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));
375 * ext4_get_branch - read the chain of indirect blocks leading to data
376 * @inode: inode in question
377 * @depth: depth of the chain (1 - direct pointer, etc.)
378 * @offsets: offsets of pointers in inode/indirect blocks
379 * @chain: place to store the result
380 * @err: here we store the error value
382 * Function fills the array of triples <key, p, bh> and returns %NULL
383 * if everything went OK or the pointer to the last filled triple
384 * (incomplete one) otherwise. Upon the return chain[i].key contains
385 * the number of (i+1)-th block in the chain (as it is stored in memory,
386 * i.e. little-endian 32-bit), chain[i].p contains the address of that
387 * number (it points into struct inode for i==0 and into the bh->b_data
388 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
389 * block for i>0 and NULL for i==0. In other words, it holds the block
390 * numbers of the chain, addresses they were taken from (and where we can
391 * verify that chain did not change) and buffer_heads hosting these
394 * Function stops when it stumbles upon zero pointer (absent block)
395 * (pointer to last triple returned, *@err == 0)
396 * or when it gets an IO error reading an indirect block
397 * (ditto, *@err == -EIO)
398 * or when it reads all @depth-1 indirect blocks successfully and finds
399 * the whole chain, all way to the data (returns %NULL, *err == 0).
401 * Need to be called with
402 * down_read(&EXT4_I(inode)->i_data_sem)
404 static Indirect *ext4_get_branch(struct inode *inode, int depth,
405 ext4_lblk_t *offsets,
406 Indirect chain[4], int *err)
408 struct super_block *sb = inode->i_sb;
410 struct buffer_head *bh;
413 /* i_data is not going away, no lock needed */
414 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
418 bh = sb_bread(sb, le32_to_cpu(p->key));
421 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
435 * ext4_find_near - find a place for allocation with sufficient locality
437 * @ind: descriptor of indirect block.
439 * This function returns the preferred place for block allocation.
440 * It is used when heuristic for sequential allocation fails.
442 * + if there is a block to the left of our position - allocate near it.
443 * + if pointer will live in indirect block - allocate near that block.
444 * + if pointer will live in inode - allocate in the same
447 * In the latter case we colour the starting block by the callers PID to
448 * prevent it from clashing with concurrent allocations for a different inode
449 * in the same block group. The PID is used here so that functionally related
450 * files will be close-by on-disk.
452 * Caller must make sure that @ind is valid and will stay that way.
454 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
456 struct ext4_inode_info *ei = EXT4_I(inode);
457 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
459 ext4_fsblk_t bg_start;
460 ext4_fsblk_t last_block;
461 ext4_grpblk_t colour;
463 /* Try to find previous block */
464 for (p = ind->p - 1; p >= start; p--) {
466 return le32_to_cpu(*p);
469 /* No such thing, so let's try location of indirect block */
471 return ind->bh->b_blocknr;
474 * It is going to be referred to from the inode itself? OK, just put it
475 * into the same cylinder group then.
477 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
478 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
480 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
481 colour = (current->pid % 16) *
482 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
484 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
485 return bg_start + colour;
489 * ext4_find_goal - find a preferred place for allocation.
491 * @block: block we want
492 * @partial: pointer to the last triple within a chain
494 * Normally this function find the preferred place for block allocation,
497 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
501 * XXX need to get goal block from mballoc's data structures
504 return ext4_find_near(inode, partial);
508 * ext4_blks_to_allocate: Look up the block map and count the number
509 * of direct blocks need to be allocated for the given branch.
511 * @branch: chain of indirect blocks
512 * @k: number of blocks need for indirect blocks
513 * @blks: number of data blocks to be mapped.
514 * @blocks_to_boundary: the offset in the indirect block
516 * return the total number of blocks to be allocate, including the
517 * direct and indirect blocks.
519 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
520 int blocks_to_boundary)
522 unsigned int count = 0;
525 * Simple case, [t,d]Indirect block(s) has not allocated yet
526 * then it's clear blocks on that path have not allocated
529 /* right now we don't handle cross boundary allocation */
530 if (blks < blocks_to_boundary + 1)
533 count += blocks_to_boundary + 1;
538 while (count < blks && count <= blocks_to_boundary &&
539 le32_to_cpu(*(branch[0].p + count)) == 0) {
546 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
547 * @indirect_blks: the number of blocks need to allocate for indirect
550 * @new_blocks: on return it will store the new block numbers for
551 * the indirect blocks(if needed) and the first direct block,
552 * @blks: on return it will store the total number of allocated
555 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
556 ext4_lblk_t iblock, ext4_fsblk_t goal,
557 int indirect_blks, int blks,
558 ext4_fsblk_t new_blocks[4], int *err)
560 struct ext4_allocation_request ar;
562 unsigned long count = 0, blk_allocated = 0;
564 ext4_fsblk_t current_block = 0;
568 * Here we try to allocate the requested multiple blocks at once,
569 * on a best-effort basis.
570 * To build a branch, we should allocate blocks for
571 * the indirect blocks(if not allocated yet), and at least
572 * the first direct block of this branch. That's the
573 * minimum number of blocks need to allocate(required)
575 /* first we try to allocate the indirect blocks */
576 target = indirect_blks;
579 /* allocating blocks for indirect blocks and direct blocks */
580 current_block = ext4_new_meta_blocks(handle, inode,
586 /* allocate blocks for indirect blocks */
587 while (index < indirect_blks && count) {
588 new_blocks[index++] = current_block++;
593 * save the new block number
594 * for the first direct block
596 new_blocks[index] = current_block;
597 printk(KERN_INFO "%s returned more blocks than "
598 "requested\n", __func__);
604 target = blks - count ;
605 blk_allocated = count;
608 /* Now allocate data blocks */
609 memset(&ar, 0, sizeof(ar));
614 if (S_ISREG(inode->i_mode))
615 /* enable in-core preallocation only for regular files */
616 ar.flags = EXT4_MB_HINT_DATA;
618 current_block = ext4_mb_new_blocks(handle, &ar, err);
620 if (*err && (target == blks)) {
622 * if the allocation failed and we didn't allocate
628 if (target == blks) {
630 * save the new block number
631 * for the first direct block
633 new_blocks[index] = current_block;
635 blk_allocated += ar.len;
638 /* total number of blocks allocated for direct blocks */
643 for (i = 0; i < index; i++)
644 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
649 * ext4_alloc_branch - allocate and set up a chain of blocks.
651 * @indirect_blks: number of allocated indirect blocks
652 * @blks: number of allocated direct blocks
653 * @offsets: offsets (in the blocks) to store the pointers to next.
654 * @branch: place to store the chain in.
656 * This function allocates blocks, zeroes out all but the last one,
657 * links them into chain and (if we are synchronous) writes them to disk.
658 * In other words, it prepares a branch that can be spliced onto the
659 * inode. It stores the information about that chain in the branch[], in
660 * the same format as ext4_get_branch() would do. We are calling it after
661 * we had read the existing part of chain and partial points to the last
662 * triple of that (one with zero ->key). Upon the exit we have the same
663 * picture as after the successful ext4_get_block(), except that in one
664 * place chain is disconnected - *branch->p is still zero (we did not
665 * set the last link), but branch->key contains the number that should
666 * be placed into *branch->p to fill that gap.
668 * If allocation fails we free all blocks we've allocated (and forget
669 * their buffer_heads) and return the error value the from failed
670 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
671 * as described above and return 0.
673 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
674 ext4_lblk_t iblock, int indirect_blks,
675 int *blks, ext4_fsblk_t goal,
676 ext4_lblk_t *offsets, Indirect *branch)
678 int blocksize = inode->i_sb->s_blocksize;
681 struct buffer_head *bh;
683 ext4_fsblk_t new_blocks[4];
684 ext4_fsblk_t current_block;
686 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
687 *blks, new_blocks, &err);
691 branch[0].key = cpu_to_le32(new_blocks[0]);
693 * metadata blocks and data blocks are allocated.
695 for (n = 1; n <= indirect_blks; n++) {
697 * Get buffer_head for parent block, zero it out
698 * and set the pointer to new one, then send
701 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
704 BUFFER_TRACE(bh, "call get_create_access");
705 err = ext4_journal_get_create_access(handle, bh);
712 memset(bh->b_data, 0, blocksize);
713 branch[n].p = (__le32 *) bh->b_data + offsets[n];
714 branch[n].key = cpu_to_le32(new_blocks[n]);
715 *branch[n].p = branch[n].key;
716 if (n == indirect_blks) {
717 current_block = new_blocks[n];
719 * End of chain, update the last new metablock of
720 * the chain to point to the new allocated
721 * data blocks numbers
723 for (i=1; i < num; i++)
724 *(branch[n].p + i) = cpu_to_le32(++current_block);
726 BUFFER_TRACE(bh, "marking uptodate");
727 set_buffer_uptodate(bh);
730 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
731 err = ext4_handle_dirty_metadata(handle, inode, bh);
738 /* Allocation failed, free what we already allocated */
739 for (i = 1; i <= n ; i++) {
740 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
741 ext4_journal_forget(handle, branch[i].bh);
743 for (i = 0; i < indirect_blks; i++)
744 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
746 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
752 * ext4_splice_branch - splice the allocated branch onto inode.
754 * @block: (logical) number of block we are adding
755 * @chain: chain of indirect blocks (with a missing link - see
757 * @where: location of missing link
758 * @num: number of indirect blocks we are adding
759 * @blks: number of direct blocks we are adding
761 * This function fills the missing link and does all housekeeping needed in
762 * inode (->i_blocks, etc.). In case of success we end up with the full
763 * chain to new block and return 0.
765 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
766 ext4_lblk_t block, Indirect *where, int num, int blks)
770 ext4_fsblk_t current_block;
773 * If we're splicing into a [td]indirect block (as opposed to the
774 * inode) then we need to get write access to the [td]indirect block
778 BUFFER_TRACE(where->bh, "get_write_access");
779 err = ext4_journal_get_write_access(handle, where->bh);
785 *where->p = where->key;
788 * Update the host buffer_head or inode to point to more just allocated
789 * direct blocks blocks
791 if (num == 0 && blks > 1) {
792 current_block = le32_to_cpu(where->key) + 1;
793 for (i = 1; i < blks; i++)
794 *(where->p + i) = cpu_to_le32(current_block++);
797 /* We are done with atomic stuff, now do the rest of housekeeping */
799 inode->i_ctime = ext4_current_time(inode);
800 ext4_mark_inode_dirty(handle, inode);
802 /* had we spliced it onto indirect block? */
805 * If we spliced it onto an indirect block, we haven't
806 * altered the inode. Note however that if it is being spliced
807 * onto an indirect block at the very end of the file (the
808 * file is growing) then we *will* alter the inode to reflect
809 * the new i_size. But that is not done here - it is done in
810 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
812 jbd_debug(5, "splicing indirect only\n");
813 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
814 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
819 * OK, we spliced it into the inode itself on a direct block.
820 * Inode was dirtied above.
822 jbd_debug(5, "splicing direct\n");
827 for (i = 1; i <= num; i++) {
828 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
829 ext4_journal_forget(handle, where[i].bh);
830 ext4_free_blocks(handle, inode,
831 le32_to_cpu(where[i-1].key), 1, 0);
833 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
839 * Allocation strategy is simple: if we have to allocate something, we will
840 * have to go the whole way to leaf. So let's do it before attaching anything
841 * to tree, set linkage between the newborn blocks, write them if sync is
842 * required, recheck the path, free and repeat if check fails, otherwise
843 * set the last missing link (that will protect us from any truncate-generated
844 * removals - all blocks on the path are immune now) and possibly force the
845 * write on the parent block.
846 * That has a nice additional property: no special recovery from the failed
847 * allocations is needed - we simply release blocks and do not touch anything
848 * reachable from inode.
850 * `handle' can be NULL if create == 0.
852 * return > 0, # of blocks mapped or allocated.
853 * return = 0, if plain lookup failed.
854 * return < 0, error case.
857 * Need to be called with
858 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
859 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
861 static int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
862 ext4_lblk_t iblock, unsigned int maxblocks,
863 struct buffer_head *bh_result,
864 int create, int extend_disksize)
867 ext4_lblk_t offsets[4];
872 int blocks_to_boundary = 0;
874 struct ext4_inode_info *ei = EXT4_I(inode);
876 ext4_fsblk_t first_block = 0;
880 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
881 J_ASSERT(handle != NULL || create == 0);
882 depth = ext4_block_to_path(inode, iblock, offsets,
883 &blocks_to_boundary);
888 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
890 /* Simplest case - block found, no allocation needed */
892 first_block = le32_to_cpu(chain[depth - 1].key);
893 clear_buffer_new(bh_result);
896 while (count < maxblocks && count <= blocks_to_boundary) {
899 blk = le32_to_cpu(*(chain[depth-1].p + count));
901 if (blk == first_block + count)
909 /* Next simple case - plain lookup or failed read of indirect block */
910 if (!create || err == -EIO)
914 * Okay, we need to do block allocation.
916 goal = ext4_find_goal(inode, iblock, partial);
918 /* the number of blocks need to allocate for [d,t]indirect blocks */
919 indirect_blks = (chain + depth) - partial - 1;
922 * Next look up the indirect map to count the totoal number of
923 * direct blocks to allocate for this branch.
925 count = ext4_blks_to_allocate(partial, indirect_blks,
926 maxblocks, blocks_to_boundary);
928 * Block out ext4_truncate while we alter the tree
930 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
932 offsets + (partial - chain), partial);
935 * The ext4_splice_branch call will free and forget any buffers
936 * on the new chain if there is a failure, but that risks using
937 * up transaction credits, especially for bitmaps where the
938 * credits cannot be returned. Can we handle this somehow? We
939 * may need to return -EAGAIN upwards in the worst case. --sct
942 err = ext4_splice_branch(handle, inode, iblock,
943 partial, indirect_blks, count);
945 * i_disksize growing is protected by i_data_sem. Don't forget to
946 * protect it if you're about to implement concurrent
947 * ext4_get_block() -bzzz
949 if (!err && extend_disksize) {
950 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
951 if (disksize > i_size_read(inode))
952 disksize = i_size_read(inode);
953 if (disksize > ei->i_disksize)
954 ei->i_disksize = disksize;
959 set_buffer_new(bh_result);
961 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
962 if (count > blocks_to_boundary)
963 set_buffer_boundary(bh_result);
965 /* Clean up and exit */
966 partial = chain + depth - 1; /* the whole chain */
968 while (partial > chain) {
969 BUFFER_TRACE(partial->bh, "call brelse");
973 BUFFER_TRACE(bh_result, "returned");
979 * Calculate the number of metadata blocks need to reserve
980 * to allocate @blocks for non extent file based file
982 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
984 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
985 int ind_blks, dind_blks, tind_blks;
987 /* number of new indirect blocks needed */
988 ind_blks = (blocks + icap - 1) / icap;
990 dind_blks = (ind_blks + icap - 1) / icap;
994 return ind_blks + dind_blks + tind_blks;
998 * Calculate the number of metadata blocks need to reserve
999 * to allocate given number of blocks
1001 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1006 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1007 return ext4_ext_calc_metadata_amount(inode, blocks);
1009 return ext4_indirect_calc_metadata_amount(inode, blocks);
1012 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1014 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1015 int total, mdb, mdb_free;
1017 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1018 /* recalculate the number of metablocks still need to be reserved */
1019 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1020 mdb = ext4_calc_metadata_amount(inode, total);
1022 /* figure out how many metablocks to release */
1023 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1024 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1027 /* Account for allocated meta_blocks */
1028 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1030 /* update fs dirty blocks counter */
1031 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1032 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1033 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1036 /* update per-inode reservations */
1037 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1038 EXT4_I(inode)->i_reserved_data_blocks -= used;
1040 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1044 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1045 * and returns if the blocks are already mapped.
1047 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1048 * and store the allocated blocks in the result buffer head and mark it
1051 * If file type is extents based, it will call ext4_ext_get_blocks(),
1052 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1055 * On success, it returns the number of blocks being mapped or allocate.
1056 * if create==0 and the blocks are pre-allocated and uninitialized block,
1057 * the result buffer head is unmapped. If the create ==1, it will make sure
1058 * the buffer head is mapped.
1060 * It returns 0 if plain look up failed (blocks have not been allocated), in
1061 * that casem, buffer head is unmapped
1063 * It returns the error in case of allocation failure.
1065 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1066 unsigned int max_blocks, struct buffer_head *bh,
1067 int create, int extend_disksize, int flag)
1071 clear_buffer_mapped(bh);
1074 * Try to see if we can get the block without requesting
1075 * for new file system block.
1077 down_read((&EXT4_I(inode)->i_data_sem));
1078 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1079 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1082 retval = ext4_get_blocks_handle(handle,
1083 inode, block, max_blocks, bh, 0, 0);
1085 up_read((&EXT4_I(inode)->i_data_sem));
1087 /* If it is only a block(s) look up */
1092 * Returns if the blocks have already allocated
1094 * Note that if blocks have been preallocated
1095 * ext4_ext_get_block() returns th create = 0
1096 * with buffer head unmapped.
1098 if (retval > 0 && buffer_mapped(bh))
1102 * New blocks allocate and/or writing to uninitialized extent
1103 * will possibly result in updating i_data, so we take
1104 * the write lock of i_data_sem, and call get_blocks()
1105 * with create == 1 flag.
1107 down_write((&EXT4_I(inode)->i_data_sem));
1110 * if the caller is from delayed allocation writeout path
1111 * we have already reserved fs blocks for allocation
1112 * let the underlying get_block() function know to
1113 * avoid double accounting
1116 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1118 * We need to check for EXT4 here because migrate
1119 * could have changed the inode type in between
1121 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1122 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1123 bh, create, extend_disksize);
1125 retval = ext4_get_blocks_handle(handle, inode, block,
1126 max_blocks, bh, create, extend_disksize);
1128 if (retval > 0 && buffer_new(bh)) {
1130 * We allocated new blocks which will result in
1131 * i_data's format changing. Force the migrate
1132 * to fail by clearing migrate flags
1134 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1140 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1142 * Update reserved blocks/metadata blocks
1143 * after successful block allocation
1144 * which were deferred till now
1146 if ((retval > 0) && buffer_delay(bh))
1147 ext4_da_update_reserve_space(inode, retval);
1150 up_write((&EXT4_I(inode)->i_data_sem));
1154 /* Maximum number of blocks we map for direct IO at once. */
1155 #define DIO_MAX_BLOCKS 4096
1157 int ext4_get_block(struct inode *inode, sector_t iblock,
1158 struct buffer_head *bh_result, int create)
1160 handle_t *handle = ext4_journal_current_handle();
1161 int ret = 0, started = 0;
1162 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1165 if (create && !handle) {
1166 /* Direct IO write... */
1167 if (max_blocks > DIO_MAX_BLOCKS)
1168 max_blocks = DIO_MAX_BLOCKS;
1169 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1170 handle = ext4_journal_start(inode, dio_credits);
1171 if (IS_ERR(handle)) {
1172 ret = PTR_ERR(handle);
1178 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1179 max_blocks, bh_result, create, 0, 0);
1181 bh_result->b_size = (ret << inode->i_blkbits);
1185 ext4_journal_stop(handle);
1191 * `handle' can be NULL if create is zero
1193 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1194 ext4_lblk_t block, int create, int *errp)
1196 struct buffer_head dummy;
1199 J_ASSERT(handle != NULL || create == 0);
1202 dummy.b_blocknr = -1000;
1203 buffer_trace_init(&dummy.b_history);
1204 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1205 &dummy, create, 1, 0);
1207 * ext4_get_blocks_handle() returns number of blocks
1208 * mapped. 0 in case of a HOLE.
1216 if (!err && buffer_mapped(&dummy)) {
1217 struct buffer_head *bh;
1218 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1223 if (buffer_new(&dummy)) {
1224 J_ASSERT(create != 0);
1225 J_ASSERT(handle != NULL);
1228 * Now that we do not always journal data, we should
1229 * keep in mind whether this should always journal the
1230 * new buffer as metadata. For now, regular file
1231 * writes use ext4_get_block instead, so it's not a
1235 BUFFER_TRACE(bh, "call get_create_access");
1236 fatal = ext4_journal_get_create_access(handle, bh);
1237 if (!fatal && !buffer_uptodate(bh)) {
1238 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1239 set_buffer_uptodate(bh);
1242 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1243 err = ext4_handle_dirty_metadata(handle, inode, bh);
1247 BUFFER_TRACE(bh, "not a new buffer");
1260 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1261 ext4_lblk_t block, int create, int *err)
1263 struct buffer_head *bh;
1265 bh = ext4_getblk(handle, inode, block, create, err);
1268 if (buffer_uptodate(bh))
1270 ll_rw_block(READ_META, 1, &bh);
1272 if (buffer_uptodate(bh))
1279 static int walk_page_buffers(handle_t *handle,
1280 struct buffer_head *head,
1284 int (*fn)(handle_t *handle,
1285 struct buffer_head *bh))
1287 struct buffer_head *bh;
1288 unsigned block_start, block_end;
1289 unsigned blocksize = head->b_size;
1291 struct buffer_head *next;
1293 for (bh = head, block_start = 0;
1294 ret == 0 && (bh != head || !block_start);
1295 block_start = block_end, bh = next)
1297 next = bh->b_this_page;
1298 block_end = block_start + blocksize;
1299 if (block_end <= from || block_start >= to) {
1300 if (partial && !buffer_uptodate(bh))
1304 err = (*fn)(handle, bh);
1312 * To preserve ordering, it is essential that the hole instantiation and
1313 * the data write be encapsulated in a single transaction. We cannot
1314 * close off a transaction and start a new one between the ext4_get_block()
1315 * and the commit_write(). So doing the jbd2_journal_start at the start of
1316 * prepare_write() is the right place.
1318 * Also, this function can nest inside ext4_writepage() ->
1319 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1320 * has generated enough buffer credits to do the whole page. So we won't
1321 * block on the journal in that case, which is good, because the caller may
1324 * By accident, ext4 can be reentered when a transaction is open via
1325 * quota file writes. If we were to commit the transaction while thus
1326 * reentered, there can be a deadlock - we would be holding a quota
1327 * lock, and the commit would never complete if another thread had a
1328 * transaction open and was blocking on the quota lock - a ranking
1331 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1332 * will _not_ run commit under these circumstances because handle->h_ref
1333 * is elevated. We'll still have enough credits for the tiny quotafile
1336 static int do_journal_get_write_access(handle_t *handle,
1337 struct buffer_head *bh)
1339 if (!buffer_mapped(bh) || buffer_freed(bh))
1341 return ext4_journal_get_write_access(handle, bh);
1344 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1345 loff_t pos, unsigned len, unsigned flags,
1346 struct page **pagep, void **fsdata)
1348 struct inode *inode = mapping->host;
1349 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1356 trace_mark(ext4_write_begin,
1357 "dev %s ino %lu pos %llu len %u flags %u",
1358 inode->i_sb->s_id, inode->i_ino,
1359 (unsigned long long) pos, len, flags);
1360 index = pos >> PAGE_CACHE_SHIFT;
1361 from = pos & (PAGE_CACHE_SIZE - 1);
1365 handle = ext4_journal_start(inode, needed_blocks);
1366 if (IS_ERR(handle)) {
1367 ret = PTR_ERR(handle);
1371 page = grab_cache_page_write_begin(mapping, index, flags);
1373 ext4_journal_stop(handle);
1379 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1382 if (!ret && ext4_should_journal_data(inode)) {
1383 ret = walk_page_buffers(handle, page_buffers(page),
1384 from, to, NULL, do_journal_get_write_access);
1389 ext4_journal_stop(handle);
1390 page_cache_release(page);
1392 * block_write_begin may have instantiated a few blocks
1393 * outside i_size. Trim these off again. Don't need
1394 * i_size_read because we hold i_mutex.
1396 if (pos + len > inode->i_size)
1397 vmtruncate(inode, inode->i_size);
1400 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1406 /* For write_end() in data=journal mode */
1407 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1409 if (!buffer_mapped(bh) || buffer_freed(bh))
1411 set_buffer_uptodate(bh);
1412 return ext4_handle_dirty_metadata(handle, NULL, bh);
1416 * We need to pick up the new inode size which generic_commit_write gave us
1417 * `file' can be NULL - eg, when called from page_symlink().
1419 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1420 * buffers are managed internally.
1422 static int ext4_ordered_write_end(struct file *file,
1423 struct address_space *mapping,
1424 loff_t pos, unsigned len, unsigned copied,
1425 struct page *page, void *fsdata)
1427 handle_t *handle = ext4_journal_current_handle();
1428 struct inode *inode = mapping->host;
1431 trace_mark(ext4_ordered_write_end,
1432 "dev %s ino %lu pos %llu len %u copied %u",
1433 inode->i_sb->s_id, inode->i_ino,
1434 (unsigned long long) pos, len, copied);
1435 ret = ext4_jbd2_file_inode(handle, inode);
1440 new_i_size = pos + copied;
1441 if (new_i_size > EXT4_I(inode)->i_disksize) {
1442 ext4_update_i_disksize(inode, new_i_size);
1443 /* We need to mark inode dirty even if
1444 * new_i_size is less that inode->i_size
1445 * bu greater than i_disksize.(hint delalloc)
1447 ext4_mark_inode_dirty(handle, inode);
1450 ret2 = generic_write_end(file, mapping, pos, len, copied,
1456 ret2 = ext4_journal_stop(handle);
1460 return ret ? ret : copied;
1463 static int ext4_writeback_write_end(struct file *file,
1464 struct address_space *mapping,
1465 loff_t pos, unsigned len, unsigned copied,
1466 struct page *page, void *fsdata)
1468 handle_t *handle = ext4_journal_current_handle();
1469 struct inode *inode = mapping->host;
1473 trace_mark(ext4_writeback_write_end,
1474 "dev %s ino %lu pos %llu len %u copied %u",
1475 inode->i_sb->s_id, inode->i_ino,
1476 (unsigned long long) pos, len, copied);
1477 new_i_size = pos + copied;
1478 if (new_i_size > EXT4_I(inode)->i_disksize) {
1479 ext4_update_i_disksize(inode, new_i_size);
1480 /* We need to mark inode dirty even if
1481 * new_i_size is less that inode->i_size
1482 * bu greater than i_disksize.(hint delalloc)
1484 ext4_mark_inode_dirty(handle, inode);
1487 ret2 = generic_write_end(file, mapping, pos, len, copied,
1493 ret2 = ext4_journal_stop(handle);
1497 return ret ? ret : copied;
1500 static int ext4_journalled_write_end(struct file *file,
1501 struct address_space *mapping,
1502 loff_t pos, unsigned len, unsigned copied,
1503 struct page *page, void *fsdata)
1505 handle_t *handle = ext4_journal_current_handle();
1506 struct inode *inode = mapping->host;
1512 trace_mark(ext4_journalled_write_end,
1513 "dev %s ino %lu pos %llu len %u copied %u",
1514 inode->i_sb->s_id, inode->i_ino,
1515 (unsigned long long) pos, len, copied);
1516 from = pos & (PAGE_CACHE_SIZE - 1);
1520 if (!PageUptodate(page))
1522 page_zero_new_buffers(page, from+copied, to);
1525 ret = walk_page_buffers(handle, page_buffers(page), from,
1526 to, &partial, write_end_fn);
1528 SetPageUptodate(page);
1529 new_i_size = pos + copied;
1530 if (new_i_size > inode->i_size)
1531 i_size_write(inode, pos+copied);
1532 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1533 if (new_i_size > EXT4_I(inode)->i_disksize) {
1534 ext4_update_i_disksize(inode, new_i_size);
1535 ret2 = ext4_mark_inode_dirty(handle, inode);
1541 ret2 = ext4_journal_stop(handle);
1544 page_cache_release(page);
1546 return ret ? ret : copied;
1549 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1552 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1553 unsigned long md_needed, mdblocks, total = 0;
1556 * recalculate the amount of metadata blocks to reserve
1557 * in order to allocate nrblocks
1558 * worse case is one extent per block
1561 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1562 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1563 mdblocks = ext4_calc_metadata_amount(inode, total);
1564 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1566 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1567 total = md_needed + nrblocks;
1569 if (ext4_claim_free_blocks(sbi, total)) {
1570 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1571 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1577 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1578 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1580 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1581 return 0; /* success */
1584 static void ext4_da_release_space(struct inode *inode, int to_free)
1586 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1587 int total, mdb, mdb_free, release;
1590 return; /* Nothing to release, exit */
1592 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1594 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1596 * if there is no reserved blocks, but we try to free some
1597 * then the counter is messed up somewhere.
1598 * but since this function is called from invalidate
1599 * page, it's harmless to return without any action
1601 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1602 "blocks for inode %lu, but there is no reserved "
1603 "data blocks\n", to_free, inode->i_ino);
1604 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1608 /* recalculate the number of metablocks still need to be reserved */
1609 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1610 mdb = ext4_calc_metadata_amount(inode, total);
1612 /* figure out how many metablocks to release */
1613 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1614 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1616 release = to_free + mdb_free;
1618 /* update fs dirty blocks counter for truncate case */
1619 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1621 /* update per-inode reservations */
1622 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1623 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1625 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1626 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1627 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1630 static void ext4_da_page_release_reservation(struct page *page,
1631 unsigned long offset)
1634 struct buffer_head *head, *bh;
1635 unsigned int curr_off = 0;
1637 head = page_buffers(page);
1640 unsigned int next_off = curr_off + bh->b_size;
1642 if ((offset <= curr_off) && (buffer_delay(bh))) {
1644 clear_buffer_delay(bh);
1646 curr_off = next_off;
1647 } while ((bh = bh->b_this_page) != head);
1648 ext4_da_release_space(page->mapping->host, to_release);
1652 * Delayed allocation stuff
1655 struct mpage_da_data {
1656 struct inode *inode;
1657 struct buffer_head lbh; /* extent of blocks */
1658 unsigned long first_page, next_page; /* extent of pages */
1659 get_block_t *get_block;
1660 struct writeback_control *wbc;
1667 * mpage_da_submit_io - walks through extent of pages and try to write
1668 * them with writepage() call back
1670 * @mpd->inode: inode
1671 * @mpd->first_page: first page of the extent
1672 * @mpd->next_page: page after the last page of the extent
1673 * @mpd->get_block: the filesystem's block mapper function
1675 * By the time mpage_da_submit_io() is called we expect all blocks
1676 * to be allocated. this may be wrong if allocation failed.
1678 * As pages are already locked by write_cache_pages(), we can't use it
1680 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1683 struct pagevec pvec;
1684 unsigned long index, end;
1685 int ret = 0, err, nr_pages, i;
1686 struct inode *inode = mpd->inode;
1687 struct address_space *mapping = inode->i_mapping;
1689 BUG_ON(mpd->next_page <= mpd->first_page);
1691 * We need to start from the first_page to the next_page - 1
1692 * to make sure we also write the mapped dirty buffer_heads.
1693 * If we look at mpd->lbh.b_blocknr we would only be looking
1694 * at the currently mapped buffer_heads.
1696 index = mpd->first_page;
1697 end = mpd->next_page - 1;
1699 pagevec_init(&pvec, 0);
1700 while (index <= end) {
1701 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1704 for (i = 0; i < nr_pages; i++) {
1705 struct page *page = pvec.pages[i];
1707 index = page->index;
1712 BUG_ON(!PageLocked(page));
1713 BUG_ON(PageWriteback(page));
1715 pages_skipped = mpd->wbc->pages_skipped;
1716 err = mapping->a_ops->writepage(page, mpd->wbc);
1717 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1719 * have successfully written the page
1720 * without skipping the same
1722 mpd->pages_written++;
1724 * In error case, we have to continue because
1725 * remaining pages are still locked
1726 * XXX: unlock and re-dirty them?
1731 pagevec_release(&pvec);
1737 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1739 * @mpd->inode - inode to walk through
1740 * @exbh->b_blocknr - first block on a disk
1741 * @exbh->b_size - amount of space in bytes
1742 * @logical - first logical block to start assignment with
1744 * the function goes through all passed space and put actual disk
1745 * block numbers into buffer heads, dropping BH_Delay
1747 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1748 struct buffer_head *exbh)
1750 struct inode *inode = mpd->inode;
1751 struct address_space *mapping = inode->i_mapping;
1752 int blocks = exbh->b_size >> inode->i_blkbits;
1753 sector_t pblock = exbh->b_blocknr, cur_logical;
1754 struct buffer_head *head, *bh;
1756 struct pagevec pvec;
1759 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1760 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1761 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1763 pagevec_init(&pvec, 0);
1765 while (index <= end) {
1766 /* XXX: optimize tail */
1767 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1770 for (i = 0; i < nr_pages; i++) {
1771 struct page *page = pvec.pages[i];
1773 index = page->index;
1778 BUG_ON(!PageLocked(page));
1779 BUG_ON(PageWriteback(page));
1780 BUG_ON(!page_has_buffers(page));
1782 bh = page_buffers(page);
1785 /* skip blocks out of the range */
1787 if (cur_logical >= logical)
1790 } while ((bh = bh->b_this_page) != head);
1793 if (cur_logical >= logical + blocks)
1795 if (buffer_delay(bh)) {
1796 bh->b_blocknr = pblock;
1797 clear_buffer_delay(bh);
1798 bh->b_bdev = inode->i_sb->s_bdev;
1799 } else if (buffer_unwritten(bh)) {
1800 bh->b_blocknr = pblock;
1801 clear_buffer_unwritten(bh);
1802 set_buffer_mapped(bh);
1804 bh->b_bdev = inode->i_sb->s_bdev;
1805 } else if (buffer_mapped(bh))
1806 BUG_ON(bh->b_blocknr != pblock);
1810 } while ((bh = bh->b_this_page) != head);
1812 pagevec_release(&pvec);
1818 * __unmap_underlying_blocks - just a helper function to unmap
1819 * set of blocks described by @bh
1821 static inline void __unmap_underlying_blocks(struct inode *inode,
1822 struct buffer_head *bh)
1824 struct block_device *bdev = inode->i_sb->s_bdev;
1827 blocks = bh->b_size >> inode->i_blkbits;
1828 for (i = 0; i < blocks; i++)
1829 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1832 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1833 sector_t logical, long blk_cnt)
1837 struct pagevec pvec;
1838 struct inode *inode = mpd->inode;
1839 struct address_space *mapping = inode->i_mapping;
1841 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1842 end = (logical + blk_cnt - 1) >>
1843 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1844 while (index <= end) {
1845 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1848 for (i = 0; i < nr_pages; i++) {
1849 struct page *page = pvec.pages[i];
1850 index = page->index;
1855 BUG_ON(!PageLocked(page));
1856 BUG_ON(PageWriteback(page));
1857 block_invalidatepage(page, 0);
1858 ClearPageUptodate(page);
1865 static void ext4_print_free_blocks(struct inode *inode)
1867 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1868 printk(KERN_EMERG "Total free blocks count %lld\n",
1869 ext4_count_free_blocks(inode->i_sb));
1870 printk(KERN_EMERG "Free/Dirty block details\n");
1871 printk(KERN_EMERG "free_blocks=%lld\n",
1872 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1873 printk(KERN_EMERG "dirty_blocks=%lld\n",
1874 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1875 printk(KERN_EMERG "Block reservation details\n");
1876 printk(KERN_EMERG "i_reserved_data_blocks=%u\n",
1877 EXT4_I(inode)->i_reserved_data_blocks);
1878 printk(KERN_EMERG "i_reserved_meta_blocks=%u\n",
1879 EXT4_I(inode)->i_reserved_meta_blocks);
1884 * mpage_da_map_blocks - go through given space
1886 * @mpd->lbh - bh describing space
1887 * @mpd->get_block - the filesystem's block mapper function
1889 * The function skips space we know is already mapped to disk blocks.
1892 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1895 struct buffer_head new;
1896 struct buffer_head *lbh = &mpd->lbh;
1900 * We consider only non-mapped and non-allocated blocks
1902 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1904 new.b_state = lbh->b_state;
1906 new.b_size = lbh->b_size;
1907 next = lbh->b_blocknr;
1909 * If we didn't accumulate anything
1910 * to write simply return
1914 err = mpd->get_block(mpd->inode, next, &new, 1);
1917 /* If get block returns with error
1918 * we simply return. Later writepage
1919 * will redirty the page and writepages
1920 * will find the dirty page again
1925 if (err == -ENOSPC &&
1926 ext4_count_free_blocks(mpd->inode->i_sb)) {
1932 * get block failure will cause us
1933 * to loop in writepages. Because
1934 * a_ops->writepage won't be able to
1935 * make progress. The page will be redirtied
1936 * by writepage and writepages will again
1937 * try to write the same.
1939 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1940 "at logical offset %llu with max blocks "
1941 "%zd with error %d\n",
1942 __func__, mpd->inode->i_ino,
1943 (unsigned long long)next,
1944 lbh->b_size >> mpd->inode->i_blkbits, err);
1945 printk(KERN_EMERG "This should not happen.!! "
1946 "Data will be lost\n");
1947 if (err == -ENOSPC) {
1948 ext4_print_free_blocks(mpd->inode);
1950 /* invlaidate all the pages */
1951 ext4_da_block_invalidatepages(mpd, next,
1952 lbh->b_size >> mpd->inode->i_blkbits);
1955 BUG_ON(new.b_size == 0);
1957 if (buffer_new(&new))
1958 __unmap_underlying_blocks(mpd->inode, &new);
1961 * If blocks are delayed marked, we need to
1962 * put actual blocknr and drop delayed bit
1964 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1965 mpage_put_bnr_to_bhs(mpd, next, &new);
1970 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1971 (1 << BH_Delay) | (1 << BH_Unwritten))
1974 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1976 * @mpd->lbh - extent of blocks
1977 * @logical - logical number of the block in the file
1978 * @bh - bh of the block (used to access block's state)
1980 * the function is used to collect contig. blocks in same state
1982 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1983 sector_t logical, struct buffer_head *bh)
1986 size_t b_size = bh->b_size;
1987 struct buffer_head *lbh = &mpd->lbh;
1988 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1990 /* check if thereserved journal credits might overflow */
1991 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1992 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1994 * With non-extent format we are limited by the journal
1995 * credit available. Total credit needed to insert
1996 * nrblocks contiguous blocks is dependent on the
1997 * nrblocks. So limit nrblocks.
2000 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2001 EXT4_MAX_TRANS_DATA) {
2003 * Adding the new buffer_head would make it cross the
2004 * allowed limit for which we have journal credit
2005 * reserved. So limit the new bh->b_size
2007 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2008 mpd->inode->i_blkbits;
2009 /* we will do mpage_da_submit_io in the next loop */
2013 * First block in the extent
2015 if (lbh->b_size == 0) {
2016 lbh->b_blocknr = logical;
2017 lbh->b_size = b_size;
2018 lbh->b_state = bh->b_state & BH_FLAGS;
2022 next = lbh->b_blocknr + nrblocks;
2024 * Can we merge the block to our big extent?
2026 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
2027 lbh->b_size += b_size;
2033 * We couldn't merge the block to our extent, so we
2034 * need to flush current extent and start new one
2036 if (mpage_da_map_blocks(mpd) == 0)
2037 mpage_da_submit_io(mpd);
2043 * __mpage_da_writepage - finds extent of pages and blocks
2045 * @page: page to consider
2046 * @wbc: not used, we just follow rules
2049 * The function finds extents of pages and scan them for all blocks.
2051 static int __mpage_da_writepage(struct page *page,
2052 struct writeback_control *wbc, void *data)
2054 struct mpage_da_data *mpd = data;
2055 struct inode *inode = mpd->inode;
2056 struct buffer_head *bh, *head, fake;
2061 * Rest of the page in the page_vec
2062 * redirty then and skip then. We will
2063 * try to to write them again after
2064 * starting a new transaction
2066 redirty_page_for_writepage(wbc, page);
2068 return MPAGE_DA_EXTENT_TAIL;
2071 * Can we merge this page to current extent?
2073 if (mpd->next_page != page->index) {
2075 * Nope, we can't. So, we map non-allocated blocks
2076 * and start IO on them using writepage()
2078 if (mpd->next_page != mpd->first_page) {
2079 if (mpage_da_map_blocks(mpd) == 0)
2080 mpage_da_submit_io(mpd);
2082 * skip rest of the page in the page_vec
2085 redirty_page_for_writepage(wbc, page);
2087 return MPAGE_DA_EXTENT_TAIL;
2091 * Start next extent of pages ...
2093 mpd->first_page = page->index;
2098 mpd->lbh.b_size = 0;
2099 mpd->lbh.b_state = 0;
2100 mpd->lbh.b_blocknr = 0;
2103 mpd->next_page = page->index + 1;
2104 logical = (sector_t) page->index <<
2105 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2107 if (!page_has_buffers(page)) {
2109 * There is no attached buffer heads yet (mmap?)
2110 * we treat the page asfull of dirty blocks
2113 bh->b_size = PAGE_CACHE_SIZE;
2115 set_buffer_dirty(bh);
2116 set_buffer_uptodate(bh);
2117 mpage_add_bh_to_extent(mpd, logical, bh);
2119 return MPAGE_DA_EXTENT_TAIL;
2122 * Page with regular buffer heads, just add all dirty ones
2124 head = page_buffers(page);
2127 BUG_ON(buffer_locked(bh));
2129 * We need to try to allocate
2130 * unmapped blocks in the same page.
2131 * Otherwise we won't make progress
2132 * with the page in ext4_da_writepage
2134 if (buffer_dirty(bh) &&
2135 (!buffer_mapped(bh) || buffer_delay(bh))) {
2136 mpage_add_bh_to_extent(mpd, logical, bh);
2138 return MPAGE_DA_EXTENT_TAIL;
2139 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2141 * mapped dirty buffer. We need to update
2142 * the b_state because we look at
2143 * b_state in mpage_da_map_blocks. We don't
2144 * update b_size because if we find an
2145 * unmapped buffer_head later we need to
2146 * use the b_state flag of that buffer_head.
2148 if (mpd->lbh.b_size == 0)
2150 bh->b_state & BH_FLAGS;
2153 } while ((bh = bh->b_this_page) != head);
2160 * mpage_da_writepages - walk the list of dirty pages of the given
2161 * address space, allocates non-allocated blocks, maps newly-allocated
2162 * blocks to existing bhs and issue IO them
2164 * @mapping: address space structure to write
2165 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2166 * @get_block: the filesystem's block mapper function.
2168 * This is a library function, which implements the writepages()
2169 * address_space_operation.
2171 static int mpage_da_writepages(struct address_space *mapping,
2172 struct writeback_control *wbc,
2173 struct mpage_da_data *mpd)
2177 if (!mpd->get_block)
2178 return generic_writepages(mapping, wbc);
2180 mpd->lbh.b_size = 0;
2181 mpd->lbh.b_state = 0;
2182 mpd->lbh.b_blocknr = 0;
2183 mpd->first_page = 0;
2186 mpd->pages_written = 0;
2189 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2191 * Handle last extent of pages
2193 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2194 if (mpage_da_map_blocks(mpd) == 0)
2195 mpage_da_submit_io(mpd);
2198 ret = MPAGE_DA_EXTENT_TAIL;
2200 wbc->nr_to_write -= mpd->pages_written;
2205 * this is a special callback for ->write_begin() only
2206 * it's intention is to return mapped block or reserve space
2208 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2209 struct buffer_head *bh_result, int create)
2213 BUG_ON(create == 0);
2214 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2217 * first, we need to know whether the block is allocated already
2218 * preallocated blocks are unmapped but should treated
2219 * the same as allocated blocks.
2221 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2222 if ((ret == 0) && !buffer_delay(bh_result)) {
2223 /* the block isn't (pre)allocated yet, let's reserve space */
2225 * XXX: __block_prepare_write() unmaps passed block,
2228 ret = ext4_da_reserve_space(inode, 1);
2230 /* not enough space to reserve */
2233 map_bh(bh_result, inode->i_sb, 0);
2234 set_buffer_new(bh_result);
2235 set_buffer_delay(bh_result);
2236 } else if (ret > 0) {
2237 bh_result->b_size = (ret << inode->i_blkbits);
2243 #define EXT4_DELALLOC_RSVED 1
2244 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2245 struct buffer_head *bh_result, int create)
2248 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2249 loff_t disksize = EXT4_I(inode)->i_disksize;
2250 handle_t *handle = NULL;
2252 handle = ext4_journal_current_handle();
2254 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2255 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2258 bh_result->b_size = (ret << inode->i_blkbits);
2260 if (ext4_should_order_data(inode)) {
2262 retval = ext4_jbd2_file_inode(handle, inode);
2265 * Failed to add inode for ordered
2266 * mode. Don't update file size
2272 * Update on-disk size along with block allocation
2273 * we don't use 'extend_disksize' as size may change
2274 * within already allocated block -bzzz
2276 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2277 if (disksize > i_size_read(inode))
2278 disksize = i_size_read(inode);
2279 if (disksize > EXT4_I(inode)->i_disksize) {
2280 ext4_update_i_disksize(inode, disksize);
2281 ret = ext4_mark_inode_dirty(handle, inode);
2289 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2292 * unmapped buffer is possible for holes.
2293 * delay buffer is possible with delayed allocation
2295 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2298 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2299 struct buffer_head *bh_result, int create)
2302 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2305 * we don't want to do block allocation in writepage
2306 * so call get_block_wrap with create = 0
2308 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2309 bh_result, 0, 0, 0);
2311 bh_result->b_size = (ret << inode->i_blkbits);
2318 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2319 * get called via journal_submit_inode_data_buffers (no journal handle)
2320 * get called via shrink_page_list via pdflush (no journal handle)
2321 * or grab_page_cache when doing write_begin (have journal handle)
2323 static int ext4_da_writepage(struct page *page,
2324 struct writeback_control *wbc)
2329 struct buffer_head *page_bufs;
2330 struct inode *inode = page->mapping->host;
2332 trace_mark(ext4_da_writepage,
2333 "dev %s ino %lu page_index %lu",
2334 inode->i_sb->s_id, inode->i_ino, page->index);
2335 size = i_size_read(inode);
2336 if (page->index == size >> PAGE_CACHE_SHIFT)
2337 len = size & ~PAGE_CACHE_MASK;
2339 len = PAGE_CACHE_SIZE;
2341 if (page_has_buffers(page)) {
2342 page_bufs = page_buffers(page);
2343 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2344 ext4_bh_unmapped_or_delay)) {
2346 * We don't want to do block allocation
2347 * So redirty the page and return
2348 * We may reach here when we do a journal commit
2349 * via journal_submit_inode_data_buffers.
2350 * If we don't have mapping block we just ignore
2351 * them. We can also reach here via shrink_page_list
2353 redirty_page_for_writepage(wbc, page);
2359 * The test for page_has_buffers() is subtle:
2360 * We know the page is dirty but it lost buffers. That means
2361 * that at some moment in time after write_begin()/write_end()
2362 * has been called all buffers have been clean and thus they
2363 * must have been written at least once. So they are all
2364 * mapped and we can happily proceed with mapping them
2365 * and writing the page.
2367 * Try to initialize the buffer_heads and check whether
2368 * all are mapped and non delay. We don't want to
2369 * do block allocation here.
2371 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2372 ext4_normal_get_block_write);
2374 page_bufs = page_buffers(page);
2375 /* check whether all are mapped and non delay */
2376 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2377 ext4_bh_unmapped_or_delay)) {
2378 redirty_page_for_writepage(wbc, page);
2384 * We can't do block allocation here
2385 * so just redity the page and unlock
2388 redirty_page_for_writepage(wbc, page);
2392 /* now mark the buffer_heads as dirty and uptodate */
2393 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2396 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2397 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2399 ret = block_write_full_page(page,
2400 ext4_normal_get_block_write,
2407 * This is called via ext4_da_writepages() to
2408 * calulate the total number of credits to reserve to fit
2409 * a single extent allocation into a single transaction,
2410 * ext4_da_writpeages() will loop calling this before
2411 * the block allocation.
2414 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2416 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2419 * With non-extent format the journal credit needed to
2420 * insert nrblocks contiguous block is dependent on
2421 * number of contiguous block. So we will limit
2422 * number of contiguous block to a sane value
2424 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2425 (max_blocks > EXT4_MAX_TRANS_DATA))
2426 max_blocks = EXT4_MAX_TRANS_DATA;
2428 return ext4_chunk_trans_blocks(inode, max_blocks);
2431 static int ext4_da_writepages(struct address_space *mapping,
2432 struct writeback_control *wbc)
2435 int range_whole = 0;
2436 handle_t *handle = NULL;
2437 struct mpage_da_data mpd;
2438 struct inode *inode = mapping->host;
2439 int no_nrwrite_index_update;
2440 int pages_written = 0;
2442 int needed_blocks, ret = 0, nr_to_writebump = 0;
2443 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2445 trace_mark(ext4_da_writepages,
2446 "dev %s ino %lu nr_t_write %ld "
2447 "pages_skipped %ld range_start %llu "
2448 "range_end %llu nonblocking %d "
2449 "for_kupdate %d for_reclaim %d "
2450 "for_writepages %d range_cyclic %d",
2451 inode->i_sb->s_id, inode->i_ino,
2452 wbc->nr_to_write, wbc->pages_skipped,
2453 (unsigned long long) wbc->range_start,
2454 (unsigned long long) wbc->range_end,
2455 wbc->nonblocking, wbc->for_kupdate,
2456 wbc->for_reclaim, wbc->for_writepages,
2460 * No pages to write? This is mainly a kludge to avoid starting
2461 * a transaction for special inodes like journal inode on last iput()
2462 * because that could violate lock ordering on umount
2464 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2468 * If the filesystem has aborted, it is read-only, so return
2469 * right away instead of dumping stack traces later on that
2470 * will obscure the real source of the problem. We test
2471 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2472 * the latter could be true if the filesystem is mounted
2473 * read-only, and in that case, ext4_da_writepages should
2474 * *never* be called, so if that ever happens, we would want
2477 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2481 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2482 * This make sure small files blocks are allocated in
2483 * single attempt. This ensure that small files
2484 * get less fragmented.
2486 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2487 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2488 wbc->nr_to_write = sbi->s_mb_stream_request;
2490 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2493 if (wbc->range_cyclic)
2494 index = mapping->writeback_index;
2496 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2499 mpd.inode = mapping->host;
2502 * we don't want write_cache_pages to update
2503 * nr_to_write and writeback_index
2505 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2506 wbc->no_nrwrite_index_update = 1;
2507 pages_skipped = wbc->pages_skipped;
2509 while (!ret && wbc->nr_to_write > 0) {
2512 * we insert one extent at a time. So we need
2513 * credit needed for single extent allocation.
2514 * journalled mode is currently not supported
2517 BUG_ON(ext4_should_journal_data(inode));
2518 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2520 /* start a new transaction*/
2521 handle = ext4_journal_start(inode, needed_blocks);
2522 if (IS_ERR(handle)) {
2523 ret = PTR_ERR(handle);
2524 printk(KERN_CRIT "%s: jbd2_start: "
2525 "%ld pages, ino %lu; err %d\n", __func__,
2526 wbc->nr_to_write, inode->i_ino, ret);
2528 goto out_writepages;
2530 mpd.get_block = ext4_da_get_block_write;
2531 ret = mpage_da_writepages(mapping, wbc, &mpd);
2533 ext4_journal_stop(handle);
2535 if (mpd.retval == -ENOSPC) {
2536 /* commit the transaction which would
2537 * free blocks released in the transaction
2540 jbd2_journal_force_commit_nested(sbi->s_journal);
2541 wbc->pages_skipped = pages_skipped;
2543 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2545 * got one extent now try with
2548 pages_written += mpd.pages_written;
2549 wbc->pages_skipped = pages_skipped;
2551 } else if (wbc->nr_to_write)
2553 * There is no more writeout needed
2554 * or we requested for a noblocking writeout
2555 * and we found the device congested
2559 if (pages_skipped != wbc->pages_skipped)
2560 printk(KERN_EMERG "This should not happen leaving %s "
2561 "with nr_to_write = %ld ret = %d\n",
2562 __func__, wbc->nr_to_write, ret);
2565 index += pages_written;
2566 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2568 * set the writeback_index so that range_cyclic
2569 * mode will write it back later
2571 mapping->writeback_index = index;
2574 if (!no_nrwrite_index_update)
2575 wbc->no_nrwrite_index_update = 0;
2576 wbc->nr_to_write -= nr_to_writebump;
2577 trace_mark(ext4_da_writepage_result,
2578 "dev %s ino %lu ret %d pages_written %d "
2579 "pages_skipped %ld congestion %d "
2580 "more_io %d no_nrwrite_index_update %d",
2581 inode->i_sb->s_id, inode->i_ino, ret,
2582 pages_written, wbc->pages_skipped,
2583 wbc->encountered_congestion, wbc->more_io,
2584 wbc->no_nrwrite_index_update);
2588 #define FALL_BACK_TO_NONDELALLOC 1
2589 static int ext4_nonda_switch(struct super_block *sb)
2591 s64 free_blocks, dirty_blocks;
2592 struct ext4_sb_info *sbi = EXT4_SB(sb);
2595 * switch to non delalloc mode if we are running low
2596 * on free block. The free block accounting via percpu
2597 * counters can get slightly wrong with percpu_counter_batch getting
2598 * accumulated on each CPU without updating global counters
2599 * Delalloc need an accurate free block accounting. So switch
2600 * to non delalloc when we are near to error range.
2602 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2603 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2604 if (2 * free_blocks < 3 * dirty_blocks ||
2605 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2607 * free block count is less that 150% of dirty blocks
2608 * or free blocks is less that watermark
2615 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2616 loff_t pos, unsigned len, unsigned flags,
2617 struct page **pagep, void **fsdata)
2619 int ret, retries = 0;
2623 struct inode *inode = mapping->host;
2626 index = pos >> PAGE_CACHE_SHIFT;
2627 from = pos & (PAGE_CACHE_SIZE - 1);
2630 if (ext4_nonda_switch(inode->i_sb)) {
2631 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2632 return ext4_write_begin(file, mapping, pos,
2633 len, flags, pagep, fsdata);
2635 *fsdata = (void *)0;
2637 trace_mark(ext4_da_write_begin,
2638 "dev %s ino %lu pos %llu len %u flags %u",
2639 inode->i_sb->s_id, inode->i_ino,
2640 (unsigned long long) pos, len, flags);
2643 * With delayed allocation, we don't log the i_disksize update
2644 * if there is delayed block allocation. But we still need
2645 * to journalling the i_disksize update if writes to the end
2646 * of file which has an already mapped buffer.
2648 handle = ext4_journal_start(inode, 1);
2649 if (IS_ERR(handle)) {
2650 ret = PTR_ERR(handle);
2654 page = grab_cache_page_write_begin(mapping, index, flags);
2656 ext4_journal_stop(handle);
2662 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2663 ext4_da_get_block_prep);
2666 ext4_journal_stop(handle);
2667 page_cache_release(page);
2669 * block_write_begin may have instantiated a few blocks
2670 * outside i_size. Trim these off again. Don't need
2671 * i_size_read because we hold i_mutex.
2673 if (pos + len > inode->i_size)
2674 vmtruncate(inode, inode->i_size);
2677 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2684 * Check if we should update i_disksize
2685 * when write to the end of file but not require block allocation
2687 static int ext4_da_should_update_i_disksize(struct page *page,
2688 unsigned long offset)
2690 struct buffer_head *bh;
2691 struct inode *inode = page->mapping->host;
2695 bh = page_buffers(page);
2696 idx = offset >> inode->i_blkbits;
2698 for (i = 0; i < idx; i++)
2699 bh = bh->b_this_page;
2701 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2706 static int ext4_da_write_end(struct file *file,
2707 struct address_space *mapping,
2708 loff_t pos, unsigned len, unsigned copied,
2709 struct page *page, void *fsdata)
2711 struct inode *inode = mapping->host;
2713 handle_t *handle = ext4_journal_current_handle();
2715 unsigned long start, end;
2716 int write_mode = (int)(unsigned long)fsdata;
2718 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2719 if (ext4_should_order_data(inode)) {
2720 return ext4_ordered_write_end(file, mapping, pos,
2721 len, copied, page, fsdata);
2722 } else if (ext4_should_writeback_data(inode)) {
2723 return ext4_writeback_write_end(file, mapping, pos,
2724 len, copied, page, fsdata);
2730 trace_mark(ext4_da_write_end,
2731 "dev %s ino %lu pos %llu len %u copied %u",
2732 inode->i_sb->s_id, inode->i_ino,
2733 (unsigned long long) pos, len, copied);
2734 start = pos & (PAGE_CACHE_SIZE - 1);
2735 end = start + copied - 1;
2738 * generic_write_end() will run mark_inode_dirty() if i_size
2739 * changes. So let's piggyback the i_disksize mark_inode_dirty
2743 new_i_size = pos + copied;
2744 if (new_i_size > EXT4_I(inode)->i_disksize) {
2745 if (ext4_da_should_update_i_disksize(page, end)) {
2746 down_write(&EXT4_I(inode)->i_data_sem);
2747 if (new_i_size > EXT4_I(inode)->i_disksize) {
2749 * Updating i_disksize when extending file
2750 * without needing block allocation
2752 if (ext4_should_order_data(inode))
2753 ret = ext4_jbd2_file_inode(handle,
2756 EXT4_I(inode)->i_disksize = new_i_size;
2758 up_write(&EXT4_I(inode)->i_data_sem);
2759 /* We need to mark inode dirty even if
2760 * new_i_size is less that inode->i_size
2761 * bu greater than i_disksize.(hint delalloc)
2763 ext4_mark_inode_dirty(handle, inode);
2766 ret2 = generic_write_end(file, mapping, pos, len, copied,
2771 ret2 = ext4_journal_stop(handle);
2775 return ret ? ret : copied;
2778 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2781 * Drop reserved blocks
2783 BUG_ON(!PageLocked(page));
2784 if (!page_has_buffers(page))
2787 ext4_da_page_release_reservation(page, offset);
2790 ext4_invalidatepage(page, offset);
2797 * bmap() is special. It gets used by applications such as lilo and by
2798 * the swapper to find the on-disk block of a specific piece of data.
2800 * Naturally, this is dangerous if the block concerned is still in the
2801 * journal. If somebody makes a swapfile on an ext4 data-journaling
2802 * filesystem and enables swap, then they may get a nasty shock when the
2803 * data getting swapped to that swapfile suddenly gets overwritten by
2804 * the original zero's written out previously to the journal and
2805 * awaiting writeback in the kernel's buffer cache.
2807 * So, if we see any bmap calls here on a modified, data-journaled file,
2808 * take extra steps to flush any blocks which might be in the cache.
2810 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2812 struct inode *inode = mapping->host;
2816 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2817 test_opt(inode->i_sb, DELALLOC)) {
2819 * With delalloc we want to sync the file
2820 * so that we can make sure we allocate
2823 filemap_write_and_wait(mapping);
2826 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2828 * This is a REALLY heavyweight approach, but the use of
2829 * bmap on dirty files is expected to be extremely rare:
2830 * only if we run lilo or swapon on a freshly made file
2831 * do we expect this to happen.
2833 * (bmap requires CAP_SYS_RAWIO so this does not
2834 * represent an unprivileged user DOS attack --- we'd be
2835 * in trouble if mortal users could trigger this path at
2838 * NB. EXT4_STATE_JDATA is not set on files other than
2839 * regular files. If somebody wants to bmap a directory
2840 * or symlink and gets confused because the buffer
2841 * hasn't yet been flushed to disk, they deserve
2842 * everything they get.
2845 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2846 journal = EXT4_JOURNAL(inode);
2847 jbd2_journal_lock_updates(journal);
2848 err = jbd2_journal_flush(journal);
2849 jbd2_journal_unlock_updates(journal);
2855 return generic_block_bmap(mapping, block, ext4_get_block);
2858 static int bget_one(handle_t *handle, struct buffer_head *bh)
2864 static int bput_one(handle_t *handle, struct buffer_head *bh)
2871 * Note that we don't need to start a transaction unless we're journaling data
2872 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2873 * need to file the inode to the transaction's list in ordered mode because if
2874 * we are writing back data added by write(), the inode is already there and if
2875 * we are writing back data modified via mmap(), noone guarantees in which
2876 * transaction the data will hit the disk. In case we are journaling data, we
2877 * cannot start transaction directly because transaction start ranks above page
2878 * lock so we have to do some magic.
2880 * In all journaling modes block_write_full_page() will start the I/O.
2884 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2889 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2891 * Same applies to ext4_get_block(). We will deadlock on various things like
2892 * lock_journal and i_data_sem
2894 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2897 * 16May01: If we're reentered then journal_current_handle() will be
2898 * non-zero. We simply *return*.
2900 * 1 July 2001: @@@ FIXME:
2901 * In journalled data mode, a data buffer may be metadata against the
2902 * current transaction. But the same file is part of a shared mapping
2903 * and someone does a writepage() on it.
2905 * We will move the buffer onto the async_data list, but *after* it has
2906 * been dirtied. So there's a small window where we have dirty data on
2909 * Note that this only applies to the last partial page in the file. The
2910 * bit which block_write_full_page() uses prepare/commit for. (That's
2911 * broken code anyway: it's wrong for msync()).
2913 * It's a rare case: affects the final partial page, for journalled data
2914 * where the file is subject to bith write() and writepage() in the same
2915 * transction. To fix it we'll need a custom block_write_full_page().
2916 * We'll probably need that anyway for journalling writepage() output.
2918 * We don't honour synchronous mounts for writepage(). That would be
2919 * disastrous. Any write() or metadata operation will sync the fs for
2923 static int __ext4_normal_writepage(struct page *page,
2924 struct writeback_control *wbc)
2926 struct inode *inode = page->mapping->host;
2928 if (test_opt(inode->i_sb, NOBH))
2929 return nobh_writepage(page,
2930 ext4_normal_get_block_write, wbc);
2932 return block_write_full_page(page,
2933 ext4_normal_get_block_write,
2937 static int ext4_normal_writepage(struct page *page,
2938 struct writeback_control *wbc)
2940 struct inode *inode = page->mapping->host;
2941 loff_t size = i_size_read(inode);
2944 trace_mark(ext4_normal_writepage,
2945 "dev %s ino %lu page_index %lu",
2946 inode->i_sb->s_id, inode->i_ino, page->index);
2947 J_ASSERT(PageLocked(page));
2948 if (page->index == size >> PAGE_CACHE_SHIFT)
2949 len = size & ~PAGE_CACHE_MASK;
2951 len = PAGE_CACHE_SIZE;
2953 if (page_has_buffers(page)) {
2954 /* if page has buffers it should all be mapped
2955 * and allocated. If there are not buffers attached
2956 * to the page we know the page is dirty but it lost
2957 * buffers. That means that at some moment in time
2958 * after write_begin() / write_end() has been called
2959 * all buffers have been clean and thus they must have been
2960 * written at least once. So they are all mapped and we can
2961 * happily proceed with mapping them and writing the page.
2963 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2964 ext4_bh_unmapped_or_delay));
2967 if (!ext4_journal_current_handle())
2968 return __ext4_normal_writepage(page, wbc);
2970 redirty_page_for_writepage(wbc, page);
2975 static int __ext4_journalled_writepage(struct page *page,
2976 struct writeback_control *wbc)
2978 struct address_space *mapping = page->mapping;
2979 struct inode *inode = mapping->host;
2980 struct buffer_head *page_bufs;
2981 handle_t *handle = NULL;
2985 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2986 ext4_normal_get_block_write);
2990 page_bufs = page_buffers(page);
2991 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2993 /* As soon as we unlock the page, it can go away, but we have
2994 * references to buffers so we are safe */
2997 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2998 if (IS_ERR(handle)) {
2999 ret = PTR_ERR(handle);
3003 ret = walk_page_buffers(handle, page_bufs, 0,
3004 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
3006 err = walk_page_buffers(handle, page_bufs, 0,
3007 PAGE_CACHE_SIZE, NULL, write_end_fn);
3010 err = ext4_journal_stop(handle);
3014 walk_page_buffers(handle, page_bufs, 0,
3015 PAGE_CACHE_SIZE, NULL, bput_one);
3016 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
3025 static int ext4_journalled_writepage(struct page *page,
3026 struct writeback_control *wbc)
3028 struct inode *inode = page->mapping->host;
3029 loff_t size = i_size_read(inode);
3032 trace_mark(ext4_journalled_writepage,
3033 "dev %s ino %lu page_index %lu",
3034 inode->i_sb->s_id, inode->i_ino, page->index);
3035 J_ASSERT(PageLocked(page));
3036 if (page->index == size >> PAGE_CACHE_SHIFT)
3037 len = size & ~PAGE_CACHE_MASK;
3039 len = PAGE_CACHE_SIZE;
3041 if (page_has_buffers(page)) {
3042 /* if page has buffers it should all be mapped
3043 * and allocated. If there are not buffers attached
3044 * to the page we know the page is dirty but it lost
3045 * buffers. That means that at some moment in time
3046 * after write_begin() / write_end() has been called
3047 * all buffers have been clean and thus they must have been
3048 * written at least once. So they are all mapped and we can
3049 * happily proceed with mapping them and writing the page.
3051 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
3052 ext4_bh_unmapped_or_delay));
3055 if (ext4_journal_current_handle())
3058 if (PageChecked(page)) {
3060 * It's mmapped pagecache. Add buffers and journal it. There
3061 * doesn't seem much point in redirtying the page here.
3063 ClearPageChecked(page);
3064 return __ext4_journalled_writepage(page, wbc);
3067 * It may be a page full of checkpoint-mode buffers. We don't
3068 * really know unless we go poke around in the buffer_heads.
3069 * But block_write_full_page will do the right thing.
3071 return block_write_full_page(page,
3072 ext4_normal_get_block_write,
3076 redirty_page_for_writepage(wbc, page);
3081 static int ext4_readpage(struct file *file, struct page *page)
3083 return mpage_readpage(page, ext4_get_block);
3087 ext4_readpages(struct file *file, struct address_space *mapping,
3088 struct list_head *pages, unsigned nr_pages)
3090 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3093 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3095 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3098 * If it's a full truncate we just forget about the pending dirtying
3101 ClearPageChecked(page);
3104 jbd2_journal_invalidatepage(journal, page, offset);
3106 block_invalidatepage(page, offset);
3109 static int ext4_releasepage(struct page *page, gfp_t wait)
3111 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3113 WARN_ON(PageChecked(page));
3114 if (!page_has_buffers(page))
3117 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3119 return try_to_free_buffers(page);
3123 * If the O_DIRECT write will extend the file then add this inode to the
3124 * orphan list. So recovery will truncate it back to the original size
3125 * if the machine crashes during the write.
3127 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3128 * crashes then stale disk data _may_ be exposed inside the file. But current
3129 * VFS code falls back into buffered path in that case so we are safe.
3131 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3132 const struct iovec *iov, loff_t offset,
3133 unsigned long nr_segs)
3135 struct file *file = iocb->ki_filp;
3136 struct inode *inode = file->f_mapping->host;
3137 struct ext4_inode_info *ei = EXT4_I(inode);
3141 size_t count = iov_length(iov, nr_segs);
3144 loff_t final_size = offset + count;
3146 if (final_size > inode->i_size) {
3147 /* Credits for sb + inode write */
3148 handle = ext4_journal_start(inode, 2);
3149 if (IS_ERR(handle)) {
3150 ret = PTR_ERR(handle);
3153 ret = ext4_orphan_add(handle, inode);
3155 ext4_journal_stop(handle);
3159 ei->i_disksize = inode->i_size;
3160 ext4_journal_stop(handle);
3164 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3166 ext4_get_block, NULL);
3171 /* Credits for sb + inode write */
3172 handle = ext4_journal_start(inode, 2);
3173 if (IS_ERR(handle)) {
3174 /* This is really bad luck. We've written the data
3175 * but cannot extend i_size. Bail out and pretend
3176 * the write failed... */
3177 ret = PTR_ERR(handle);
3181 ext4_orphan_del(handle, inode);
3183 loff_t end = offset + ret;
3184 if (end > inode->i_size) {
3185 ei->i_disksize = end;
3186 i_size_write(inode, end);
3188 * We're going to return a positive `ret'
3189 * here due to non-zero-length I/O, so there's
3190 * no way of reporting error returns from
3191 * ext4_mark_inode_dirty() to userspace. So
3194 ext4_mark_inode_dirty(handle, inode);
3197 err = ext4_journal_stop(handle);
3206 * Pages can be marked dirty completely asynchronously from ext4's journalling
3207 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3208 * much here because ->set_page_dirty is called under VFS locks. The page is
3209 * not necessarily locked.
3211 * We cannot just dirty the page and leave attached buffers clean, because the
3212 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3213 * or jbddirty because all the journalling code will explode.
3215 * So what we do is to mark the page "pending dirty" and next time writepage
3216 * is called, propagate that into the buffers appropriately.
3218 static int ext4_journalled_set_page_dirty(struct page *page)
3220 SetPageChecked(page);
3221 return __set_page_dirty_nobuffers(page);
3224 static const struct address_space_operations ext4_ordered_aops = {
3225 .readpage = ext4_readpage,
3226 .readpages = ext4_readpages,
3227 .writepage = ext4_normal_writepage,
3228 .sync_page = block_sync_page,
3229 .write_begin = ext4_write_begin,
3230 .write_end = ext4_ordered_write_end,
3232 .invalidatepage = ext4_invalidatepage,
3233 .releasepage = ext4_releasepage,
3234 .direct_IO = ext4_direct_IO,
3235 .migratepage = buffer_migrate_page,
3236 .is_partially_uptodate = block_is_partially_uptodate,
3239 static const struct address_space_operations ext4_writeback_aops = {
3240 .readpage = ext4_readpage,
3241 .readpages = ext4_readpages,
3242 .writepage = ext4_normal_writepage,
3243 .sync_page = block_sync_page,
3244 .write_begin = ext4_write_begin,
3245 .write_end = ext4_writeback_write_end,
3247 .invalidatepage = ext4_invalidatepage,
3248 .releasepage = ext4_releasepage,
3249 .direct_IO = ext4_direct_IO,
3250 .migratepage = buffer_migrate_page,
3251 .is_partially_uptodate = block_is_partially_uptodate,
3254 static const struct address_space_operations ext4_journalled_aops = {
3255 .readpage = ext4_readpage,
3256 .readpages = ext4_readpages,
3257 .writepage = ext4_journalled_writepage,
3258 .sync_page = block_sync_page,
3259 .write_begin = ext4_write_begin,
3260 .write_end = ext4_journalled_write_end,
3261 .set_page_dirty = ext4_journalled_set_page_dirty,
3263 .invalidatepage = ext4_invalidatepage,
3264 .releasepage = ext4_releasepage,
3265 .is_partially_uptodate = block_is_partially_uptodate,
3268 static const struct address_space_operations ext4_da_aops = {
3269 .readpage = ext4_readpage,
3270 .readpages = ext4_readpages,
3271 .writepage = ext4_da_writepage,
3272 .writepages = ext4_da_writepages,
3273 .sync_page = block_sync_page,
3274 .write_begin = ext4_da_write_begin,
3275 .write_end = ext4_da_write_end,
3277 .invalidatepage = ext4_da_invalidatepage,
3278 .releasepage = ext4_releasepage,
3279 .direct_IO = ext4_direct_IO,
3280 .migratepage = buffer_migrate_page,
3281 .is_partially_uptodate = block_is_partially_uptodate,
3284 void ext4_set_aops(struct inode *inode)
3286 if (ext4_should_order_data(inode) &&
3287 test_opt(inode->i_sb, DELALLOC))
3288 inode->i_mapping->a_ops = &ext4_da_aops;
3289 else if (ext4_should_order_data(inode))
3290 inode->i_mapping->a_ops = &ext4_ordered_aops;
3291 else if (ext4_should_writeback_data(inode) &&
3292 test_opt(inode->i_sb, DELALLOC))
3293 inode->i_mapping->a_ops = &ext4_da_aops;
3294 else if (ext4_should_writeback_data(inode))
3295 inode->i_mapping->a_ops = &ext4_writeback_aops;
3297 inode->i_mapping->a_ops = &ext4_journalled_aops;
3301 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3302 * up to the end of the block which corresponds to `from'.
3303 * This required during truncate. We need to physically zero the tail end
3304 * of that block so it doesn't yield old data if the file is later grown.
3306 int ext4_block_truncate_page(handle_t *handle,
3307 struct address_space *mapping, loff_t from)
3309 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3310 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3311 unsigned blocksize, length, pos;
3313 struct inode *inode = mapping->host;
3314 struct buffer_head *bh;
3318 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3322 blocksize = inode->i_sb->s_blocksize;
3323 length = blocksize - (offset & (blocksize - 1));
3324 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3327 * For "nobh" option, we can only work if we don't need to
3328 * read-in the page - otherwise we create buffers to do the IO.
3330 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3331 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3332 zero_user(page, offset, length);
3333 set_page_dirty(page);
3337 if (!page_has_buffers(page))
3338 create_empty_buffers(page, blocksize, 0);
3340 /* Find the buffer that contains "offset" */
3341 bh = page_buffers(page);
3343 while (offset >= pos) {
3344 bh = bh->b_this_page;
3350 if (buffer_freed(bh)) {
3351 BUFFER_TRACE(bh, "freed: skip");
3355 if (!buffer_mapped(bh)) {
3356 BUFFER_TRACE(bh, "unmapped");
3357 ext4_get_block(inode, iblock, bh, 0);
3358 /* unmapped? It's a hole - nothing to do */
3359 if (!buffer_mapped(bh)) {
3360 BUFFER_TRACE(bh, "still unmapped");
3365 /* Ok, it's mapped. Make sure it's up-to-date */
3366 if (PageUptodate(page))
3367 set_buffer_uptodate(bh);
3369 if (!buffer_uptodate(bh)) {
3371 ll_rw_block(READ, 1, &bh);
3373 /* Uhhuh. Read error. Complain and punt. */
3374 if (!buffer_uptodate(bh))
3378 if (ext4_should_journal_data(inode)) {
3379 BUFFER_TRACE(bh, "get write access");
3380 err = ext4_journal_get_write_access(handle, bh);
3385 zero_user(page, offset, length);
3387 BUFFER_TRACE(bh, "zeroed end of block");
3390 if (ext4_should_journal_data(inode)) {
3391 err = ext4_handle_dirty_metadata(handle, inode, bh);
3393 if (ext4_should_order_data(inode))
3394 err = ext4_jbd2_file_inode(handle, inode);
3395 mark_buffer_dirty(bh);
3400 page_cache_release(page);
3405 * Probably it should be a library function... search for first non-zero word
3406 * or memcmp with zero_page, whatever is better for particular architecture.
3409 static inline int all_zeroes(__le32 *p, __le32 *q)
3418 * ext4_find_shared - find the indirect blocks for partial truncation.
3419 * @inode: inode in question
3420 * @depth: depth of the affected branch
3421 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3422 * @chain: place to store the pointers to partial indirect blocks
3423 * @top: place to the (detached) top of branch
3425 * This is a helper function used by ext4_truncate().
3427 * When we do truncate() we may have to clean the ends of several
3428 * indirect blocks but leave the blocks themselves alive. Block is
3429 * partially truncated if some data below the new i_size is refered
3430 * from it (and it is on the path to the first completely truncated
3431 * data block, indeed). We have to free the top of that path along
3432 * with everything to the right of the path. Since no allocation
3433 * past the truncation point is possible until ext4_truncate()
3434 * finishes, we may safely do the latter, but top of branch may
3435 * require special attention - pageout below the truncation point
3436 * might try to populate it.
3438 * We atomically detach the top of branch from the tree, store the
3439 * block number of its root in *@top, pointers to buffer_heads of
3440 * partially truncated blocks - in @chain[].bh and pointers to
3441 * their last elements that should not be removed - in
3442 * @chain[].p. Return value is the pointer to last filled element
3445 * The work left to caller to do the actual freeing of subtrees:
3446 * a) free the subtree starting from *@top
3447 * b) free the subtrees whose roots are stored in
3448 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3449 * c) free the subtrees growing from the inode past the @chain[0].
3450 * (no partially truncated stuff there). */
3452 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3453 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3455 Indirect *partial, *p;
3459 /* Make k index the deepest non-null offest + 1 */
3460 for (k = depth; k > 1 && !offsets[k-1]; k--)
3462 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3463 /* Writer: pointers */
3465 partial = chain + k-1;
3467 * If the branch acquired continuation since we've looked at it -
3468 * fine, it should all survive and (new) top doesn't belong to us.
3470 if (!partial->key && *partial->p)
3473 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3476 * OK, we've found the last block that must survive. The rest of our
3477 * branch should be detached before unlocking. However, if that rest
3478 * of branch is all ours and does not grow immediately from the inode
3479 * it's easier to cheat and just decrement partial->p.
3481 if (p == chain + k - 1 && p > chain) {
3485 /* Nope, don't do this in ext4. Must leave the tree intact */
3492 while (partial > p) {
3493 brelse(partial->bh);
3501 * Zero a number of block pointers in either an inode or an indirect block.
3502 * If we restart the transaction we must again get write access to the
3503 * indirect block for further modification.
3505 * We release `count' blocks on disk, but (last - first) may be greater
3506 * than `count' because there can be holes in there.
3508 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3509 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3510 unsigned long count, __le32 *first, __le32 *last)
3513 if (try_to_extend_transaction(handle, inode)) {
3515 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3516 ext4_handle_dirty_metadata(handle, inode, bh);
3518 ext4_mark_inode_dirty(handle, inode);
3519 ext4_journal_test_restart(handle, inode);
3521 BUFFER_TRACE(bh, "retaking write access");
3522 ext4_journal_get_write_access(handle, bh);
3527 * Any buffers which are on the journal will be in memory. We find
3528 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3529 * on them. We've already detached each block from the file, so
3530 * bforget() in jbd2_journal_forget() should be safe.
3532 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3534 for (p = first; p < last; p++) {
3535 u32 nr = le32_to_cpu(*p);
3537 struct buffer_head *tbh;
3540 tbh = sb_find_get_block(inode->i_sb, nr);
3541 ext4_forget(handle, 0, inode, tbh, nr);
3545 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3549 * ext4_free_data - free a list of data blocks
3550 * @handle: handle for this transaction
3551 * @inode: inode we are dealing with
3552 * @this_bh: indirect buffer_head which contains *@first and *@last
3553 * @first: array of block numbers
3554 * @last: points immediately past the end of array
3556 * We are freeing all blocks refered from that array (numbers are stored as
3557 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3559 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3560 * blocks are contiguous then releasing them at one time will only affect one
3561 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3562 * actually use a lot of journal space.
3564 * @this_bh will be %NULL if @first and @last point into the inode's direct
3567 static void ext4_free_data(handle_t *handle, struct inode *inode,
3568 struct buffer_head *this_bh,
3569 __le32 *first, __le32 *last)
3571 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3572 unsigned long count = 0; /* Number of blocks in the run */
3573 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3576 ext4_fsblk_t nr; /* Current block # */
3577 __le32 *p; /* Pointer into inode/ind
3578 for current block */
3581 if (this_bh) { /* For indirect block */
3582 BUFFER_TRACE(this_bh, "get_write_access");
3583 err = ext4_journal_get_write_access(handle, this_bh);
3584 /* Important: if we can't update the indirect pointers
3585 * to the blocks, we can't free them. */
3590 for (p = first; p < last; p++) {
3591 nr = le32_to_cpu(*p);
3593 /* accumulate blocks to free if they're contiguous */
3596 block_to_free_p = p;
3598 } else if (nr == block_to_free + count) {
3601 ext4_clear_blocks(handle, inode, this_bh,
3603 count, block_to_free_p, p);
3605 block_to_free_p = p;
3612 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3613 count, block_to_free_p, p);
3616 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3619 * The buffer head should have an attached journal head at this
3620 * point. However, if the data is corrupted and an indirect
3621 * block pointed to itself, it would have been detached when
3622 * the block was cleared. Check for this instead of OOPSing.
3624 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
3625 ext4_handle_dirty_metadata(handle, inode, this_bh);
3627 ext4_error(inode->i_sb, __func__,
3628 "circular indirect block detected, "
3629 "inode=%lu, block=%llu",
3631 (unsigned long long) this_bh->b_blocknr);
3636 * ext4_free_branches - free an array of branches
3637 * @handle: JBD handle for this transaction
3638 * @inode: inode we are dealing with
3639 * @parent_bh: the buffer_head which contains *@first and *@last
3640 * @first: array of block numbers
3641 * @last: pointer immediately past the end of array
3642 * @depth: depth of the branches to free
3644 * We are freeing all blocks refered from these branches (numbers are
3645 * stored as little-endian 32-bit) and updating @inode->i_blocks
3648 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3649 struct buffer_head *parent_bh,
3650 __le32 *first, __le32 *last, int depth)
3655 if (ext4_handle_is_aborted(handle))
3659 struct buffer_head *bh;
3660 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3662 while (--p >= first) {
3663 nr = le32_to_cpu(*p);
3665 continue; /* A hole */
3667 /* Go read the buffer for the next level down */
3668 bh = sb_bread(inode->i_sb, nr);
3671 * A read failure? Report error and clear slot
3675 ext4_error(inode->i_sb, "ext4_free_branches",
3676 "Read failure, inode=%lu, block=%llu",
3681 /* This zaps the entire block. Bottom up. */
3682 BUFFER_TRACE(bh, "free child branches");
3683 ext4_free_branches(handle, inode, bh,
3684 (__le32 *) bh->b_data,
3685 (__le32 *) bh->b_data + addr_per_block,
3689 * We've probably journalled the indirect block several
3690 * times during the truncate. But it's no longer
3691 * needed and we now drop it from the transaction via
3692 * jbd2_journal_revoke().
3694 * That's easy if it's exclusively part of this
3695 * transaction. But if it's part of the committing
3696 * transaction then jbd2_journal_forget() will simply
3697 * brelse() it. That means that if the underlying
3698 * block is reallocated in ext4_get_block(),
3699 * unmap_underlying_metadata() will find this block
3700 * and will try to get rid of it. damn, damn.
3702 * If this block has already been committed to the
3703 * journal, a revoke record will be written. And
3704 * revoke records must be emitted *before* clearing
3705 * this block's bit in the bitmaps.
3707 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3710 * Everything below this this pointer has been
3711 * released. Now let this top-of-subtree go.
3713 * We want the freeing of this indirect block to be
3714 * atomic in the journal with the updating of the
3715 * bitmap block which owns it. So make some room in
3718 * We zero the parent pointer *after* freeing its
3719 * pointee in the bitmaps, so if extend_transaction()
3720 * for some reason fails to put the bitmap changes and
3721 * the release into the same transaction, recovery
3722 * will merely complain about releasing a free block,
3723 * rather than leaking blocks.
3725 if (ext4_handle_is_aborted(handle))
3727 if (try_to_extend_transaction(handle, inode)) {
3728 ext4_mark_inode_dirty(handle, inode);
3729 ext4_journal_test_restart(handle, inode);
3732 ext4_free_blocks(handle, inode, nr, 1, 1);
3736 * The block which we have just freed is
3737 * pointed to by an indirect block: journal it
3739 BUFFER_TRACE(parent_bh, "get_write_access");
3740 if (!ext4_journal_get_write_access(handle,
3743 BUFFER_TRACE(parent_bh,
3744 "call ext4_handle_dirty_metadata");
3745 ext4_handle_dirty_metadata(handle,
3752 /* We have reached the bottom of the tree. */
3753 BUFFER_TRACE(parent_bh, "free data blocks");
3754 ext4_free_data(handle, inode, parent_bh, first, last);
3758 int ext4_can_truncate(struct inode *inode)
3760 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3762 if (S_ISREG(inode->i_mode))
3764 if (S_ISDIR(inode->i_mode))
3766 if (S_ISLNK(inode->i_mode))
3767 return !ext4_inode_is_fast_symlink(inode);
3774 * We block out ext4_get_block() block instantiations across the entire
3775 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3776 * simultaneously on behalf of the same inode.
3778 * As we work through the truncate and commmit bits of it to the journal there
3779 * is one core, guiding principle: the file's tree must always be consistent on
3780 * disk. We must be able to restart the truncate after a crash.
3782 * The file's tree may be transiently inconsistent in memory (although it
3783 * probably isn't), but whenever we close off and commit a journal transaction,
3784 * the contents of (the filesystem + the journal) must be consistent and
3785 * restartable. It's pretty simple, really: bottom up, right to left (although
3786 * left-to-right works OK too).
3788 * Note that at recovery time, journal replay occurs *before* the restart of
3789 * truncate against the orphan inode list.
3791 * The committed inode has the new, desired i_size (which is the same as
3792 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3793 * that this inode's truncate did not complete and it will again call
3794 * ext4_truncate() to have another go. So there will be instantiated blocks
3795 * to the right of the truncation point in a crashed ext4 filesystem. But
3796 * that's fine - as long as they are linked from the inode, the post-crash
3797 * ext4_truncate() run will find them and release them.
3799 void ext4_truncate(struct inode *inode)
3802 struct ext4_inode_info *ei = EXT4_I(inode);
3803 __le32 *i_data = ei->i_data;
3804 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3805 struct address_space *mapping = inode->i_mapping;
3806 ext4_lblk_t offsets[4];
3811 ext4_lblk_t last_block;
3812 unsigned blocksize = inode->i_sb->s_blocksize;
3814 if (!ext4_can_truncate(inode))
3817 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3818 ext4_ext_truncate(inode);
3822 handle = start_transaction(inode);
3824 return; /* AKPM: return what? */
3826 last_block = (inode->i_size + blocksize-1)
3827 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3829 if (inode->i_size & (blocksize - 1))
3830 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3833 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3835 goto out_stop; /* error */
3838 * OK. This truncate is going to happen. We add the inode to the
3839 * orphan list, so that if this truncate spans multiple transactions,
3840 * and we crash, we will resume the truncate when the filesystem
3841 * recovers. It also marks the inode dirty, to catch the new size.
3843 * Implication: the file must always be in a sane, consistent
3844 * truncatable state while each transaction commits.
3846 if (ext4_orphan_add(handle, inode))
3850 * From here we block out all ext4_get_block() callers who want to
3851 * modify the block allocation tree.
3853 down_write(&ei->i_data_sem);
3855 ext4_discard_preallocations(inode);
3858 * The orphan list entry will now protect us from any crash which
3859 * occurs before the truncate completes, so it is now safe to propagate
3860 * the new, shorter inode size (held for now in i_size) into the
3861 * on-disk inode. We do this via i_disksize, which is the value which
3862 * ext4 *really* writes onto the disk inode.
3864 ei->i_disksize = inode->i_size;
3866 if (n == 1) { /* direct blocks */
3867 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3868 i_data + EXT4_NDIR_BLOCKS);
3872 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3873 /* Kill the top of shared branch (not detached) */
3875 if (partial == chain) {
3876 /* Shared branch grows from the inode */
3877 ext4_free_branches(handle, inode, NULL,
3878 &nr, &nr+1, (chain+n-1) - partial);
3881 * We mark the inode dirty prior to restart,
3882 * and prior to stop. No need for it here.
3885 /* Shared branch grows from an indirect block */
3886 BUFFER_TRACE(partial->bh, "get_write_access");
3887 ext4_free_branches(handle, inode, partial->bh,
3889 partial->p+1, (chain+n-1) - partial);
3892 /* Clear the ends of indirect blocks on the shared branch */
3893 while (partial > chain) {
3894 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3895 (__le32*)partial->bh->b_data+addr_per_block,
3896 (chain+n-1) - partial);
3897 BUFFER_TRACE(partial->bh, "call brelse");
3898 brelse (partial->bh);
3902 /* Kill the remaining (whole) subtrees */
3903 switch (offsets[0]) {
3905 nr = i_data[EXT4_IND_BLOCK];
3907 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3908 i_data[EXT4_IND_BLOCK] = 0;
3910 case EXT4_IND_BLOCK:
3911 nr = i_data[EXT4_DIND_BLOCK];
3913 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3914 i_data[EXT4_DIND_BLOCK] = 0;
3916 case EXT4_DIND_BLOCK:
3917 nr = i_data[EXT4_TIND_BLOCK];
3919 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3920 i_data[EXT4_TIND_BLOCK] = 0;
3922 case EXT4_TIND_BLOCK:
3926 up_write(&ei->i_data_sem);
3927 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3928 ext4_mark_inode_dirty(handle, inode);
3931 * In a multi-transaction truncate, we only make the final transaction
3935 ext4_handle_sync(handle);
3938 * If this was a simple ftruncate(), and the file will remain alive
3939 * then we need to clear up the orphan record which we created above.
3940 * However, if this was a real unlink then we were called by
3941 * ext4_delete_inode(), and we allow that function to clean up the
3942 * orphan info for us.
3945 ext4_orphan_del(handle, inode);
3947 ext4_journal_stop(handle);
3951 * ext4_get_inode_loc returns with an extra refcount against the inode's
3952 * underlying buffer_head on success. If 'in_mem' is true, we have all
3953 * data in memory that is needed to recreate the on-disk version of this
3956 static int __ext4_get_inode_loc(struct inode *inode,
3957 struct ext4_iloc *iloc, int in_mem)
3959 struct ext4_group_desc *gdp;
3960 struct buffer_head *bh;
3961 struct super_block *sb = inode->i_sb;
3963 int inodes_per_block, inode_offset;
3966 if (!ext4_valid_inum(sb, inode->i_ino))
3969 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3970 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3975 * Figure out the offset within the block group inode table
3977 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3978 inode_offset = ((inode->i_ino - 1) %
3979 EXT4_INODES_PER_GROUP(sb));
3980 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3981 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3983 bh = sb_getblk(sb, block);
3985 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3986 "inode block - inode=%lu, block=%llu",
3987 inode->i_ino, block);
3990 if (!buffer_uptodate(bh)) {
3994 * If the buffer has the write error flag, we have failed
3995 * to write out another inode in the same block. In this
3996 * case, we don't have to read the block because we may
3997 * read the old inode data successfully.
3999 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4000 set_buffer_uptodate(bh);
4002 if (buffer_uptodate(bh)) {
4003 /* someone brought it uptodate while we waited */
4009 * If we have all information of the inode in memory and this
4010 * is the only valid inode in the block, we need not read the
4014 struct buffer_head *bitmap_bh;
4017 start = inode_offset & ~(inodes_per_block - 1);
4019 /* Is the inode bitmap in cache? */
4020 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4025 * If the inode bitmap isn't in cache then the
4026 * optimisation may end up performing two reads instead
4027 * of one, so skip it.
4029 if (!buffer_uptodate(bitmap_bh)) {
4033 for (i = start; i < start + inodes_per_block; i++) {
4034 if (i == inode_offset)
4036 if (ext4_test_bit(i, bitmap_bh->b_data))
4040 if (i == start + inodes_per_block) {
4041 /* all other inodes are free, so skip I/O */
4042 memset(bh->b_data, 0, bh->b_size);
4043 set_buffer_uptodate(bh);
4051 * If we need to do any I/O, try to pre-readahead extra
4052 * blocks from the inode table.
4054 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4055 ext4_fsblk_t b, end, table;
4058 table = ext4_inode_table(sb, gdp);
4059 /* Make sure s_inode_readahead_blks is a power of 2 */
4060 while (EXT4_SB(sb)->s_inode_readahead_blks &
4061 (EXT4_SB(sb)->s_inode_readahead_blks-1))
4062 EXT4_SB(sb)->s_inode_readahead_blks =
4063 (EXT4_SB(sb)->s_inode_readahead_blks &
4064 (EXT4_SB(sb)->s_inode_readahead_blks-1));
4065 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4068 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4069 num = EXT4_INODES_PER_GROUP(sb);
4070 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4071 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4072 num -= ext4_itable_unused_count(sb, gdp);
4073 table += num / inodes_per_block;
4077 sb_breadahead(sb, b++);
4081 * There are other valid inodes in the buffer, this inode
4082 * has in-inode xattrs, or we don't have this inode in memory.
4083 * Read the block from disk.
4086 bh->b_end_io = end_buffer_read_sync;
4087 submit_bh(READ_META, bh);
4089 if (!buffer_uptodate(bh)) {
4090 ext4_error(sb, __func__,
4091 "unable to read inode block - inode=%lu, "
4092 "block=%llu", inode->i_ino, block);
4102 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4104 /* We have all inode data except xattrs in memory here. */
4105 return __ext4_get_inode_loc(inode, iloc,
4106 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4109 void ext4_set_inode_flags(struct inode *inode)
4111 unsigned int flags = EXT4_I(inode)->i_flags;
4113 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4114 if (flags & EXT4_SYNC_FL)
4115 inode->i_flags |= S_SYNC;
4116 if (flags & EXT4_APPEND_FL)
4117 inode->i_flags |= S_APPEND;
4118 if (flags & EXT4_IMMUTABLE_FL)
4119 inode->i_flags |= S_IMMUTABLE;
4120 if (flags & EXT4_NOATIME_FL)
4121 inode->i_flags |= S_NOATIME;
4122 if (flags & EXT4_DIRSYNC_FL)
4123 inode->i_flags |= S_DIRSYNC;
4126 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4127 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4129 unsigned int flags = ei->vfs_inode.i_flags;
4131 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4132 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4134 ei->i_flags |= EXT4_SYNC_FL;
4135 if (flags & S_APPEND)
4136 ei->i_flags |= EXT4_APPEND_FL;
4137 if (flags & S_IMMUTABLE)
4138 ei->i_flags |= EXT4_IMMUTABLE_FL;
4139 if (flags & S_NOATIME)
4140 ei->i_flags |= EXT4_NOATIME_FL;
4141 if (flags & S_DIRSYNC)
4142 ei->i_flags |= EXT4_DIRSYNC_FL;
4144 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4145 struct ext4_inode_info *ei)
4148 struct inode *inode = &(ei->vfs_inode);
4149 struct super_block *sb = inode->i_sb;
4151 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4152 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4153 /* we are using combined 48 bit field */
4154 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4155 le32_to_cpu(raw_inode->i_blocks_lo);
4156 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4157 /* i_blocks represent file system block size */
4158 return i_blocks << (inode->i_blkbits - 9);
4163 return le32_to_cpu(raw_inode->i_blocks_lo);
4167 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4169 struct ext4_iloc iloc;
4170 struct ext4_inode *raw_inode;
4171 struct ext4_inode_info *ei;
4172 struct buffer_head *bh;
4173 struct inode *inode;
4177 inode = iget_locked(sb, ino);
4179 return ERR_PTR(-ENOMEM);
4180 if (!(inode->i_state & I_NEW))
4184 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4185 ei->i_acl = EXT4_ACL_NOT_CACHED;
4186 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4189 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4193 raw_inode = ext4_raw_inode(&iloc);
4194 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4195 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4196 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4197 if (!(test_opt(inode->i_sb, NO_UID32))) {
4198 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4199 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4201 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4204 ei->i_dir_start_lookup = 0;
4205 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4206 /* We now have enough fields to check if the inode was active or not.
4207 * This is needed because nfsd might try to access dead inodes
4208 * the test is that same one that e2fsck uses
4209 * NeilBrown 1999oct15
4211 if (inode->i_nlink == 0) {
4212 if (inode->i_mode == 0 ||
4213 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4214 /* this inode is deleted */
4219 /* The only unlinked inodes we let through here have
4220 * valid i_mode and are being read by the orphan
4221 * recovery code: that's fine, we're about to complete
4222 * the process of deleting those. */
4224 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4225 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4226 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4227 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4228 cpu_to_le32(EXT4_OS_HURD)) {
4230 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4232 inode->i_size = ext4_isize(raw_inode);
4233 ei->i_disksize = inode->i_size;
4234 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4235 ei->i_block_group = iloc.block_group;
4237 * NOTE! The in-memory inode i_data array is in little-endian order
4238 * even on big-endian machines: we do NOT byteswap the block numbers!
4240 for (block = 0; block < EXT4_N_BLOCKS; block++)
4241 ei->i_data[block] = raw_inode->i_block[block];
4242 INIT_LIST_HEAD(&ei->i_orphan);
4244 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4245 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4246 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4247 EXT4_INODE_SIZE(inode->i_sb)) {
4252 if (ei->i_extra_isize == 0) {
4253 /* The extra space is currently unused. Use it. */
4254 ei->i_extra_isize = sizeof(struct ext4_inode) -
4255 EXT4_GOOD_OLD_INODE_SIZE;
4257 __le32 *magic = (void *)raw_inode +
4258 EXT4_GOOD_OLD_INODE_SIZE +
4260 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4261 ei->i_state |= EXT4_STATE_XATTR;
4264 ei->i_extra_isize = 0;
4266 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4267 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4268 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4269 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4271 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4272 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4273 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4275 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4278 if (S_ISREG(inode->i_mode)) {
4279 inode->i_op = &ext4_file_inode_operations;
4280 inode->i_fop = &ext4_file_operations;
4281 ext4_set_aops(inode);
4282 } else if (S_ISDIR(inode->i_mode)) {
4283 inode->i_op = &ext4_dir_inode_operations;
4284 inode->i_fop = &ext4_dir_operations;
4285 } else if (S_ISLNK(inode->i_mode)) {
4286 if (ext4_inode_is_fast_symlink(inode)) {
4287 inode->i_op = &ext4_fast_symlink_inode_operations;
4288 nd_terminate_link(ei->i_data, inode->i_size,
4289 sizeof(ei->i_data) - 1);
4291 inode->i_op = &ext4_symlink_inode_operations;
4292 ext4_set_aops(inode);
4295 inode->i_op = &ext4_special_inode_operations;
4296 if (raw_inode->i_block[0])
4297 init_special_inode(inode, inode->i_mode,
4298 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4300 init_special_inode(inode, inode->i_mode,
4301 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4304 ext4_set_inode_flags(inode);
4305 unlock_new_inode(inode);
4310 return ERR_PTR(ret);
4313 static int ext4_inode_blocks_set(handle_t *handle,
4314 struct ext4_inode *raw_inode,
4315 struct ext4_inode_info *ei)
4317 struct inode *inode = &(ei->vfs_inode);
4318 u64 i_blocks = inode->i_blocks;
4319 struct super_block *sb = inode->i_sb;
4321 if (i_blocks <= ~0U) {
4323 * i_blocks can be represnted in a 32 bit variable
4324 * as multiple of 512 bytes
4326 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4327 raw_inode->i_blocks_high = 0;
4328 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4331 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4334 if (i_blocks <= 0xffffffffffffULL) {
4336 * i_blocks can be represented in a 48 bit variable
4337 * as multiple of 512 bytes
4339 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4340 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4341 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4343 ei->i_flags |= EXT4_HUGE_FILE_FL;
4344 /* i_block is stored in file system block size */
4345 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4346 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4347 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4353 * Post the struct inode info into an on-disk inode location in the
4354 * buffer-cache. This gobbles the caller's reference to the
4355 * buffer_head in the inode location struct.
4357 * The caller must have write access to iloc->bh.
4359 static int ext4_do_update_inode(handle_t *handle,
4360 struct inode *inode,
4361 struct ext4_iloc *iloc)
4363 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4364 struct ext4_inode_info *ei = EXT4_I(inode);
4365 struct buffer_head *bh = iloc->bh;
4366 int err = 0, rc, block;
4368 /* For fields not not tracking in the in-memory inode,
4369 * initialise them to zero for new inodes. */
4370 if (ei->i_state & EXT4_STATE_NEW)
4371 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4373 ext4_get_inode_flags(ei);
4374 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4375 if (!(test_opt(inode->i_sb, NO_UID32))) {
4376 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4377 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4379 * Fix up interoperability with old kernels. Otherwise, old inodes get
4380 * re-used with the upper 16 bits of the uid/gid intact
4383 raw_inode->i_uid_high =
4384 cpu_to_le16(high_16_bits(inode->i_uid));
4385 raw_inode->i_gid_high =
4386 cpu_to_le16(high_16_bits(inode->i_gid));
4388 raw_inode->i_uid_high = 0;
4389 raw_inode->i_gid_high = 0;
4392 raw_inode->i_uid_low =
4393 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4394 raw_inode->i_gid_low =
4395 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4396 raw_inode->i_uid_high = 0;
4397 raw_inode->i_gid_high = 0;
4399 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4401 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4402 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4403 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4404 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4406 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4408 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4409 /* clear the migrate flag in the raw_inode */
4410 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4411 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4412 cpu_to_le32(EXT4_OS_HURD))
4413 raw_inode->i_file_acl_high =
4414 cpu_to_le16(ei->i_file_acl >> 32);
4415 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4416 ext4_isize_set(raw_inode, ei->i_disksize);
4417 if (ei->i_disksize > 0x7fffffffULL) {
4418 struct super_block *sb = inode->i_sb;
4419 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4420 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4421 EXT4_SB(sb)->s_es->s_rev_level ==
4422 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4423 /* If this is the first large file
4424 * created, add a flag to the superblock.
4426 err = ext4_journal_get_write_access(handle,
4427 EXT4_SB(sb)->s_sbh);
4430 ext4_update_dynamic_rev(sb);
4431 EXT4_SET_RO_COMPAT_FEATURE(sb,
4432 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4434 ext4_handle_sync(handle);
4435 err = ext4_handle_dirty_metadata(handle, inode,
4436 EXT4_SB(sb)->s_sbh);
4439 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4440 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4441 if (old_valid_dev(inode->i_rdev)) {
4442 raw_inode->i_block[0] =
4443 cpu_to_le32(old_encode_dev(inode->i_rdev));
4444 raw_inode->i_block[1] = 0;
4446 raw_inode->i_block[0] = 0;
4447 raw_inode->i_block[1] =
4448 cpu_to_le32(new_encode_dev(inode->i_rdev));
4449 raw_inode->i_block[2] = 0;
4451 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4452 raw_inode->i_block[block] = ei->i_data[block];
4454 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4455 if (ei->i_extra_isize) {
4456 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4457 raw_inode->i_version_hi =
4458 cpu_to_le32(inode->i_version >> 32);
4459 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4462 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4463 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4466 ei->i_state &= ~EXT4_STATE_NEW;
4470 ext4_std_error(inode->i_sb, err);
4475 * ext4_write_inode()
4477 * We are called from a few places:
4479 * - Within generic_file_write() for O_SYNC files.
4480 * Here, there will be no transaction running. We wait for any running
4481 * trasnaction to commit.
4483 * - Within sys_sync(), kupdate and such.
4484 * We wait on commit, if tol to.
4486 * - Within prune_icache() (PF_MEMALLOC == true)
4487 * Here we simply return. We can't afford to block kswapd on the
4490 * In all cases it is actually safe for us to return without doing anything,
4491 * because the inode has been copied into a raw inode buffer in
4492 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4495 * Note that we are absolutely dependent upon all inode dirtiers doing the
4496 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4497 * which we are interested.
4499 * It would be a bug for them to not do this. The code:
4501 * mark_inode_dirty(inode)
4503 * inode->i_size = expr;
4505 * is in error because a kswapd-driven write_inode() could occur while
4506 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4507 * will no longer be on the superblock's dirty inode list.
4509 int ext4_write_inode(struct inode *inode, int wait)
4511 if (current->flags & PF_MEMALLOC)
4514 if (ext4_journal_current_handle()) {
4515 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4523 return ext4_force_commit(inode->i_sb);
4526 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4530 mark_buffer_dirty(bh);
4531 if (inode && inode_needs_sync(inode)) {
4532 sync_dirty_buffer(bh);
4533 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4534 ext4_error(inode->i_sb, __func__,
4535 "IO error syncing inode, "
4536 "inode=%lu, block=%llu",
4538 (unsigned long long)bh->b_blocknr);
4548 * Called from notify_change.
4550 * We want to trap VFS attempts to truncate the file as soon as
4551 * possible. In particular, we want to make sure that when the VFS
4552 * shrinks i_size, we put the inode on the orphan list and modify
4553 * i_disksize immediately, so that during the subsequent flushing of
4554 * dirty pages and freeing of disk blocks, we can guarantee that any
4555 * commit will leave the blocks being flushed in an unused state on
4556 * disk. (On recovery, the inode will get truncated and the blocks will
4557 * be freed, so we have a strong guarantee that no future commit will
4558 * leave these blocks visible to the user.)
4560 * Another thing we have to assure is that if we are in ordered mode
4561 * and inode is still attached to the committing transaction, we must
4562 * we start writeout of all the dirty pages which are being truncated.
4563 * This way we are sure that all the data written in the previous
4564 * transaction are already on disk (truncate waits for pages under
4567 * Called with inode->i_mutex down.
4569 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4571 struct inode *inode = dentry->d_inode;
4573 const unsigned int ia_valid = attr->ia_valid;
4575 error = inode_change_ok(inode, attr);
4579 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4580 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4583 /* (user+group)*(old+new) structure, inode write (sb,
4584 * inode block, ? - but truncate inode update has it) */
4585 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4586 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4587 if (IS_ERR(handle)) {
4588 error = PTR_ERR(handle);
4591 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4593 ext4_journal_stop(handle);
4596 /* Update corresponding info in inode so that everything is in
4597 * one transaction */
4598 if (attr->ia_valid & ATTR_UID)
4599 inode->i_uid = attr->ia_uid;
4600 if (attr->ia_valid & ATTR_GID)
4601 inode->i_gid = attr->ia_gid;
4602 error = ext4_mark_inode_dirty(handle, inode);
4603 ext4_journal_stop(handle);
4606 if (attr->ia_valid & ATTR_SIZE) {
4607 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4608 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4610 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4617 if (S_ISREG(inode->i_mode) &&
4618 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4621 handle = ext4_journal_start(inode, 3);
4622 if (IS_ERR(handle)) {
4623 error = PTR_ERR(handle);
4627 error = ext4_orphan_add(handle, inode);
4628 EXT4_I(inode)->i_disksize = attr->ia_size;
4629 rc = ext4_mark_inode_dirty(handle, inode);
4632 ext4_journal_stop(handle);
4634 if (ext4_should_order_data(inode)) {
4635 error = ext4_begin_ordered_truncate(inode,
4638 /* Do as much error cleanup as possible */
4639 handle = ext4_journal_start(inode, 3);
4640 if (IS_ERR(handle)) {
4641 ext4_orphan_del(NULL, inode);
4644 ext4_orphan_del(handle, inode);
4645 ext4_journal_stop(handle);
4651 rc = inode_setattr(inode, attr);
4653 /* If inode_setattr's call to ext4_truncate failed to get a
4654 * transaction handle at all, we need to clean up the in-core
4655 * orphan list manually. */
4657 ext4_orphan_del(NULL, inode);
4659 if (!rc && (ia_valid & ATTR_MODE))
4660 rc = ext4_acl_chmod(inode);
4663 ext4_std_error(inode->i_sb, error);
4669 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4672 struct inode *inode;
4673 unsigned long delalloc_blocks;
4675 inode = dentry->d_inode;
4676 generic_fillattr(inode, stat);
4679 * We can't update i_blocks if the block allocation is delayed
4680 * otherwise in the case of system crash before the real block
4681 * allocation is done, we will have i_blocks inconsistent with
4682 * on-disk file blocks.
4683 * We always keep i_blocks updated together with real
4684 * allocation. But to not confuse with user, stat
4685 * will return the blocks that include the delayed allocation
4686 * blocks for this file.
4688 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4689 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4690 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4692 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4696 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4701 /* if nrblocks are contiguous */
4704 * With N contiguous data blocks, it need at most
4705 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4706 * 2 dindirect blocks
4709 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4710 return indirects + 3;
4713 * if nrblocks are not contiguous, worse case, each block touch
4714 * a indirect block, and each indirect block touch a double indirect
4715 * block, plus a triple indirect block
4717 indirects = nrblocks * 2 + 1;
4721 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4723 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4724 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4725 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4729 * Account for index blocks, block groups bitmaps and block group
4730 * descriptor blocks if modify datablocks and index blocks
4731 * worse case, the indexs blocks spread over different block groups
4733 * If datablocks are discontiguous, they are possible to spread over
4734 * different block groups too. If they are contiugous, with flexbg,
4735 * they could still across block group boundary.
4737 * Also account for superblock, inode, quota and xattr blocks
4739 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4741 int groups, gdpblocks;
4746 * How many index blocks need to touch to modify nrblocks?
4747 * The "Chunk" flag indicating whether the nrblocks is
4748 * physically contiguous on disk
4750 * For Direct IO and fallocate, they calls get_block to allocate
4751 * one single extent at a time, so they could set the "Chunk" flag
4753 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4758 * Now let's see how many group bitmaps and group descriptors need
4768 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4769 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4770 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4771 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4773 /* bitmaps and block group descriptor blocks */
4774 ret += groups + gdpblocks;
4776 /* Blocks for super block, inode, quota and xattr blocks */
4777 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4783 * Calulate the total number of credits to reserve to fit
4784 * the modification of a single pages into a single transaction,
4785 * which may include multiple chunks of block allocations.
4787 * This could be called via ext4_write_begin()
4789 * We need to consider the worse case, when
4790 * one new block per extent.
4792 int ext4_writepage_trans_blocks(struct inode *inode)
4794 int bpp = ext4_journal_blocks_per_page(inode);
4797 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4799 /* Account for data blocks for journalled mode */
4800 if (ext4_should_journal_data(inode))
4806 * Calculate the journal credits for a chunk of data modification.
4808 * This is called from DIO, fallocate or whoever calling
4809 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4811 * journal buffers for data blocks are not included here, as DIO
4812 * and fallocate do no need to journal data buffers.
4814 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4816 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4820 * The caller must have previously called ext4_reserve_inode_write().
4821 * Give this, we know that the caller already has write access to iloc->bh.
4823 int ext4_mark_iloc_dirty(handle_t *handle,
4824 struct inode *inode, struct ext4_iloc *iloc)
4828 if (test_opt(inode->i_sb, I_VERSION))
4829 inode_inc_iversion(inode);
4831 /* the do_update_inode consumes one bh->b_count */
4834 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4835 err = ext4_do_update_inode(handle, inode, iloc);
4841 * On success, We end up with an outstanding reference count against
4842 * iloc->bh. This _must_ be cleaned up later.
4846 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4847 struct ext4_iloc *iloc)
4851 err = ext4_get_inode_loc(inode, iloc);
4853 BUFFER_TRACE(iloc->bh, "get_write_access");
4854 err = ext4_journal_get_write_access(handle, iloc->bh);
4860 ext4_std_error(inode->i_sb, err);
4865 * Expand an inode by new_extra_isize bytes.
4866 * Returns 0 on success or negative error number on failure.
4868 static int ext4_expand_extra_isize(struct inode *inode,
4869 unsigned int new_extra_isize,
4870 struct ext4_iloc iloc,
4873 struct ext4_inode *raw_inode;
4874 struct ext4_xattr_ibody_header *header;
4875 struct ext4_xattr_entry *entry;
4877 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4880 raw_inode = ext4_raw_inode(&iloc);
4882 header = IHDR(inode, raw_inode);
4883 entry = IFIRST(header);
4885 /* No extended attributes present */
4886 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4887 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4888 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4890 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4894 /* try to expand with EAs present */
4895 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4900 * What we do here is to mark the in-core inode as clean with respect to inode
4901 * dirtiness (it may still be data-dirty).
4902 * This means that the in-core inode may be reaped by prune_icache
4903 * without having to perform any I/O. This is a very good thing,
4904 * because *any* task may call prune_icache - even ones which
4905 * have a transaction open against a different journal.
4907 * Is this cheating? Not really. Sure, we haven't written the
4908 * inode out, but prune_icache isn't a user-visible syncing function.
4909 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4910 * we start and wait on commits.
4912 * Is this efficient/effective? Well, we're being nice to the system
4913 * by cleaning up our inodes proactively so they can be reaped
4914 * without I/O. But we are potentially leaving up to five seconds'
4915 * worth of inodes floating about which prune_icache wants us to
4916 * write out. One way to fix that would be to get prune_icache()
4917 * to do a write_super() to free up some memory. It has the desired
4920 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4922 struct ext4_iloc iloc;
4923 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4924 static unsigned int mnt_count;
4928 err = ext4_reserve_inode_write(handle, inode, &iloc);
4929 if (ext4_handle_valid(handle) &&
4930 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4931 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4933 * We need extra buffer credits since we may write into EA block
4934 * with this same handle. If journal_extend fails, then it will
4935 * only result in a minor loss of functionality for that inode.
4936 * If this is felt to be critical, then e2fsck should be run to
4937 * force a large enough s_min_extra_isize.
4939 if ((jbd2_journal_extend(handle,
4940 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4941 ret = ext4_expand_extra_isize(inode,
4942 sbi->s_want_extra_isize,
4945 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4947 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4948 ext4_warning(inode->i_sb, __func__,
4949 "Unable to expand inode %lu. Delete"
4950 " some EAs or run e2fsck.",
4953 le16_to_cpu(sbi->s_es->s_mnt_count);
4959 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4964 * ext4_dirty_inode() is called from __mark_inode_dirty()
4966 * We're really interested in the case where a file is being extended.
4967 * i_size has been changed by generic_commit_write() and we thus need
4968 * to include the updated inode in the current transaction.
4970 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4971 * are allocated to the file.
4973 * If the inode is marked synchronous, we don't honour that here - doing
4974 * so would cause a commit on atime updates, which we don't bother doing.
4975 * We handle synchronous inodes at the highest possible level.
4977 void ext4_dirty_inode(struct inode *inode)
4979 handle_t *current_handle = ext4_journal_current_handle();
4982 if (!ext4_handle_valid(current_handle)) {
4983 ext4_mark_inode_dirty(current_handle, inode);
4987 handle = ext4_journal_start(inode, 2);
4990 if (current_handle &&
4991 current_handle->h_transaction != handle->h_transaction) {
4992 /* This task has a transaction open against a different fs */
4993 printk(KERN_EMERG "%s: transactions do not match!\n",
4996 jbd_debug(5, "marking dirty. outer handle=%p\n",
4998 ext4_mark_inode_dirty(handle, inode);
5000 ext4_journal_stop(handle);
5007 * Bind an inode's backing buffer_head into this transaction, to prevent
5008 * it from being flushed to disk early. Unlike
5009 * ext4_reserve_inode_write, this leaves behind no bh reference and
5010 * returns no iloc structure, so the caller needs to repeat the iloc
5011 * lookup to mark the inode dirty later.
5013 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5015 struct ext4_iloc iloc;
5019 err = ext4_get_inode_loc(inode, &iloc);
5021 BUFFER_TRACE(iloc.bh, "get_write_access");
5022 err = jbd2_journal_get_write_access(handle, iloc.bh);
5024 err = ext4_handle_dirty_metadata(handle,
5030 ext4_std_error(inode->i_sb, err);
5035 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5042 * We have to be very careful here: changing a data block's
5043 * journaling status dynamically is dangerous. If we write a
5044 * data block to the journal, change the status and then delete
5045 * that block, we risk forgetting to revoke the old log record
5046 * from the journal and so a subsequent replay can corrupt data.
5047 * So, first we make sure that the journal is empty and that
5048 * nobody is changing anything.
5051 journal = EXT4_JOURNAL(inode);
5054 if (is_journal_aborted(journal))
5057 jbd2_journal_lock_updates(journal);
5058 jbd2_journal_flush(journal);
5061 * OK, there are no updates running now, and all cached data is
5062 * synced to disk. We are now in a completely consistent state
5063 * which doesn't have anything in the journal, and we know that
5064 * no filesystem updates are running, so it is safe to modify
5065 * the inode's in-core data-journaling state flag now.
5069 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5071 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5072 ext4_set_aops(inode);
5074 jbd2_journal_unlock_updates(journal);
5076 /* Finally we can mark the inode as dirty. */
5078 handle = ext4_journal_start(inode, 1);
5080 return PTR_ERR(handle);
5082 err = ext4_mark_inode_dirty(handle, inode);
5083 ext4_handle_sync(handle);
5084 ext4_journal_stop(handle);
5085 ext4_std_error(inode->i_sb, err);
5090 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5092 return !buffer_mapped(bh);
5095 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
5101 struct file *file = vma->vm_file;
5102 struct inode *inode = file->f_path.dentry->d_inode;
5103 struct address_space *mapping = inode->i_mapping;
5106 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5107 * get i_mutex because we are already holding mmap_sem.
5109 down_read(&inode->i_alloc_sem);
5110 size = i_size_read(inode);
5111 if (page->mapping != mapping || size <= page_offset(page)
5112 || !PageUptodate(page)) {
5113 /* page got truncated from under us? */
5117 if (PageMappedToDisk(page))
5120 if (page->index == size >> PAGE_CACHE_SHIFT)
5121 len = size & ~PAGE_CACHE_MASK;
5123 len = PAGE_CACHE_SIZE;
5125 if (page_has_buffers(page)) {
5126 /* return if we have all the buffers mapped */
5127 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5132 * OK, we need to fill the hole... Do write_begin write_end
5133 * to do block allocation/reservation.We are not holding
5134 * inode.i__mutex here. That allow * parallel write_begin,
5135 * write_end call. lock_page prevent this from happening
5136 * on the same page though
5138 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5139 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5142 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5143 len, len, page, fsdata);
5148 up_read(&inode->i_alloc_sem);