2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode *inode,
50 return jbd2_journal_begin_ordered_truncate(
51 EXT4_SB(inode->i_sb)->s_journal,
52 &EXT4_I(inode)->jinode,
56 static void ext4_invalidatepage(struct page *page, unsigned long offset);
59 * Test whether an inode is a fast symlink.
61 static int ext4_inode_is_fast_symlink(struct inode *inode)
63 int ea_blocks = EXT4_I(inode)->i_file_acl ?
64 (inode->i_sb->s_blocksize >> 9) : 0;
66 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
70 * The ext4 forget function must perform a revoke if we are freeing data
71 * which has been journaled. Metadata (eg. indirect blocks) must be
72 * revoked in all cases.
74 * "bh" may be NULL: a metadata block may have been freed from memory
75 * but there may still be a record of it in the journal, and that record
76 * still needs to be revoked.
78 * If the handle isn't valid we're not journaling so there's nothing to do.
80 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
81 struct buffer_head *bh, ext4_fsblk_t blocknr)
85 if (!ext4_handle_valid(handle))
90 BUFFER_TRACE(bh, "enter");
92 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
94 bh, is_metadata, inode->i_mode,
95 test_opt(inode->i_sb, DATA_FLAGS));
97 /* Never use the revoke function if we are doing full data
98 * journaling: there is no need to, and a V1 superblock won't
99 * support it. Otherwise, only skip the revoke on un-journaled
102 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
103 (!is_metadata && !ext4_should_journal_data(inode))) {
105 BUFFER_TRACE(bh, "call jbd2_journal_forget");
106 return ext4_journal_forget(handle, bh);
112 * data!=journal && (is_metadata || should_journal_data(inode))
114 BUFFER_TRACE(bh, "call ext4_journal_revoke");
115 err = ext4_journal_revoke(handle, blocknr, bh);
117 ext4_abort(inode->i_sb, __func__,
118 "error %d when attempting revoke", err);
119 BUFFER_TRACE(bh, "exit");
124 * Work out how many blocks we need to proceed with the next chunk of a
125 * truncate transaction.
127 static unsigned long blocks_for_truncate(struct inode *inode)
131 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
133 /* Give ourselves just enough room to cope with inodes in which
134 * i_blocks is corrupt: we've seen disk corruptions in the past
135 * which resulted in random data in an inode which looked enough
136 * like a regular file for ext4 to try to delete it. Things
137 * will go a bit crazy if that happens, but at least we should
138 * try not to panic the whole kernel. */
142 /* But we need to bound the transaction so we don't overflow the
144 if (needed > EXT4_MAX_TRANS_DATA)
145 needed = EXT4_MAX_TRANS_DATA;
147 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
151 * Truncate transactions can be complex and absolutely huge. So we need to
152 * be able to restart the transaction at a conventient checkpoint to make
153 * sure we don't overflow the journal.
155 * start_transaction gets us a new handle for a truncate transaction,
156 * and extend_transaction tries to extend the existing one a bit. If
157 * extend fails, we need to propagate the failure up and restart the
158 * transaction in the top-level truncate loop. --sct
160 static handle_t *start_transaction(struct inode *inode)
164 result = ext4_journal_start(inode, blocks_for_truncate(inode));
168 ext4_std_error(inode->i_sb, PTR_ERR(result));
173 * Try to extend this transaction for the purposes of truncation.
175 * Returns 0 if we managed to create more room. If we can't create more
176 * room, and the transaction must be restarted we return 1.
178 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
180 if (!ext4_handle_valid(handle))
182 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
184 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
190 * Restart the transaction associated with *handle. This does a commit,
191 * so before we call here everything must be consistently dirtied against
194 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
196 BUG_ON(EXT4_JOURNAL(inode) == NULL);
197 jbd_debug(2, "restarting handle %p\n", handle);
198 return ext4_journal_restart(handle, blocks_for_truncate(inode));
202 * Called at the last iput() if i_nlink is zero.
204 void ext4_delete_inode(struct inode *inode)
209 if (ext4_should_order_data(inode))
210 ext4_begin_ordered_truncate(inode, 0);
211 truncate_inode_pages(&inode->i_data, 0);
213 if (is_bad_inode(inode))
216 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
217 if (IS_ERR(handle)) {
218 ext4_std_error(inode->i_sb, PTR_ERR(handle));
220 * If we're going to skip the normal cleanup, we still need to
221 * make sure that the in-core orphan linked list is properly
224 ext4_orphan_del(NULL, inode);
229 ext4_handle_sync(handle);
231 err = ext4_mark_inode_dirty(handle, inode);
233 ext4_warning(inode->i_sb, __func__,
234 "couldn't mark inode dirty (err %d)", err);
238 ext4_truncate(inode);
241 * ext4_ext_truncate() doesn't reserve any slop when it
242 * restarts journal transactions; therefore there may not be
243 * enough credits left in the handle to remove the inode from
244 * the orphan list and set the dtime field.
246 if (!ext4_handle_has_enough_credits(handle, 3)) {
247 err = ext4_journal_extend(handle, 3);
249 err = ext4_journal_restart(handle, 3);
251 ext4_warning(inode->i_sb, __func__,
252 "couldn't extend journal (err %d)", err);
254 ext4_journal_stop(handle);
260 * Kill off the orphan record which ext4_truncate created.
261 * AKPM: I think this can be inside the above `if'.
262 * Note that ext4_orphan_del() has to be able to cope with the
263 * deletion of a non-existent orphan - this is because we don't
264 * know if ext4_truncate() actually created an orphan record.
265 * (Well, we could do this if we need to, but heck - it works)
267 ext4_orphan_del(handle, inode);
268 EXT4_I(inode)->i_dtime = get_seconds();
271 * One subtle ordering requirement: if anything has gone wrong
272 * (transaction abort, IO errors, whatever), then we can still
273 * do these next steps (the fs will already have been marked as
274 * having errors), but we can't free the inode if the mark_dirty
277 if (ext4_mark_inode_dirty(handle, inode))
278 /* If that failed, just do the required in-core inode clear. */
281 ext4_free_inode(handle, inode);
282 ext4_journal_stop(handle);
285 clear_inode(inode); /* We must guarantee clearing of inode... */
291 struct buffer_head *bh;
294 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
296 p->key = *(p->p = v);
301 * ext4_block_to_path - parse the block number into array of offsets
302 * @inode: inode in question (we are only interested in its superblock)
303 * @i_block: block number to be parsed
304 * @offsets: array to store the offsets in
305 * @boundary: set this non-zero if the referred-to block is likely to be
306 * followed (on disk) by an indirect block.
308 * To store the locations of file's data ext4 uses a data structure common
309 * for UNIX filesystems - tree of pointers anchored in the inode, with
310 * data blocks at leaves and indirect blocks in intermediate nodes.
311 * This function translates the block number into path in that tree -
312 * return value is the path length and @offsets[n] is the offset of
313 * pointer to (n+1)th node in the nth one. If @block is out of range
314 * (negative or too large) warning is printed and zero returned.
316 * Note: function doesn't find node addresses, so no IO is needed. All
317 * we need to know is the capacity of indirect blocks (taken from the
322 * Portability note: the last comparison (check that we fit into triple
323 * indirect block) is spelled differently, because otherwise on an
324 * architecture with 32-bit longs and 8Kb pages we might get into trouble
325 * if our filesystem had 8Kb blocks. We might use long long, but that would
326 * kill us on x86. Oh, well, at least the sign propagation does not matter -
327 * i_block would have to be negative in the very beginning, so we would not
331 static int ext4_block_to_path(struct inode *inode,
333 ext4_lblk_t offsets[4], int *boundary)
335 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
336 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
337 const long direct_blocks = EXT4_NDIR_BLOCKS,
338 indirect_blocks = ptrs,
339 double_blocks = (1 << (ptrs_bits * 2));
344 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
345 } else if (i_block < direct_blocks) {
346 offsets[n++] = i_block;
347 final = direct_blocks;
348 } else if ((i_block -= direct_blocks) < indirect_blocks) {
349 offsets[n++] = EXT4_IND_BLOCK;
350 offsets[n++] = i_block;
352 } else if ((i_block -= indirect_blocks) < double_blocks) {
353 offsets[n++] = EXT4_DIND_BLOCK;
354 offsets[n++] = i_block >> ptrs_bits;
355 offsets[n++] = i_block & (ptrs - 1);
357 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
358 offsets[n++] = EXT4_TIND_BLOCK;
359 offsets[n++] = i_block >> (ptrs_bits * 2);
360 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
361 offsets[n++] = i_block & (ptrs - 1);
364 ext4_warning(inode->i_sb, "ext4_block_to_path",
365 "block %lu > max in inode %lu",
366 i_block + direct_blocks +
367 indirect_blocks + double_blocks, inode->i_ino);
370 *boundary = final - 1 - (i_block & (ptrs - 1));
374 static int __ext4_check_blockref(const char *function, struct inode *inode,
375 __le32 *p, unsigned int max) {
377 unsigned int maxblocks = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es);
379 while (bref < p+max) {
380 if (unlikely(le32_to_cpu(*bref) >= maxblocks)) {
381 ext4_error(inode->i_sb, function,
382 "block reference %u >= max (%u) "
383 "in inode #%lu, offset=%d",
384 le32_to_cpu(*bref), maxblocks,
385 inode->i_ino, (int)(bref-p));
394 #define ext4_check_indirect_blockref(inode, bh) \
395 __ext4_check_blockref(__func__, inode, (__le32 *)(bh)->b_data, \
396 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
398 #define ext4_check_inode_blockref(inode) \
399 __ext4_check_blockref(__func__, inode, EXT4_I(inode)->i_data, \
403 * ext4_get_branch - read the chain of indirect blocks leading to data
404 * @inode: inode in question
405 * @depth: depth of the chain (1 - direct pointer, etc.)
406 * @offsets: offsets of pointers in inode/indirect blocks
407 * @chain: place to store the result
408 * @err: here we store the error value
410 * Function fills the array of triples <key, p, bh> and returns %NULL
411 * if everything went OK or the pointer to the last filled triple
412 * (incomplete one) otherwise. Upon the return chain[i].key contains
413 * the number of (i+1)-th block in the chain (as it is stored in memory,
414 * i.e. little-endian 32-bit), chain[i].p contains the address of that
415 * number (it points into struct inode for i==0 and into the bh->b_data
416 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
417 * block for i>0 and NULL for i==0. In other words, it holds the block
418 * numbers of the chain, addresses they were taken from (and where we can
419 * verify that chain did not change) and buffer_heads hosting these
422 * Function stops when it stumbles upon zero pointer (absent block)
423 * (pointer to last triple returned, *@err == 0)
424 * or when it gets an IO error reading an indirect block
425 * (ditto, *@err == -EIO)
426 * or when it reads all @depth-1 indirect blocks successfully and finds
427 * the whole chain, all way to the data (returns %NULL, *err == 0).
429 * Need to be called with
430 * down_read(&EXT4_I(inode)->i_data_sem)
432 static Indirect *ext4_get_branch(struct inode *inode, int depth,
433 ext4_lblk_t *offsets,
434 Indirect chain[4], int *err)
436 struct super_block *sb = inode->i_sb;
438 struct buffer_head *bh;
441 /* i_data is not going away, no lock needed */
442 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
446 bh = sb_getblk(sb, le32_to_cpu(p->key));
450 if (!bh_uptodate_or_lock(bh)) {
451 if (bh_submit_read(bh) < 0) {
455 /* validate block references */
456 if (ext4_check_indirect_blockref(inode, bh)) {
462 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
476 * ext4_find_near - find a place for allocation with sufficient locality
478 * @ind: descriptor of indirect block.
480 * This function returns the preferred place for block allocation.
481 * It is used when heuristic for sequential allocation fails.
483 * + if there is a block to the left of our position - allocate near it.
484 * + if pointer will live in indirect block - allocate near that block.
485 * + if pointer will live in inode - allocate in the same
488 * In the latter case we colour the starting block by the callers PID to
489 * prevent it from clashing with concurrent allocations for a different inode
490 * in the same block group. The PID is used here so that functionally related
491 * files will be close-by on-disk.
493 * Caller must make sure that @ind is valid and will stay that way.
495 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
497 struct ext4_inode_info *ei = EXT4_I(inode);
498 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
500 ext4_fsblk_t bg_start;
501 ext4_fsblk_t last_block;
502 ext4_grpblk_t colour;
503 ext4_group_t block_group;
504 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
506 /* Try to find previous block */
507 for (p = ind->p - 1; p >= start; p--) {
509 return le32_to_cpu(*p);
512 /* No such thing, so let's try location of indirect block */
514 return ind->bh->b_blocknr;
517 * It is going to be referred to from the inode itself? OK, just put it
518 * into the same cylinder group then.
520 block_group = ei->i_block_group;
521 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
522 block_group &= ~(flex_size-1);
523 if (S_ISREG(inode->i_mode))
526 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
527 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
530 * If we are doing delayed allocation, we don't need take
531 * colour into account.
533 if (test_opt(inode->i_sb, DELALLOC))
536 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
537 colour = (current->pid % 16) *
538 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
540 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
541 return bg_start + colour;
545 * ext4_find_goal - find a preferred place for allocation.
547 * @block: block we want
548 * @partial: pointer to the last triple within a chain
550 * Normally this function find the preferred place for block allocation,
553 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
557 * XXX need to get goal block from mballoc's data structures
560 return ext4_find_near(inode, partial);
564 * ext4_blks_to_allocate: Look up the block map and count the number
565 * of direct blocks need to be allocated for the given branch.
567 * @branch: chain of indirect blocks
568 * @k: number of blocks need for indirect blocks
569 * @blks: number of data blocks to be mapped.
570 * @blocks_to_boundary: the offset in the indirect block
572 * return the total number of blocks to be allocate, including the
573 * direct and indirect blocks.
575 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
576 int blocks_to_boundary)
578 unsigned int count = 0;
581 * Simple case, [t,d]Indirect block(s) has not allocated yet
582 * then it's clear blocks on that path have not allocated
585 /* right now we don't handle cross boundary allocation */
586 if (blks < blocks_to_boundary + 1)
589 count += blocks_to_boundary + 1;
594 while (count < blks && count <= blocks_to_boundary &&
595 le32_to_cpu(*(branch[0].p + count)) == 0) {
602 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
603 * @indirect_blks: the number of blocks need to allocate for indirect
606 * @new_blocks: on return it will store the new block numbers for
607 * the indirect blocks(if needed) and the first direct block,
608 * @blks: on return it will store the total number of allocated
611 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
612 ext4_lblk_t iblock, ext4_fsblk_t goal,
613 int indirect_blks, int blks,
614 ext4_fsblk_t new_blocks[4], int *err)
616 struct ext4_allocation_request ar;
618 unsigned long count = 0, blk_allocated = 0;
620 ext4_fsblk_t current_block = 0;
624 * Here we try to allocate the requested multiple blocks at once,
625 * on a best-effort basis.
626 * To build a branch, we should allocate blocks for
627 * the indirect blocks(if not allocated yet), and at least
628 * the first direct block of this branch. That's the
629 * minimum number of blocks need to allocate(required)
631 /* first we try to allocate the indirect blocks */
632 target = indirect_blks;
635 /* allocating blocks for indirect blocks and direct blocks */
636 current_block = ext4_new_meta_blocks(handle, inode,
642 /* allocate blocks for indirect blocks */
643 while (index < indirect_blks && count) {
644 new_blocks[index++] = current_block++;
649 * save the new block number
650 * for the first direct block
652 new_blocks[index] = current_block;
653 printk(KERN_INFO "%s returned more blocks than "
654 "requested\n", __func__);
660 target = blks - count ;
661 blk_allocated = count;
664 /* Now allocate data blocks */
665 memset(&ar, 0, sizeof(ar));
670 if (S_ISREG(inode->i_mode))
671 /* enable in-core preallocation only for regular files */
672 ar.flags = EXT4_MB_HINT_DATA;
674 current_block = ext4_mb_new_blocks(handle, &ar, err);
676 if (*err && (target == blks)) {
678 * if the allocation failed and we didn't allocate
684 if (target == blks) {
686 * save the new block number
687 * for the first direct block
689 new_blocks[index] = current_block;
691 blk_allocated += ar.len;
694 /* total number of blocks allocated for direct blocks */
699 for (i = 0; i < index; i++)
700 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
705 * ext4_alloc_branch - allocate and set up a chain of blocks.
707 * @indirect_blks: number of allocated indirect blocks
708 * @blks: number of allocated direct blocks
709 * @offsets: offsets (in the blocks) to store the pointers to next.
710 * @branch: place to store the chain in.
712 * This function allocates blocks, zeroes out all but the last one,
713 * links them into chain and (if we are synchronous) writes them to disk.
714 * In other words, it prepares a branch that can be spliced onto the
715 * inode. It stores the information about that chain in the branch[], in
716 * the same format as ext4_get_branch() would do. We are calling it after
717 * we had read the existing part of chain and partial points to the last
718 * triple of that (one with zero ->key). Upon the exit we have the same
719 * picture as after the successful ext4_get_block(), except that in one
720 * place chain is disconnected - *branch->p is still zero (we did not
721 * set the last link), but branch->key contains the number that should
722 * be placed into *branch->p to fill that gap.
724 * If allocation fails we free all blocks we've allocated (and forget
725 * their buffer_heads) and return the error value the from failed
726 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
727 * as described above and return 0.
729 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
730 ext4_lblk_t iblock, int indirect_blks,
731 int *blks, ext4_fsblk_t goal,
732 ext4_lblk_t *offsets, Indirect *branch)
734 int blocksize = inode->i_sb->s_blocksize;
737 struct buffer_head *bh;
739 ext4_fsblk_t new_blocks[4];
740 ext4_fsblk_t current_block;
742 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
743 *blks, new_blocks, &err);
747 branch[0].key = cpu_to_le32(new_blocks[0]);
749 * metadata blocks and data blocks are allocated.
751 for (n = 1; n <= indirect_blks; n++) {
753 * Get buffer_head for parent block, zero it out
754 * and set the pointer to new one, then send
757 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
760 BUFFER_TRACE(bh, "call get_create_access");
761 err = ext4_journal_get_create_access(handle, bh);
768 memset(bh->b_data, 0, blocksize);
769 branch[n].p = (__le32 *) bh->b_data + offsets[n];
770 branch[n].key = cpu_to_le32(new_blocks[n]);
771 *branch[n].p = branch[n].key;
772 if (n == indirect_blks) {
773 current_block = new_blocks[n];
775 * End of chain, update the last new metablock of
776 * the chain to point to the new allocated
777 * data blocks numbers
779 for (i=1; i < num; i++)
780 *(branch[n].p + i) = cpu_to_le32(++current_block);
782 BUFFER_TRACE(bh, "marking uptodate");
783 set_buffer_uptodate(bh);
786 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
787 err = ext4_handle_dirty_metadata(handle, inode, bh);
794 /* Allocation failed, free what we already allocated */
795 for (i = 1; i <= n ; i++) {
796 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
797 ext4_journal_forget(handle, branch[i].bh);
799 for (i = 0; i < indirect_blks; i++)
800 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
802 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
808 * ext4_splice_branch - splice the allocated branch onto inode.
810 * @block: (logical) number of block we are adding
811 * @chain: chain of indirect blocks (with a missing link - see
813 * @where: location of missing link
814 * @num: number of indirect blocks we are adding
815 * @blks: number of direct blocks we are adding
817 * This function fills the missing link and does all housekeeping needed in
818 * inode (->i_blocks, etc.). In case of success we end up with the full
819 * chain to new block and return 0.
821 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
822 ext4_lblk_t block, Indirect *where, int num, int blks)
826 ext4_fsblk_t current_block;
829 * If we're splicing into a [td]indirect block (as opposed to the
830 * inode) then we need to get write access to the [td]indirect block
834 BUFFER_TRACE(where->bh, "get_write_access");
835 err = ext4_journal_get_write_access(handle, where->bh);
841 *where->p = where->key;
844 * Update the host buffer_head or inode to point to more just allocated
845 * direct blocks blocks
847 if (num == 0 && blks > 1) {
848 current_block = le32_to_cpu(where->key) + 1;
849 for (i = 1; i < blks; i++)
850 *(where->p + i) = cpu_to_le32(current_block++);
853 /* We are done with atomic stuff, now do the rest of housekeeping */
855 inode->i_ctime = ext4_current_time(inode);
856 ext4_mark_inode_dirty(handle, inode);
858 /* had we spliced it onto indirect block? */
861 * If we spliced it onto an indirect block, we haven't
862 * altered the inode. Note however that if it is being spliced
863 * onto an indirect block at the very end of the file (the
864 * file is growing) then we *will* alter the inode to reflect
865 * the new i_size. But that is not done here - it is done in
866 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
868 jbd_debug(5, "splicing indirect only\n");
869 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
870 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
875 * OK, we spliced it into the inode itself on a direct block.
876 * Inode was dirtied above.
878 jbd_debug(5, "splicing direct\n");
883 for (i = 1; i <= num; i++) {
884 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
885 ext4_journal_forget(handle, where[i].bh);
886 ext4_free_blocks(handle, inode,
887 le32_to_cpu(where[i-1].key), 1, 0);
889 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
895 * Allocation strategy is simple: if we have to allocate something, we will
896 * have to go the whole way to leaf. So let's do it before attaching anything
897 * to tree, set linkage between the newborn blocks, write them if sync is
898 * required, recheck the path, free and repeat if check fails, otherwise
899 * set the last missing link (that will protect us from any truncate-generated
900 * removals - all blocks on the path are immune now) and possibly force the
901 * write on the parent block.
902 * That has a nice additional property: no special recovery from the failed
903 * allocations is needed - we simply release blocks and do not touch anything
904 * reachable from inode.
906 * `handle' can be NULL if create == 0.
908 * return > 0, # of blocks mapped or allocated.
909 * return = 0, if plain lookup failed.
910 * return < 0, error case.
913 * Need to be called with
914 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
915 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
917 static int ext4_ind_get_blocks(handle_t *handle, struct inode *inode,
918 ext4_lblk_t iblock, unsigned int maxblocks,
919 struct buffer_head *bh_result,
923 ext4_lblk_t offsets[4];
928 int blocks_to_boundary = 0;
930 struct ext4_inode_info *ei = EXT4_I(inode);
932 ext4_fsblk_t first_block = 0;
936 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
937 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
938 depth = ext4_block_to_path(inode, iblock, offsets,
939 &blocks_to_boundary);
944 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
946 /* Simplest case - block found, no allocation needed */
948 first_block = le32_to_cpu(chain[depth - 1].key);
949 clear_buffer_new(bh_result);
952 while (count < maxblocks && count <= blocks_to_boundary) {
955 blk = le32_to_cpu(*(chain[depth-1].p + count));
957 if (blk == first_block + count)
965 /* Next simple case - plain lookup or failed read of indirect block */
966 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
970 * Okay, we need to do block allocation.
972 goal = ext4_find_goal(inode, iblock, partial);
974 /* the number of blocks need to allocate for [d,t]indirect blocks */
975 indirect_blks = (chain + depth) - partial - 1;
978 * Next look up the indirect map to count the totoal number of
979 * direct blocks to allocate for this branch.
981 count = ext4_blks_to_allocate(partial, indirect_blks,
982 maxblocks, blocks_to_boundary);
984 * Block out ext4_truncate while we alter the tree
986 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
988 offsets + (partial - chain), partial);
991 * The ext4_splice_branch call will free and forget any buffers
992 * on the new chain if there is a failure, but that risks using
993 * up transaction credits, especially for bitmaps where the
994 * credits cannot be returned. Can we handle this somehow? We
995 * may need to return -EAGAIN upwards in the worst case. --sct
998 err = ext4_splice_branch(handle, inode, iblock,
999 partial, indirect_blks, count);
1001 * i_disksize growing is protected by i_data_sem. Don't forget to
1002 * protect it if you're about to implement concurrent
1003 * ext4_get_block() -bzzz
1005 if (!err && (flags & EXT4_GET_BLOCKS_EXTEND_DISKSIZE)) {
1006 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
1007 if (disksize > i_size_read(inode))
1008 disksize = i_size_read(inode);
1009 if (disksize > ei->i_disksize)
1010 ei->i_disksize = disksize;
1015 set_buffer_new(bh_result);
1017 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
1018 if (count > blocks_to_boundary)
1019 set_buffer_boundary(bh_result);
1021 /* Clean up and exit */
1022 partial = chain + depth - 1; /* the whole chain */
1024 while (partial > chain) {
1025 BUFFER_TRACE(partial->bh, "call brelse");
1026 brelse(partial->bh);
1029 BUFFER_TRACE(bh_result, "returned");
1034 qsize_t ext4_get_reserved_space(struct inode *inode)
1036 unsigned long long total;
1038 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1039 total = EXT4_I(inode)->i_reserved_data_blocks +
1040 EXT4_I(inode)->i_reserved_meta_blocks;
1041 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1046 * Calculate the number of metadata blocks need to reserve
1047 * to allocate @blocks for non extent file based file
1049 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1051 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1052 int ind_blks, dind_blks, tind_blks;
1054 /* number of new indirect blocks needed */
1055 ind_blks = (blocks + icap - 1) / icap;
1057 dind_blks = (ind_blks + icap - 1) / icap;
1061 return ind_blks + dind_blks + tind_blks;
1065 * Calculate the number of metadata blocks need to reserve
1066 * to allocate given number of blocks
1068 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1073 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1074 return ext4_ext_calc_metadata_amount(inode, blocks);
1076 return ext4_indirect_calc_metadata_amount(inode, blocks);
1079 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1081 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1082 int total, mdb, mdb_free;
1084 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1085 /* recalculate the number of metablocks still need to be reserved */
1086 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1087 mdb = ext4_calc_metadata_amount(inode, total);
1089 /* figure out how many metablocks to release */
1090 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1091 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1094 /* Account for allocated meta_blocks */
1095 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1097 /* update fs dirty blocks counter */
1098 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1099 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1100 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1103 /* update per-inode reservations */
1104 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1105 EXT4_I(inode)->i_reserved_data_blocks -= used;
1106 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1109 * free those over-booking quota for metadata blocks
1112 vfs_dq_release_reservation_block(inode, mdb_free);
1115 * If we have done all the pending block allocations and if
1116 * there aren't any writers on the inode, we can discard the
1117 * inode's preallocations.
1119 if (!total && (atomic_read(&inode->i_writecount) == 0))
1120 ext4_discard_preallocations(inode);
1124 * The ext4_get_blocks() function tries to look up the requested blocks,
1125 * and returns if the blocks are already mapped.
1127 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1128 * and store the allocated blocks in the result buffer head and mark it
1131 * If file type is extents based, it will call ext4_ext_get_blocks(),
1132 * Otherwise, call with ext4_ind_get_blocks() to handle indirect mapping
1135 * On success, it returns the number of blocks being mapped or allocate.
1136 * if create==0 and the blocks are pre-allocated and uninitialized block,
1137 * the result buffer head is unmapped. If the create ==1, it will make sure
1138 * the buffer head is mapped.
1140 * It returns 0 if plain look up failed (blocks have not been allocated), in
1141 * that casem, buffer head is unmapped
1143 * It returns the error in case of allocation failure.
1145 int ext4_get_blocks(handle_t *handle, struct inode *inode, sector_t block,
1146 unsigned int max_blocks, struct buffer_head *bh,
1151 clear_buffer_mapped(bh);
1152 clear_buffer_unwritten(bh);
1155 * Try to see if we can get the block without requesting
1156 * for new file system block.
1158 down_read((&EXT4_I(inode)->i_data_sem));
1159 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1160 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1163 retval = ext4_ind_get_blocks(handle, inode, block, max_blocks,
1166 up_read((&EXT4_I(inode)->i_data_sem));
1168 /* If it is only a block(s) look up */
1169 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1173 * Returns if the blocks have already allocated
1175 * Note that if blocks have been preallocated
1176 * ext4_ext_get_block() returns th create = 0
1177 * with buffer head unmapped.
1179 if (retval > 0 && buffer_mapped(bh))
1183 * When we call get_blocks without the create flag, the
1184 * BH_Unwritten flag could have gotten set if the blocks
1185 * requested were part of a uninitialized extent. We need to
1186 * clear this flag now that we are committed to convert all or
1187 * part of the uninitialized extent to be an initialized
1188 * extent. This is because we need to avoid the combination
1189 * of BH_Unwritten and BH_Mapped flags being simultaneously
1190 * set on the buffer_head.
1192 clear_buffer_unwritten(bh);
1195 * New blocks allocate and/or writing to uninitialized extent
1196 * will possibly result in updating i_data, so we take
1197 * the write lock of i_data_sem, and call get_blocks()
1198 * with create == 1 flag.
1200 down_write((&EXT4_I(inode)->i_data_sem));
1203 * if the caller is from delayed allocation writeout path
1204 * we have already reserved fs blocks for allocation
1205 * let the underlying get_block() function know to
1206 * avoid double accounting
1208 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1209 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1211 * We need to check for EXT4 here because migrate
1212 * could have changed the inode type in between
1214 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1215 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1218 retval = ext4_ind_get_blocks(handle, inode, block,
1219 max_blocks, bh, flags);
1221 if (retval > 0 && buffer_new(bh)) {
1223 * We allocated new blocks which will result in
1224 * i_data's format changing. Force the migrate
1225 * to fail by clearing migrate flags
1227 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1232 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
1233 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1235 * Update reserved blocks/metadata blocks
1236 * after successful block allocation
1237 * which were deferred till now
1239 if ((retval > 0) && buffer_delay(bh))
1240 ext4_da_update_reserve_space(inode, retval);
1243 up_write((&EXT4_I(inode)->i_data_sem));
1247 /* Maximum number of blocks we map for direct IO at once. */
1248 #define DIO_MAX_BLOCKS 4096
1250 int ext4_get_block(struct inode *inode, sector_t iblock,
1251 struct buffer_head *bh_result, int create)
1253 handle_t *handle = ext4_journal_current_handle();
1254 int ret = 0, started = 0;
1255 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1258 if (create && !handle) {
1259 /* Direct IO write... */
1260 if (max_blocks > DIO_MAX_BLOCKS)
1261 max_blocks = DIO_MAX_BLOCKS;
1262 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1263 handle = ext4_journal_start(inode, dio_credits);
1264 if (IS_ERR(handle)) {
1265 ret = PTR_ERR(handle);
1271 ret = ext4_get_blocks(handle, inode, iblock, max_blocks, bh_result,
1272 create ? EXT4_GET_BLOCKS_CREATE : 0);
1274 bh_result->b_size = (ret << inode->i_blkbits);
1278 ext4_journal_stop(handle);
1284 * `handle' can be NULL if create is zero
1286 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1287 ext4_lblk_t block, int create, int *errp)
1289 struct buffer_head dummy;
1291 int flags = EXT4_GET_BLOCKS_EXTEND_DISKSIZE;
1293 J_ASSERT(handle != NULL || create == 0);
1296 dummy.b_blocknr = -1000;
1297 buffer_trace_init(&dummy.b_history);
1299 flags |= EXT4_GET_BLOCKS_CREATE;
1300 err = ext4_get_blocks(handle, inode, block, 1, &dummy, flags);
1302 * ext4_get_blocks() returns number of blocks mapped. 0 in
1311 if (!err && buffer_mapped(&dummy)) {
1312 struct buffer_head *bh;
1313 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1318 if (buffer_new(&dummy)) {
1319 J_ASSERT(create != 0);
1320 J_ASSERT(handle != NULL);
1323 * Now that we do not always journal data, we should
1324 * keep in mind whether this should always journal the
1325 * new buffer as metadata. For now, regular file
1326 * writes use ext4_get_block instead, so it's not a
1330 BUFFER_TRACE(bh, "call get_create_access");
1331 fatal = ext4_journal_get_create_access(handle, bh);
1332 if (!fatal && !buffer_uptodate(bh)) {
1333 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1334 set_buffer_uptodate(bh);
1337 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1338 err = ext4_handle_dirty_metadata(handle, inode, bh);
1342 BUFFER_TRACE(bh, "not a new buffer");
1355 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1356 ext4_lblk_t block, int create, int *err)
1358 struct buffer_head *bh;
1360 bh = ext4_getblk(handle, inode, block, create, err);
1363 if (buffer_uptodate(bh))
1365 ll_rw_block(READ_META, 1, &bh);
1367 if (buffer_uptodate(bh))
1374 static int walk_page_buffers(handle_t *handle,
1375 struct buffer_head *head,
1379 int (*fn)(handle_t *handle,
1380 struct buffer_head *bh))
1382 struct buffer_head *bh;
1383 unsigned block_start, block_end;
1384 unsigned blocksize = head->b_size;
1386 struct buffer_head *next;
1388 for (bh = head, block_start = 0;
1389 ret == 0 && (bh != head || !block_start);
1390 block_start = block_end, bh = next)
1392 next = bh->b_this_page;
1393 block_end = block_start + blocksize;
1394 if (block_end <= from || block_start >= to) {
1395 if (partial && !buffer_uptodate(bh))
1399 err = (*fn)(handle, bh);
1407 * To preserve ordering, it is essential that the hole instantiation and
1408 * the data write be encapsulated in a single transaction. We cannot
1409 * close off a transaction and start a new one between the ext4_get_block()
1410 * and the commit_write(). So doing the jbd2_journal_start at the start of
1411 * prepare_write() is the right place.
1413 * Also, this function can nest inside ext4_writepage() ->
1414 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1415 * has generated enough buffer credits to do the whole page. So we won't
1416 * block on the journal in that case, which is good, because the caller may
1419 * By accident, ext4 can be reentered when a transaction is open via
1420 * quota file writes. If we were to commit the transaction while thus
1421 * reentered, there can be a deadlock - we would be holding a quota
1422 * lock, and the commit would never complete if another thread had a
1423 * transaction open and was blocking on the quota lock - a ranking
1426 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1427 * will _not_ run commit under these circumstances because handle->h_ref
1428 * is elevated. We'll still have enough credits for the tiny quotafile
1431 static int do_journal_get_write_access(handle_t *handle,
1432 struct buffer_head *bh)
1434 if (!buffer_mapped(bh) || buffer_freed(bh))
1436 return ext4_journal_get_write_access(handle, bh);
1439 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1440 loff_t pos, unsigned len, unsigned flags,
1441 struct page **pagep, void **fsdata)
1443 struct inode *inode = mapping->host;
1444 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1451 trace_mark(ext4_write_begin,
1452 "dev %s ino %lu pos %llu len %u flags %u",
1453 inode->i_sb->s_id, inode->i_ino,
1454 (unsigned long long) pos, len, flags);
1455 index = pos >> PAGE_CACHE_SHIFT;
1456 from = pos & (PAGE_CACHE_SIZE - 1);
1460 handle = ext4_journal_start(inode, needed_blocks);
1461 if (IS_ERR(handle)) {
1462 ret = PTR_ERR(handle);
1466 /* We cannot recurse into the filesystem as the transaction is already
1468 flags |= AOP_FLAG_NOFS;
1470 page = grab_cache_page_write_begin(mapping, index, flags);
1472 ext4_journal_stop(handle);
1478 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1481 if (!ret && ext4_should_journal_data(inode)) {
1482 ret = walk_page_buffers(handle, page_buffers(page),
1483 from, to, NULL, do_journal_get_write_access);
1488 ext4_journal_stop(handle);
1489 page_cache_release(page);
1491 * block_write_begin may have instantiated a few blocks
1492 * outside i_size. Trim these off again. Don't need
1493 * i_size_read because we hold i_mutex.
1495 if (pos + len > inode->i_size)
1496 vmtruncate(inode, inode->i_size);
1499 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1505 /* For write_end() in data=journal mode */
1506 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1508 if (!buffer_mapped(bh) || buffer_freed(bh))
1510 set_buffer_uptodate(bh);
1511 return ext4_handle_dirty_metadata(handle, NULL, bh);
1515 * We need to pick up the new inode size which generic_commit_write gave us
1516 * `file' can be NULL - eg, when called from page_symlink().
1518 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1519 * buffers are managed internally.
1521 static int ext4_ordered_write_end(struct file *file,
1522 struct address_space *mapping,
1523 loff_t pos, unsigned len, unsigned copied,
1524 struct page *page, void *fsdata)
1526 handle_t *handle = ext4_journal_current_handle();
1527 struct inode *inode = mapping->host;
1530 trace_mark(ext4_ordered_write_end,
1531 "dev %s ino %lu pos %llu len %u copied %u",
1532 inode->i_sb->s_id, inode->i_ino,
1533 (unsigned long long) pos, len, copied);
1534 ret = ext4_jbd2_file_inode(handle, inode);
1539 new_i_size = pos + copied;
1540 if (new_i_size > EXT4_I(inode)->i_disksize) {
1541 ext4_update_i_disksize(inode, new_i_size);
1542 /* We need to mark inode dirty even if
1543 * new_i_size is less that inode->i_size
1544 * bu greater than i_disksize.(hint delalloc)
1546 ext4_mark_inode_dirty(handle, inode);
1549 ret2 = generic_write_end(file, mapping, pos, len, copied,
1555 ret2 = ext4_journal_stop(handle);
1559 return ret ? ret : copied;
1562 static int ext4_writeback_write_end(struct file *file,
1563 struct address_space *mapping,
1564 loff_t pos, unsigned len, unsigned copied,
1565 struct page *page, void *fsdata)
1567 handle_t *handle = ext4_journal_current_handle();
1568 struct inode *inode = mapping->host;
1572 trace_mark(ext4_writeback_write_end,
1573 "dev %s ino %lu pos %llu len %u copied %u",
1574 inode->i_sb->s_id, inode->i_ino,
1575 (unsigned long long) pos, len, copied);
1576 new_i_size = pos + copied;
1577 if (new_i_size > EXT4_I(inode)->i_disksize) {
1578 ext4_update_i_disksize(inode, new_i_size);
1579 /* We need to mark inode dirty even if
1580 * new_i_size is less that inode->i_size
1581 * bu greater than i_disksize.(hint delalloc)
1583 ext4_mark_inode_dirty(handle, inode);
1586 ret2 = generic_write_end(file, mapping, pos, len, copied,
1592 ret2 = ext4_journal_stop(handle);
1596 return ret ? ret : copied;
1599 static int ext4_journalled_write_end(struct file *file,
1600 struct address_space *mapping,
1601 loff_t pos, unsigned len, unsigned copied,
1602 struct page *page, void *fsdata)
1604 handle_t *handle = ext4_journal_current_handle();
1605 struct inode *inode = mapping->host;
1611 trace_mark(ext4_journalled_write_end,
1612 "dev %s ino %lu pos %llu len %u copied %u",
1613 inode->i_sb->s_id, inode->i_ino,
1614 (unsigned long long) pos, len, copied);
1615 from = pos & (PAGE_CACHE_SIZE - 1);
1619 if (!PageUptodate(page))
1621 page_zero_new_buffers(page, from+copied, to);
1624 ret = walk_page_buffers(handle, page_buffers(page), from,
1625 to, &partial, write_end_fn);
1627 SetPageUptodate(page);
1628 new_i_size = pos + copied;
1629 if (new_i_size > inode->i_size)
1630 i_size_write(inode, pos+copied);
1631 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1632 if (new_i_size > EXT4_I(inode)->i_disksize) {
1633 ext4_update_i_disksize(inode, new_i_size);
1634 ret2 = ext4_mark_inode_dirty(handle, inode);
1640 ret2 = ext4_journal_stop(handle);
1643 page_cache_release(page);
1645 return ret ? ret : copied;
1648 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1651 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1652 unsigned long md_needed, mdblocks, total = 0;
1655 * recalculate the amount of metadata blocks to reserve
1656 * in order to allocate nrblocks
1657 * worse case is one extent per block
1660 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1661 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1662 mdblocks = ext4_calc_metadata_amount(inode, total);
1663 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1665 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1666 total = md_needed + nrblocks;
1669 * Make quota reservation here to prevent quota overflow
1670 * later. Real quota accounting is done at pages writeout
1673 if (vfs_dq_reserve_block(inode, total)) {
1674 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1678 if (ext4_claim_free_blocks(sbi, total)) {
1679 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1680 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1684 vfs_dq_release_reservation_block(inode, total);
1687 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1688 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1690 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1691 return 0; /* success */
1694 static void ext4_da_release_space(struct inode *inode, int to_free)
1696 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1697 int total, mdb, mdb_free, release;
1700 return; /* Nothing to release, exit */
1702 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1704 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1706 * if there is no reserved blocks, but we try to free some
1707 * then the counter is messed up somewhere.
1708 * but since this function is called from invalidate
1709 * page, it's harmless to return without any action
1711 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1712 "blocks for inode %lu, but there is no reserved "
1713 "data blocks\n", to_free, inode->i_ino);
1714 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1718 /* recalculate the number of metablocks still need to be reserved */
1719 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1720 mdb = ext4_calc_metadata_amount(inode, total);
1722 /* figure out how many metablocks to release */
1723 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1724 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1726 release = to_free + mdb_free;
1728 /* update fs dirty blocks counter for truncate case */
1729 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1731 /* update per-inode reservations */
1732 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1733 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1735 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1736 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1737 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1739 vfs_dq_release_reservation_block(inode, release);
1742 static void ext4_da_page_release_reservation(struct page *page,
1743 unsigned long offset)
1746 struct buffer_head *head, *bh;
1747 unsigned int curr_off = 0;
1749 head = page_buffers(page);
1752 unsigned int next_off = curr_off + bh->b_size;
1754 if ((offset <= curr_off) && (buffer_delay(bh))) {
1756 clear_buffer_delay(bh);
1758 curr_off = next_off;
1759 } while ((bh = bh->b_this_page) != head);
1760 ext4_da_release_space(page->mapping->host, to_release);
1764 * Delayed allocation stuff
1767 struct mpage_da_data {
1768 struct inode *inode;
1769 sector_t b_blocknr; /* start block number of extent */
1770 size_t b_size; /* size of extent */
1771 unsigned long b_state; /* state of the extent */
1772 unsigned long first_page, next_page; /* extent of pages */
1773 struct writeback_control *wbc;
1780 * mpage_da_submit_io - walks through extent of pages and try to write
1781 * them with writepage() call back
1783 * @mpd->inode: inode
1784 * @mpd->first_page: first page of the extent
1785 * @mpd->next_page: page after the last page of the extent
1787 * By the time mpage_da_submit_io() is called we expect all blocks
1788 * to be allocated. this may be wrong if allocation failed.
1790 * As pages are already locked by write_cache_pages(), we can't use it
1792 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1795 struct pagevec pvec;
1796 unsigned long index, end;
1797 int ret = 0, err, nr_pages, i;
1798 struct inode *inode = mpd->inode;
1799 struct address_space *mapping = inode->i_mapping;
1801 BUG_ON(mpd->next_page <= mpd->first_page);
1803 * We need to start from the first_page to the next_page - 1
1804 * to make sure we also write the mapped dirty buffer_heads.
1805 * If we look at mpd->b_blocknr we would only be looking
1806 * at the currently mapped buffer_heads.
1808 index = mpd->first_page;
1809 end = mpd->next_page - 1;
1811 pagevec_init(&pvec, 0);
1812 while (index <= end) {
1813 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1816 for (i = 0; i < nr_pages; i++) {
1817 struct page *page = pvec.pages[i];
1819 index = page->index;
1824 BUG_ON(!PageLocked(page));
1825 BUG_ON(PageWriteback(page));
1827 pages_skipped = mpd->wbc->pages_skipped;
1828 err = mapping->a_ops->writepage(page, mpd->wbc);
1829 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1831 * have successfully written the page
1832 * without skipping the same
1834 mpd->pages_written++;
1836 * In error case, we have to continue because
1837 * remaining pages are still locked
1838 * XXX: unlock and re-dirty them?
1843 pagevec_release(&pvec);
1849 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1851 * @mpd->inode - inode to walk through
1852 * @exbh->b_blocknr - first block on a disk
1853 * @exbh->b_size - amount of space in bytes
1854 * @logical - first logical block to start assignment with
1856 * the function goes through all passed space and put actual disk
1857 * block numbers into buffer heads, dropping BH_Delay and BH_Unwritten
1859 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1860 struct buffer_head *exbh)
1862 struct inode *inode = mpd->inode;
1863 struct address_space *mapping = inode->i_mapping;
1864 int blocks = exbh->b_size >> inode->i_blkbits;
1865 sector_t pblock = exbh->b_blocknr, cur_logical;
1866 struct buffer_head *head, *bh;
1868 struct pagevec pvec;
1871 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1872 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1873 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1875 pagevec_init(&pvec, 0);
1877 while (index <= end) {
1878 /* XXX: optimize tail */
1879 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1882 for (i = 0; i < nr_pages; i++) {
1883 struct page *page = pvec.pages[i];
1885 index = page->index;
1890 BUG_ON(!PageLocked(page));
1891 BUG_ON(PageWriteback(page));
1892 BUG_ON(!page_has_buffers(page));
1894 bh = page_buffers(page);
1897 /* skip blocks out of the range */
1899 if (cur_logical >= logical)
1902 } while ((bh = bh->b_this_page) != head);
1905 if (cur_logical >= logical + blocks)
1908 if (buffer_delay(bh) ||
1909 buffer_unwritten(bh)) {
1911 BUG_ON(bh->b_bdev != inode->i_sb->s_bdev);
1913 if (buffer_delay(bh)) {
1914 clear_buffer_delay(bh);
1915 bh->b_blocknr = pblock;
1918 * unwritten already should have
1919 * blocknr assigned. Verify that
1921 clear_buffer_unwritten(bh);
1922 BUG_ON(bh->b_blocknr != pblock);
1925 } else if (buffer_mapped(bh))
1926 BUG_ON(bh->b_blocknr != pblock);
1930 } while ((bh = bh->b_this_page) != head);
1932 pagevec_release(&pvec);
1938 * __unmap_underlying_blocks - just a helper function to unmap
1939 * set of blocks described by @bh
1941 static inline void __unmap_underlying_blocks(struct inode *inode,
1942 struct buffer_head *bh)
1944 struct block_device *bdev = inode->i_sb->s_bdev;
1947 blocks = bh->b_size >> inode->i_blkbits;
1948 for (i = 0; i < blocks; i++)
1949 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1952 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1953 sector_t logical, long blk_cnt)
1957 struct pagevec pvec;
1958 struct inode *inode = mpd->inode;
1959 struct address_space *mapping = inode->i_mapping;
1961 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1962 end = (logical + blk_cnt - 1) >>
1963 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1964 while (index <= end) {
1965 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1968 for (i = 0; i < nr_pages; i++) {
1969 struct page *page = pvec.pages[i];
1970 index = page->index;
1975 BUG_ON(!PageLocked(page));
1976 BUG_ON(PageWriteback(page));
1977 block_invalidatepage(page, 0);
1978 ClearPageUptodate(page);
1985 static void ext4_print_free_blocks(struct inode *inode)
1987 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1988 printk(KERN_EMERG "Total free blocks count %lld\n",
1989 ext4_count_free_blocks(inode->i_sb));
1990 printk(KERN_EMERG "Free/Dirty block details\n");
1991 printk(KERN_EMERG "free_blocks=%lld\n",
1992 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1993 printk(KERN_EMERG "dirty_blocks=%lld\n",
1994 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1995 printk(KERN_EMERG "Block reservation details\n");
1996 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1997 EXT4_I(inode)->i_reserved_data_blocks);
1998 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1999 EXT4_I(inode)->i_reserved_meta_blocks);
2003 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2004 struct buffer_head *bh_result)
2007 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2008 loff_t disksize = EXT4_I(inode)->i_disksize;
2009 handle_t *handle = NULL;
2011 handle = ext4_journal_current_handle();
2013 ret = ext4_get_blocks(handle, inode, iblock, max_blocks,
2014 bh_result, EXT4_GET_BLOCKS_CREATE|
2015 EXT4_GET_BLOCKS_DELALLOC_RESERVE);
2019 bh_result->b_size = (ret << inode->i_blkbits);
2021 if (ext4_should_order_data(inode)) {
2023 retval = ext4_jbd2_file_inode(handle, inode);
2026 * Failed to add inode for ordered mode. Don't
2033 * Update on-disk size along with block allocation we don't
2034 * use 'extend_disksize' as size may change within already
2035 * allocated block -bzzz
2037 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2038 if (disksize > i_size_read(inode))
2039 disksize = i_size_read(inode);
2040 if (disksize > EXT4_I(inode)->i_disksize) {
2041 ext4_update_i_disksize(inode, disksize);
2042 ret = ext4_mark_inode_dirty(handle, inode);
2049 * mpage_da_map_blocks - go through given space
2051 * @mpd - bh describing space
2053 * The function skips space we know is already mapped to disk blocks.
2056 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
2059 struct buffer_head new;
2063 * We consider only non-mapped and non-allocated blocks
2065 if ((mpd->b_state & (1 << BH_Mapped)) &&
2066 !(mpd->b_state & (1 << BH_Delay)) &&
2067 !(mpd->b_state & (1 << BH_Unwritten)))
2070 * We need to make sure the BH_Delay flag is passed down to
2071 * ext4_da_get_block_write(), since it calls ext4_get_blocks()
2072 * with the EXT4_GET_BLOCKS_DELALLOC_RESERVE flag. This flag
2073 * causes ext4_get_blocks() to call
2074 * ext4_da_update_reserve_space() if the passed buffer head
2075 * has the BH_Delay flag set. In the future, once we clean up
2076 * the interfaces to ext4_get_blocks(), we should pass in a
2077 * separate flag which requests that the delayed allocation
2078 * statistics should be updated, instead of depending on the
2079 * state information getting passed down via the map_bh's
2080 * state bitmasks plus the magic
2081 * EXT4_GET_BLOCKS_DELALLOC_RESERVE flag.
2083 new.b_state = mpd->b_state & (1 << BH_Delay);
2085 new.b_size = mpd->b_size;
2086 next = mpd->b_blocknr;
2088 * If we didn't accumulate anything
2089 * to write simply return
2094 err = ext4_da_get_block_write(mpd->inode, next, &new);
2097 * If get block returns with error we simply
2098 * return. Later writepage will redirty the page and
2099 * writepages will find the dirty page again
2104 if (err == -ENOSPC &&
2105 ext4_count_free_blocks(mpd->inode->i_sb)) {
2111 * get block failure will cause us to loop in
2112 * writepages, because a_ops->writepage won't be able
2113 * to make progress. The page will be redirtied by
2114 * writepage and writepages will again try to write
2117 printk(KERN_EMERG "%s block allocation failed for inode %lu "
2118 "at logical offset %llu with max blocks "
2119 "%zd with error %d\n",
2120 __func__, mpd->inode->i_ino,
2121 (unsigned long long)next,
2122 mpd->b_size >> mpd->inode->i_blkbits, err);
2123 printk(KERN_EMERG "This should not happen.!! "
2124 "Data will be lost\n");
2125 if (err == -ENOSPC) {
2126 ext4_print_free_blocks(mpd->inode);
2128 /* invlaidate all the pages */
2129 ext4_da_block_invalidatepages(mpd, next,
2130 mpd->b_size >> mpd->inode->i_blkbits);
2133 BUG_ON(new.b_size == 0);
2135 if (buffer_new(&new))
2136 __unmap_underlying_blocks(mpd->inode, &new);
2139 * If blocks are delayed marked, we need to
2140 * put actual blocknr and drop delayed bit
2142 if ((mpd->b_state & (1 << BH_Delay)) ||
2143 (mpd->b_state & (1 << BH_Unwritten)))
2144 mpage_put_bnr_to_bhs(mpd, next, &new);
2149 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2150 (1 << BH_Delay) | (1 << BH_Unwritten))
2153 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2155 * @mpd->lbh - extent of blocks
2156 * @logical - logical number of the block in the file
2157 * @bh - bh of the block (used to access block's state)
2159 * the function is used to collect contig. blocks in same state
2161 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2162 sector_t logical, size_t b_size,
2163 unsigned long b_state)
2166 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2168 /* check if thereserved journal credits might overflow */
2169 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
2170 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2172 * With non-extent format we are limited by the journal
2173 * credit available. Total credit needed to insert
2174 * nrblocks contiguous blocks is dependent on the
2175 * nrblocks. So limit nrblocks.
2178 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2179 EXT4_MAX_TRANS_DATA) {
2181 * Adding the new buffer_head would make it cross the
2182 * allowed limit for which we have journal credit
2183 * reserved. So limit the new bh->b_size
2185 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2186 mpd->inode->i_blkbits;
2187 /* we will do mpage_da_submit_io in the next loop */
2191 * First block in the extent
2193 if (mpd->b_size == 0) {
2194 mpd->b_blocknr = logical;
2195 mpd->b_size = b_size;
2196 mpd->b_state = b_state & BH_FLAGS;
2200 next = mpd->b_blocknr + nrblocks;
2202 * Can we merge the block to our big extent?
2204 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2205 mpd->b_size += b_size;
2211 * We couldn't merge the block to our extent, so we
2212 * need to flush current extent and start new one
2214 if (mpage_da_map_blocks(mpd) == 0)
2215 mpage_da_submit_io(mpd);
2220 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2223 * unmapped buffer is possible for holes.
2224 * delay buffer is possible with delayed allocation.
2225 * We also need to consider unwritten buffer as unmapped.
2227 return (!buffer_mapped(bh) || buffer_delay(bh) ||
2228 buffer_unwritten(bh)) && buffer_dirty(bh);
2232 * __mpage_da_writepage - finds extent of pages and blocks
2234 * @page: page to consider
2235 * @wbc: not used, we just follow rules
2238 * The function finds extents of pages and scan them for all blocks.
2240 static int __mpage_da_writepage(struct page *page,
2241 struct writeback_control *wbc, void *data)
2243 struct mpage_da_data *mpd = data;
2244 struct inode *inode = mpd->inode;
2245 struct buffer_head *bh, *head;
2250 * Rest of the page in the page_vec
2251 * redirty then and skip then. We will
2252 * try to to write them again after
2253 * starting a new transaction
2255 redirty_page_for_writepage(wbc, page);
2257 return MPAGE_DA_EXTENT_TAIL;
2260 * Can we merge this page to current extent?
2262 if (mpd->next_page != page->index) {
2264 * Nope, we can't. So, we map non-allocated blocks
2265 * and start IO on them using writepage()
2267 if (mpd->next_page != mpd->first_page) {
2268 if (mpage_da_map_blocks(mpd) == 0)
2269 mpage_da_submit_io(mpd);
2271 * skip rest of the page in the page_vec
2274 redirty_page_for_writepage(wbc, page);
2276 return MPAGE_DA_EXTENT_TAIL;
2280 * Start next extent of pages ...
2282 mpd->first_page = page->index;
2292 mpd->next_page = page->index + 1;
2293 logical = (sector_t) page->index <<
2294 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2296 if (!page_has_buffers(page)) {
2297 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2298 (1 << BH_Dirty) | (1 << BH_Uptodate));
2300 return MPAGE_DA_EXTENT_TAIL;
2303 * Page with regular buffer heads, just add all dirty ones
2305 head = page_buffers(page);
2308 BUG_ON(buffer_locked(bh));
2310 * We need to try to allocate
2311 * unmapped blocks in the same page.
2312 * Otherwise we won't make progress
2313 * with the page in ext4_da_writepage
2315 if (ext4_bh_unmapped_or_delay(NULL, bh)) {
2316 mpage_add_bh_to_extent(mpd, logical,
2320 return MPAGE_DA_EXTENT_TAIL;
2321 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2323 * mapped dirty buffer. We need to update
2324 * the b_state because we look at
2325 * b_state in mpage_da_map_blocks. We don't
2326 * update b_size because if we find an
2327 * unmapped buffer_head later we need to
2328 * use the b_state flag of that buffer_head.
2330 if (mpd->b_size == 0)
2331 mpd->b_state = bh->b_state & BH_FLAGS;
2334 } while ((bh = bh->b_this_page) != head);
2341 * this is a special callback for ->write_begin() only
2342 * it's intention is to return mapped block or reserve space
2344 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2345 * We also have b_blocknr = -1 and b_bdev initialized properly
2347 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2348 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2349 * initialized properly.
2352 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2353 struct buffer_head *bh_result, int create)
2356 sector_t invalid_block = ~((sector_t) 0xffff);
2358 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2361 BUG_ON(create == 0);
2362 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2365 * first, we need to know whether the block is allocated already
2366 * preallocated blocks are unmapped but should treated
2367 * the same as allocated blocks.
2369 ret = ext4_get_blocks(NULL, inode, iblock, 1, bh_result, 0);
2370 if ((ret == 0) && !buffer_delay(bh_result)) {
2371 /* the block isn't (pre)allocated yet, let's reserve space */
2373 * XXX: __block_prepare_write() unmaps passed block,
2376 ret = ext4_da_reserve_space(inode, 1);
2378 /* not enough space to reserve */
2381 map_bh(bh_result, inode->i_sb, invalid_block);
2382 set_buffer_new(bh_result);
2383 set_buffer_delay(bh_result);
2384 } else if (ret > 0) {
2385 bh_result->b_size = (ret << inode->i_blkbits);
2386 if (buffer_unwritten(bh_result)) {
2387 /* A delayed write to unwritten bh should
2388 * be marked new and mapped. Mapped ensures
2389 * that we don't do get_block multiple times
2390 * when we write to the same offset and new
2391 * ensures that we do proper zero out for
2394 set_buffer_new(bh_result);
2395 set_buffer_mapped(bh_result);
2403 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2404 struct buffer_head *bh_result, int create)
2407 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2410 * we don't want to do block allocation in writepage
2411 * so call get_block_wrap with create = 0
2413 ret = ext4_get_blocks(NULL, inode, iblock, max_blocks, bh_result, 0);
2415 bh_result->b_size = (ret << inode->i_blkbits);
2422 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2423 * get called via journal_submit_inode_data_buffers (no journal handle)
2424 * get called via shrink_page_list via pdflush (no journal handle)
2425 * or grab_page_cache when doing write_begin (have journal handle)
2427 static int ext4_da_writepage(struct page *page,
2428 struct writeback_control *wbc)
2433 struct buffer_head *page_bufs;
2434 struct inode *inode = page->mapping->host;
2436 trace_mark(ext4_da_writepage,
2437 "dev %s ino %lu page_index %lu",
2438 inode->i_sb->s_id, inode->i_ino, page->index);
2439 size = i_size_read(inode);
2440 if (page->index == size >> PAGE_CACHE_SHIFT)
2441 len = size & ~PAGE_CACHE_MASK;
2443 len = PAGE_CACHE_SIZE;
2445 if (page_has_buffers(page)) {
2446 page_bufs = page_buffers(page);
2447 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2448 ext4_bh_unmapped_or_delay)) {
2450 * We don't want to do block allocation
2451 * So redirty the page and return
2452 * We may reach here when we do a journal commit
2453 * via journal_submit_inode_data_buffers.
2454 * If we don't have mapping block we just ignore
2455 * them. We can also reach here via shrink_page_list
2457 redirty_page_for_writepage(wbc, page);
2463 * The test for page_has_buffers() is subtle:
2464 * We know the page is dirty but it lost buffers. That means
2465 * that at some moment in time after write_begin()/write_end()
2466 * has been called all buffers have been clean and thus they
2467 * must have been written at least once. So they are all
2468 * mapped and we can happily proceed with mapping them
2469 * and writing the page.
2471 * Try to initialize the buffer_heads and check whether
2472 * all are mapped and non delay. We don't want to
2473 * do block allocation here.
2475 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2476 ext4_normal_get_block_write);
2478 page_bufs = page_buffers(page);
2479 /* check whether all are mapped and non delay */
2480 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2481 ext4_bh_unmapped_or_delay)) {
2482 redirty_page_for_writepage(wbc, page);
2488 * We can't do block allocation here
2489 * so just redity the page and unlock
2492 redirty_page_for_writepage(wbc, page);
2496 /* now mark the buffer_heads as dirty and uptodate */
2497 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2500 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2501 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2503 ret = block_write_full_page(page,
2504 ext4_normal_get_block_write,
2511 * This is called via ext4_da_writepages() to
2512 * calulate the total number of credits to reserve to fit
2513 * a single extent allocation into a single transaction,
2514 * ext4_da_writpeages() will loop calling this before
2515 * the block allocation.
2518 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2520 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2523 * With non-extent format the journal credit needed to
2524 * insert nrblocks contiguous block is dependent on
2525 * number of contiguous block. So we will limit
2526 * number of contiguous block to a sane value
2528 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2529 (max_blocks > EXT4_MAX_TRANS_DATA))
2530 max_blocks = EXT4_MAX_TRANS_DATA;
2532 return ext4_chunk_trans_blocks(inode, max_blocks);
2535 static int ext4_da_writepages(struct address_space *mapping,
2536 struct writeback_control *wbc)
2539 int range_whole = 0;
2540 handle_t *handle = NULL;
2541 struct mpage_da_data mpd;
2542 struct inode *inode = mapping->host;
2543 int no_nrwrite_index_update;
2544 int pages_written = 0;
2546 int range_cyclic, cycled = 1, io_done = 0;
2547 int needed_blocks, ret = 0, nr_to_writebump = 0;
2548 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2550 trace_mark(ext4_da_writepages,
2551 "dev %s ino %lu nr_t_write %ld "
2552 "pages_skipped %ld range_start %llu "
2553 "range_end %llu nonblocking %d "
2554 "for_kupdate %d for_reclaim %d "
2555 "for_writepages %d range_cyclic %d",
2556 inode->i_sb->s_id, inode->i_ino,
2557 wbc->nr_to_write, wbc->pages_skipped,
2558 (unsigned long long) wbc->range_start,
2559 (unsigned long long) wbc->range_end,
2560 wbc->nonblocking, wbc->for_kupdate,
2561 wbc->for_reclaim, wbc->for_writepages,
2565 * No pages to write? This is mainly a kludge to avoid starting
2566 * a transaction for special inodes like journal inode on last iput()
2567 * because that could violate lock ordering on umount
2569 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2573 * If the filesystem has aborted, it is read-only, so return
2574 * right away instead of dumping stack traces later on that
2575 * will obscure the real source of the problem. We test
2576 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2577 * the latter could be true if the filesystem is mounted
2578 * read-only, and in that case, ext4_da_writepages should
2579 * *never* be called, so if that ever happens, we would want
2582 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2586 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2587 * This make sure small files blocks are allocated in
2588 * single attempt. This ensure that small files
2589 * get less fragmented.
2591 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2592 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2593 wbc->nr_to_write = sbi->s_mb_stream_request;
2595 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2598 range_cyclic = wbc->range_cyclic;
2599 if (wbc->range_cyclic) {
2600 index = mapping->writeback_index;
2603 wbc->range_start = index << PAGE_CACHE_SHIFT;
2604 wbc->range_end = LLONG_MAX;
2605 wbc->range_cyclic = 0;
2607 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2610 mpd.inode = mapping->host;
2613 * we don't want write_cache_pages to update
2614 * nr_to_write and writeback_index
2616 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2617 wbc->no_nrwrite_index_update = 1;
2618 pages_skipped = wbc->pages_skipped;
2621 while (!ret && wbc->nr_to_write > 0) {
2624 * we insert one extent at a time. So we need
2625 * credit needed for single extent allocation.
2626 * journalled mode is currently not supported
2629 BUG_ON(ext4_should_journal_data(inode));
2630 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2632 /* start a new transaction*/
2633 handle = ext4_journal_start(inode, needed_blocks);
2634 if (IS_ERR(handle)) {
2635 ret = PTR_ERR(handle);
2636 printk(KERN_CRIT "%s: jbd2_start: "
2637 "%ld pages, ino %lu; err %d\n", __func__,
2638 wbc->nr_to_write, inode->i_ino, ret);
2640 goto out_writepages;
2644 * Now call __mpage_da_writepage to find the next
2645 * contiguous region of logical blocks that need
2646 * blocks to be allocated by ext4. We don't actually
2647 * submit the blocks for I/O here, even though
2648 * write_cache_pages thinks it will, and will set the
2649 * pages as clean for write before calling
2650 * __mpage_da_writepage().
2658 mpd.pages_written = 0;
2660 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage,
2663 * If we have a contigous extent of pages and we
2664 * haven't done the I/O yet, map the blocks and submit
2667 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
2668 if (mpage_da_map_blocks(&mpd) == 0)
2669 mpage_da_submit_io(&mpd);
2671 ret = MPAGE_DA_EXTENT_TAIL;
2673 wbc->nr_to_write -= mpd.pages_written;
2675 ext4_journal_stop(handle);
2677 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
2678 /* commit the transaction which would
2679 * free blocks released in the transaction
2682 jbd2_journal_force_commit_nested(sbi->s_journal);
2683 wbc->pages_skipped = pages_skipped;
2685 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2687 * got one extent now try with
2690 pages_written += mpd.pages_written;
2691 wbc->pages_skipped = pages_skipped;
2694 } else if (wbc->nr_to_write)
2696 * There is no more writeout needed
2697 * or we requested for a noblocking writeout
2698 * and we found the device congested
2702 if (!io_done && !cycled) {
2705 wbc->range_start = index << PAGE_CACHE_SHIFT;
2706 wbc->range_end = mapping->writeback_index - 1;
2709 if (pages_skipped != wbc->pages_skipped)
2710 printk(KERN_EMERG "This should not happen leaving %s "
2711 "with nr_to_write = %ld ret = %d\n",
2712 __func__, wbc->nr_to_write, ret);
2715 index += pages_written;
2716 wbc->range_cyclic = range_cyclic;
2717 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2719 * set the writeback_index so that range_cyclic
2720 * mode will write it back later
2722 mapping->writeback_index = index;
2725 if (!no_nrwrite_index_update)
2726 wbc->no_nrwrite_index_update = 0;
2727 wbc->nr_to_write -= nr_to_writebump;
2728 trace_mark(ext4_da_writepage_result,
2729 "dev %s ino %lu ret %d pages_written %d "
2730 "pages_skipped %ld congestion %d "
2731 "more_io %d no_nrwrite_index_update %d",
2732 inode->i_sb->s_id, inode->i_ino, ret,
2733 pages_written, wbc->pages_skipped,
2734 wbc->encountered_congestion, wbc->more_io,
2735 wbc->no_nrwrite_index_update);
2739 #define FALL_BACK_TO_NONDELALLOC 1
2740 static int ext4_nonda_switch(struct super_block *sb)
2742 s64 free_blocks, dirty_blocks;
2743 struct ext4_sb_info *sbi = EXT4_SB(sb);
2746 * switch to non delalloc mode if we are running low
2747 * on free block. The free block accounting via percpu
2748 * counters can get slightly wrong with percpu_counter_batch getting
2749 * accumulated on each CPU without updating global counters
2750 * Delalloc need an accurate free block accounting. So switch
2751 * to non delalloc when we are near to error range.
2753 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2754 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2755 if (2 * free_blocks < 3 * dirty_blocks ||
2756 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2758 * free block count is less that 150% of dirty blocks
2759 * or free blocks is less that watermark
2766 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2767 loff_t pos, unsigned len, unsigned flags,
2768 struct page **pagep, void **fsdata)
2770 int ret, retries = 0;
2774 struct inode *inode = mapping->host;
2777 index = pos >> PAGE_CACHE_SHIFT;
2778 from = pos & (PAGE_CACHE_SIZE - 1);
2781 if (ext4_nonda_switch(inode->i_sb)) {
2782 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2783 return ext4_write_begin(file, mapping, pos,
2784 len, flags, pagep, fsdata);
2786 *fsdata = (void *)0;
2788 trace_mark(ext4_da_write_begin,
2789 "dev %s ino %lu pos %llu len %u flags %u",
2790 inode->i_sb->s_id, inode->i_ino,
2791 (unsigned long long) pos, len, flags);
2794 * With delayed allocation, we don't log the i_disksize update
2795 * if there is delayed block allocation. But we still need
2796 * to journalling the i_disksize update if writes to the end
2797 * of file which has an already mapped buffer.
2799 handle = ext4_journal_start(inode, 1);
2800 if (IS_ERR(handle)) {
2801 ret = PTR_ERR(handle);
2804 /* We cannot recurse into the filesystem as the transaction is already
2806 flags |= AOP_FLAG_NOFS;
2808 page = grab_cache_page_write_begin(mapping, index, flags);
2810 ext4_journal_stop(handle);
2816 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2817 ext4_da_get_block_prep);
2820 ext4_journal_stop(handle);
2821 page_cache_release(page);
2823 * block_write_begin may have instantiated a few blocks
2824 * outside i_size. Trim these off again. Don't need
2825 * i_size_read because we hold i_mutex.
2827 if (pos + len > inode->i_size)
2828 vmtruncate(inode, inode->i_size);
2831 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2838 * Check if we should update i_disksize
2839 * when write to the end of file but not require block allocation
2841 static int ext4_da_should_update_i_disksize(struct page *page,
2842 unsigned long offset)
2844 struct buffer_head *bh;
2845 struct inode *inode = page->mapping->host;
2849 bh = page_buffers(page);
2850 idx = offset >> inode->i_blkbits;
2852 for (i = 0; i < idx; i++)
2853 bh = bh->b_this_page;
2855 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2860 static int ext4_da_write_end(struct file *file,
2861 struct address_space *mapping,
2862 loff_t pos, unsigned len, unsigned copied,
2863 struct page *page, void *fsdata)
2865 struct inode *inode = mapping->host;
2867 handle_t *handle = ext4_journal_current_handle();
2869 unsigned long start, end;
2870 int write_mode = (int)(unsigned long)fsdata;
2872 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2873 if (ext4_should_order_data(inode)) {
2874 return ext4_ordered_write_end(file, mapping, pos,
2875 len, copied, page, fsdata);
2876 } else if (ext4_should_writeback_data(inode)) {
2877 return ext4_writeback_write_end(file, mapping, pos,
2878 len, copied, page, fsdata);
2884 trace_mark(ext4_da_write_end,
2885 "dev %s ino %lu pos %llu len %u copied %u",
2886 inode->i_sb->s_id, inode->i_ino,
2887 (unsigned long long) pos, len, copied);
2888 start = pos & (PAGE_CACHE_SIZE - 1);
2889 end = start + copied - 1;
2892 * generic_write_end() will run mark_inode_dirty() if i_size
2893 * changes. So let's piggyback the i_disksize mark_inode_dirty
2897 new_i_size = pos + copied;
2898 if (new_i_size > EXT4_I(inode)->i_disksize) {
2899 if (ext4_da_should_update_i_disksize(page, end)) {
2900 down_write(&EXT4_I(inode)->i_data_sem);
2901 if (new_i_size > EXT4_I(inode)->i_disksize) {
2903 * Updating i_disksize when extending file
2904 * without needing block allocation
2906 if (ext4_should_order_data(inode))
2907 ret = ext4_jbd2_file_inode(handle,
2910 EXT4_I(inode)->i_disksize = new_i_size;
2912 up_write(&EXT4_I(inode)->i_data_sem);
2913 /* We need to mark inode dirty even if
2914 * new_i_size is less that inode->i_size
2915 * bu greater than i_disksize.(hint delalloc)
2917 ext4_mark_inode_dirty(handle, inode);
2920 ret2 = generic_write_end(file, mapping, pos, len, copied,
2925 ret2 = ext4_journal_stop(handle);
2929 return ret ? ret : copied;
2932 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2935 * Drop reserved blocks
2937 BUG_ON(!PageLocked(page));
2938 if (!page_has_buffers(page))
2941 ext4_da_page_release_reservation(page, offset);
2944 ext4_invalidatepage(page, offset);
2950 * Force all delayed allocation blocks to be allocated for a given inode.
2952 int ext4_alloc_da_blocks(struct inode *inode)
2954 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2955 !EXT4_I(inode)->i_reserved_meta_blocks)
2959 * We do something simple for now. The filemap_flush() will
2960 * also start triggering a write of the data blocks, which is
2961 * not strictly speaking necessary (and for users of
2962 * laptop_mode, not even desirable). However, to do otherwise
2963 * would require replicating code paths in:
2965 * ext4_da_writepages() ->
2966 * write_cache_pages() ---> (via passed in callback function)
2967 * __mpage_da_writepage() -->
2968 * mpage_add_bh_to_extent()
2969 * mpage_da_map_blocks()
2971 * The problem is that write_cache_pages(), located in
2972 * mm/page-writeback.c, marks pages clean in preparation for
2973 * doing I/O, which is not desirable if we're not planning on
2976 * We could call write_cache_pages(), and then redirty all of
2977 * the pages by calling redirty_page_for_writeback() but that
2978 * would be ugly in the extreme. So instead we would need to
2979 * replicate parts of the code in the above functions,
2980 * simplifying them becuase we wouldn't actually intend to
2981 * write out the pages, but rather only collect contiguous
2982 * logical block extents, call the multi-block allocator, and
2983 * then update the buffer heads with the block allocations.
2985 * For now, though, we'll cheat by calling filemap_flush(),
2986 * which will map the blocks, and start the I/O, but not
2987 * actually wait for the I/O to complete.
2989 return filemap_flush(inode->i_mapping);
2993 * bmap() is special. It gets used by applications such as lilo and by
2994 * the swapper to find the on-disk block of a specific piece of data.
2996 * Naturally, this is dangerous if the block concerned is still in the
2997 * journal. If somebody makes a swapfile on an ext4 data-journaling
2998 * filesystem and enables swap, then they may get a nasty shock when the
2999 * data getting swapped to that swapfile suddenly gets overwritten by
3000 * the original zero's written out previously to the journal and
3001 * awaiting writeback in the kernel's buffer cache.
3003 * So, if we see any bmap calls here on a modified, data-journaled file,
3004 * take extra steps to flush any blocks which might be in the cache.
3006 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3008 struct inode *inode = mapping->host;
3012 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3013 test_opt(inode->i_sb, DELALLOC)) {
3015 * With delalloc we want to sync the file
3016 * so that we can make sure we allocate
3019 filemap_write_and_wait(mapping);
3022 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
3024 * This is a REALLY heavyweight approach, but the use of
3025 * bmap on dirty files is expected to be extremely rare:
3026 * only if we run lilo or swapon on a freshly made file
3027 * do we expect this to happen.
3029 * (bmap requires CAP_SYS_RAWIO so this does not
3030 * represent an unprivileged user DOS attack --- we'd be
3031 * in trouble if mortal users could trigger this path at
3034 * NB. EXT4_STATE_JDATA is not set on files other than
3035 * regular files. If somebody wants to bmap a directory
3036 * or symlink and gets confused because the buffer
3037 * hasn't yet been flushed to disk, they deserve
3038 * everything they get.
3041 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
3042 journal = EXT4_JOURNAL(inode);
3043 jbd2_journal_lock_updates(journal);
3044 err = jbd2_journal_flush(journal);
3045 jbd2_journal_unlock_updates(journal);
3051 return generic_block_bmap(mapping, block, ext4_get_block);
3054 static int bget_one(handle_t *handle, struct buffer_head *bh)
3060 static int bput_one(handle_t *handle, struct buffer_head *bh)
3067 * Note that we don't need to start a transaction unless we're journaling data
3068 * because we should have holes filled from ext4_page_mkwrite(). We even don't
3069 * need to file the inode to the transaction's list in ordered mode because if
3070 * we are writing back data added by write(), the inode is already there and if
3071 * we are writing back data modified via mmap(), noone guarantees in which
3072 * transaction the data will hit the disk. In case we are journaling data, we
3073 * cannot start transaction directly because transaction start ranks above page
3074 * lock so we have to do some magic.
3076 * In all journaling modes block_write_full_page() will start the I/O.
3080 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
3085 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
3087 * Same applies to ext4_get_block(). We will deadlock on various things like
3088 * lock_journal and i_data_sem
3090 * Setting PF_MEMALLOC here doesn't work - too many internal memory
3093 * 16May01: If we're reentered then journal_current_handle() will be
3094 * non-zero. We simply *return*.
3096 * 1 July 2001: @@@ FIXME:
3097 * In journalled data mode, a data buffer may be metadata against the
3098 * current transaction. But the same file is part of a shared mapping
3099 * and someone does a writepage() on it.
3101 * We will move the buffer onto the async_data list, but *after* it has
3102 * been dirtied. So there's a small window where we have dirty data on
3105 * Note that this only applies to the last partial page in the file. The
3106 * bit which block_write_full_page() uses prepare/commit for. (That's
3107 * broken code anyway: it's wrong for msync()).
3109 * It's a rare case: affects the final partial page, for journalled data
3110 * where the file is subject to bith write() and writepage() in the same
3111 * transction. To fix it we'll need a custom block_write_full_page().
3112 * We'll probably need that anyway for journalling writepage() output.
3114 * We don't honour synchronous mounts for writepage(). That would be
3115 * disastrous. Any write() or metadata operation will sync the fs for
3119 static int __ext4_normal_writepage(struct page *page,
3120 struct writeback_control *wbc)
3122 struct inode *inode = page->mapping->host;
3124 if (test_opt(inode->i_sb, NOBH))
3125 return nobh_writepage(page,
3126 ext4_normal_get_block_write, wbc);
3128 return block_write_full_page(page,
3129 ext4_normal_get_block_write,
3133 static int ext4_normal_writepage(struct page *page,
3134 struct writeback_control *wbc)
3136 struct inode *inode = page->mapping->host;
3137 loff_t size = i_size_read(inode);
3140 trace_mark(ext4_normal_writepage,
3141 "dev %s ino %lu page_index %lu",
3142 inode->i_sb->s_id, inode->i_ino, page->index);
3143 J_ASSERT(PageLocked(page));
3144 if (page->index == size >> PAGE_CACHE_SHIFT)
3145 len = size & ~PAGE_CACHE_MASK;
3147 len = PAGE_CACHE_SIZE;
3149 if (page_has_buffers(page)) {
3150 /* if page has buffers it should all be mapped
3151 * and allocated. If there are not buffers attached
3152 * to the page we know the page is dirty but it lost
3153 * buffers. That means that at some moment in time
3154 * after write_begin() / write_end() has been called
3155 * all buffers have been clean and thus they must have been
3156 * written at least once. So they are all mapped and we can
3157 * happily proceed with mapping them and writing the page.
3159 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3160 ext4_bh_unmapped_or_delay));
3163 if (!ext4_journal_current_handle())
3164 return __ext4_normal_writepage(page, wbc);
3166 redirty_page_for_writepage(wbc, page);
3171 static int __ext4_journalled_writepage(struct page *page,
3172 struct writeback_control *wbc)
3174 struct address_space *mapping = page->mapping;
3175 struct inode *inode = mapping->host;
3176 struct buffer_head *page_bufs;
3177 handle_t *handle = NULL;
3181 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
3182 ext4_normal_get_block_write);
3186 page_bufs = page_buffers(page);
3187 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
3189 /* As soon as we unlock the page, it can go away, but we have
3190 * references to buffers so we are safe */
3193 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
3194 if (IS_ERR(handle)) {
3195 ret = PTR_ERR(handle);
3199 ret = walk_page_buffers(handle, page_bufs, 0,
3200 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3202 err = walk_page_buffers(handle, page_bufs, 0,
3203 PAGE_CACHE_SIZE, NULL, write_end_fn);
3206 err = ext4_journal_stop(handle);
3210 walk_page_buffers(handle, page_bufs, 0,
3211 PAGE_CACHE_SIZE, NULL, bput_one);
3212 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3221 static int ext4_journalled_writepage(struct page *page,
3222 struct writeback_control *wbc)
3224 struct inode *inode = page->mapping->host;
3225 loff_t size = i_size_read(inode);
3228 trace_mark(ext4_journalled_writepage,
3229 "dev %s ino %lu page_index %lu",
3230 inode->i_sb->s_id, inode->i_ino, page->index);
3231 J_ASSERT(PageLocked(page));
3232 if (page->index == size >> PAGE_CACHE_SHIFT)
3233 len = size & ~PAGE_CACHE_MASK;
3235 len = PAGE_CACHE_SIZE;
3237 if (page_has_buffers(page)) {
3238 /* if page has buffers it should all be mapped
3239 * and allocated. If there are not buffers attached
3240 * to the page we know the page is dirty but it lost
3241 * buffers. That means that at some moment in time
3242 * after write_begin() / write_end() has been called
3243 * all buffers have been clean and thus they must have been
3244 * written at least once. So they are all mapped and we can
3245 * happily proceed with mapping them and writing the page.
3247 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3248 ext4_bh_unmapped_or_delay));
3251 if (ext4_journal_current_handle())
3254 if (PageChecked(page)) {
3256 * It's mmapped pagecache. Add buffers and journal it. There
3257 * doesn't seem much point in redirtying the page here.
3259 ClearPageChecked(page);
3260 return __ext4_journalled_writepage(page, wbc);
3263 * It may be a page full of checkpoint-mode buffers. We don't
3264 * really know unless we go poke around in the buffer_heads.
3265 * But block_write_full_page will do the right thing.
3267 return block_write_full_page(page,
3268 ext4_normal_get_block_write,
3272 redirty_page_for_writepage(wbc, page);
3277 static int ext4_readpage(struct file *file, struct page *page)
3279 return mpage_readpage(page, ext4_get_block);
3283 ext4_readpages(struct file *file, struct address_space *mapping,
3284 struct list_head *pages, unsigned nr_pages)
3286 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3289 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3291 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3294 * If it's a full truncate we just forget about the pending dirtying
3297 ClearPageChecked(page);
3300 jbd2_journal_invalidatepage(journal, page, offset);
3302 block_invalidatepage(page, offset);
3305 static int ext4_releasepage(struct page *page, gfp_t wait)
3307 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3309 WARN_ON(PageChecked(page));
3310 if (!page_has_buffers(page))
3313 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3315 return try_to_free_buffers(page);
3319 * If the O_DIRECT write will extend the file then add this inode to the
3320 * orphan list. So recovery will truncate it back to the original size
3321 * if the machine crashes during the write.
3323 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3324 * crashes then stale disk data _may_ be exposed inside the file. But current
3325 * VFS code falls back into buffered path in that case so we are safe.
3327 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3328 const struct iovec *iov, loff_t offset,
3329 unsigned long nr_segs)
3331 struct file *file = iocb->ki_filp;
3332 struct inode *inode = file->f_mapping->host;
3333 struct ext4_inode_info *ei = EXT4_I(inode);
3337 size_t count = iov_length(iov, nr_segs);
3340 loff_t final_size = offset + count;
3342 if (final_size > inode->i_size) {
3343 /* Credits for sb + inode write */
3344 handle = ext4_journal_start(inode, 2);
3345 if (IS_ERR(handle)) {
3346 ret = PTR_ERR(handle);
3349 ret = ext4_orphan_add(handle, inode);
3351 ext4_journal_stop(handle);
3355 ei->i_disksize = inode->i_size;
3356 ext4_journal_stop(handle);
3360 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3362 ext4_get_block, NULL);
3367 /* Credits for sb + inode write */
3368 handle = ext4_journal_start(inode, 2);
3369 if (IS_ERR(handle)) {
3370 /* This is really bad luck. We've written the data
3371 * but cannot extend i_size. Bail out and pretend
3372 * the write failed... */
3373 ret = PTR_ERR(handle);
3377 ext4_orphan_del(handle, inode);
3379 loff_t end = offset + ret;
3380 if (end > inode->i_size) {
3381 ei->i_disksize = end;
3382 i_size_write(inode, end);
3384 * We're going to return a positive `ret'
3385 * here due to non-zero-length I/O, so there's
3386 * no way of reporting error returns from
3387 * ext4_mark_inode_dirty() to userspace. So
3390 ext4_mark_inode_dirty(handle, inode);
3393 err = ext4_journal_stop(handle);
3402 * Pages can be marked dirty completely asynchronously from ext4's journalling
3403 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3404 * much here because ->set_page_dirty is called under VFS locks. The page is
3405 * not necessarily locked.
3407 * We cannot just dirty the page and leave attached buffers clean, because the
3408 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3409 * or jbddirty because all the journalling code will explode.
3411 * So what we do is to mark the page "pending dirty" and next time writepage
3412 * is called, propagate that into the buffers appropriately.
3414 static int ext4_journalled_set_page_dirty(struct page *page)
3416 SetPageChecked(page);
3417 return __set_page_dirty_nobuffers(page);
3420 static const struct address_space_operations ext4_ordered_aops = {
3421 .readpage = ext4_readpage,
3422 .readpages = ext4_readpages,
3423 .writepage = ext4_normal_writepage,
3424 .sync_page = block_sync_page,
3425 .write_begin = ext4_write_begin,
3426 .write_end = ext4_ordered_write_end,
3428 .invalidatepage = ext4_invalidatepage,
3429 .releasepage = ext4_releasepage,
3430 .direct_IO = ext4_direct_IO,
3431 .migratepage = buffer_migrate_page,
3432 .is_partially_uptodate = block_is_partially_uptodate,
3435 static const struct address_space_operations ext4_writeback_aops = {
3436 .readpage = ext4_readpage,
3437 .readpages = ext4_readpages,
3438 .writepage = ext4_normal_writepage,
3439 .sync_page = block_sync_page,
3440 .write_begin = ext4_write_begin,
3441 .write_end = ext4_writeback_write_end,
3443 .invalidatepage = ext4_invalidatepage,
3444 .releasepage = ext4_releasepage,
3445 .direct_IO = ext4_direct_IO,
3446 .migratepage = buffer_migrate_page,
3447 .is_partially_uptodate = block_is_partially_uptodate,
3450 static const struct address_space_operations ext4_journalled_aops = {
3451 .readpage = ext4_readpage,
3452 .readpages = ext4_readpages,
3453 .writepage = ext4_journalled_writepage,
3454 .sync_page = block_sync_page,
3455 .write_begin = ext4_write_begin,
3456 .write_end = ext4_journalled_write_end,
3457 .set_page_dirty = ext4_journalled_set_page_dirty,
3459 .invalidatepage = ext4_invalidatepage,
3460 .releasepage = ext4_releasepage,
3461 .is_partially_uptodate = block_is_partially_uptodate,
3464 static const struct address_space_operations ext4_da_aops = {
3465 .readpage = ext4_readpage,
3466 .readpages = ext4_readpages,
3467 .writepage = ext4_da_writepage,
3468 .writepages = ext4_da_writepages,
3469 .sync_page = block_sync_page,
3470 .write_begin = ext4_da_write_begin,
3471 .write_end = ext4_da_write_end,
3473 .invalidatepage = ext4_da_invalidatepage,
3474 .releasepage = ext4_releasepage,
3475 .direct_IO = ext4_direct_IO,
3476 .migratepage = buffer_migrate_page,
3477 .is_partially_uptodate = block_is_partially_uptodate,
3480 void ext4_set_aops(struct inode *inode)
3482 if (ext4_should_order_data(inode) &&
3483 test_opt(inode->i_sb, DELALLOC))
3484 inode->i_mapping->a_ops = &ext4_da_aops;
3485 else if (ext4_should_order_data(inode))
3486 inode->i_mapping->a_ops = &ext4_ordered_aops;
3487 else if (ext4_should_writeback_data(inode) &&
3488 test_opt(inode->i_sb, DELALLOC))
3489 inode->i_mapping->a_ops = &ext4_da_aops;
3490 else if (ext4_should_writeback_data(inode))
3491 inode->i_mapping->a_ops = &ext4_writeback_aops;
3493 inode->i_mapping->a_ops = &ext4_journalled_aops;
3497 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3498 * up to the end of the block which corresponds to `from'.
3499 * This required during truncate. We need to physically zero the tail end
3500 * of that block so it doesn't yield old data if the file is later grown.
3502 int ext4_block_truncate_page(handle_t *handle,
3503 struct address_space *mapping, loff_t from)
3505 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3506 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3507 unsigned blocksize, length, pos;
3509 struct inode *inode = mapping->host;
3510 struct buffer_head *bh;
3514 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3518 blocksize = inode->i_sb->s_blocksize;
3519 length = blocksize - (offset & (blocksize - 1));
3520 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3523 * For "nobh" option, we can only work if we don't need to
3524 * read-in the page - otherwise we create buffers to do the IO.
3526 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3527 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3528 zero_user(page, offset, length);
3529 set_page_dirty(page);
3533 if (!page_has_buffers(page))
3534 create_empty_buffers(page, blocksize, 0);
3536 /* Find the buffer that contains "offset" */
3537 bh = page_buffers(page);
3539 while (offset >= pos) {
3540 bh = bh->b_this_page;
3546 if (buffer_freed(bh)) {
3547 BUFFER_TRACE(bh, "freed: skip");
3551 if (!buffer_mapped(bh)) {
3552 BUFFER_TRACE(bh, "unmapped");
3553 ext4_get_block(inode, iblock, bh, 0);
3554 /* unmapped? It's a hole - nothing to do */
3555 if (!buffer_mapped(bh)) {
3556 BUFFER_TRACE(bh, "still unmapped");
3561 /* Ok, it's mapped. Make sure it's up-to-date */
3562 if (PageUptodate(page))
3563 set_buffer_uptodate(bh);
3565 if (!buffer_uptodate(bh)) {
3567 ll_rw_block(READ, 1, &bh);
3569 /* Uhhuh. Read error. Complain and punt. */
3570 if (!buffer_uptodate(bh))
3574 if (ext4_should_journal_data(inode)) {
3575 BUFFER_TRACE(bh, "get write access");
3576 err = ext4_journal_get_write_access(handle, bh);
3581 zero_user(page, offset, length);
3583 BUFFER_TRACE(bh, "zeroed end of block");
3586 if (ext4_should_journal_data(inode)) {
3587 err = ext4_handle_dirty_metadata(handle, inode, bh);
3589 if (ext4_should_order_data(inode))
3590 err = ext4_jbd2_file_inode(handle, inode);
3591 mark_buffer_dirty(bh);
3596 page_cache_release(page);
3601 * Probably it should be a library function... search for first non-zero word
3602 * or memcmp with zero_page, whatever is better for particular architecture.
3605 static inline int all_zeroes(__le32 *p, __le32 *q)
3614 * ext4_find_shared - find the indirect blocks for partial truncation.
3615 * @inode: inode in question
3616 * @depth: depth of the affected branch
3617 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3618 * @chain: place to store the pointers to partial indirect blocks
3619 * @top: place to the (detached) top of branch
3621 * This is a helper function used by ext4_truncate().
3623 * When we do truncate() we may have to clean the ends of several
3624 * indirect blocks but leave the blocks themselves alive. Block is
3625 * partially truncated if some data below the new i_size is refered
3626 * from it (and it is on the path to the first completely truncated
3627 * data block, indeed). We have to free the top of that path along
3628 * with everything to the right of the path. Since no allocation
3629 * past the truncation point is possible until ext4_truncate()
3630 * finishes, we may safely do the latter, but top of branch may
3631 * require special attention - pageout below the truncation point
3632 * might try to populate it.
3634 * We atomically detach the top of branch from the tree, store the
3635 * block number of its root in *@top, pointers to buffer_heads of
3636 * partially truncated blocks - in @chain[].bh and pointers to
3637 * their last elements that should not be removed - in
3638 * @chain[].p. Return value is the pointer to last filled element
3641 * The work left to caller to do the actual freeing of subtrees:
3642 * a) free the subtree starting from *@top
3643 * b) free the subtrees whose roots are stored in
3644 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3645 * c) free the subtrees growing from the inode past the @chain[0].
3646 * (no partially truncated stuff there). */
3648 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3649 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3651 Indirect *partial, *p;
3655 /* Make k index the deepest non-null offest + 1 */
3656 for (k = depth; k > 1 && !offsets[k-1]; k--)
3658 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3659 /* Writer: pointers */
3661 partial = chain + k-1;
3663 * If the branch acquired continuation since we've looked at it -
3664 * fine, it should all survive and (new) top doesn't belong to us.
3666 if (!partial->key && *partial->p)
3669 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3672 * OK, we've found the last block that must survive. The rest of our
3673 * branch should be detached before unlocking. However, if that rest
3674 * of branch is all ours and does not grow immediately from the inode
3675 * it's easier to cheat and just decrement partial->p.
3677 if (p == chain + k - 1 && p > chain) {
3681 /* Nope, don't do this in ext4. Must leave the tree intact */
3688 while (partial > p) {
3689 brelse(partial->bh);
3697 * Zero a number of block pointers in either an inode or an indirect block.
3698 * If we restart the transaction we must again get write access to the
3699 * indirect block for further modification.
3701 * We release `count' blocks on disk, but (last - first) may be greater
3702 * than `count' because there can be holes in there.
3704 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3705 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3706 unsigned long count, __le32 *first, __le32 *last)
3709 if (try_to_extend_transaction(handle, inode)) {
3711 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3712 ext4_handle_dirty_metadata(handle, inode, bh);
3714 ext4_mark_inode_dirty(handle, inode);
3715 ext4_journal_test_restart(handle, inode);
3717 BUFFER_TRACE(bh, "retaking write access");
3718 ext4_journal_get_write_access(handle, bh);
3723 * Any buffers which are on the journal will be in memory. We find
3724 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3725 * on them. We've already detached each block from the file, so
3726 * bforget() in jbd2_journal_forget() should be safe.
3728 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3730 for (p = first; p < last; p++) {
3731 u32 nr = le32_to_cpu(*p);
3733 struct buffer_head *tbh;
3736 tbh = sb_find_get_block(inode->i_sb, nr);
3737 ext4_forget(handle, 0, inode, tbh, nr);
3741 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3745 * ext4_free_data - free a list of data blocks
3746 * @handle: handle for this transaction
3747 * @inode: inode we are dealing with
3748 * @this_bh: indirect buffer_head which contains *@first and *@last
3749 * @first: array of block numbers
3750 * @last: points immediately past the end of array
3752 * We are freeing all blocks refered from that array (numbers are stored as
3753 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3755 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3756 * blocks are contiguous then releasing them at one time will only affect one
3757 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3758 * actually use a lot of journal space.
3760 * @this_bh will be %NULL if @first and @last point into the inode's direct
3763 static void ext4_free_data(handle_t *handle, struct inode *inode,
3764 struct buffer_head *this_bh,
3765 __le32 *first, __le32 *last)
3767 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3768 unsigned long count = 0; /* Number of blocks in the run */
3769 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3772 ext4_fsblk_t nr; /* Current block # */
3773 __le32 *p; /* Pointer into inode/ind
3774 for current block */
3777 if (this_bh) { /* For indirect block */
3778 BUFFER_TRACE(this_bh, "get_write_access");
3779 err = ext4_journal_get_write_access(handle, this_bh);
3780 /* Important: if we can't update the indirect pointers
3781 * to the blocks, we can't free them. */
3786 for (p = first; p < last; p++) {
3787 nr = le32_to_cpu(*p);
3789 /* accumulate blocks to free if they're contiguous */
3792 block_to_free_p = p;
3794 } else if (nr == block_to_free + count) {
3797 ext4_clear_blocks(handle, inode, this_bh,
3799 count, block_to_free_p, p);
3801 block_to_free_p = p;
3808 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3809 count, block_to_free_p, p);
3812 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3815 * The buffer head should have an attached journal head at this
3816 * point. However, if the data is corrupted and an indirect
3817 * block pointed to itself, it would have been detached when
3818 * the block was cleared. Check for this instead of OOPSing.
3820 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3821 ext4_handle_dirty_metadata(handle, inode, this_bh);
3823 ext4_error(inode->i_sb, __func__,
3824 "circular indirect block detected, "
3825 "inode=%lu, block=%llu",
3827 (unsigned long long) this_bh->b_blocknr);
3832 * ext4_free_branches - free an array of branches
3833 * @handle: JBD handle for this transaction
3834 * @inode: inode we are dealing with
3835 * @parent_bh: the buffer_head which contains *@first and *@last
3836 * @first: array of block numbers
3837 * @last: pointer immediately past the end of array
3838 * @depth: depth of the branches to free
3840 * We are freeing all blocks refered from these branches (numbers are
3841 * stored as little-endian 32-bit) and updating @inode->i_blocks
3844 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3845 struct buffer_head *parent_bh,
3846 __le32 *first, __le32 *last, int depth)
3851 if (ext4_handle_is_aborted(handle))
3855 struct buffer_head *bh;
3856 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3858 while (--p >= first) {
3859 nr = le32_to_cpu(*p);
3861 continue; /* A hole */
3863 /* Go read the buffer for the next level down */
3864 bh = sb_bread(inode->i_sb, nr);
3867 * A read failure? Report error and clear slot
3871 ext4_error(inode->i_sb, "ext4_free_branches",
3872 "Read failure, inode=%lu, block=%llu",
3877 /* This zaps the entire block. Bottom up. */
3878 BUFFER_TRACE(bh, "free child branches");
3879 ext4_free_branches(handle, inode, bh,
3880 (__le32 *) bh->b_data,
3881 (__le32 *) bh->b_data + addr_per_block,
3885 * We've probably journalled the indirect block several
3886 * times during the truncate. But it's no longer
3887 * needed and we now drop it from the transaction via
3888 * jbd2_journal_revoke().
3890 * That's easy if it's exclusively part of this
3891 * transaction. But if it's part of the committing
3892 * transaction then jbd2_journal_forget() will simply
3893 * brelse() it. That means that if the underlying
3894 * block is reallocated in ext4_get_block(),
3895 * unmap_underlying_metadata() will find this block
3896 * and will try to get rid of it. damn, damn.
3898 * If this block has already been committed to the
3899 * journal, a revoke record will be written. And
3900 * revoke records must be emitted *before* clearing
3901 * this block's bit in the bitmaps.
3903 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3906 * Everything below this this pointer has been
3907 * released. Now let this top-of-subtree go.
3909 * We want the freeing of this indirect block to be
3910 * atomic in the journal with the updating of the
3911 * bitmap block which owns it. So make some room in
3914 * We zero the parent pointer *after* freeing its
3915 * pointee in the bitmaps, so if extend_transaction()
3916 * for some reason fails to put the bitmap changes and
3917 * the release into the same transaction, recovery
3918 * will merely complain about releasing a free block,
3919 * rather than leaking blocks.
3921 if (ext4_handle_is_aborted(handle))
3923 if (try_to_extend_transaction(handle, inode)) {
3924 ext4_mark_inode_dirty(handle, inode);
3925 ext4_journal_test_restart(handle, inode);
3928 ext4_free_blocks(handle, inode, nr, 1, 1);
3932 * The block which we have just freed is
3933 * pointed to by an indirect block: journal it
3935 BUFFER_TRACE(parent_bh, "get_write_access");
3936 if (!ext4_journal_get_write_access(handle,
3939 BUFFER_TRACE(parent_bh,
3940 "call ext4_handle_dirty_metadata");
3941 ext4_handle_dirty_metadata(handle,
3948 /* We have reached the bottom of the tree. */
3949 BUFFER_TRACE(parent_bh, "free data blocks");
3950 ext4_free_data(handle, inode, parent_bh, first, last);
3954 int ext4_can_truncate(struct inode *inode)
3956 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3958 if (S_ISREG(inode->i_mode))
3960 if (S_ISDIR(inode->i_mode))
3962 if (S_ISLNK(inode->i_mode))
3963 return !ext4_inode_is_fast_symlink(inode);
3970 * We block out ext4_get_block() block instantiations across the entire
3971 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3972 * simultaneously on behalf of the same inode.
3974 * As we work through the truncate and commmit bits of it to the journal there
3975 * is one core, guiding principle: the file's tree must always be consistent on
3976 * disk. We must be able to restart the truncate after a crash.
3978 * The file's tree may be transiently inconsistent in memory (although it
3979 * probably isn't), but whenever we close off and commit a journal transaction,
3980 * the contents of (the filesystem + the journal) must be consistent and
3981 * restartable. It's pretty simple, really: bottom up, right to left (although
3982 * left-to-right works OK too).
3984 * Note that at recovery time, journal replay occurs *before* the restart of
3985 * truncate against the orphan inode list.
3987 * The committed inode has the new, desired i_size (which is the same as
3988 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3989 * that this inode's truncate did not complete and it will again call
3990 * ext4_truncate() to have another go. So there will be instantiated blocks
3991 * to the right of the truncation point in a crashed ext4 filesystem. But
3992 * that's fine - as long as they are linked from the inode, the post-crash
3993 * ext4_truncate() run will find them and release them.
3995 void ext4_truncate(struct inode *inode)
3998 struct ext4_inode_info *ei = EXT4_I(inode);
3999 __le32 *i_data = ei->i_data;
4000 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4001 struct address_space *mapping = inode->i_mapping;
4002 ext4_lblk_t offsets[4];
4007 ext4_lblk_t last_block;
4008 unsigned blocksize = inode->i_sb->s_blocksize;
4010 if (!ext4_can_truncate(inode))
4013 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4014 ei->i_state |= EXT4_STATE_DA_ALLOC_CLOSE;
4016 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
4017 ext4_ext_truncate(inode);
4021 handle = start_transaction(inode);
4023 return; /* AKPM: return what? */
4025 last_block = (inode->i_size + blocksize-1)
4026 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4028 if (inode->i_size & (blocksize - 1))
4029 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4032 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4034 goto out_stop; /* error */
4037 * OK. This truncate is going to happen. We add the inode to the
4038 * orphan list, so that if this truncate spans multiple transactions,
4039 * and we crash, we will resume the truncate when the filesystem
4040 * recovers. It also marks the inode dirty, to catch the new size.
4042 * Implication: the file must always be in a sane, consistent
4043 * truncatable state while each transaction commits.
4045 if (ext4_orphan_add(handle, inode))
4049 * From here we block out all ext4_get_block() callers who want to
4050 * modify the block allocation tree.
4052 down_write(&ei->i_data_sem);
4054 ext4_discard_preallocations(inode);
4057 * The orphan list entry will now protect us from any crash which
4058 * occurs before the truncate completes, so it is now safe to propagate
4059 * the new, shorter inode size (held for now in i_size) into the
4060 * on-disk inode. We do this via i_disksize, which is the value which
4061 * ext4 *really* writes onto the disk inode.
4063 ei->i_disksize = inode->i_size;
4065 if (n == 1) { /* direct blocks */
4066 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4067 i_data + EXT4_NDIR_BLOCKS);
4071 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4072 /* Kill the top of shared branch (not detached) */
4074 if (partial == chain) {
4075 /* Shared branch grows from the inode */
4076 ext4_free_branches(handle, inode, NULL,
4077 &nr, &nr+1, (chain+n-1) - partial);
4080 * We mark the inode dirty prior to restart,
4081 * and prior to stop. No need for it here.
4084 /* Shared branch grows from an indirect block */
4085 BUFFER_TRACE(partial->bh, "get_write_access");
4086 ext4_free_branches(handle, inode, partial->bh,
4088 partial->p+1, (chain+n-1) - partial);
4091 /* Clear the ends of indirect blocks on the shared branch */
4092 while (partial > chain) {
4093 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4094 (__le32*)partial->bh->b_data+addr_per_block,
4095 (chain+n-1) - partial);
4096 BUFFER_TRACE(partial->bh, "call brelse");
4097 brelse (partial->bh);
4101 /* Kill the remaining (whole) subtrees */
4102 switch (offsets[0]) {
4104 nr = i_data[EXT4_IND_BLOCK];
4106 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4107 i_data[EXT4_IND_BLOCK] = 0;
4109 case EXT4_IND_BLOCK:
4110 nr = i_data[EXT4_DIND_BLOCK];
4112 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4113 i_data[EXT4_DIND_BLOCK] = 0;
4115 case EXT4_DIND_BLOCK:
4116 nr = i_data[EXT4_TIND_BLOCK];
4118 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4119 i_data[EXT4_TIND_BLOCK] = 0;
4121 case EXT4_TIND_BLOCK:
4125 up_write(&ei->i_data_sem);
4126 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4127 ext4_mark_inode_dirty(handle, inode);
4130 * In a multi-transaction truncate, we only make the final transaction
4134 ext4_handle_sync(handle);
4137 * If this was a simple ftruncate(), and the file will remain alive
4138 * then we need to clear up the orphan record which we created above.
4139 * However, if this was a real unlink then we were called by
4140 * ext4_delete_inode(), and we allow that function to clean up the
4141 * orphan info for us.
4144 ext4_orphan_del(handle, inode);
4146 ext4_journal_stop(handle);
4150 * ext4_get_inode_loc returns with an extra refcount against the inode's
4151 * underlying buffer_head on success. If 'in_mem' is true, we have all
4152 * data in memory that is needed to recreate the on-disk version of this
4155 static int __ext4_get_inode_loc(struct inode *inode,
4156 struct ext4_iloc *iloc, int in_mem)
4158 struct ext4_group_desc *gdp;
4159 struct buffer_head *bh;
4160 struct super_block *sb = inode->i_sb;
4162 int inodes_per_block, inode_offset;
4165 if (!ext4_valid_inum(sb, inode->i_ino))
4168 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4169 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4174 * Figure out the offset within the block group inode table
4176 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4177 inode_offset = ((inode->i_ino - 1) %
4178 EXT4_INODES_PER_GROUP(sb));
4179 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4180 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4182 bh = sb_getblk(sb, block);
4184 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
4185 "inode block - inode=%lu, block=%llu",
4186 inode->i_ino, block);
4189 if (!buffer_uptodate(bh)) {
4193 * If the buffer has the write error flag, we have failed
4194 * to write out another inode in the same block. In this
4195 * case, we don't have to read the block because we may
4196 * read the old inode data successfully.
4198 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4199 set_buffer_uptodate(bh);
4201 if (buffer_uptodate(bh)) {
4202 /* someone brought it uptodate while we waited */
4208 * If we have all information of the inode in memory and this
4209 * is the only valid inode in the block, we need not read the
4213 struct buffer_head *bitmap_bh;
4216 start = inode_offset & ~(inodes_per_block - 1);
4218 /* Is the inode bitmap in cache? */
4219 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4224 * If the inode bitmap isn't in cache then the
4225 * optimisation may end up performing two reads instead
4226 * of one, so skip it.
4228 if (!buffer_uptodate(bitmap_bh)) {
4232 for (i = start; i < start + inodes_per_block; i++) {
4233 if (i == inode_offset)
4235 if (ext4_test_bit(i, bitmap_bh->b_data))
4239 if (i == start + inodes_per_block) {
4240 /* all other inodes are free, so skip I/O */
4241 memset(bh->b_data, 0, bh->b_size);
4242 set_buffer_uptodate(bh);
4250 * If we need to do any I/O, try to pre-readahead extra
4251 * blocks from the inode table.
4253 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4254 ext4_fsblk_t b, end, table;
4257 table = ext4_inode_table(sb, gdp);
4258 /* s_inode_readahead_blks is always a power of 2 */
4259 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4262 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4263 num = EXT4_INODES_PER_GROUP(sb);
4264 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4265 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4266 num -= ext4_itable_unused_count(sb, gdp);
4267 table += num / inodes_per_block;
4271 sb_breadahead(sb, b++);
4275 * There are other valid inodes in the buffer, this inode
4276 * has in-inode xattrs, or we don't have this inode in memory.
4277 * Read the block from disk.
4280 bh->b_end_io = end_buffer_read_sync;
4281 submit_bh(READ_META, bh);
4283 if (!buffer_uptodate(bh)) {
4284 ext4_error(sb, __func__,
4285 "unable to read inode block - inode=%lu, "
4286 "block=%llu", inode->i_ino, block);
4296 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4298 /* We have all inode data except xattrs in memory here. */
4299 return __ext4_get_inode_loc(inode, iloc,
4300 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4303 void ext4_set_inode_flags(struct inode *inode)
4305 unsigned int flags = EXT4_I(inode)->i_flags;
4307 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4308 if (flags & EXT4_SYNC_FL)
4309 inode->i_flags |= S_SYNC;
4310 if (flags & EXT4_APPEND_FL)
4311 inode->i_flags |= S_APPEND;
4312 if (flags & EXT4_IMMUTABLE_FL)
4313 inode->i_flags |= S_IMMUTABLE;
4314 if (flags & EXT4_NOATIME_FL)
4315 inode->i_flags |= S_NOATIME;
4316 if (flags & EXT4_DIRSYNC_FL)
4317 inode->i_flags |= S_DIRSYNC;
4320 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4321 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4323 unsigned int flags = ei->vfs_inode.i_flags;
4325 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4326 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4328 ei->i_flags |= EXT4_SYNC_FL;
4329 if (flags & S_APPEND)
4330 ei->i_flags |= EXT4_APPEND_FL;
4331 if (flags & S_IMMUTABLE)
4332 ei->i_flags |= EXT4_IMMUTABLE_FL;
4333 if (flags & S_NOATIME)
4334 ei->i_flags |= EXT4_NOATIME_FL;
4335 if (flags & S_DIRSYNC)
4336 ei->i_flags |= EXT4_DIRSYNC_FL;
4338 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4339 struct ext4_inode_info *ei)
4342 struct inode *inode = &(ei->vfs_inode);
4343 struct super_block *sb = inode->i_sb;
4345 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4346 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4347 /* we are using combined 48 bit field */
4348 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4349 le32_to_cpu(raw_inode->i_blocks_lo);
4350 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4351 /* i_blocks represent file system block size */
4352 return i_blocks << (inode->i_blkbits - 9);
4357 return le32_to_cpu(raw_inode->i_blocks_lo);
4361 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4363 struct ext4_iloc iloc;
4364 struct ext4_inode *raw_inode;
4365 struct ext4_inode_info *ei;
4366 struct buffer_head *bh;
4367 struct inode *inode;
4371 inode = iget_locked(sb, ino);
4373 return ERR_PTR(-ENOMEM);
4374 if (!(inode->i_state & I_NEW))
4378 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4379 ei->i_acl = EXT4_ACL_NOT_CACHED;
4380 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4383 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4387 raw_inode = ext4_raw_inode(&iloc);
4388 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4389 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4390 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4391 if (!(test_opt(inode->i_sb, NO_UID32))) {
4392 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4393 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4395 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4398 ei->i_dir_start_lookup = 0;
4399 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4400 /* We now have enough fields to check if the inode was active or not.
4401 * This is needed because nfsd might try to access dead inodes
4402 * the test is that same one that e2fsck uses
4403 * NeilBrown 1999oct15
4405 if (inode->i_nlink == 0) {
4406 if (inode->i_mode == 0 ||
4407 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4408 /* this inode is deleted */
4413 /* The only unlinked inodes we let through here have
4414 * valid i_mode and are being read by the orphan
4415 * recovery code: that's fine, we're about to complete
4416 * the process of deleting those. */
4418 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4419 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4420 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4421 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4423 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4424 inode->i_size = ext4_isize(raw_inode);
4425 ei->i_disksize = inode->i_size;
4426 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4427 ei->i_block_group = iloc.block_group;
4428 ei->i_last_alloc_group = ~0;
4430 * NOTE! The in-memory inode i_data array is in little-endian order
4431 * even on big-endian machines: we do NOT byteswap the block numbers!
4433 for (block = 0; block < EXT4_N_BLOCKS; block++)
4434 ei->i_data[block] = raw_inode->i_block[block];
4435 INIT_LIST_HEAD(&ei->i_orphan);
4437 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4438 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4439 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4440 EXT4_INODE_SIZE(inode->i_sb)) {
4445 if (ei->i_extra_isize == 0) {
4446 /* The extra space is currently unused. Use it. */
4447 ei->i_extra_isize = sizeof(struct ext4_inode) -
4448 EXT4_GOOD_OLD_INODE_SIZE;
4450 __le32 *magic = (void *)raw_inode +
4451 EXT4_GOOD_OLD_INODE_SIZE +
4453 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4454 ei->i_state |= EXT4_STATE_XATTR;
4457 ei->i_extra_isize = 0;
4459 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4460 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4461 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4462 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4464 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4465 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4466 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4468 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4472 if (ei->i_file_acl &&
4474 (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
4475 EXT4_SB(sb)->s_gdb_count)) ||
4476 (ei->i_file_acl >= ext4_blocks_count(EXT4_SB(sb)->s_es)))) {
4477 ext4_error(sb, __func__,
4478 "bad extended attribute block %llu in inode #%lu",
4479 ei->i_file_acl, inode->i_ino);
4482 } else if (ei->i_flags & EXT4_EXTENTS_FL) {
4483 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4484 (S_ISLNK(inode->i_mode) &&
4485 !ext4_inode_is_fast_symlink(inode)))
4486 /* Validate extent which is part of inode */
4487 ret = ext4_ext_check_inode(inode);
4488 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4489 (S_ISLNK(inode->i_mode) &&
4490 !ext4_inode_is_fast_symlink(inode))) {
4491 /* Validate block references which are part of inode */
4492 ret = ext4_check_inode_blockref(inode);
4499 if (S_ISREG(inode->i_mode)) {
4500 inode->i_op = &ext4_file_inode_operations;
4501 inode->i_fop = &ext4_file_operations;
4502 ext4_set_aops(inode);
4503 } else if (S_ISDIR(inode->i_mode)) {
4504 inode->i_op = &ext4_dir_inode_operations;
4505 inode->i_fop = &ext4_dir_operations;
4506 } else if (S_ISLNK(inode->i_mode)) {
4507 if (ext4_inode_is_fast_symlink(inode)) {
4508 inode->i_op = &ext4_fast_symlink_inode_operations;
4509 nd_terminate_link(ei->i_data, inode->i_size,
4510 sizeof(ei->i_data) - 1);
4512 inode->i_op = &ext4_symlink_inode_operations;
4513 ext4_set_aops(inode);
4515 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4516 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4517 inode->i_op = &ext4_special_inode_operations;
4518 if (raw_inode->i_block[0])
4519 init_special_inode(inode, inode->i_mode,
4520 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4522 init_special_inode(inode, inode->i_mode,
4523 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4527 ext4_error(inode->i_sb, __func__,
4528 "bogus i_mode (%o) for inode=%lu",
4529 inode->i_mode, inode->i_ino);
4533 ext4_set_inode_flags(inode);
4534 unlock_new_inode(inode);
4539 return ERR_PTR(ret);
4542 static int ext4_inode_blocks_set(handle_t *handle,
4543 struct ext4_inode *raw_inode,
4544 struct ext4_inode_info *ei)
4546 struct inode *inode = &(ei->vfs_inode);
4547 u64 i_blocks = inode->i_blocks;
4548 struct super_block *sb = inode->i_sb;
4550 if (i_blocks <= ~0U) {
4552 * i_blocks can be represnted in a 32 bit variable
4553 * as multiple of 512 bytes
4555 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4556 raw_inode->i_blocks_high = 0;
4557 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4560 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4563 if (i_blocks <= 0xffffffffffffULL) {
4565 * i_blocks can be represented in a 48 bit variable
4566 * as multiple of 512 bytes
4568 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4569 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4570 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4572 ei->i_flags |= EXT4_HUGE_FILE_FL;
4573 /* i_block is stored in file system block size */
4574 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4575 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4576 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4582 * Post the struct inode info into an on-disk inode location in the
4583 * buffer-cache. This gobbles the caller's reference to the
4584 * buffer_head in the inode location struct.
4586 * The caller must have write access to iloc->bh.
4588 static int ext4_do_update_inode(handle_t *handle,
4589 struct inode *inode,
4590 struct ext4_iloc *iloc)
4592 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4593 struct ext4_inode_info *ei = EXT4_I(inode);
4594 struct buffer_head *bh = iloc->bh;
4595 int err = 0, rc, block;
4597 /* For fields not not tracking in the in-memory inode,
4598 * initialise them to zero for new inodes. */
4599 if (ei->i_state & EXT4_STATE_NEW)
4600 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4602 ext4_get_inode_flags(ei);
4603 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4604 if (!(test_opt(inode->i_sb, NO_UID32))) {
4605 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4606 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4608 * Fix up interoperability with old kernels. Otherwise, old inodes get
4609 * re-used with the upper 16 bits of the uid/gid intact
4612 raw_inode->i_uid_high =
4613 cpu_to_le16(high_16_bits(inode->i_uid));
4614 raw_inode->i_gid_high =
4615 cpu_to_le16(high_16_bits(inode->i_gid));
4617 raw_inode->i_uid_high = 0;
4618 raw_inode->i_gid_high = 0;
4621 raw_inode->i_uid_low =
4622 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4623 raw_inode->i_gid_low =
4624 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4625 raw_inode->i_uid_high = 0;
4626 raw_inode->i_gid_high = 0;
4628 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4630 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4631 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4632 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4633 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4635 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4637 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4638 /* clear the migrate flag in the raw_inode */
4639 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4640 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4641 cpu_to_le32(EXT4_OS_HURD))
4642 raw_inode->i_file_acl_high =
4643 cpu_to_le16(ei->i_file_acl >> 32);
4644 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4645 ext4_isize_set(raw_inode, ei->i_disksize);
4646 if (ei->i_disksize > 0x7fffffffULL) {
4647 struct super_block *sb = inode->i_sb;
4648 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4649 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4650 EXT4_SB(sb)->s_es->s_rev_level ==
4651 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4652 /* If this is the first large file
4653 * created, add a flag to the superblock.
4655 err = ext4_journal_get_write_access(handle,
4656 EXT4_SB(sb)->s_sbh);
4659 ext4_update_dynamic_rev(sb);
4660 EXT4_SET_RO_COMPAT_FEATURE(sb,
4661 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4663 ext4_handle_sync(handle);
4664 err = ext4_handle_dirty_metadata(handle, inode,
4665 EXT4_SB(sb)->s_sbh);
4668 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4669 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4670 if (old_valid_dev(inode->i_rdev)) {
4671 raw_inode->i_block[0] =
4672 cpu_to_le32(old_encode_dev(inode->i_rdev));
4673 raw_inode->i_block[1] = 0;
4675 raw_inode->i_block[0] = 0;
4676 raw_inode->i_block[1] =
4677 cpu_to_le32(new_encode_dev(inode->i_rdev));
4678 raw_inode->i_block[2] = 0;
4680 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4681 raw_inode->i_block[block] = ei->i_data[block];
4683 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4684 if (ei->i_extra_isize) {
4685 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4686 raw_inode->i_version_hi =
4687 cpu_to_le32(inode->i_version >> 32);
4688 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4691 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4692 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4695 ei->i_state &= ~EXT4_STATE_NEW;
4699 ext4_std_error(inode->i_sb, err);
4704 * ext4_write_inode()
4706 * We are called from a few places:
4708 * - Within generic_file_write() for O_SYNC files.
4709 * Here, there will be no transaction running. We wait for any running
4710 * trasnaction to commit.
4712 * - Within sys_sync(), kupdate and such.
4713 * We wait on commit, if tol to.
4715 * - Within prune_icache() (PF_MEMALLOC == true)
4716 * Here we simply return. We can't afford to block kswapd on the
4719 * In all cases it is actually safe for us to return without doing anything,
4720 * because the inode has been copied into a raw inode buffer in
4721 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4724 * Note that we are absolutely dependent upon all inode dirtiers doing the
4725 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4726 * which we are interested.
4728 * It would be a bug for them to not do this. The code:
4730 * mark_inode_dirty(inode)
4732 * inode->i_size = expr;
4734 * is in error because a kswapd-driven write_inode() could occur while
4735 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4736 * will no longer be on the superblock's dirty inode list.
4738 int ext4_write_inode(struct inode *inode, int wait)
4740 if (current->flags & PF_MEMALLOC)
4743 if (ext4_journal_current_handle()) {
4744 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4752 return ext4_force_commit(inode->i_sb);
4755 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4759 mark_buffer_dirty(bh);
4760 if (inode && inode_needs_sync(inode)) {
4761 sync_dirty_buffer(bh);
4762 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4763 ext4_error(inode->i_sb, __func__,
4764 "IO error syncing inode, "
4765 "inode=%lu, block=%llu",
4767 (unsigned long long)bh->b_blocknr);
4777 * Called from notify_change.
4779 * We want to trap VFS attempts to truncate the file as soon as
4780 * possible. In particular, we want to make sure that when the VFS
4781 * shrinks i_size, we put the inode on the orphan list and modify
4782 * i_disksize immediately, so that during the subsequent flushing of
4783 * dirty pages and freeing of disk blocks, we can guarantee that any
4784 * commit will leave the blocks being flushed in an unused state on
4785 * disk. (On recovery, the inode will get truncated and the blocks will
4786 * be freed, so we have a strong guarantee that no future commit will
4787 * leave these blocks visible to the user.)
4789 * Another thing we have to assure is that if we are in ordered mode
4790 * and inode is still attached to the committing transaction, we must
4791 * we start writeout of all the dirty pages which are being truncated.
4792 * This way we are sure that all the data written in the previous
4793 * transaction are already on disk (truncate waits for pages under
4796 * Called with inode->i_mutex down.
4798 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4800 struct inode *inode = dentry->d_inode;
4802 const unsigned int ia_valid = attr->ia_valid;
4804 error = inode_change_ok(inode, attr);
4808 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4809 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4812 /* (user+group)*(old+new) structure, inode write (sb,
4813 * inode block, ? - but truncate inode update has it) */
4814 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4815 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4816 if (IS_ERR(handle)) {
4817 error = PTR_ERR(handle);
4820 error = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
4822 ext4_journal_stop(handle);
4825 /* Update corresponding info in inode so that everything is in
4826 * one transaction */
4827 if (attr->ia_valid & ATTR_UID)
4828 inode->i_uid = attr->ia_uid;
4829 if (attr->ia_valid & ATTR_GID)
4830 inode->i_gid = attr->ia_gid;
4831 error = ext4_mark_inode_dirty(handle, inode);
4832 ext4_journal_stop(handle);
4835 if (attr->ia_valid & ATTR_SIZE) {
4836 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4837 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4839 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4846 if (S_ISREG(inode->i_mode) &&
4847 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4850 handle = ext4_journal_start(inode, 3);
4851 if (IS_ERR(handle)) {
4852 error = PTR_ERR(handle);
4856 error = ext4_orphan_add(handle, inode);
4857 EXT4_I(inode)->i_disksize = attr->ia_size;
4858 rc = ext4_mark_inode_dirty(handle, inode);
4861 ext4_journal_stop(handle);
4863 if (ext4_should_order_data(inode)) {
4864 error = ext4_begin_ordered_truncate(inode,
4867 /* Do as much error cleanup as possible */
4868 handle = ext4_journal_start(inode, 3);
4869 if (IS_ERR(handle)) {
4870 ext4_orphan_del(NULL, inode);
4873 ext4_orphan_del(handle, inode);
4874 ext4_journal_stop(handle);
4880 rc = inode_setattr(inode, attr);
4882 /* If inode_setattr's call to ext4_truncate failed to get a
4883 * transaction handle at all, we need to clean up the in-core
4884 * orphan list manually. */
4886 ext4_orphan_del(NULL, inode);
4888 if (!rc && (ia_valid & ATTR_MODE))
4889 rc = ext4_acl_chmod(inode);
4892 ext4_std_error(inode->i_sb, error);
4898 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4901 struct inode *inode;
4902 unsigned long delalloc_blocks;
4904 inode = dentry->d_inode;
4905 generic_fillattr(inode, stat);
4908 * We can't update i_blocks if the block allocation is delayed
4909 * otherwise in the case of system crash before the real block
4910 * allocation is done, we will have i_blocks inconsistent with
4911 * on-disk file blocks.
4912 * We always keep i_blocks updated together with real
4913 * allocation. But to not confuse with user, stat
4914 * will return the blocks that include the delayed allocation
4915 * blocks for this file.
4917 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4918 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4919 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4921 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4925 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4930 /* if nrblocks are contiguous */
4933 * With N contiguous data blocks, it need at most
4934 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4935 * 2 dindirect blocks
4938 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4939 return indirects + 3;
4942 * if nrblocks are not contiguous, worse case, each block touch
4943 * a indirect block, and each indirect block touch a double indirect
4944 * block, plus a triple indirect block
4946 indirects = nrblocks * 2 + 1;
4950 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4952 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4953 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4954 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4958 * Account for index blocks, block groups bitmaps and block group
4959 * descriptor blocks if modify datablocks and index blocks
4960 * worse case, the indexs blocks spread over different block groups
4962 * If datablocks are discontiguous, they are possible to spread over
4963 * different block groups too. If they are contiugous, with flexbg,
4964 * they could still across block group boundary.
4966 * Also account for superblock, inode, quota and xattr blocks
4968 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4970 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4976 * How many index blocks need to touch to modify nrblocks?
4977 * The "Chunk" flag indicating whether the nrblocks is
4978 * physically contiguous on disk
4980 * For Direct IO and fallocate, they calls get_block to allocate
4981 * one single extent at a time, so they could set the "Chunk" flag
4983 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4988 * Now let's see how many group bitmaps and group descriptors need
4998 if (groups > ngroups)
5000 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5001 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5003 /* bitmaps and block group descriptor blocks */
5004 ret += groups + gdpblocks;
5006 /* Blocks for super block, inode, quota and xattr blocks */
5007 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5013 * Calulate the total number of credits to reserve to fit
5014 * the modification of a single pages into a single transaction,
5015 * which may include multiple chunks of block allocations.
5017 * This could be called via ext4_write_begin()
5019 * We need to consider the worse case, when
5020 * one new block per extent.
5022 int ext4_writepage_trans_blocks(struct inode *inode)
5024 int bpp = ext4_journal_blocks_per_page(inode);
5027 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5029 /* Account for data blocks for journalled mode */
5030 if (ext4_should_journal_data(inode))
5036 * Calculate the journal credits for a chunk of data modification.
5038 * This is called from DIO, fallocate or whoever calling
5039 * ext4_get_blocks() to map/allocate a chunk of contigous disk blocks.
5041 * journal buffers for data blocks are not included here, as DIO
5042 * and fallocate do no need to journal data buffers.
5044 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5046 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5050 * The caller must have previously called ext4_reserve_inode_write().
5051 * Give this, we know that the caller already has write access to iloc->bh.
5053 int ext4_mark_iloc_dirty(handle_t *handle,
5054 struct inode *inode, struct ext4_iloc *iloc)
5058 if (test_opt(inode->i_sb, I_VERSION))
5059 inode_inc_iversion(inode);
5061 /* the do_update_inode consumes one bh->b_count */
5064 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5065 err = ext4_do_update_inode(handle, inode, iloc);
5071 * On success, We end up with an outstanding reference count against
5072 * iloc->bh. This _must_ be cleaned up later.
5076 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5077 struct ext4_iloc *iloc)
5081 err = ext4_get_inode_loc(inode, iloc);
5083 BUFFER_TRACE(iloc->bh, "get_write_access");
5084 err = ext4_journal_get_write_access(handle, iloc->bh);
5090 ext4_std_error(inode->i_sb, err);
5095 * Expand an inode by new_extra_isize bytes.
5096 * Returns 0 on success or negative error number on failure.
5098 static int ext4_expand_extra_isize(struct inode *inode,
5099 unsigned int new_extra_isize,
5100 struct ext4_iloc iloc,
5103 struct ext4_inode *raw_inode;
5104 struct ext4_xattr_ibody_header *header;
5105 struct ext4_xattr_entry *entry;
5107 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5110 raw_inode = ext4_raw_inode(&iloc);
5112 header = IHDR(inode, raw_inode);
5113 entry = IFIRST(header);
5115 /* No extended attributes present */
5116 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
5117 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5118 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5120 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5124 /* try to expand with EAs present */
5125 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5130 * What we do here is to mark the in-core inode as clean with respect to inode
5131 * dirtiness (it may still be data-dirty).
5132 * This means that the in-core inode may be reaped by prune_icache
5133 * without having to perform any I/O. This is a very good thing,
5134 * because *any* task may call prune_icache - even ones which
5135 * have a transaction open against a different journal.
5137 * Is this cheating? Not really. Sure, we haven't written the
5138 * inode out, but prune_icache isn't a user-visible syncing function.
5139 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5140 * we start and wait on commits.
5142 * Is this efficient/effective? Well, we're being nice to the system
5143 * by cleaning up our inodes proactively so they can be reaped
5144 * without I/O. But we are potentially leaving up to five seconds'
5145 * worth of inodes floating about which prune_icache wants us to
5146 * write out. One way to fix that would be to get prune_icache()
5147 * to do a write_super() to free up some memory. It has the desired
5150 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5152 struct ext4_iloc iloc;
5153 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5154 static unsigned int mnt_count;
5158 err = ext4_reserve_inode_write(handle, inode, &iloc);
5159 if (ext4_handle_valid(handle) &&
5160 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5161 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
5163 * We need extra buffer credits since we may write into EA block
5164 * with this same handle. If journal_extend fails, then it will
5165 * only result in a minor loss of functionality for that inode.
5166 * If this is felt to be critical, then e2fsck should be run to
5167 * force a large enough s_min_extra_isize.
5169 if ((jbd2_journal_extend(handle,
5170 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5171 ret = ext4_expand_extra_isize(inode,
5172 sbi->s_want_extra_isize,
5175 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
5177 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5178 ext4_warning(inode->i_sb, __func__,
5179 "Unable to expand inode %lu. Delete"
5180 " some EAs or run e2fsck.",
5183 le16_to_cpu(sbi->s_es->s_mnt_count);
5189 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5194 * ext4_dirty_inode() is called from __mark_inode_dirty()
5196 * We're really interested in the case where a file is being extended.
5197 * i_size has been changed by generic_commit_write() and we thus need
5198 * to include the updated inode in the current transaction.
5200 * Also, vfs_dq_alloc_block() will always dirty the inode when blocks
5201 * are allocated to the file.
5203 * If the inode is marked synchronous, we don't honour that here - doing
5204 * so would cause a commit on atime updates, which we don't bother doing.
5205 * We handle synchronous inodes at the highest possible level.
5207 void ext4_dirty_inode(struct inode *inode)
5209 handle_t *current_handle = ext4_journal_current_handle();
5212 if (!ext4_handle_valid(current_handle)) {
5213 ext4_mark_inode_dirty(current_handle, inode);
5217 handle = ext4_journal_start(inode, 2);
5220 if (current_handle &&
5221 current_handle->h_transaction != handle->h_transaction) {
5222 /* This task has a transaction open against a different fs */
5223 printk(KERN_EMERG "%s: transactions do not match!\n",
5226 jbd_debug(5, "marking dirty. outer handle=%p\n",
5228 ext4_mark_inode_dirty(handle, inode);
5230 ext4_journal_stop(handle);
5237 * Bind an inode's backing buffer_head into this transaction, to prevent
5238 * it from being flushed to disk early. Unlike
5239 * ext4_reserve_inode_write, this leaves behind no bh reference and
5240 * returns no iloc structure, so the caller needs to repeat the iloc
5241 * lookup to mark the inode dirty later.
5243 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5245 struct ext4_iloc iloc;
5249 err = ext4_get_inode_loc(inode, &iloc);
5251 BUFFER_TRACE(iloc.bh, "get_write_access");
5252 err = jbd2_journal_get_write_access(handle, iloc.bh);
5254 err = ext4_handle_dirty_metadata(handle,
5260 ext4_std_error(inode->i_sb, err);
5265 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5272 * We have to be very careful here: changing a data block's
5273 * journaling status dynamically is dangerous. If we write a
5274 * data block to the journal, change the status and then delete
5275 * that block, we risk forgetting to revoke the old log record
5276 * from the journal and so a subsequent replay can corrupt data.
5277 * So, first we make sure that the journal is empty and that
5278 * nobody is changing anything.
5281 journal = EXT4_JOURNAL(inode);
5284 if (is_journal_aborted(journal))
5287 jbd2_journal_lock_updates(journal);
5288 jbd2_journal_flush(journal);
5291 * OK, there are no updates running now, and all cached data is
5292 * synced to disk. We are now in a completely consistent state
5293 * which doesn't have anything in the journal, and we know that
5294 * no filesystem updates are running, so it is safe to modify
5295 * the inode's in-core data-journaling state flag now.
5299 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5301 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5302 ext4_set_aops(inode);
5304 jbd2_journal_unlock_updates(journal);
5306 /* Finally we can mark the inode as dirty. */
5308 handle = ext4_journal_start(inode, 1);
5310 return PTR_ERR(handle);
5312 err = ext4_mark_inode_dirty(handle, inode);
5313 ext4_handle_sync(handle);
5314 ext4_journal_stop(handle);
5315 ext4_std_error(inode->i_sb, err);
5320 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5322 return !buffer_mapped(bh);
5325 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5327 struct page *page = vmf->page;
5332 struct file *file = vma->vm_file;
5333 struct inode *inode = file->f_path.dentry->d_inode;
5334 struct address_space *mapping = inode->i_mapping;
5337 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5338 * get i_mutex because we are already holding mmap_sem.
5340 down_read(&inode->i_alloc_sem);
5341 size = i_size_read(inode);
5342 if (page->mapping != mapping || size <= page_offset(page)
5343 || !PageUptodate(page)) {
5344 /* page got truncated from under us? */
5348 if (PageMappedToDisk(page))
5351 if (page->index == size >> PAGE_CACHE_SHIFT)
5352 len = size & ~PAGE_CACHE_MASK;
5354 len = PAGE_CACHE_SIZE;
5356 if (page_has_buffers(page)) {
5357 /* return if we have all the buffers mapped */
5358 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5363 * OK, we need to fill the hole... Do write_begin write_end
5364 * to do block allocation/reservation.We are not holding
5365 * inode.i__mutex here. That allow * parallel write_begin,
5366 * write_end call. lock_page prevent this from happening
5367 * on the same page though
5369 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5370 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5373 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5374 len, len, page, fsdata);
5380 ret = VM_FAULT_SIGBUS;
5381 up_read(&inode->i_alloc_sem);