENT_TAIL;
2296 * Page with regular buffer heads, just add all dirty ones
2298 head = page_buffers(page);
2301 BUG_ON(buffer_locked(bh));
2303 * We need to try to allocate
2304 * unmapped blocks in the same page.
2305 * Otherwise we won't make progress
2306 * with the page in ext4_da_writepage
2308 if (ext4_bh_unmapped_or_delay(NULL, bh)) {
2309 mpage_add_bh_to_extent(mpd, logical,
2313 return MPAGE_DA_EXTENT_TAIL;
2314 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2316 * mapped dirty buffer. We need to update
2317 * the b_state because we look at
2318 * b_state in mpage_da_map_blocks. We don't
2319 * update b_size because if we find an
2320 * unmapped buffer_head later we need to
2321 * use the b_state flag of that buffer_head.
2323 if (mpd->b_size == 0)
2324 mpd->b_state = bh->b_state & BH_FLAGS;
2327 } while ((bh = bh->b_this_page) != head);
2334 * This is a special get_blocks_t callback which is used by
2335 * ext4_da_write_begin(). It will either return mapped block or
2336 * reserve space for a single block.
2338 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2339 * We also have b_blocknr = -1 and b_bdev initialized properly
2341 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2342 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2343 * initialized properly.
2345 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2346 struct buffer_head *bh_result, int create)
2349 sector_t invalid_block = ~((sector_t) 0xffff);
2351 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2354 BUG_ON(create == 0);
2355 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2358 * first, we need to know whether the block is allocated already
2359 * preallocated blocks are unmapped but should treated
2360 * the same as allocated blocks.
2362 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2363 if ((ret == 0) && !buffer_delay(bh_result)) {
2364 /* the block isn't (pre)allocated yet, let's reserve space */
2366 * XXX: __block_prepare_write() unmaps passed block,
2369 ret = ext4_da_reserve_space(inode, 1);
2371 /* not enough space to reserve */
2374 map_bh(bh_result, inode->i_sb, invalid_block);
2375 set_buffer_new(bh_result);
2376 set_buffer_delay(bh_result);
2377 } else if (ret > 0) {
2378 bh_result->b_size = (ret << inode->i_blkbits);
2379 if (buffer_unwritten(bh_result)) {
2380 /* A delayed write to unwritten bh should
2381 * be marked new and mapped. Mapped ensures
2382 * that we don't do get_block multiple times
2383 * when we write to the same offset and new
2384 * ensures that we do proper zero out for
2387 set_buffer_new(bh_result);
2388 set_buffer_mapped(bh_result);
2397 * This function is used as a standard get_block_t calback function
2398 * when there is no desire to allocate any blocks. It is used as a
2399 * callback function for block_prepare_write(), nobh_writepage(), and
2400 * block_write_full_page(). These functions should only try to map a
2401 * single block at a time.
2403 * Since this function doesn't do block allocations even if the caller
2404 * requests it by passing in create=1, it is critically important that
2405 * any caller checks to make sure that any buffer heads are returned
2406 * by this function are either all already mapped or marked for
2407 * delayed allocation before calling nobh_writepage() or
2408 * block_write_full_page(). Otherwise, b_blocknr could be left
2409 * unitialized, and the page write functions will be taken by
2412 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2413 struct buffer_head *bh_result, int create)
2416 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2418 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2421 * we don't want to do block allocation in writepage
2422 * so call get_block_wrap with create = 0
2424 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2425 BUG_ON(create && ret == 0);
2427 bh_result->b_size = (ret << inode->i_blkbits);
2434 * This function can get called via...
2435 * - ext4_da_writepages after taking page lock (have journal handle)
2436 * - journal_submit_inode_data_buffers (no journal handle)
2437 * - shrink_page_list via pdflush (no journal handle)
2438 * - grab_page_cache when doing write_begin (have journal handle)
2440 static int ext4_da_writepage(struct page *page,
2441 struct writeback_control *wbc)
2446 struct buffer_head *page_bufs;
2447 struct inode *inode = page->mapping->host;
2449 trace_mark(ext4_da_writepage,
2450 "dev %s ino %lu page_index %lu",
2451 inode->i_sb->s_id, inode->i_ino, page->index);
2452 size = i_size_read(inode);
2453 if (page->index == size >> PAGE_CACHE_SHIFT)
2454 len = size & ~PAGE_CACHE_MASK;
2456 len = PAGE_CACHE_SIZE;
2458 if (page_has_buffers(page)) {
2459 page_bufs = page_buffers(page);
2460 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2461 ext4_bh_unmapped_or_delay)) {
2463 * We don't want to do block allocation
2464 * So redirty the page and return
2465 * We may reach here when we do a journal commit
2466 * via journal_submit_inode_data_buffers.
2467 * If we don't have mapping block we just ignore
2468 * them. We can also reach here via shrink_page_list
2470 redirty_page_for_writepage(wbc, page);
2476 * The test for page_has_buffers() is subtle:
2477 * We know the page is dirty but it lost buffers. That means
2478 * that at some moment in time after write_begin()/write_end()
2479 * has been called all buffers have been clean and thus they
2480 * must have been written at least once. So they are all
2481 * mapped and we can happily proceed with mapping them
2482 * and writing the page.
2484 * Try to initialize the buffer_heads and check whether
2485 * all are mapped and non delay. We don't want to
2486 * do block allocation here.
2488 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2489 noalloc_get_block_write);
2491 page_bufs = page_buffers(page);
2492 /* check whether all are mapped and non delay */
2493 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2494 ext4_bh_unmapped_or_delay)) {
2495 redirty_page_for_writepage(wbc, page);
2501 * We can't do block allocation here
2502 * so just redity the page and unlock
2505 redirty_page_for_writepage(wbc, page);
2509 /* now mark the buffer_heads as dirty and uptodate */
2510 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2513 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2514 ret = nobh_writepage(page, noalloc_get_block_write, wbc);
2516 ret = block_write_full_page(page, noalloc_get_block_write,
2523 * This is called via ext4_da_writepages() to
2524 * calulate the total number of credits to reserve to fit
2525 * a single extent allocation into a single transaction,
2526 * ext4_da_writpeages() will loop calling this before
2527 * the block allocation.
2530 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2532 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2535 * With non-extent format the journal credit needed to
2536 * insert nrblocks contiguous block is dependent on
2537 * number of contiguous block. So we will limit
2538 * number of contiguous block to a sane value
2540 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2541 (max_blocks > EXT4_MAX_TRANS_DATA))
2542 max_blocks = EXT4_MAX_TRANS_DATA;
2544 return ext4_chunk_trans_blocks(inode, max_blocks);
2547 static int ext4_da_writepages(struct address_space *mapping,
2548 struct writeback_control *wbc)
2551 int range_whole = 0;
2552 handle_t *handle = NULL;
2553 struct mpage_da_data mpd;
2554 struct inode *inode = mapping->host;
2555 int no_nrwrite_index_update;
2556 int pages_written = 0;
2558 int range_cyclic, cycled = 1, io_done = 0;
2559 int needed_blocks, ret = 0, nr_to_writebump = 0;
2560 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2562 trace_mark(ext4_da_writepages,
2563 "dev %s ino %lu nr_t_write %ld "
2564 "pages_skipped %ld range_start %llu "
2565 "range_end %llu nonblocking %d "
2566 "for_kupdate %d for_reclaim %d "
2567 "for_writepages %d range_cyclic %d",
2568 inode->i_sb->s_id, inode->i_ino,
2569 wbc->nr_to_write, wbc->pages_skipped,
2570 (unsigned long long) wbc->range_start,
2571 (unsigned long long) wbc->range_end,
2572 wbc->nonblocking, wbc->for_kupdate,
2573 wbc->for_reclaim, wbc->for_writepages,
2577 * No pages to write? This is mainly a kludge to avoid starting
2578 * a transaction for special inodes like journal inode on last iput()
2579 * because that could violate lock ordering on umount
2581 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2585 * If the filesystem has aborted, it is read-only, so return
2586 * right away instead of dumping stack traces later on that
2587 * will obscure the real source of the problem. We test
2588 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2589 * the latter could be true if the filesystem is mounted
2590 * read-only, and in that case, ext4_da_writepages should
2591 * *never* be called, so if that ever happens, we would want
2594 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2598 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2599 * This make sure small files blocks are allocated in
2600 * single attempt. This ensure that small files
2601 * get less fragmented.
2603 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2604 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2605 wbc->nr_to_write = sbi->s_mb_stream_request;
2607 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2610 range_cyclic = wbc->range_cyclic;
2611 if (wbc->range_cyclic) {
2612 index = mapping->writeback_index;
2615 wbc->range_start = index << PAGE_CACHE_SHIFT;
2616 wbc->range_end = LLONG_MAX;
2617 wbc->range_cyclic = 0;
2619 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2622 mpd.inode = mapping->host;
2625 * we don't want write_cache_pages to update
2626 * nr_to_write and writeback_index
2628 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2629 wbc->no_nrwrite_index_update = 1;
2630 pages_skipped = wbc->pages_skipped;
2633 while (!ret && wbc->nr_to_write > 0) {
2636 * we insert one extent at a time. So we need
2637 * credit needed for single extent allocation.
2638 * journalled mode is currently not supported
2641 BUG_ON(ext4_should_journal_data(inode));
2642 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2644 /* start a new transaction*/
2645 handle = ext4_journal_start(inode, needed_blocks);
2646 if (IS_ERR(handle)) {
2647 ret = PTR_ERR(handle);
2648 printk(KERN_CRIT "%s: jbd2_start: "
2649 "%ld pages, ino %lu; err %d\n", __func__,
2650 wbc->nr_to_write, inode->i_ino, ret);
2652 goto out_writepages;
2656 * Now call __mpage_da_writepage to find the next
2657 * contiguous region of logical blocks that need
2658 * blocks to be allocated by ext4. We don't actually
2659 * submit the blocks for I/O here, even though
2660 * write_cache_pages thinks it will, and will set the
2661 * pages as clean for write before calling
2662 * __mpage_da_writepage().
2670 mpd.pages_written = 0;
2672 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2675 * If we have a contigous extent of pages and we
2676 * haven't done the I/O yet, map the blocks and submit
2679 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2680 if (mpage_da_map_blocks(&mpd) == 0)
2681 mpage_da_submit_io(&mpd);
2683 ret = MPAGE_DA_EXTENT_TAIL;
2685 wbc->nr_to_write -= mpd.pages_written;
2687 ext4_journal_stop(handle);
2689 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2690 /* commit the transaction which would
2691 * free blocks released in the transaction
2694 jbd2_journal_force_commit_nested(sbi->s_journal);
2695 wbc->pages_skipped = pages_skipped;
2697 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2699 * got one extent now try with
2702 pages_written += mpd.pages_written;
2703 wbc->pages_skipped = pages_skipped;
2706 } else if (wbc->nr_to_write)
2708 * There is no more writeout needed
2709 * or we requested for a noblocking writeout
2710 * and we found the device congested
2714 if (!io_done && !cycled) {
2717 wbc->range_start = index << PAGE_CACHE_SHIFT;
2718 wbc->range_end = mapping->writeback_index - 1;
2721 if (pages_skipped != wbc->pages_skipped)
2722 printk(KERN_EMERG "This should not happen leaving %s "
2723 "with nr_to_write = %ld ret = %d\n",
2724 __func__, wbc->nr_to_write, ret);
2727 index += pages_written;
2728 wbc->range_cyclic = range_cyclic;
2729 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2731 * set the writeback_index so that range_cyclic
2732 * mode will write it back later
2734 mapping->writeback_index = index;
2737 if (!no_nrwrite_index_update)
2738 wbc->no_nrwrite_index_update = 0;
2739 wbc->nr_to_write -= nr_to_writebump;
2740 trace_mark(ext4_da_writepage_result,
2741 "dev %s ino %lu ret %d pages_written %d "
2742 "pages_skipped %ld congestion %d "
2743 "more_io %d no_nrwrite_index_update %d",
2744 inode->i_sb->s_id, inode->i_ino, ret,
2745 pages_written, wbc->pages_skipped,
2746 wbc->encountered_congestion, wbc->more_io,
2747 wbc->no_nrwrite_index_update);
2751 #define FALL_BACK_TO_NONDELALLOC 1
2752 static int ext4_nonda_switch(struct super_block *sb)
2754 s64 free_blocks, dirty_blocks;
2755 struct ext4_sb_info *sbi = EXT4_SB(sb);
2758 * switch to non delalloc mode if we are running low
2759 * on free block. The free block accounting via percpu
2760 * counters can get slightly wrong with percpu_counter_batch getting
2761 * accumulated on each CPU without updating global counters
2762 * Delalloc need an accurate free block accounting. So switch
2763 * to non delalloc when we are near to error range.
2765 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2766 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2767 if (2 * free_blocks < 3 * dirty_blocks ||
2768 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2770 * free block count is less that 150% of dirty blocks
2771 * or free blocks is less that watermark
2778 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2779 loff_t pos, unsigned len, unsigned flags,
2780 struct page **pagep, void **fsdata)
2782 int ret, retries = 0;
2786 struct inode *inode = mapping->host;
2789 index = pos >> PAGE_CACHE_SHIFT;
2790 from = pos & (PAGE_CACHE_SIZE - 1);
2793 if (ext4_nonda_switch(inode->i_sb)) {
2794 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2795 return ext4_write_begin(file, mapping, pos,
2796 len, flags, pagep, fsdata);
2798 *fsdata = (void *)0;
2800 trace_mark(ext4_da_write_begin,
2801 "dev %s ino %lu pos %llu len %u flags %u",
2802 inode->i_sb->s_id, inode->i_ino,
2803 (unsigned long long) pos, len, flags);
2806 * With delayed allocation, we don't log the i_disksize update
2807 * if there is delayed block allocation. But we still need
2808 * to journalling the i_disksize update if writes to the end
2809 * of file which has an already mapped buffer.
2811 handle = ext4_journal_start(inode, 1);
2812 if (IS_ERR(handle)) {
2813 ret = PTR_ERR(handle);
2816 /* We cannot recurse into the filesystem as the transaction is already
2818 flags |= AOP_FLAG_NOFS;
2820 page = grab_cache_page_write_begin(mapping, index, flags);
2822 ext4_journal_stop(handle);
2828 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2829 ext4_da_get_block_prep);
2832 ext4_journal_stop(handle);
2833 page_cache_release(page);
2835 * block_write_begin may have instantiated a few blocks
2836 * outside i_size. Trim these off again. Don't need
2837 * i_size_read because we hold i_mutex.
2839 if (pos + len > inode->i_size)
2840 vmtruncate(inode, inode->i_size);
2843 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2850 * Check if we should update i_disksize
2851 * when write to the end of file but not require block allocation
2853 static int ext4_da_should_update_i_disksize(struct page *page,
2854 unsigned long offset)
2856 struct buffer_head *bh;
2857 struct inode *inode = page->mapping->host;
2861 bh = page_buffers(page);
2862 idx = offset >> inode->i_blkbits;
2864 for (i = 0; i < idx; i++)
2865 bh = bh->b_this_page;
2867 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2872 static int ext4_da_write_end(struct file *file,
2873 struct address_space *mapping,
2874 loff_t pos, unsigned len, unsigned copied,
2875 struct page *page, void *fsdata)
2877 struct inode *inode = mapping->host;
2879 handle_t *handle = ext4_journal_current_handle();
2881 unsigned long start, end;
2882 int write_mode = (int)(unsigned long)fsdata;
2884 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2885 if (ext4_should_order_data(inode)) {
2886 return ext4_ordered_write_end(file, mapping, pos,
2887 len, copied, page, fsdata);
2888 } else if (ext4_should_writeback_data(inode)) {
2889 return ext4_writeback_write_end(file, mapping, pos,
2890 len, copied, page, fsdata);
2896 trace_mark(ext4_da_write_end,
2897 "dev %s ino %lu pos %llu len %u copied %u",
2898 inode->i_sb->s_id, inode->i_ino,
2899 (unsigned long long) pos, len, copied);
2900 start = pos & (PAGE_CACHE_SIZE - 1);
2901 end = start + copied - 1;
2904 * generic_write_end() will run mark_inode_dirty() if i_size
2905 * changes. So let's piggyback the i_disksize mark_inode_dirty
2909 new_i_size = pos + copied;
2910 if (new_i_size > EXT4_I(inode)->i_disksize) {
2911 if (ext4_da_should_update_i_disksize(page, end)) {
2912 down_write(&EXT4_I(inode)->i_data_sem);
2913 if (new_i_size > EXT4_I(inode)->i_disksize) {
2915 * Updating i_disksize when extending file
2916 * without needing block allocation
2918 if (ext4_should_order_data(inode))
2919 ret = ext4_jbd2_file_inode(handle,
2922 EXT4_I(inode)->i_disksize = new_i_size;
2924 up_write(&EXT4_I(inode)->i_data_sem);
2925 /* We need to mark inode dirty even if
2926 * new_i_size is less that inode->i_size
2927 * bu greater than i_disksize.(hint delalloc)
2929 ext4_mark_inode_dirty(handle, inode);
2932 ret2 = generic_write_end(file, mapping, pos, len, copied,
2937 ret2 = ext4_journal_stop(handle);
2941 return ret ? ret : copied;
2944 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2947 * Drop reserved blocks
2949 BUG_ON(!PageLocked(page));
2950 if (!page_has_buffers(page))
2953 ext4_da_page_release_reservation(page, offset);
2956 ext4_invalidatepage(page, offset);
2962 * Force all delayed allocation blocks to be allocated for a given inode.
2964 int ext4_alloc_da_blocks(struct inode *inode)
2966 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2967 !EXT4_I(inode)->i_reserved_meta_blocks)
2971 * We do something simple for now. The filemap_flush() will
2972 * also start triggering a write of the data blocks, which is
2973 * not strictly speaking necessary (and for users of
2974 * laptop_mode, not even desirable). However, to do otherwise
2975 * would require replicating code paths in:
2977 * ext4_da_writepages() ->
2978 * write_cache_pages() ---> (via passed in callback function)
2979 * __mpage_da_writepage() -->
2980 * mpage_add_bh_to_extent()
2981 * mpage_da_map_blocks()
2983 * The problem is that write_cache_pages(), located in
2984 * mm/page-writeback.c, marks pages clean in preparation for
2985 * doing I/O, which is not desirable if we're not planning on
2988 * We could call write_cache_pages(), and then redirty all of
2989 * the pages by calling redirty_page_for_writeback() but that
2990 * would be ugly in the extreme. So instead we would need to
2991 * replicate parts of the code in the above functions,
2992 * simplifying them becuase we wouldn't actually intend to
2993 * write out the pages, but rather only collect contiguous
2994 * logical block extents, call the multi-block allocator, and
2995 * then update the buffer heads with the block allocations.
2997 * For now, though, we'll cheat by calling filemap_flush(),
2998 * which will map the blocks, and start the I/O, but not
2999 * actually wait for the I/O to complete.
3001 return filemap_flush(inode->i_mapping);
3005 * bmap() is special. It gets used by applications such as lilo and by
3006 * the swapper to find the on-disk block of a specific piece of data.
3008 * Naturally, this is dangerous if the block concerned is still in the
3009 * journal. If somebody makes a swapfile on an ext4 data-journaling
3010 * filesystem and enables swap, then they may get a nasty shock when the
3011 * data getting swapped to that swapfile suddenly gets overwritten by
3012 * the original zero's written out previously to the journal and
3013 * awaiting writeback in the kernel's buffer cache.
3015 * So, if we see any bmap calls here on a modified, data-journaled file,
3016 * take extra steps to flush any blocks which might be in the cache.
3018 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3020 struct inode *inode = mapping->host;
3024 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3025 test_opt(inode->i_sb, DELALLOC)) {
3027 * With delalloc we want to sync the file
3028 * so that we can make sure we allocate
3031 filemap_write_and_wait(mapping);
3034 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3036 * This is a REALLY heavyweight approach, but the use of
3037 * bmap on dirty files is expected to be extremely rare:
3038 * only if we run lilo or swapon on a freshly made file
3039 * do we expect this to happen.
3041 * (bmap requires CAP_SYS_RAWIO so this does not
3042 * represent an unprivileged user DOS attack --- we'd be
3043 * in trouble if mortal users could trigger this path at
3046 * NB. EXT4_STATE_JDATA is not set on files other than
3047 * regular files. If somebody wants to bmap a directory
3048 * or symlink and gets confused because the buffer
3049 * hasn't yet been flushed to disk, they deserve
3050 * everything they get.
3053 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3054 journal = EXT4_JOURNAL(inode);
3055 jbd2_journal_lock_updates(journal);
3056 err = jbd2_journal_flush(journal);
3057 jbd2_journal_unlock_updates(journal);
3063 return generic_block_bmap(mapping, block, ext4_get_block);
3066 static int bget_one(handle_t *handle, struct buffer_head *bh)
3072 static int bput_one(handle_t *handle, struct buffer_head *bh)
3079 * Note that we don't need to start a transaction unless we're journaling data
3080 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3081 * need to file the inode to the transaction's list in ordered mode because if
3082 * we are writing back data added by write(), the inode is already there and if
3083 * we are writing back data modified via mmap(), noone guarantees in which
3084 * transaction the data will hit the disk. In case we are journaling data, we
3085 * cannot start transaction directly because transaction start ranks above page
3086 * lock so we have to do some magic.
3088 * In all journaling modes block_write_full_page() will start the I/O.
3092 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3097 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3099 * Same applies to ext4_get_block(). We will deadlock on various things like
3100 * lock_journal and i_data_sem
3102 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3105 * 16May01: If we're reentered then journal_current_handle() will be
3106 * non-zero. We simply *return*.
3108 * 1 July 2001: @@@ FIXME:
3109 * In journalled data mode, a data buffer may be metadata against the
3110 * current transaction. But the same file is part of a shared mapping
3111 * and someone does a writepage() on it.
3113 * We will move the buffer onto the async_data list, but *after* it has
3114 * been dirtied. So there's a small window where we have dirty data on
3117 * Note that this only applies to the last partial page in the file. The
3118 * bit which block_write_full_page() uses prepare/commit for. (That's
3119 * broken code anyway: it's wrong for msync()).
3121 * It's a rare case: affects the final partial page, for journalled data
3122 * where the file is subject to bith write() and writepage() in the same
3123 * transction. To fix it we'll need a custom block_write_full_page().
3124 * We'll probably need that anyway for journalling writepage() output.
3126 * We don't honour synchronous mounts for writepage(). That would be
3127 * disastrous. Any write() or metadata operation will sync the fs for
3131 static int __ext4_normal_writepage(struct page *page,
3132 struct writeback_control *wbc)
3134 struct inode *inode = page->mapping->host;
3136 if (test_opt(inode->i_sb, NOBH))
3137 return nobh_writepage(page, noalloc_get_block_write, wbc);
3139 return block_write_full_page(page, noalloc_get_block_write,
3143 static int ext4_normal_writepage(struct page *page,
3144 struct writeback_control *wbc)
3146 struct inode *inode = page->mapping->host;
3147 loff_t size = i_size_read(inode);
3150 trace_mark(ext4_normal_writepage,
3151 "dev %s ino %lu page_index %lu",
3152 inode->i_sb->s_id, inode->i_ino, page->index);
3153 J_ASSERT(PageLocked(page));
3154 if (page->index == size >> PAGE_CACHE_SHIFT)
3155 len = size & ~PAGE_CACHE_MASK;
3157 len = PAGE_CACHE_SIZE;
3159 if (page_has_buffers(page)) {
3160 /* if page has buffers it should all be mapped
3161 * and allocated. If there are not buffers attached
3162 * to the page we know the page is dirty but it lost
3163 * buffers. That means that at some moment in time
3164 * after write_begin() / write_end() has been called
3165 * all buffers have been clean and thus they must have been
3166 * written at least once. So they are all mapped and we can
3167 * happily proceed with mapping them and writing the page.
3169 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3170 ext4_bh_unmapped_or_delay));
3173 if (!ext4_journal_current_handle())
3174 return __ext4_normal_writepage(page, wbc);
3176 redirty_page_for_writepage(wbc, page);
3181 static int __ext4_journalled_writepage(struct page *page,
3182 struct writeback_control *wbc)
3184 struct address_space *mapping = page->mapping;
3185 struct inode *inode = mapping->host;
3186 struct buffer_head *page_bufs;
3187 handle_t *handle = NULL;
3191 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3192 noalloc_get_block_write);
3196 page_bufs = page_buffers(page);
3197 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3199 /* As soon as we unlock the page, it can go away, but we have
3200 * references to buffers so we are safe */
3203 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3204 if (IS_ERR(handle)) {
3205 ret = PTR_ERR(handle);
3209 ret = walk_page_buffers(handle, page_bufs, 0,
3210 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3212 err = walk_page_buffers(handle, page_bufs, 0,
3213 PAGE_CACHE_SIZE, NULL, write_end_fn);
3216 err = ext4_journal_stop(handle);
3220 walk_page_buffers(handle, page_bufs, 0,
3221 PAGE_CACHE_SIZE, NULL, bput_one);
3222 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3231 static int ext4_journalled_writepage(struct page *page,
3232 struct writeback_control *wbc)
3234 struct inode *inode = page->mapping->host;
3235 loff_t size = i_size_read(inode);
3238 trace_mark(ext4_journalled_writepage,
3239 "dev %s ino %lu page_index %lu",
3240 inode->i_sb->s_id, inode->i_ino, page->index);
3241 J_ASSERT(PageLocked(page));
3242 if (page->index == size >> PAGE_CACHE_SHIFT)
3243 len = size & ~PAGE_CACHE_MASK;
3245 len = PAGE_CACHE_SIZE;
3247 if (page_has_buffers(page)) {
3248 /* if page has buffers it should all be mapped
3249 * and allocated. If there are not buffers attached
3250 * to the page we know the page is dirty but it lost
3251 * buffers. That means that at some moment in time
3252 * after write_begin() / write_end() has been called
3253 * all buffers have been clean and thus they must have been
3254 * written at least once. So they are all mapped and we can
3255 * happily proceed with mapping them and writing the page.
3257 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3258 ext4_bh_unmapped_or_delay));
3261 if (ext4_journal_current_handle())
3264 if (PageChecked(page)) {
3266 * It's mmapped pagecache. Add buffers and journal it. There
3267 * doesn't seem much point in redirtying the page here.
3269 ClearPageChecked(page);
3270 return __ext4_journalled_writepage(page, wbc);
3273 * It may be a page full of checkpoint-mode buffers. We don't
3274 * really know unless we go poke around in the buffer_heads.
3275 * But block_write_full_page will do the right thing.
3277 return block_write_full_page(page, noalloc_get_block_write,
3281 redirty_page_for_writepage(wbc, page);
3286 static int ext4_readpage(struct file *file, struct page *page)
3288 return mpage_readpage(page, ext4_get_block);
3292 ext4_readpages(struct file *file, struct address_space *mapping,
3293 struct list_head *pages, unsigned nr_pages)
3295 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3298 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3300 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3303 * If it's a full truncate we just forget about the pending dirtying
3306 ClearPageChecked(page);
3309 jbd2_journal_invalidatepage(journal, page, offset);
3311 block_invalidatepage(page, offset);
3314 static int ext4_releasepage(struct page *page, gfp_t wait)
3316 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3318 WARN_ON(PageChecked(page));
3319 if (!page_has_buffers(page))
3322 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3324 return try_to_free_buffers(page);
3328 * If the O_DIRECT write will extend the file then add this inode to the
3329 * orphan list. So recovery will truncate it back to the original size
3330 * if the machine crashes during the write.
3332 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3333 * crashes then stale disk data _may_ be exposed inside the file. But current
3334 * VFS code falls back into buffered path in that case so we are safe.
3336 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3337 const struct iovec *iov, loff_t offset,
3338 unsigned long nr_segs)
3340 struct file *file = iocb->ki_filp;
3341 struct inode *inode = file->f_mapping->host;
3342 struct ext4_inode_info *ei = EXT4_I(inode);
3346 size_t count = iov_length(iov, nr_segs);
3349 loff_t final_size = offset + count;
3351 if (final_size > inode->i_size) {
3352 /* Credits for sb + inode write */
3353 handle = ext4_journal_start(inode, 2);
3354 if (IS_ERR(handle)) {
3355 ret = PTR_ERR(handle);
3358 ret = ext4_orphan_add(handle, inode);
3360 ext4_journal_stop(handle);
3364 ei->i_disksize = inode->i_size;
3365 ext4_journal_stop(handle);
3369 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3371 ext4_get_block, NULL);
3376 /* Credits for sb + inode write */
3377 handle = ext4_journal_start(inode, 2);
3378 if (IS_ERR(handle)) {
3379 /* This is really bad luck. We've written the data
3380 * but cannot extend i_size. Bail out and pretend
3381 * the write failed... */
3382 ret = PTR_ERR(handle);
3386 ext4_orphan_del(handle, inode);
3388 loff_t end = offset + ret;
3389 if (end > inode->i_size) {
3390 ei->i_disksize = end;
3391 i_size_write(inode, end);
3393 * We're going to return a positive `ret'
3394 * here due to non-zero-length I/O, so there's
3395 * no way of reporting error returns from
3396 * ext4_mark_inode_dirty() to userspace. So
3399 ext4_mark_inode_dirty(handle, inode);
3402 err = ext4_journal_stop(handle);
3411 * Pages can be marked dirty completely asynchronously from ext4's journalling
3412 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3413 * much here because ->set_page_dirty is called under VFS locks. The page is
3414 * not necessarily locked.
3416 * We cannot just dirty the page and leave attached buffers clean, because the
3417 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3418 * or jbddirty because all the journalling code will explode.
3420 * So what we do is to mark the page "pending dirty" and next time writepage
3421 * is called, propagate that into the buffers appropriately.
3423 static int ext4_journalled_set_page_dirty(struct page *page)
3425 SetPageChecked(page);
3426 return __set_page_dirty_nobuffers(page);
3429 static const struct address_space_operations ext4_ordered_aops = {
3430 .readpage = ext4_readpage,
3431 .readpages = ext4_readpages,
3432 .writepage = ext4_normal_writepage,
3433 .sync_page = block_sync_page,
3434 .write_begin = ext4_write_begin,
3435 .write_end = ext4_ordered_write_end,
3437 .invalidatepage = ext4_invalidatepage,
3438 .releasepage = ext4_releasepage,
3439 .direct_IO = ext4_direct_IO,
3440 .migratepage = buffer_migrate_page,
3441 .is_partially_uptodate = block_is_partially_uptodate,
3444 static const struct address_space_operations ext4_writeback_aops = {
3445 .readpage = ext4_readpage,
3446 .readpages = ext4_readpages,
3447 .writepage = ext4_normal_writepage,
3448 .sync_page = block_sync_page,
3449 .write_begin = ext4_write_begin,
3450 .write_end = ext4_writeback_write_end,
3452 .invalidatepage = ext4_invalidatepage,
3453 .releasepage = ext4_releasepage,
3454 .direct_IO = ext4_direct_IO,
3455 .migratepage = buffer_migrate_page,
3456 .is_partially_uptodate = block_is_partially_uptodate,
3459 static const struct address_space_operations ext4_journalled_aops = {
3460 .readpage = ext4_readpage,
3461 .readpages = ext4_readpages,
3462 .writepage = ext4_journalled_writepage,
3463 .sync_page = block_sync_page,
3464 .write_begin = ext4_write_begin,
3465 .write_end = ext4_journalled_write_end,
3466 .set_page_dirty = ext4_journalled_set_page_dirty,
3468 .invalidatepage = ext4_invalidatepage,
3469 .releasepage = ext4_releasepage,
3470 .is_partially_uptodate = block_is_partially_uptodate,
3473 static const struct address_space_operations ext4_da_aops = {
3474 .readpage = ext4_readpage,
3475 .readpages = ext4_readpages,
3476 .writepage = ext4_da_writepage,
3477 .writepages = ext4_da_writepages,
3478 .sync_page = block_sync_page,
3479 .write_begin = ext4_da_write_begin,
3480 .write_end = ext4_da_write_end,
3482 .invalidatepage = ext4_da_invalidatepage,
3483 .releasepage = ext4_releasepage,
3484 .direct_IO = ext4_direct_IO,
3485 .migratepage = buffer_migrate_page,
3486 .is_partially_uptodate = block_is_partially_uptodate,
3489 void ext4_set_aops(struct inode *inode)
3491 if (ext4_should_order_data(inode) &&
3492 test_opt(inode->i_sb, DELALLOC))
3493 inode->i_mapping->a_ops = &ext4_da_aops;
3494 else if (ext4_should_order_data(inode))
3495 inode->i_mapping->a_ops = &ext4_ordered_aops;
3496 else if (ext4_should_writeback_data(inode) &&
3497 test_opt(inode->i_sb, DELALLOC))
3498 inode->i_mapping->a_ops = &ext4_da_aops;
3499 else if (ext4_should_writeback_data(inode))
3500 inode->i_mapping->a_ops = &ext4_writeback_aops;
3502 inode->i_mapping->a_ops = &ext4_journalled_aops;
3506 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3507 * up to the end of the block which corresponds to `from'.
3508 * This required during truncate. We need to physically zero the tail end
3509 * of that block so it doesn't yield old data if the file is later grown.
3511 int ext4_block_truncate_page(handle_t *handle,
3512 struct address_space *mapping, loff_t from)
3514 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3515 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3516 unsigned blocksize, length, pos;
3518 struct inode *inode = mapping->host;
3519 struct buffer_head *bh;
3523 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3527 blocksize = inode->i_sb->s_blocksize;
3528 length = blocksize - (offset & (blocksize - 1));
3529 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3532 * For "nobh" option, we can only work if we don't need to
3533 * read-in the page - otherwise we create buffers to do the IO.
3535 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3536 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3537 zero_user(page, offset, length);
3538 set_page_dirty(page);
3542 if (!page_has_buffers(page))
3543 create_empty_buffers(page, blocksize, 0);
3545 /* Find the buffer that contains "offset" */
3546 bh = page_buffers(page);
3548 while (offset >= pos) {
3549 bh = bh->b_this_page;
3555 if (buffer_freed(bh)) {
3556 BUFFER_TRACE(bh, "freed: skip");
3560 if (!buffer_mapped(bh)) {
3561 BUFFER_TRACE(bh, "unmapped");
3562 ext4_get_block(inode, iblock, bh, 0);
3563 /* unmapped? It's a hole - nothing to do */
3564 if (!buffer_mapped(bh)) {
3565 BUFFER_TRACE(bh, "still unmapped");
3570 /* Ok, it's mapped. Make sure it's up-to-date */
3571 if (PageUptodate(page))
3572 set_buffer_uptodate(bh);
3574 if (!buffer_uptodate(bh)) {
3576 ll_rw_block(READ, 1, &bh);
3578 /* Uhhuh. Read error. Complain and punt. */
3579 if (!buffer_uptodate(bh))
3583 if (ext4_should_journal_data(inode)) {
3584 BUFFER_TRACE(bh, "get write access");
3585 err = ext4_journal_get_write_access(handle, bh);
3590 zero_user(page, offset, length);
3592 BUFFER_TRACE(bh, "zeroed end of block");
3595 if (ext4_should_journal_data(inode)) {
3596 err = ext4_handle_dirty_metadata(handle, inode, bh);
3598 if (ext4_should_order_data(inode))
3599 err = ext4_jbd2_file_inode(handle, inode);
3600 mark_buffer_dirty(bh);
3605 page_cache_release(page);
3610 * Probably it should be a library function... search for first non-zero word
3611 * or memcmp with zero_page, whatever is better for particular architecture.
3614 static inline int all_zeroes(__le32 *p, __le32 *q)
3623 * ext4_find_shared - find the indirect blocks for partial truncation.
3624 * @inode: inode in question
3625 * @depth: depth of the affected branch
3626 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3627 * @chain: place to store the pointers to partial indirect blocks
3628 * @top: place to the (detached) top of branch
3630 * This is a helper function used by ext4_truncate().
3632 * When we do truncate() we may have to clean the ends of several
3633 * indirect blocks but leave the blocks themselves alive. Block is
3634 * partially truncated if some data below the new i_size is refered
3635 * from it (and it is on the path to the first completely truncated
3636 * data block, indeed). We have to free the top of that path along
3637 * with everything to the right of the path. Since no allocation
3638 * past the truncation point is possible until ext4_truncate()
3639 * finishes, we may safely do the latter, but top of branch may
3640 * require special attention - pageout below the truncation point
3641 * might try to populate it.
3643 * We atomically detach the top of branch from the tree, store the
3644 * block number of its root in *@top, pointers to buffer_heads of
3645 * partially truncated blocks - in @chain[].bh and pointers to
3646 * their last elements that should not be removed - in
3647 * @chain[].p. Return value is the pointer to last filled element
3650 * The work left to caller to do the actual freeing of subtrees:
3651 * a) free the subtree starting from *@top
3652 * b) free the subtrees whose roots are stored in
3653 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3654 * c) free the subtrees growing from the inode past the @chain[0].
3655 * (no partially truncated stuff there). */
3657 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3658 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3660 Indirect *partial, *p;
3664 /* Make k index the deepest non-null offest + 1 */
3665 for (k = depth; k > 1 && !offsets[k-1]; k--)
3667 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3668 /* Writer: pointers */
3670 partial = chain + k-1;
3672 * If the branch acquired continuation since we've looked at it -
3673 * fine, it should all survive and (new) top doesn't belong to us.
3675 if (!partial->key && *partial->p)
3678 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3681 * OK, we've found the last block that must survive. The rest of our
3682 * branch should be detached before unlocking. However, if that rest
3683 * of branch is all ours and does not grow immediately from the inode
3684 * it's easier to cheat and just decrement partial->p.
3686 if (p == chain + k - 1 && p > chain) {
3690 /* Nope, don't do this in ext4. Must leave the tree intact */
3697 while (partial > p) {
3698 brelse(partial->bh);
3706 * Zero a number of block pointers in either an inode or an indirect block.
3707 * If we restart the transaction we must again get write access to the
3708 * indirect block for further modification.
3710 * We release `count' blocks on disk, but (last - first) may be greater
3711 * than `count' because there can be holes in there.
3713 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3714 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3715 unsigned long count, __le32 *first, __le32 *last)
3718 if (try_to_extend_transaction(handle, inode)) {
3720 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3721 ext4_handle_dirty_metadata(handle, inode, bh);
3723 ext4_mark_inode_dirty(handle, inode);
3724 ext4_journal_test_restart(handle, inode);
3726 BUFFER_TRACE(bh, "retaking write access");
3727 ext4_journal_get_write_access(handle, bh);
3732 * Any buffers which are on the journal will be in memory. We find
3733 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3734 * on them. We've already detached each block from the file, so
3735 * bforget() in jbd2_journal_forget() should be safe.
3737 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3739 for (p = first; p < last; p++) {
3740 u32 nr = le32_to_cpu(*p);
3742 struct buffer_head *tbh;
3745 tbh = sb_find_get_block(inode->i_sb, nr);
3746 ext4_forget(handle, 0, inode, tbh, nr);
3750 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3754 * ext4_free_data - free a list of data blocks
3755 * @handle: handle for this transaction
3756 * @inode: inode we are dealing with
3757 * @this_bh: indirect buffer_head which contains *@first and *@last
3758 * @first: array of block numbers
3759 * @last: points immediately past the end of array
3761 * We are freeing all blocks refered from that array (numbers are stored as
3762 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3764 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3765 * blocks are contiguous then releasing them at one time will only affect one
3766 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3767 * actually use a lot of journal space.
3769 * @this_bh will be %NULL if @first and @last point into the inode's direct
3772 static void ext4_free_data(handle_t *handle, struct inode *inode,
3773 struct buffer_head *this_bh,
3774 __le32 *first, __le32 *last)
3776 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3777 unsigned long count = 0; /* Number of blocks in the run */
3778 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3781 ext4_fsblk_t nr; /* Current block # */
3782 __le32 *p; /* Pointer into inode/ind
3783 for current block */
3786 if (this_bh) { /* For indirect block */
3787 BUFFER_TRACE(this_bh, "get_write_access");
3788 err = ext4_journal_get_write_access(handle, this_bh);
3789 /* Important: if we can't update the indirect pointers
3790 * to the blocks, we can't free them. */
3795 for (p = first; p < last; p++) {
3796 nr = le32_to_cpu(*p);
3798 /* accumulate blocks to free if they're contiguous */
3801 block_to_free_p = p;
3803 } else if (nr == block_to_free + count) {
3806 ext4_clear_blocks(handle, inode, this_bh,
3808 count, block_to_free_p, p);
3810 block_to_free_p = p;
3817 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3818 count, block_to_free_p, p);
3821 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3824 * The buffer head should have an attached journal head at this
3825 * point. However, if the data is corrupted and an indirect
3826 * block pointed to itself, it would have been detached when
3827 * the block was cleared. Check for this instead of OOPSing.
3829 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3830 ext4_handle_dirty_metadata(handle, inode, this_bh);
3832 ext4_error(inode->i_sb, __func__,
3833 "circular indirect block detected, "
3834 "inode=%lu, block=%llu",
3836 (unsigned long long) this_bh->b_blocknr);
3841 * ext4_free_branches - free an array of branches
3842 * @handle: JBD handle for this transaction
3843 * @inode: inode we are dealing with
3844 * @parent_bh: the buffer_head which contains *@first and *@last
3845 * @first: array of block numbers
3846 * @last: pointer immediately past the end of array
3847 * @depth: depth of the branches to free
3849 * We are freeing all blocks refered from these branches (numbers are
3850 * stored as little-endian 32-bit) and updating @inode->i_blocks
3853 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3854 struct buffer_head *parent_bh,
3855 __le32 *first, __le32 *last, int depth)
3860 if (ext4_handle_is_aborted(handle))
3864 struct buffer_head *bh;
3865 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3867 while (--p >= first) {
3868 nr = le32_to_cpu(*p);
3870 continue; /* A hole */
3872 /* Go read the buffer for the next level down */
3873 bh = sb_bread(inode->i_sb, nr);
3876 * A read failure? Report error and clear slot
3880 ext4_error(inode->i_sb, "ext4_free_branches",
3881 "Read failure, inode=%lu, block=%llu",
3886 /* This zaps the entire block. Bottom up. */
3887 BUFFER_TRACE(bh, "free child branches");
3888 ext4_free_branches(handle, inode, bh,
3889 (__le32 *) bh->b_data,
3890 (__le32 *) bh->b_data + addr_per_block,
3894 * We've probably journalled the indirect block several
3895 * times during the truncate. But it's no longer
3896 * needed and we now drop it from the transaction via
3897 * jbd2_journal_revoke().
3899 * That's easy if it's exclusively part of this
3900 * transaction. But if it's part of the committing
3901 * transaction then jbd2_journal_forget() will simply
3902 * brelse() it. That means that if the underlying
3903 * block is reallocated in ext4_get_block(),
3904 * unmap_underlying_metadata() will find this block
3905 * and will try to get rid of it. damn, damn.
3907 * If this block has already been committed to the
3908 * journal, a revoke record will be written. And
3909 * revoke records must be emitted *before* clearing
3910 * this block's bit in the bitmaps.
3912 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3915 * Everything below this this pointer has been
3916 * released. Now let this top-of-subtree go.
3918 * We want the freeing of this indirect block to be
3919 * atomic in the journal with the updating of the
3920 * bitmap block which owns it. So make some room in
3923 * We zero the parent pointer *after* freeing its
3924 * pointee in the bitmaps, so if extend_transaction()
3925 * for some reason fails to put the bitmap changes and
3926 * the release into the same transaction, recovery
3927 * will merely complain about releasing a free block,
3928 * rather than leaking blocks.
3930 if (ext4_handle_is_aborted(handle))
3932 if (try_to_extend_transaction(handle, inode)) {
3933 ext4_mark_inode_dirty(handle, inode);
3934 ext4_journal_test_restart(handle, inode);
3937 ext4_free_blocks(handle, inode, nr, 1, 1);
3941 * The block which we have just freed is
3942 * pointed to by an indirect block: journal it
3944 BUFFER_TRACE(parent_bh, "get_write_access");
3945 if (!ext4_journal_get_write_access(handle,
3948 BUFFER_TRACE(parent_bh,
3949 "call ext4_handle_dirty_metadata");
3950 ext4_handle_dirty_metadata(handle,
3957 /* We have reached the bottom of the tree. */
3958 BUFFER_TRACE(parent_bh, "free data blocks");
3959 ext4_free_data(handle, inode, parent_bh, first, last);
3963 int ext4_can_truncate(struct inode *inode)
3965 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3967 if (S_ISREG(inode->i_mode))
3969 if (S_ISDIR(inode->i_mode))
3971 if (S_ISLNK(inode->i_mode))
3972 return !ext4_inode_is_fast_symlink(inode);
3979 * We block out ext4_get_block() block instantiations across the entire
3980 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3981 * simultaneously on behalf of the same inode.
3983 * As we work through the truncate and commmit bits of it to the journal there
3984 * is one core, guiding principle: the file's tree must always be consistent on
3985 * disk. We must be able to restart the truncate after a crash.
3987 * The file's tree may be transiently inconsistent in memory (although it
3988 * probably isn't), but whenever we close off and commit a journal transaction,
3989 * the contents of (the filesystem + the journal) must be consistent and
3990 * restartable. It's pretty simple, really: bottom up, right to left (although
3991 * left-to-right works OK too).
3993 * Note that at recovery time, journal replay occurs *before* the restart of
3994 * truncate against the orphan inode list.
3996 * The committed inode has the new, desired i_size (which is the same as
3997 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3998 * that this inode's truncate did not complete and it will again call
3999 * ext4_truncate() to have another go. So there will be instantiated blocks
4000 * to the right of the truncation point in a crashed ext4 filesystem. But
4001 * that's fine - as long as they are linked from the inode, the post-crash
4002 * ext4_truncate() run will find them and release them.
4004 void ext4_truncate(struct inode *inode)
4007 struct ext4_inode_info *ei = EXT4_I(inode);
4008 __le32 *i_data = ei->i_data;
4009 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4010 struct address_space *mapping = inode->i_mapping;
4011 ext4_lblk_t offsets[4];
4016 ext4_lblk_t last_block;
4017 unsigned blocksize = inode->i_sb->s_blocksize;
4019 if (!ext4_can_truncate(inode))
4022 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4023 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4025 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4026 ext4_ext_truncate(inode);
4030 handle = start_transaction(inode);
4032 return; /* AKPM: return what? */
4034 last_block = (inode->i_size + blocksize-1)
4035 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4037 if (inode->i_size & (blocksize - 1))
4038 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4041 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4043 goto out_stop; /* error */
4046 * OK. This truncate is going to happen. We add the inode to the
4047 * orphan list, so that if this truncate spans multiple transactions,
4048 * and we crash, we will resume the truncate when the filesystem
4049 * recovers. It also marks the inode dirty, to catch the new size.
4051 * Implication: the file must always be in a sane, consistent
4052 * truncatable state while each transaction commits.
4054 if (ext4_orphan_add(handle, inode))
4058 * From here we block out all ext4_get_block() callers who want to
4059 * modify the block allocation tree.
4061 down_write(&ei->i_data_sem);
4063 ext4_discard_preallocations(inode);
4066 * The orphan list entry will now protect us from any crash which
4067 * occurs before the truncate completes, so it is now safe to propagate
4068 * the new, shorter inode size (held for now in i_size) into the
4069 * on-disk inode. We do this via i_disksize, which is the value which
4070 * ext4 *really* writes onto the disk inode.
4072 ei->i_disksize = inode->i_size;
4074 if (n == 1) { /* direct blocks */
4075 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4076 i_data + EXT4_NDIR_BLOCKS);
4080 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4081 /* Kill the top of shared branch (not detached) */
4083 if (partial == chain) {
4084 /* Shared branch grows from the inode */
4085 ext4_free_branches(handle, inode, NULL,
4086 &nr, &nr+1, (chain+n-1) - partial);
4089 * We mark the inode dirty prior to restart,
4090 * and prior to stop. No need for it here.
4093 /* Shared branch grows from an indirect block */
4094 BUFFER_TRACE(partial->bh, "get_write_access");
4095 ext4_free_branches(handle, inode, partial->bh,
4097 partial->p+1, (chain+n-1) - partial);
4100 /* Clear the ends of indirect blocks on the shared branch */
4101 while (partial > chain) {
4102 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4103 (__le32*)partial->bh->b_data+addr_per_block,
4104 (chain+n-1) - partial);
4105 BUFFER_TRACE(partial->bh, "call brelse");
4106 brelse (partial->bh);
4110 /* Kill the remaining (whole) subtrees */
4111 switch (offsets[0]) {
4113 nr = i_data[EXT4_IND_BLOCK];
4115 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4116 i_data[EXT4_IND_BLOCK] = 0;
4118 case EXT4_IND_BLOCK:
4119 nr = i_data[EXT4_DIND_BLOCK];
4121 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4122 i_data[EXT4_DIND_BLOCK] = 0;
4124 case EXT4_DIND_BLOCK:
4125 nr = i_data[EXT4_TIND_BLOCK];
4127 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4128 i_data[EXT4_TIND_BLOCK] = 0;
4130 case EXT4_TIND_BLOCK:
4134 up_write(&ei->i_data_sem);
4135 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4136 ext4_mark_inode_dirty(handle, inode);
4139 * In a multi-transaction truncate, we only make the final transaction
4143 ext4_handle_sync(handle);
4146 * If this was a simple ftruncate(), and the file will remain alive
4147 * then we need to clear up the orphan record which we created above.
4148 * However, if this was a real unlink then we were called by
4149 * ext4_delete_inode(), and we allow that function to clean up the
4150 * orphan info for us.
4153 ext4_orphan_del(handle, inode);
4155 ext4_journal_stop(handle);
4159 * ext4_get_inode_loc returns with an extra refcount against the inode's
4160 * underlying buffer_head on success. If 'in_mem' is true, we have all
4161 * data in memory that is needed to recreate the on-disk version of this
4164 static int __ext4_get_inode_loc(struct inode *inode,
4165 struct ext4_iloc *iloc, int in_mem)
4167 struct ext4_group_desc *gdp;
4168 struct buffer_head *bh;
4169 struct super_block *sb = inode->i_sb;
4171 int inodes_per_block, inode_offset;
4174 if (!ext4_valid_inum(sb, inode->i_ino))
4177 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4178 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4183 * Figure out the offset within the block group inode table
4185 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4186 inode_offset = ((inode->i_ino - 1) %
4187 EXT4_INODES_PER_GROUP(sb));
4188 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4189 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4191 bh = sb_getblk(sb, block);
4193 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4194 "inode block - inode=%lu, block=%llu",
4195 inode->i_ino, block);
4198 if (!buffer_uptodate(bh)) {
4202 * If the buffer has the write error flag, we have failed
4203 * to write out another inode in the same block. In this
4204 * case, we don't have to read the block because we may
4205 * read the old inode data successfully.
4207 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4208 set_buffer_uptodate(bh);
4210 if (buffer_uptodate(bh)) {
4211 /* someone brought it uptodate while we waited */
4217 * If we have all information of the inode in memory and this
4218 * is the only valid inode in the block, we need not read the
4222 struct buffer_head *bitmap_bh;
4225 start = inode_offset & ~(inodes_per_block - 1);
4227 /* Is the inode bitmap in cache? */
4228 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4233 * If the inode bitmap isn't in cache then the
4234 * optimisation may end up performing two reads instead
4235 * of one, so skip it.
4237 if (!buffer_uptodate(bitmap_bh)) {
4241 for (i = start; i < start + inodes_per_block; i++) {
4242 if (i == inode_offset)
4244 if (ext4_test_bit(i, bitmap_bh->b_data))
4248 if (i == start + inodes_per_block) {
4249 /* all other inodes are free, so skip I/O */
4250 memset(bh->b_data, 0, bh->b_size);
4251 set_buffer_uptodate(bh);
4259 * If we need to do any I/O, try to pre-readahead extra
4260 * blocks from the inode table.
4262 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4263 ext4_fsblk_t b, end, table;
4266 table = ext4_inode_table(sb, gdp);
4267 /* s_inode_readahead_blks is always a power of 2 */
4268 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4271 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4272 num = EXT4_INODES_PER_GROUP(sb);
4273 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4274 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4275 num -= ext4_itable_unused_count(sb, gdp);
4276 table += num / inodes_per_block;
4280 sb_breadahead(sb, b++);
4284 * There are other valid inodes in the buffer, this inode
4285 * has in-inode xattrs, or we don't have this inode in memory.
4286 * Read the block from disk.
4289 bh->b_end_io = end_buffer_read_sync;
4290 submit_bh(READ_META, bh);
4292 if (!buffer_uptodate(bh)) {
4293 ext4_error(sb, __func__,
4294 "unable to read inode block - inode=%lu, "
4295 "block=%llu", inode->i_ino, block);
4305 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4307 /* We have all inode data except xattrs in memory here. */
4308 return __ext4_get_inode_loc(inode, iloc,
4309 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4312 void ext4_set_inode_flags(struct inode *inode)
4314 unsigned int flags = EXT4_I(inode)->i_flags;
4316 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4317 if (flags & EXT4_SYNC_FL)
4318 inode->i_flags |= S_SYNC;
4319 if (flags & EXT4_APPEND_FL)
4320 inode->i_flags |= S_APPEND;
4321 if (flags & EXT4_IMMUTABLE_FL)
4322 inode->i_flags |= S_IMMUTABLE;
4323 if (flags & EXT4_NOATIME_FL)
4324 inode->i_flags |= S_NOATIME;
4325 if (flags & EXT4_DIRSYNC_FL)
4326 inode->i_flags |= S_DIRSYNC;
4329 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4330 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4332 unsigned int flags = ei->vfs_inode.i_flags;
4334 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4335 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4337 ei->i_flags |= EXT4_SYNC_FL;
4338 if (flags & S_APPEND)
4339 ei->i_flags |= EXT4_APPEND_FL;
4340 if (flags & S_IMMUTABLE)
4341 ei->i_flags |= EXT4_IMMUTABLE_FL;
4342 if (flags & S_NOATIME)
4343 ei->i_flags |= EXT4_NOATIME_FL;
4344 if (flags & S_DIRSYNC)
4345 ei->i_flags |= EXT4_DIRSYNC_FL;
4347 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4348 struct ext4_inode_info *ei)
4351 struct inode *inode = &(ei->vfs_inode);
4352 struct super_block *sb = inode->i_sb;
4354 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4355 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4356 /* we are using combined 48 bit field */
4357 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4358 le32_to_cpu(raw_inode->i_blocks_lo);
4359 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4360 /* i_blocks represent file system block size */
4361 return i_blocks << (inode->i_blkbits - 9);
4366 return le32_to_cpu(raw_inode->i_blocks_lo);
4370 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4372 struct ext4_iloc iloc;
4373 struct ext4_inode *raw_inode;
4374 struct ext4_inode_info *ei;
4375 struct buffer_head *bh;
4376 struct inode *inode;
4380 inode = iget_locked(sb, ino);
4382 return ERR_PTR(-ENOMEM);
4383 if (!(inode->i_state & I_NEW))
4387 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4388 ei->i_acl = EXT4_ACL_NOT_CACHED;
4389 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4392 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4396 raw_inode = ext4_raw_inode(&iloc);
4397 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4398 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4399 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4400 if (!(test_opt(inode->i_sb, NO_UID32))) {
4401 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4402 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4404 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4407 ei->i_dir_start_lookup = 0;
4408 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4409 /* We now have enough fields to check if the inode was active or not.
4410 * This is needed because nfsd might try to access dead inodes
4411 * the test is that same one that e2fsck uses
4412 * NeilBrown 1999oct15
4414 if (inode->i_nlink == 0) {
4415 if (inode->i_mode == 0 ||
4416 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4417 /* this inode is deleted */
4422 /* The only unlinked inodes we let through here have
4423 * valid i_mode and are being read by the orphan
4424 * recovery code: that's fine, we're about to complete
4425 * the process of deleting those. */
4427 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4428 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4429 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4430 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4432 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4433 inode->i_size = ext4_isize(raw_inode);
4434 ei->i_disksize = inode->i_size;
4435 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4436 ei->i_block_group = iloc.block_group;
4437 ei->i_last_alloc_group = ~0;
4439 * NOTE! The in-memory inode i_data array is in little-endian order
4440 * even on big-endian machines: we do NOT byteswap the block numbers!
4442 for (block = 0; block < EXT4_N_BLOCKS; block++)
4443 ei->i_data[block] = raw_inode->i_block[block];
4444 INIT_LIST_HEAD(&ei->i_orphan);
4446 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4447 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4448 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4449 EXT4_INODE_SIZE(inode->i_sb)) {
4454 if (ei->i_extra_isize == 0) {
4455 /* The extra space is currently unused. Use it. */
4456 ei->i_extra_isize = sizeof(struct ext4_inode) -
4457 EXT4_GOOD_OLD_INODE_SIZE;
4459 __le32 *magic = (void *)raw_inode +
4460 EXT4_GOOD_OLD_INODE_SIZE +
4462 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4463 ei->i_state |= EXT4_STATE_XATTR;
4466 ei->i_extra_isize = 0;
4468 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4469 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4470 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4471 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4473 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4474 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4475 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4477 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4481 if (ei->i_file_acl &&
4483 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4484 EXT4_SB(sb)->s_gdb_count)) ||
4485 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4486 ext4_error(sb, __func__,
4487 "bad extended attribute block %llu in inode #%lu",
4488 ei->i_file_acl, inode->i_ino);
4491 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4492 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4493 (S_ISLNK(inode->i_mode) &&
4494 !ext4_inode_is_fast_symlink(inode)))
4495 /* Validate extent which is part of inode */
4496 ret = ext4_ext_check_inode(inode);
4497 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4498 (S_ISLNK(inode->i_mode) &&
4499 !ext4_inode_is_fast_symlink(inode))) {
4500 /* Validate block references which are part of inode */
4501 ret = ext4_check_inode_blockref(inode);
4508 if (S_ISREG(inode->i_mode)) {
4509 inode->i_op = &ext4_file_inode_operations;
4510 inode->i_fop = &ext4_file_operations;
4511 ext4_set_aops(inode);
4512 } else if (S_ISDIR(inode->i_mode)) {
4513 inode->i_op = &ext4_dir_inode_operations;
4514 inode->i_fop = &ext4_dir_operations;
4515 } else if (S_ISLNK(inode->i_mode)) {
4516 if (ext4_inode_is_fast_symlink(inode)) {
4517 inode->i_op = &ext4_fast_symlink_inode_operations;
4518 nd_terminate_link(ei->i_data, inode->i_size,
4519 sizeof(ei->i_data) - 1);
4521 inode->i_op = &ext4_symlink_inode_operations;
4522 ext4_set_aops(inode);
4524 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4525 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4526 inode->i_op = &ext4_special_inode_operations;
4527 if (raw_inode->i_block[0])
4528 init_special_inode(inode, inode->i_mode,
4529 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4531 init_special_inode(inode, inode->i_mode,
4532 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4536 ext4_error(inode->i_sb, __func__,
4537 "bogus i_mode (%o) for inode=%lu",
4538 inode->i_mode, inode->i_ino);
4542 ext4_set_inode_flags(inode);
4543 unlock_new_inode(inode);
4548 return ERR_PTR(ret);
4551 static int ext4_inode_blocks_set(handle_t *handle,
4552 struct ext4_inode *raw_inode,
4553 struct ext4_inode_info *ei)
4555 struct inode *inode = &(ei->vfs_inode);
4556 u64 i_blocks = inode->i_blocks;
4557 struct super_block *sb = inode->i_sb;
4559 if (i_blocks <= ~0U) {
4561 * i_blocks can be represnted in a 32 bit variable
4562 * as multiple of 512 bytes
4564 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4565 raw_inode->i_blocks_high = 0;
4566 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4569 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4572 if (i_blocks <= 0xffffffffffffULL) {
4574 * i_blocks can be represented in a 48 bit variable
4575 * as multiple of 512 bytes
4577 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4578 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4579 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4581 ei->i_flags |= EXT4_HUGE_FILE_FL;
4582 /* i_block is stored in file system block size */
4583 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4584 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4585 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4591 * Post the struct inode info into an on-disk inode location in the
4592 * buffer-cache. This gobbles the caller's reference to the
4593 * buffer_head in the inode location struct.
4595 * The caller must have write access to iloc->bh.
4597 static int ext4_do_update_inode(handle_t *handle,
4598 struct inode *inode,
4599 struct ext4_iloc *iloc)
4601 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4602 struct ext4_inode_info *ei = EXT4_I(inode);
4603 struct buffer_head *bh = iloc->bh;
4604 int err = 0, rc, block;
4606 /* For fields not not tracking in the in-memory inode,
4607 * initialise them to zero for new inodes. */
4608 if (ei->i_state & EXT4_STATE_NEW)
4609 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4611 ext4_get_inode_flags(ei);
4612 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4613 if (!(test_opt(inode->i_sb, NO_UID32))) {
4614 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4615 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4617 * Fix up interoperability with old kernels. Otherwise, old inodes get
4618 * re-used with the upper 16 bits of the uid/gid intact
4621 raw_inode->i_uid_high =
4622 cpu_to_le16(high_16_bits(inode->i_uid));
4623 raw_inode->i_gid_high =
4624 cpu_to_le16(high_16_bits(inode->i_gid));
4626 raw_inode->i_uid_high = 0;
4627 raw_inode->i_gid_high = 0;
4630 raw_inode->i_uid_low =
4631 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4632 raw_inode->i_gid_low =
4633 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4634 raw_inode->i_uid_high = 0;
4635 raw_inode->i_gid_high = 0;
4637 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4639 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4640 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4641 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4642 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4644 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4646 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4647 /* clear the migrate flag in the raw_inode */
4648 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4649 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4650 cpu_to_le32(EXT4_OS_HURD))
4651 raw_inode->i_file_acl_high =
4652 cpu_to_le16(ei->i_file_acl >> 32);
4653 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4654 ext4_isize_set(raw_inode, ei->i_disksize);
4655 if (ei->i_disksize > 0x7fffffffULL) {
4656 struct super_block *sb = inode->i_sb;
4657 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4658 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4659 EXT4_SB(sb)->s_es->s_rev_level ==
4660 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4661 /* If this is the first large file
4662 * created, add a flag to the superblock.
4664 err = ext4_journal_get_write_access(handle,
4665 EXT4_SB(sb)->s_sbh);
4668 ext4_update_dynamic_rev(sb);
4669 EXT4_SET_RO_COMPAT_FEATURE(sb,
4670 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4672 ext4_handle_sync(handle);
4673 err = ext4_handle_dirty_metadata(handle, inode,
4674 EXT4_SB(sb)->s_sbh);
4677 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4678 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4679 if (old_valid_dev(inode->i_rdev)) {
4680 raw_inode->i_block[0] =
4681 cpu_to_le32(old_encode_dev(inode->i_rdev));
4682 raw_inode->i_block[1] = 0;
4684 raw_inode->i_block[0] = 0;
4685 raw_inode->i_block[1] =
4686 cpu_to_le32(new_encode_dev(inode->i_rdev));
4687 raw_inode->i_block[2] = 0;
4689 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4690 raw_inode->i_block[block] = ei->i_data[block];
4692 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4693 if (ei->i_extra_isize) {
4694 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4695 raw_inode->i_version_hi =
4696 cpu_to_le32(inode->i_version >> 32);
4697 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4700 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4701 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4704 ei->i_state &= ~EXT4_STATE_NEW;
4708 ext4_std_error(inode->i_sb, err);
4713 * ext4_write_inode()
4715 * We are called from a few places:
4717 * - Within generic_file_write() for O_SYNC files.
4718 * Here, there will be no transaction running. We wait for any running
4719 * trasnaction to commit.
4721 * - Within sys_sync(), kupdate and such.
4722 * We wait on commit, if tol to.
4724 * - Within prune_icache() (PF_MEMALLOC == true)
4725 * Here we simply return. We can't afford to block kswapd on the
4728 * In all cases it is actually safe for us to return without doing anything,
4729 * because the inode has been copied into a raw inode buffer in
4730 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4733 * Note that we are absolutely dependent upon all inode dirtiers doing the
4734 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4735 * which we are interested.
4737 * It would be a bug for them to not do this. The code:
4739 * mark_inode_dirty(inode)
4741 * inode->i_size = expr;
4743 * is in error because a kswapd-driven write_inode() could occur while
4744 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4745 * will no longer be on the superblock's dirty inode list.
4747 int ext4_write_inode(struct inode *inode, int wait)
4749 if (current->flags & PF_MEMALLOC)
4752 if (ext4_journal_current_handle()) {
4753 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4761 return ext4_force_commit(inode->i_sb);
4764 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4768 mark_buffer_dirty(bh);
4769 if (inode && inode_needs_sync(inode)) {
4770 sync_dirty_buffer(bh);
4771 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4772 ext4_error(inode->i_sb, __func__,
4773 "IO error syncing inode, "
4774 "inode=%lu, block=%llu",
4776 (unsigned long long)bh->b_blocknr);
4786 * Called from notify_change.
4788 * We want to trap VFS attempts to truncate the file as soon as
4789 * possible. In particular, we want to make sure that when the VFS
4790 * shrinks i_size, we put the inode on the orphan list and modify
4791 * i_disksize immediately, so that during the subsequent flushing of
4792 * dirty pages and freeing of disk blocks, we can guarantee that any
4793 * commit will leave the blocks being flushed in an unused state on
4794 * disk. (On recovery, the inode will get truncated and the blocks will
4795 * be freed, so we have a strong guarantee that no future commit will
4796 * leave these blocks visible to the user.)
4798 * Another thing we have to assure is that if we are in ordered mode
4799 * and inode is still attached to the committing transaction, we must
4800 * we start writeout of all the dirty pages which are being truncated.
4801 * This way we are sure that all the data written in the previous
4802 * transaction are already on disk (truncate waits for pages under
4805 * Called with inode->i_mutex down.
4807 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4809 struct inode *inode = dentry->d_inode;
4811 const unsigned int ia_valid = attr->ia_valid;
4813 error = inode_change_ok(inode, attr);
4817 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4818 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4821 /* (user+group)*(old+new) structure, inode write (sb,
4822 * inode block, ? - but truncate inode update has it) */
4823 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4824 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4825 if (IS_ERR(handle)) {
4826 error = PTR_ERR(handle);
4829 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4831 ext4_journal_stop(handle);
4834 /* Update corresponding info in inode so that everything is in
4835 * one transaction */
4836 if (attr->ia_valid & ATTR_UID)
4837 inode->i_uid = attr->ia_uid;
4838 if (attr->ia_valid & ATTR_GID)
4839 inode->i_gid = attr->ia_gid;
4840 error = ext4_mark_inode_dirty(handle, inode);
4841 ext4_journal_stop(handle);
4844 if (attr->ia_valid & ATTR_SIZE) {
4845 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4846 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4848 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4855 if (S_ISREG(inode->i_mode) &&
4856 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4859 handle = ext4_journal_start(inode, 3);
4860 if (IS_ERR(handle)) {
4861 error = PTR_ERR(handle);
4865 error = ext4_orphan_add(handle, inode);
4866 EXT4_I(inode)->i_disksize = attr->ia_size;
4867 rc = ext4_mark_inode_dirty(handle, inode);
4870 ext4_journal_stop(handle);
4872 if (ext4_should_order_data(inode)) {
4873 error = ext4_begin_ordered_truncate(inode,
4876 /* Do as much error cleanup as possible */
4877 handle = ext4_journal_start(inode, 3);
4878 if (IS_ERR(handle)) {
4879 ext4_orphan_del(NULL, inode);
4882 ext4_orphan_del(handle, inode);
4883 ext4_journal_stop(handle);
4889 rc = inode_setattr(inode, attr);
4891 /* If inode_setattr's call to ext4_truncate failed to get a
4892 * transaction handle at all, we need to clean up the in-core
4893 * orphan list manually. */
4895 ext4_orphan_del(NULL, inode);
4897 if (!rc && (ia_valid & ATTR_MODE))
4898 rc = ext4_acl_chmod(inode);
4901 ext4_std_error(inode->i_sb, error);
4907 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4910 struct inode *inode;
4911 unsigned long delalloc_blocks;
4913 inode = dentry->d_inode;
4914 generic_fillattr(inode, stat);
4917 * We can't update i_blocks if the block allocation is delayed
4918 * otherwise in the case of system crash before the real block
4919 * allocation is done, we will have i_blocks inconsistent with
4920 * on-disk file blocks.
4921 * We always keep i_blocks updated together with real
4922 * allocation. But to not confuse with user, stat
4923 * will return the blocks that include the delayed allocation
4924 * blocks for this file.
4926 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4927 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4928 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4930 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4934 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4939 /* if nrblocks are contiguous */
4942 * With N contiguous data blocks, it need at most
4943 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4944 * 2 dindirect blocks
4947 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4948 return indirects + 3;
4951 * if nrblocks are not contiguous, worse case, each block touch
4952 * a indirect block, and each indirect block touch a double indirect
4953 * block, plus a triple indirect block
4955 indirects = nrblocks * 2 + 1;
4959 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4961 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4962 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4963 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4967 * Account for index blocks, block groups bitmaps and block group
4968 * descriptor blocks if modify datablocks and index blocks
4969 * worse case, the indexs blocks spread over different block groups
4971 * If datablocks are discontiguous, they are possible to spread over
4972 * different block groups too. If they are contiugous, with flexbg,
4973 * they could still across block group boundary.
4975 * Also account for superblock, inode, quota and xattr blocks
4977 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4979 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4985 * How many index blocks need to touch to modify nrblocks?
4986 * The "Chunk" flag indicating whether the nrblocks is
4987 * physically contiguous on disk
4989 * For Direct IO and fallocate, they calls get_block to allocate
4990 * one single extent at a time, so they could set the "Chunk" flag
4992 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4997 * Now let's see how many group bitmaps and group descriptors need
5007 if (groups > ngroups)
5009 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5010 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5012 /* bitmaps and block group descriptor blocks */
5013 ret += groups + gdpblocks;
5015 /* Blocks for super block, inode, quota and xattr blocks */
5016 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5022 * Calulate the total number of credits to reserve to fit
5023 * the modification of a single pages into a single transaction,
5024 * which may include multiple chunks of block allocations.
5026 * This could be called via ext4_write_begin()
5028 * We need to consider the worse case, when
5029 * one new block per extent.
5031 int ext4_writepage_trans_blocks(struct inode *inode)
5033 int bpp = ext4_journal_blocks_per_page(inode);
5036 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5038 /* Account for data blocks for journalled mode */
5039 if (ext4_should_journal_data(inode))
5045 * Calculate the journal credits for a chunk of data modification.
5047 * This is called from DIO, fallocate or whoever calling
5048 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5050 * journal buffers for data blocks are not included here, as DIO
5051 * and fallocate do no need to journal data buffers.
5053 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5055 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5059 * The caller must have previously called ext4_reserve_inode_write().
5060 * Give this, we know that the caller already has write access to iloc->bh.
5062 int ext4_mark_iloc_dirty(handle_t *handle,
5063 struct inode *inode, struct ext4_iloc *iloc)
5067 if (test_opt(inode->i_sb, I_VERSION))
5068 inode_inc_iversion(inode);
5070 /* the do_update_inode consumes one bh->b_count */
5073 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5074 err = ext4_do_update_inode(handle, inode, iloc);
5080 * On success, We end up with an outstanding reference count against
5081 * iloc->bh. This _must_ be cleaned up later.
5085 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5086 struct ext4_iloc *iloc)
5090 err = ext4_get_inode_loc(inode, iloc);
5092 BUFFER_TRACE(iloc->bh, "get_write_access");
5093 err = ext4_journal_get_write_access(handle, iloc->bh);
5099 ext4_std_error(inode->i_sb, err);
5104 * Expand an inode by new_extra_isize bytes.
5105 * Returns 0 on success or negative error number on failure.
5107 static int ext4_expand_extra_isize(struct inode *inode,
5108 unsigned int new_extra_isize,
5109 struct ext4_iloc iloc,
5112 struct ext4_inode *raw_inode;
5113 struct ext4_xattr_ibody_header *header;
5114 struct ext4_xattr_entry *entry;
5116 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5119 raw_inode = ext4_raw_inode(&iloc);
5121 header = IHDR(inode, raw_inode);
5122 entry = IFIRST(header);
5124 /* No extended attributes present */
5125 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5126 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5127 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5129 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5133 /* try to expand with EAs present */
5134 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5139 * What we do here is to mark the in-core inode as clean with respect to inode
5140 * dirtiness (it may still be data-dirty).
5141 * This means that the in-core inode may be reaped by prune_icache
5142 * without having to perform any I/O. This is a very good thing,
5143 * because *any* task may call prune_icache - even ones which
5144 * have a transaction open against a different journal.
5146 * Is this cheating? Not really. Sure, we haven't written the
5147 * inode out, but prune_icache isn't a user-visible syncing function.
5148 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5149 * we start and wait on commits.
5151 * Is this efficient/effective? Well, we're being nice to the system
5152 * by cleaning up our inodes proactively so they can be reaped
5153 * without I/O. But we are potentially leaving up to five seconds'
5154 * worth of inodes floating about which prune_icache wants us to
5155 * write out. One way to fix that would be to get prune_icache()
5156 * to do a write_super() to free up some memory. It has the desired
5159 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5161 struct ext4_iloc iloc;
5162 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5163 static unsigned int mnt_count;
5167 err = ext4_reserve_inode_write(handle, inode, &iloc);
5168 if (ext4_handle_valid(handle) &&
5169 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5170 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5172 * We need extra buffer credits since we may write into EA block
5173 * with this same handle. If journal_extend fails, then it will
5174 * only result in a minor loss of functionality for that inode.
5175 * If this is felt to be critical, then e2fsck should be run to
5176 * force a large enough s_min_extra_isize.
5178 if ((jbd2_journal_extend(handle,
5179 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5180 ret = ext4_expand_extra_isize(inode,
5181 sbi->s_want_extra_isize,
5184 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5186 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5187 ext4_warning(inode->i_sb, __func__,
5188 "Unable to expand inode %lu. Delete"
5189 " some EAs or run e2fsck.",
5192 le16_to_cpu(sbi->s_es->s_mnt_count);
5198 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5203 * ext4_dirty_inode() is called from __mark_inode_dirty()
5205 * We're really interested in the case where a file is being extended.
5206 * i_size has been changed by generic_commit_write() and we thus need
5207 * to include the updated inode in the current transaction.
5209 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5210 * are allocated to the file.
5212 * If the inode is marked synchronous, we don't honour that here - doing
5213 * so would cause a commit on atime updates, which we don't bother doing.
5214 * We handle synchronous inodes at the highest possible level.
5216 void ext4_dirty_inode(struct inode *inode)
5218 handle_t *current_handle = ext4_journal_current_handle();
5221 if (!ext4_handle_valid(current_handle)) {
5222 ext4_mark_inode_dirty(current_handle, inode);
5226 handle = ext4_journal_start(inode, 2);
5229 if (current_handle &&
5230 current_handle->h_transaction != handle->h_transaction) {
5231 /* This task has a transaction open against a different fs */
5232 printk(KERN_EMERG "%s: transactions do not match!\n",
5235 jbd_debug(5, "marking dirty. outer handle=%p\n",
5237 ext4_mark_inode_dirty(handle, inode);
5239 ext4_journal_stop(handle);
5246 * Bind an inode's backing buffer_head into this transaction, to prevent
5247 * it from being flushed to disk early. Unlike
5248 * ext4_reserve_inode_write, this leaves behind no bh reference and
5249 * returns no iloc structure, so the caller needs to repeat the iloc
5250 * lookup to mark the inode dirty later.
5252 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5254 struct ext4_iloc iloc;
5258 err = ext4_get_inode_loc(inode, &iloc);
5260 BUFFER_TRACE(iloc.bh, "get_write_access");
5261 err = jbd2_journal_get_write_access(handle, iloc.bh);
5263 err = ext4_handle_dirty_metadata(handle,
5269 ext4_std_error(inode->i_sb, err);
5274 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5281 * We have to be very careful here: changing a data block's
5282 * journaling status dynamically is dangerous. If we write a
5283 * data block to the journal, change the status and then delete
5284 * that block, we risk forgetting to revoke the old log record
5285 * from the journal and so a subsequent replay can corrupt data.
5286 * So, first we make sure that the journal is empty and that
5287 * nobody is changing anything.
5290 journal = EXT4_JOURNAL(inode);
5293 if (is_journal_aborted(journal))
5296 jbd2_journal_lock_updates(journal);
5297 jbd2_journal_flush(journal);
5300 * OK, there are no updates running now, and all cached data is
5301 * synced to disk. We are now in a completely consistent state
5302 * which doesn't have anything in the journal, and we know that
5303 * no filesystem updates are running, so it is safe to modify
5304 * the inode's in-core data-journaling state flag now.
5308 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5310 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5311 ext4_set_aops(inode);
5313 jbd2_journal_unlock_updates(journal);
5315 /* Finally we can mark the inode as dirty. */
5317 handle = ext4_journal_start(inode, 1);
5319 return PTR_ERR(handle);
5321 err = ext4_mark_inode_dirty(handle, inode);
5322 ext4_handle_sync(handle);
5323 ext4_journal_stop(handle);
5324 ext4_std_error(inode->i_sb, err);
5329 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5331 return !buffer_mapped(bh);
5334 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5336 struct page *page = vmf->page;
5341 struct file *file = vma->vm_file;
5342 struct inode *inode = file->f_path.dentry->d_inode;
5343 struct address_space *mapping = inode->i_mapping;
5346 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5347 * get i_mutex because we are already holding mmap_sem.
5349 down_read(&inode->i_alloc_sem);
5350 size = i_size_read(inode);
5351 if (page->mapping != mapping || size <= page_offset(page)
5352 || !PageUptodate(page)) {
5353 /* page got truncated from under us? */
5357 if (PageMappedToDisk(page))
5360 if (page->index == size >> PAGE_CACHE_SHIFT)
5361 len = size & ~PAGE_CACHE_MASK;
5363 len = PAGE_CACHE_SIZE;
5365 if (page_has_buffers(page)) {
5366 /* return if we have all the buffers mapped */
5367 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5372 * OK, we need to fill the hole... Do write_begin write_end
5373 * to do block allocation/reservation.We are not holding
5374 * inode.i__mutex here. That allow * parallel write_begin,
5375 * write_end call. lock_page prevent this from happening
5376 * on the same page though
5378 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5379 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5382 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5383 len, len, page, fsdata);
5389 ret = VM_FAULT_SIGBUS;
5390 up_read(&inode->i_alloc_sem);