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(&EXT4_I(inode)->jinode,
54 static void ext4_invalidatepage(struct page *page, unsigned long offset);
57 * Test whether an inode is a fast symlink.
59 static int ext4_inode_is_fast_symlink(struct inode *inode)
61 int ea_blocks = EXT4_I(inode)->i_file_acl ?
62 (inode->i_sb->s_blocksize >> 9) : 0;
64 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
68 * The ext4 forget function must perform a revoke if we are freeing data
69 * which has been journaled. Metadata (eg. indirect blocks) must be
70 * revoked in all cases.
72 * "bh" may be NULL: a metadata block may have been freed from memory
73 * but there may still be a record of it in the journal, and that record
74 * still needs to be revoked.
76 * If the handle isn't valid we're not journaling so there's nothing to do.
78 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
79 struct buffer_head *bh, ext4_fsblk_t blocknr)
83 if (!ext4_handle_valid(handle))
88 BUFFER_TRACE(bh, "enter");
90 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
92 bh, is_metadata, inode->i_mode,
93 test_opt(inode->i_sb, DATA_FLAGS));
95 /* Never use the revoke function if we are doing full data
96 * journaling: there is no need to, and a V1 superblock won't
97 * support it. Otherwise, only skip the revoke on un-journaled
100 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
101 (!is_metadata && !ext4_should_journal_data(inode))) {
103 BUFFER_TRACE(bh, "call jbd2_journal_forget");
104 return ext4_journal_forget(handle, bh);
110 * data!=journal && (is_metadata || should_journal_data(inode))
112 BUFFER_TRACE(bh, "call ext4_journal_revoke");
113 err = ext4_journal_revoke(handle, blocknr, bh);
115 ext4_abort(inode->i_sb, __func__,
116 "error %d when attempting revoke", err);
117 BUFFER_TRACE(bh, "exit");
122 * Work out how many blocks we need to proceed with the next chunk of a
123 * truncate transaction.
125 static unsigned long blocks_for_truncate(struct inode *inode)
129 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
131 /* Give ourselves just enough room to cope with inodes in which
132 * i_blocks is corrupt: we've seen disk corruptions in the past
133 * which resulted in random data in an inode which looked enough
134 * like a regular file for ext4 to try to delete it. Things
135 * will go a bit crazy if that happens, but at least we should
136 * try not to panic the whole kernel. */
140 /* But we need to bound the transaction so we don't overflow the
142 if (needed > EXT4_MAX_TRANS_DATA)
143 needed = EXT4_MAX_TRANS_DATA;
145 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
149 * Truncate transactions can be complex and absolutely huge. So we need to
150 * be able to restart the transaction at a conventient checkpoint to make
151 * sure we don't overflow the journal.
153 * start_transaction gets us a new handle for a truncate transaction,
154 * and extend_transaction tries to extend the existing one a bit. If
155 * extend fails, we need to propagate the failure up and restart the
156 * transaction in the top-level truncate loop. --sct
158 static handle_t *start_transaction(struct inode *inode)
162 result = ext4_journal_start(inode, blocks_for_truncate(inode));
166 ext4_std_error(inode->i_sb, PTR_ERR(result));
171 * Try to extend this transaction for the purposes of truncation.
173 * Returns 0 if we managed to create more room. If we can't create more
174 * room, and the transaction must be restarted we return 1.
176 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
178 if (!ext4_handle_valid(handle))
180 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
182 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
188 * Restart the transaction associated with *handle. This does a commit,
189 * so before we call here everything must be consistently dirtied against
192 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
194 BUG_ON(EXT4_JOURNAL(inode) == NULL);
195 jbd_debug(2, "restarting handle %p\n", handle);
196 return ext4_journal_restart(handle, blocks_for_truncate(inode));
200 * Called at the last iput() if i_nlink is zero.
202 void ext4_delete_inode(struct inode *inode)
207 if (ext4_should_order_data(inode))
208 ext4_begin_ordered_truncate(inode, 0);
209 truncate_inode_pages(&inode->i_data, 0);
211 if (is_bad_inode(inode))
214 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
215 if (IS_ERR(handle)) {
216 ext4_std_error(inode->i_sb, PTR_ERR(handle));
218 * If we're going to skip the normal cleanup, we still need to
219 * make sure that the in-core orphan linked list is properly
222 ext4_orphan_del(NULL, inode);
227 ext4_handle_sync(handle);
229 err = ext4_mark_inode_dirty(handle, inode);
231 ext4_warning(inode->i_sb, __func__,
232 "couldn't mark inode dirty (err %d)", err);
236 ext4_truncate(inode);
239 * ext4_ext_truncate() doesn't reserve any slop when it
240 * restarts journal transactions; therefore there may not be
241 * enough credits left in the handle to remove the inode from
242 * the orphan list and set the dtime field.
244 if (!ext4_handle_has_enough_credits(handle, 3)) {
245 err = ext4_journal_extend(handle, 3);
247 err = ext4_journal_restart(handle, 3);
249 ext4_warning(inode->i_sb, __func__,
250 "couldn't extend journal (err %d)", err);
252 ext4_journal_stop(handle);
258 * Kill off the orphan record which ext4_truncate created.
259 * AKPM: I think this can be inside the above `if'.
260 * Note that ext4_orphan_del() has to be able to cope with the
261 * deletion of a non-existent orphan - this is because we don't
262 * know if ext4_truncate() actually created an orphan record.
263 * (Well, we could do this if we need to, but heck - it works)
265 ext4_orphan_del(handle, inode);
266 EXT4_I(inode)->i_dtime = get_seconds();
269 * One subtle ordering requirement: if anything has gone wrong
270 * (transaction abort, IO errors, whatever), then we can still
271 * do these next steps (the fs will already have been marked as
272 * having errors), but we can't free the inode if the mark_dirty
275 if (ext4_mark_inode_dirty(handle, inode))
276 /* If that failed, just do the required in-core inode clear. */
279 ext4_free_inode(handle, inode);
280 ext4_journal_stop(handle);
283 clear_inode(inode); /* We must guarantee clearing of inode... */
289 struct buffer_head *bh;
292 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
294 p->key = *(p->p = v);
299 * ext4_block_to_path - parse the block number into array of offsets
300 * @inode: inode in question (we are only interested in its superblock)
301 * @i_block: block number to be parsed
302 * @offsets: array to store the offsets in
303 * @boundary: set this non-zero if the referred-to block is likely to be
304 * followed (on disk) by an indirect block.
306 * To store the locations of file's data ext4 uses a data structure common
307 * for UNIX filesystems - tree of pointers anchored in the inode, with
308 * data blocks at leaves and indirect blocks in intermediate nodes.
309 * This function translates the block number into path in that tree -
310 * return value is the path length and @offsets[n] is the offset of
311 * pointer to (n+1)th node in the nth one. If @block is out of range
312 * (negative or too large) warning is printed and zero returned.
314 * Note: function doesn't find node addresses, so no IO is needed. All
315 * we need to know is the capacity of indirect blocks (taken from the
320 * Portability note: the last comparison (check that we fit into triple
321 * indirect block) is spelled differently, because otherwise on an
322 * architecture with 32-bit longs and 8Kb pages we might get into trouble
323 * if our filesystem had 8Kb blocks. We might use long long, but that would
324 * kill us on x86. Oh, well, at least the sign propagation does not matter -
325 * i_block would have to be negative in the very beginning, so we would not
329 static int ext4_block_to_path(struct inode *inode,
331 ext4_lblk_t offsets[4], int *boundary)
333 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
334 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
335 const long direct_blocks = EXT4_NDIR_BLOCKS,
336 indirect_blocks = ptrs,
337 double_blocks = (1 << (ptrs_bits * 2));
342 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
343 } else if (i_block < direct_blocks) {
344 offsets[n++] = i_block;
345 final = direct_blocks;
346 } else if ((i_block -= direct_blocks) < indirect_blocks) {
347 offsets[n++] = EXT4_IND_BLOCK;
348 offsets[n++] = i_block;
350 } else if ((i_block -= indirect_blocks) < double_blocks) {
351 offsets[n++] = EXT4_DIND_BLOCK;
352 offsets[n++] = i_block >> ptrs_bits;
353 offsets[n++] = i_block & (ptrs - 1);
355 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
356 offsets[n++] = EXT4_TIND_BLOCK;
357 offsets[n++] = i_block >> (ptrs_bits * 2);
358 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
359 offsets[n++] = i_block & (ptrs - 1);
362 ext4_warning(inode->i_sb, "ext4_block_to_path",
364 i_block + direct_blocks +
365 indirect_blocks + double_blocks);
368 *boundary = final - 1 - (i_block & (ptrs - 1));
373 * ext4_get_branch - read the chain of indirect blocks leading to data
374 * @inode: inode in question
375 * @depth: depth of the chain (1 - direct pointer, etc.)
376 * @offsets: offsets of pointers in inode/indirect blocks
377 * @chain: place to store the result
378 * @err: here we store the error value
380 * Function fills the array of triples <key, p, bh> and returns %NULL
381 * if everything went OK or the pointer to the last filled triple
382 * (incomplete one) otherwise. Upon the return chain[i].key contains
383 * the number of (i+1)-th block in the chain (as it is stored in memory,
384 * i.e. little-endian 32-bit), chain[i].p contains the address of that
385 * number (it points into struct inode for i==0 and into the bh->b_data
386 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
387 * block for i>0 and NULL for i==0. In other words, it holds the block
388 * numbers of the chain, addresses they were taken from (and where we can
389 * verify that chain did not change) and buffer_heads hosting these
392 * Function stops when it stumbles upon zero pointer (absent block)
393 * (pointer to last triple returned, *@err == 0)
394 * or when it gets an IO error reading an indirect block
395 * (ditto, *@err == -EIO)
396 * or when it reads all @depth-1 indirect blocks successfully and finds
397 * the whole chain, all way to the data (returns %NULL, *err == 0).
399 * Need to be called with
400 * down_read(&EXT4_I(inode)->i_data_sem)
402 static Indirect *ext4_get_branch(struct inode *inode, int depth,
403 ext4_lblk_t *offsets,
404 Indirect chain[4], int *err)
406 struct super_block *sb = inode->i_sb;
408 struct buffer_head *bh;
411 /* i_data is not going away, no lock needed */
412 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
416 bh = sb_bread(sb, le32_to_cpu(p->key));
419 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
433 * ext4_find_near - find a place for allocation with sufficient locality
435 * @ind: descriptor of indirect block.
437 * This function returns the preferred place for block allocation.
438 * It is used when heuristic for sequential allocation fails.
440 * + if there is a block to the left of our position - allocate near it.
441 * + if pointer will live in indirect block - allocate near that block.
442 * + if pointer will live in inode - allocate in the same
445 * In the latter case we colour the starting block by the callers PID to
446 * prevent it from clashing with concurrent allocations for a different inode
447 * in the same block group. The PID is used here so that functionally related
448 * files will be close-by on-disk.
450 * Caller must make sure that @ind is valid and will stay that way.
452 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
454 struct ext4_inode_info *ei = EXT4_I(inode);
455 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
457 ext4_fsblk_t bg_start;
458 ext4_fsblk_t last_block;
459 ext4_grpblk_t colour;
461 /* Try to find previous block */
462 for (p = ind->p - 1; p >= start; p--) {
464 return le32_to_cpu(*p);
467 /* No such thing, so let's try location of indirect block */
469 return ind->bh->b_blocknr;
472 * It is going to be referred to from the inode itself? OK, just put it
473 * into the same cylinder group then.
475 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
476 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
478 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
479 colour = (current->pid % 16) *
480 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
482 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
483 return bg_start + colour;
487 * ext4_find_goal - find a preferred place for allocation.
489 * @block: block we want
490 * @partial: pointer to the last triple within a chain
492 * Normally this function find the preferred place for block allocation,
495 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
499 * XXX need to get goal block from mballoc's data structures
502 return ext4_find_near(inode, partial);
506 * ext4_blks_to_allocate: Look up the block map and count the number
507 * of direct blocks need to be allocated for the given branch.
509 * @branch: chain of indirect blocks
510 * @k: number of blocks need for indirect blocks
511 * @blks: number of data blocks to be mapped.
512 * @blocks_to_boundary: the offset in the indirect block
514 * return the total number of blocks to be allocate, including the
515 * direct and indirect blocks.
517 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
518 int blocks_to_boundary)
520 unsigned long count = 0;
523 * Simple case, [t,d]Indirect block(s) has not allocated yet
524 * then it's clear blocks on that path have not allocated
527 /* right now we don't handle cross boundary allocation */
528 if (blks < blocks_to_boundary + 1)
531 count += blocks_to_boundary + 1;
536 while (count < blks && count <= blocks_to_boundary &&
537 le32_to_cpu(*(branch[0].p + count)) == 0) {
544 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
545 * @indirect_blks: the number of blocks need to allocate for indirect
548 * @new_blocks: on return it will store the new block numbers for
549 * the indirect blocks(if needed) and the first direct block,
550 * @blks: on return it will store the total number of allocated
553 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
554 ext4_lblk_t iblock, ext4_fsblk_t goal,
555 int indirect_blks, int blks,
556 ext4_fsblk_t new_blocks[4], int *err)
558 struct ext4_allocation_request ar;
560 unsigned long count = 0, blk_allocated = 0;
562 ext4_fsblk_t current_block = 0;
566 * Here we try to allocate the requested multiple blocks at once,
567 * on a best-effort basis.
568 * To build a branch, we should allocate blocks for
569 * the indirect blocks(if not allocated yet), and at least
570 * the first direct block of this branch. That's the
571 * minimum number of blocks need to allocate(required)
573 /* first we try to allocate the indirect blocks */
574 target = indirect_blks;
577 /* allocating blocks for indirect blocks and direct blocks */
578 current_block = ext4_new_meta_blocks(handle, inode,
584 /* allocate blocks for indirect blocks */
585 while (index < indirect_blks && count) {
586 new_blocks[index++] = current_block++;
591 * save the new block number
592 * for the first direct block
594 new_blocks[index] = current_block;
595 printk(KERN_INFO "%s returned more blocks than "
596 "requested\n", __func__);
602 target = blks - count ;
603 blk_allocated = count;
606 /* Now allocate data blocks */
607 memset(&ar, 0, sizeof(ar));
612 if (S_ISREG(inode->i_mode))
613 /* enable in-core preallocation only for regular files */
614 ar.flags = EXT4_MB_HINT_DATA;
616 current_block = ext4_mb_new_blocks(handle, &ar, err);
618 if (*err && (target == blks)) {
620 * if the allocation failed and we didn't allocate
626 if (target == blks) {
628 * save the new block number
629 * for the first direct block
631 new_blocks[index] = current_block;
633 blk_allocated += ar.len;
636 /* total number of blocks allocated for direct blocks */
641 for (i = 0; i < index; i++)
642 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
647 * ext4_alloc_branch - allocate and set up a chain of blocks.
649 * @indirect_blks: number of allocated indirect blocks
650 * @blks: number of allocated direct blocks
651 * @offsets: offsets (in the blocks) to store the pointers to next.
652 * @branch: place to store the chain in.
654 * This function allocates blocks, zeroes out all but the last one,
655 * links them into chain and (if we are synchronous) writes them to disk.
656 * In other words, it prepares a branch that can be spliced onto the
657 * inode. It stores the information about that chain in the branch[], in
658 * the same format as ext4_get_branch() would do. We are calling it after
659 * we had read the existing part of chain and partial points to the last
660 * triple of that (one with zero ->key). Upon the exit we have the same
661 * picture as after the successful ext4_get_block(), except that in one
662 * place chain is disconnected - *branch->p is still zero (we did not
663 * set the last link), but branch->key contains the number that should
664 * be placed into *branch->p to fill that gap.
666 * If allocation fails we free all blocks we've allocated (and forget
667 * their buffer_heads) and return the error value the from failed
668 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
669 * as described above and return 0.
671 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
672 ext4_lblk_t iblock, int indirect_blks,
673 int *blks, ext4_fsblk_t goal,
674 ext4_lblk_t *offsets, Indirect *branch)
676 int blocksize = inode->i_sb->s_blocksize;
679 struct buffer_head *bh;
681 ext4_fsblk_t new_blocks[4];
682 ext4_fsblk_t current_block;
684 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
685 *blks, new_blocks, &err);
689 branch[0].key = cpu_to_le32(new_blocks[0]);
691 * metadata blocks and data blocks are allocated.
693 for (n = 1; n <= indirect_blks; n++) {
695 * Get buffer_head for parent block, zero it out
696 * and set the pointer to new one, then send
699 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
702 BUFFER_TRACE(bh, "call get_create_access");
703 err = ext4_journal_get_create_access(handle, bh);
710 memset(bh->b_data, 0, blocksize);
711 branch[n].p = (__le32 *) bh->b_data + offsets[n];
712 branch[n].key = cpu_to_le32(new_blocks[n]);
713 *branch[n].p = branch[n].key;
714 if (n == indirect_blks) {
715 current_block = new_blocks[n];
717 * End of chain, update the last new metablock of
718 * the chain to point to the new allocated
719 * data blocks numbers
721 for (i=1; i < num; i++)
722 *(branch[n].p + i) = cpu_to_le32(++current_block);
724 BUFFER_TRACE(bh, "marking uptodate");
725 set_buffer_uptodate(bh);
728 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
729 err = ext4_handle_dirty_metadata(handle, inode, bh);
736 /* Allocation failed, free what we already allocated */
737 for (i = 1; i <= n ; i++) {
738 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
739 ext4_journal_forget(handle, branch[i].bh);
741 for (i = 0; i < indirect_blks; i++)
742 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
744 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
750 * ext4_splice_branch - splice the allocated branch onto inode.
752 * @block: (logical) number of block we are adding
753 * @chain: chain of indirect blocks (with a missing link - see
755 * @where: location of missing link
756 * @num: number of indirect blocks we are adding
757 * @blks: number of direct blocks we are adding
759 * This function fills the missing link and does all housekeeping needed in
760 * inode (->i_blocks, etc.). In case of success we end up with the full
761 * chain to new block and return 0.
763 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
764 ext4_lblk_t block, Indirect *where, int num, int blks)
768 ext4_fsblk_t current_block;
771 * If we're splicing into a [td]indirect block (as opposed to the
772 * inode) then we need to get write access to the [td]indirect block
776 BUFFER_TRACE(where->bh, "get_write_access");
777 err = ext4_journal_get_write_access(handle, where->bh);
783 *where->p = where->key;
786 * Update the host buffer_head or inode to point to more just allocated
787 * direct blocks blocks
789 if (num == 0 && blks > 1) {
790 current_block = le32_to_cpu(where->key) + 1;
791 for (i = 1; i < blks; i++)
792 *(where->p + i) = cpu_to_le32(current_block++);
795 /* We are done with atomic stuff, now do the rest of housekeeping */
797 inode->i_ctime = ext4_current_time(inode);
798 ext4_mark_inode_dirty(handle, inode);
800 /* had we spliced it onto indirect block? */
803 * If we spliced it onto an indirect block, we haven't
804 * altered the inode. Note however that if it is being spliced
805 * onto an indirect block at the very end of the file (the
806 * file is growing) then we *will* alter the inode to reflect
807 * the new i_size. But that is not done here - it is done in
808 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
810 jbd_debug(5, "splicing indirect only\n");
811 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
812 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
817 * OK, we spliced it into the inode itself on a direct block.
818 * Inode was dirtied above.
820 jbd_debug(5, "splicing direct\n");
825 for (i = 1; i <= num; i++) {
826 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
827 ext4_journal_forget(handle, where[i].bh);
828 ext4_free_blocks(handle, inode,
829 le32_to_cpu(where[i-1].key), 1, 0);
831 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
837 * Allocation strategy is simple: if we have to allocate something, we will
838 * have to go the whole way to leaf. So let's do it before attaching anything
839 * to tree, set linkage between the newborn blocks, write them if sync is
840 * required, recheck the path, free and repeat if check fails, otherwise
841 * set the last missing link (that will protect us from any truncate-generated
842 * removals - all blocks on the path are immune now) and possibly force the
843 * write on the parent block.
844 * That has a nice additional property: no special recovery from the failed
845 * allocations is needed - we simply release blocks and do not touch anything
846 * reachable from inode.
848 * `handle' can be NULL if create == 0.
850 * return > 0, # of blocks mapped or allocated.
851 * return = 0, if plain lookup failed.
852 * return < 0, error case.
855 * Need to be called with
856 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
857 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
859 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
860 ext4_lblk_t iblock, unsigned long maxblocks,
861 struct buffer_head *bh_result,
862 int create, int extend_disksize)
865 ext4_lblk_t offsets[4];
870 int blocks_to_boundary = 0;
872 struct ext4_inode_info *ei = EXT4_I(inode);
874 ext4_fsblk_t first_block = 0;
878 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
879 J_ASSERT(handle != NULL || create == 0);
880 depth = ext4_block_to_path(inode, iblock, offsets,
881 &blocks_to_boundary);
886 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
888 /* Simplest case - block found, no allocation needed */
890 first_block = le32_to_cpu(chain[depth - 1].key);
891 clear_buffer_new(bh_result);
894 while (count < maxblocks && count <= blocks_to_boundary) {
897 blk = le32_to_cpu(*(chain[depth-1].p + count));
899 if (blk == first_block + count)
907 /* Next simple case - plain lookup or failed read of indirect block */
908 if (!create || err == -EIO)
912 * Okay, we need to do block allocation.
914 goal = ext4_find_goal(inode, iblock, partial);
916 /* the number of blocks need to allocate for [d,t]indirect blocks */
917 indirect_blks = (chain + depth) - partial - 1;
920 * Next look up the indirect map to count the totoal number of
921 * direct blocks to allocate for this branch.
923 count = ext4_blks_to_allocate(partial, indirect_blks,
924 maxblocks, blocks_to_boundary);
926 * Block out ext4_truncate while we alter the tree
928 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
930 offsets + (partial - chain), partial);
933 * The ext4_splice_branch call will free and forget any buffers
934 * on the new chain if there is a failure, but that risks using
935 * up transaction credits, especially for bitmaps where the
936 * credits cannot be returned. Can we handle this somehow? We
937 * may need to return -EAGAIN upwards in the worst case. --sct
940 err = ext4_splice_branch(handle, inode, iblock,
941 partial, indirect_blks, count);
943 * i_disksize growing is protected by i_data_sem. Don't forget to
944 * protect it if you're about to implement concurrent
945 * ext4_get_block() -bzzz
947 if (!err && extend_disksize) {
948 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
949 if (disksize > i_size_read(inode))
950 disksize = i_size_read(inode);
951 if (disksize > ei->i_disksize)
952 ei->i_disksize = disksize;
957 set_buffer_new(bh_result);
959 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
960 if (count > blocks_to_boundary)
961 set_buffer_boundary(bh_result);
963 /* Clean up and exit */
964 partial = chain + depth - 1; /* the whole chain */
966 while (partial > chain) {
967 BUFFER_TRACE(partial->bh, "call brelse");
971 BUFFER_TRACE(bh_result, "returned");
977 * Calculate the number of metadata blocks need to reserve
978 * to allocate @blocks for non extent file based file
980 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
982 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
983 int ind_blks, dind_blks, tind_blks;
985 /* number of new indirect blocks needed */
986 ind_blks = (blocks + icap - 1) / icap;
988 dind_blks = (ind_blks + icap - 1) / icap;
992 return ind_blks + dind_blks + tind_blks;
996 * Calculate the number of metadata blocks need to reserve
997 * to allocate given number of blocks
999 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1004 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1005 return ext4_ext_calc_metadata_amount(inode, blocks);
1007 return ext4_indirect_calc_metadata_amount(inode, blocks);
1010 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1012 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1013 int total, mdb, mdb_free;
1015 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1016 /* recalculate the number of metablocks still need to be reserved */
1017 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1018 mdb = ext4_calc_metadata_amount(inode, total);
1020 /* figure out how many metablocks to release */
1021 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1022 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1025 /* Account for allocated meta_blocks */
1026 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1028 /* update fs dirty blocks counter */
1029 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1030 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1031 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1034 /* update per-inode reservations */
1035 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1036 EXT4_I(inode)->i_reserved_data_blocks -= used;
1038 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1042 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1043 * and returns if the blocks are already mapped.
1045 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1046 * and store the allocated blocks in the result buffer head and mark it
1049 * If file type is extents based, it will call ext4_ext_get_blocks(),
1050 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1053 * On success, it returns the number of blocks being mapped or allocate.
1054 * if create==0 and the blocks are pre-allocated and uninitialized block,
1055 * the result buffer head is unmapped. If the create ==1, it will make sure
1056 * the buffer head is mapped.
1058 * It returns 0 if plain look up failed (blocks have not been allocated), in
1059 * that casem, buffer head is unmapped
1061 * It returns the error in case of allocation failure.
1063 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1064 unsigned long max_blocks, struct buffer_head *bh,
1065 int create, int extend_disksize, int flag)
1069 clear_buffer_mapped(bh);
1072 * Try to see if we can get the block without requesting
1073 * for new file system block.
1075 down_read((&EXT4_I(inode)->i_data_sem));
1076 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1077 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1080 retval = ext4_get_blocks_handle(handle,
1081 inode, block, max_blocks, bh, 0, 0);
1083 up_read((&EXT4_I(inode)->i_data_sem));
1085 /* If it is only a block(s) look up */
1090 * Returns if the blocks have already allocated
1092 * Note that if blocks have been preallocated
1093 * ext4_ext_get_block() returns th create = 0
1094 * with buffer head unmapped.
1096 if (retval > 0 && buffer_mapped(bh))
1100 * New blocks allocate and/or writing to uninitialized extent
1101 * will possibly result in updating i_data, so we take
1102 * the write lock of i_data_sem, and call get_blocks()
1103 * with create == 1 flag.
1105 down_write((&EXT4_I(inode)->i_data_sem));
1108 * if the caller is from delayed allocation writeout path
1109 * we have already reserved fs blocks for allocation
1110 * let the underlying get_block() function know to
1111 * avoid double accounting
1114 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1116 * We need to check for EXT4 here because migrate
1117 * could have changed the inode type in between
1119 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1120 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1121 bh, create, extend_disksize);
1123 retval = ext4_get_blocks_handle(handle, inode, block,
1124 max_blocks, bh, create, extend_disksize);
1126 if (retval > 0 && buffer_new(bh)) {
1128 * We allocated new blocks which will result in
1129 * i_data's format changing. Force the migrate
1130 * to fail by clearing migrate flags
1132 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1138 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1140 * Update reserved blocks/metadata blocks
1141 * after successful block allocation
1142 * which were deferred till now
1144 if ((retval > 0) && buffer_delay(bh))
1145 ext4_da_update_reserve_space(inode, retval);
1148 up_write((&EXT4_I(inode)->i_data_sem));
1152 /* Maximum number of blocks we map for direct IO at once. */
1153 #define DIO_MAX_BLOCKS 4096
1155 int ext4_get_block(struct inode *inode, sector_t iblock,
1156 struct buffer_head *bh_result, int create)
1158 handle_t *handle = ext4_journal_current_handle();
1159 int ret = 0, started = 0;
1160 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1163 if (create && !handle) {
1164 /* Direct IO write... */
1165 if (max_blocks > DIO_MAX_BLOCKS)
1166 max_blocks = DIO_MAX_BLOCKS;
1167 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1168 handle = ext4_journal_start(inode, dio_credits);
1169 if (IS_ERR(handle)) {
1170 ret = PTR_ERR(handle);
1176 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1177 max_blocks, bh_result, create, 0, 0);
1179 bh_result->b_size = (ret << inode->i_blkbits);
1183 ext4_journal_stop(handle);
1189 * `handle' can be NULL if create is zero
1191 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1192 ext4_lblk_t block, int create, int *errp)
1194 struct buffer_head dummy;
1197 J_ASSERT(handle != NULL || create == 0);
1200 dummy.b_blocknr = -1000;
1201 buffer_trace_init(&dummy.b_history);
1202 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1203 &dummy, create, 1, 0);
1205 * ext4_get_blocks_handle() returns number of blocks
1206 * mapped. 0 in case of a HOLE.
1214 if (!err && buffer_mapped(&dummy)) {
1215 struct buffer_head *bh;
1216 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1221 if (buffer_new(&dummy)) {
1222 J_ASSERT(create != 0);
1223 J_ASSERT(handle != NULL);
1226 * Now that we do not always journal data, we should
1227 * keep in mind whether this should always journal the
1228 * new buffer as metadata. For now, regular file
1229 * writes use ext4_get_block instead, so it's not a
1233 BUFFER_TRACE(bh, "call get_create_access");
1234 fatal = ext4_journal_get_create_access(handle, bh);
1235 if (!fatal && !buffer_uptodate(bh)) {
1236 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1237 set_buffer_uptodate(bh);
1240 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1241 err = ext4_handle_dirty_metadata(handle, inode, bh);
1245 BUFFER_TRACE(bh, "not a new buffer");
1258 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1259 ext4_lblk_t block, int create, int *err)
1261 struct buffer_head *bh;
1263 bh = ext4_getblk(handle, inode, block, create, err);
1266 if (buffer_uptodate(bh))
1268 ll_rw_block(READ_META, 1, &bh);
1270 if (buffer_uptodate(bh))
1277 static int walk_page_buffers(handle_t *handle,
1278 struct buffer_head *head,
1282 int (*fn)(handle_t *handle,
1283 struct buffer_head *bh))
1285 struct buffer_head *bh;
1286 unsigned block_start, block_end;
1287 unsigned blocksize = head->b_size;
1289 struct buffer_head *next;
1291 for (bh = head, block_start = 0;
1292 ret == 0 && (bh != head || !block_start);
1293 block_start = block_end, bh = next)
1295 next = bh->b_this_page;
1296 block_end = block_start + blocksize;
1297 if (block_end <= from || block_start >= to) {
1298 if (partial && !buffer_uptodate(bh))
1302 err = (*fn)(handle, bh);
1310 * To preserve ordering, it is essential that the hole instantiation and
1311 * the data write be encapsulated in a single transaction. We cannot
1312 * close off a transaction and start a new one between the ext4_get_block()
1313 * and the commit_write(). So doing the jbd2_journal_start at the start of
1314 * prepare_write() is the right place.
1316 * Also, this function can nest inside ext4_writepage() ->
1317 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1318 * has generated enough buffer credits to do the whole page. So we won't
1319 * block on the journal in that case, which is good, because the caller may
1322 * By accident, ext4 can be reentered when a transaction is open via
1323 * quota file writes. If we were to commit the transaction while thus
1324 * reentered, there can be a deadlock - we would be holding a quota
1325 * lock, and the commit would never complete if another thread had a
1326 * transaction open and was blocking on the quota lock - a ranking
1329 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1330 * will _not_ run commit under these circumstances because handle->h_ref
1331 * is elevated. We'll still have enough credits for the tiny quotafile
1334 static int do_journal_get_write_access(handle_t *handle,
1335 struct buffer_head *bh)
1337 if (!buffer_mapped(bh) || buffer_freed(bh))
1339 return ext4_journal_get_write_access(handle, bh);
1342 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1343 loff_t pos, unsigned len, unsigned flags,
1344 struct page **pagep, void **fsdata)
1346 struct inode *inode = mapping->host;
1347 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1354 index = pos >> PAGE_CACHE_SHIFT;
1355 from = pos & (PAGE_CACHE_SIZE - 1);
1359 handle = ext4_journal_start(inode, needed_blocks);
1360 if (IS_ERR(handle)) {
1361 ret = PTR_ERR(handle);
1365 page = grab_cache_page_write_begin(mapping, index, flags);
1367 ext4_journal_stop(handle);
1373 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1376 if (!ret && ext4_should_journal_data(inode)) {
1377 ret = walk_page_buffers(handle, page_buffers(page),
1378 from, to, NULL, do_journal_get_write_access);
1383 ext4_journal_stop(handle);
1384 page_cache_release(page);
1386 * block_write_begin may have instantiated a few blocks
1387 * outside i_size. Trim these off again. Don't need
1388 * i_size_read because we hold i_mutex.
1390 if (pos + len > inode->i_size)
1391 vmtruncate(inode, inode->i_size);
1394 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1400 /* For write_end() in data=journal mode */
1401 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1403 if (!buffer_mapped(bh) || buffer_freed(bh))
1405 set_buffer_uptodate(bh);
1406 return ext4_handle_dirty_metadata(handle, NULL, bh);
1410 * We need to pick up the new inode size which generic_commit_write gave us
1411 * `file' can be NULL - eg, when called from page_symlink().
1413 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1414 * buffers are managed internally.
1416 static int ext4_ordered_write_end(struct file *file,
1417 struct address_space *mapping,
1418 loff_t pos, unsigned len, unsigned copied,
1419 struct page *page, void *fsdata)
1421 handle_t *handle = ext4_journal_current_handle();
1422 struct inode *inode = mapping->host;
1425 ret = ext4_jbd2_file_inode(handle, inode);
1430 new_i_size = pos + copied;
1431 if (new_i_size > EXT4_I(inode)->i_disksize) {
1432 ext4_update_i_disksize(inode, new_i_size);
1433 /* We need to mark inode dirty even if
1434 * new_i_size is less that inode->i_size
1435 * bu greater than i_disksize.(hint delalloc)
1437 ext4_mark_inode_dirty(handle, inode);
1440 ret2 = generic_write_end(file, mapping, pos, len, copied,
1446 ret2 = ext4_journal_stop(handle);
1450 return ret ? ret : copied;
1453 static int ext4_writeback_write_end(struct file *file,
1454 struct address_space *mapping,
1455 loff_t pos, unsigned len, unsigned copied,
1456 struct page *page, void *fsdata)
1458 handle_t *handle = ext4_journal_current_handle();
1459 struct inode *inode = mapping->host;
1463 new_i_size = pos + copied;
1464 if (new_i_size > EXT4_I(inode)->i_disksize) {
1465 ext4_update_i_disksize(inode, new_i_size);
1466 /* We need to mark inode dirty even if
1467 * new_i_size is less that inode->i_size
1468 * bu greater than i_disksize.(hint delalloc)
1470 ext4_mark_inode_dirty(handle, inode);
1473 ret2 = generic_write_end(file, mapping, pos, len, copied,
1479 ret2 = ext4_journal_stop(handle);
1483 return ret ? ret : copied;
1486 static int ext4_journalled_write_end(struct file *file,
1487 struct address_space *mapping,
1488 loff_t pos, unsigned len, unsigned copied,
1489 struct page *page, void *fsdata)
1491 handle_t *handle = ext4_journal_current_handle();
1492 struct inode *inode = mapping->host;
1498 from = pos & (PAGE_CACHE_SIZE - 1);
1502 if (!PageUptodate(page))
1504 page_zero_new_buffers(page, from+copied, to);
1507 ret = walk_page_buffers(handle, page_buffers(page), from,
1508 to, &partial, write_end_fn);
1510 SetPageUptodate(page);
1511 new_i_size = pos + copied;
1512 if (new_i_size > inode->i_size)
1513 i_size_write(inode, pos+copied);
1514 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1515 if (new_i_size > EXT4_I(inode)->i_disksize) {
1516 ext4_update_i_disksize(inode, new_i_size);
1517 ret2 = ext4_mark_inode_dirty(handle, inode);
1523 ret2 = ext4_journal_stop(handle);
1526 page_cache_release(page);
1528 return ret ? ret : copied;
1531 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1534 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1535 unsigned long md_needed, mdblocks, total = 0;
1538 * recalculate the amount of metadata blocks to reserve
1539 * in order to allocate nrblocks
1540 * worse case is one extent per block
1543 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1544 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1545 mdblocks = ext4_calc_metadata_amount(inode, total);
1546 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1548 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1549 total = md_needed + nrblocks;
1551 if (ext4_claim_free_blocks(sbi, total)) {
1552 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1553 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1559 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1560 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1562 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1563 return 0; /* success */
1566 static void ext4_da_release_space(struct inode *inode, int to_free)
1568 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1569 int total, mdb, mdb_free, release;
1572 return; /* Nothing to release, exit */
1574 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1576 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1578 * if there is no reserved blocks, but we try to free some
1579 * then the counter is messed up somewhere.
1580 * but since this function is called from invalidate
1581 * page, it's harmless to return without any action
1583 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1584 "blocks for inode %lu, but there is no reserved "
1585 "data blocks\n", to_free, inode->i_ino);
1586 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1590 /* recalculate the number of metablocks still need to be reserved */
1591 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1592 mdb = ext4_calc_metadata_amount(inode, total);
1594 /* figure out how many metablocks to release */
1595 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1596 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1598 release = to_free + mdb_free;
1600 /* update fs dirty blocks counter for truncate case */
1601 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1603 /* update per-inode reservations */
1604 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1605 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1607 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1608 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1609 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1612 static void ext4_da_page_release_reservation(struct page *page,
1613 unsigned long offset)
1616 struct buffer_head *head, *bh;
1617 unsigned int curr_off = 0;
1619 head = page_buffers(page);
1622 unsigned int next_off = curr_off + bh->b_size;
1624 if ((offset <= curr_off) && (buffer_delay(bh))) {
1626 clear_buffer_delay(bh);
1628 curr_off = next_off;
1629 } while ((bh = bh->b_this_page) != head);
1630 ext4_da_release_space(page->mapping->host, to_release);
1634 * Delayed allocation stuff
1637 struct mpage_da_data {
1638 struct inode *inode;
1639 struct buffer_head lbh; /* extent of blocks */
1640 unsigned long first_page, next_page; /* extent of pages */
1641 get_block_t *get_block;
1642 struct writeback_control *wbc;
1649 * mpage_da_submit_io - walks through extent of pages and try to write
1650 * them with writepage() call back
1652 * @mpd->inode: inode
1653 * @mpd->first_page: first page of the extent
1654 * @mpd->next_page: page after the last page of the extent
1655 * @mpd->get_block: the filesystem's block mapper function
1657 * By the time mpage_da_submit_io() is called we expect all blocks
1658 * to be allocated. this may be wrong if allocation failed.
1660 * As pages are already locked by write_cache_pages(), we can't use it
1662 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1665 struct pagevec pvec;
1666 unsigned long index, end;
1667 int ret = 0, err, nr_pages, i;
1668 struct inode *inode = mpd->inode;
1669 struct address_space *mapping = inode->i_mapping;
1671 BUG_ON(mpd->next_page <= mpd->first_page);
1673 * We need to start from the first_page to the next_page - 1
1674 * to make sure we also write the mapped dirty buffer_heads.
1675 * If we look at mpd->lbh.b_blocknr we would only be looking
1676 * at the currently mapped buffer_heads.
1678 index = mpd->first_page;
1679 end = mpd->next_page - 1;
1681 pagevec_init(&pvec, 0);
1682 while (index <= end) {
1683 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1686 for (i = 0; i < nr_pages; i++) {
1687 struct page *page = pvec.pages[i];
1689 index = page->index;
1694 BUG_ON(!PageLocked(page));
1695 BUG_ON(PageWriteback(page));
1697 pages_skipped = mpd->wbc->pages_skipped;
1698 err = mapping->a_ops->writepage(page, mpd->wbc);
1699 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1701 * have successfully written the page
1702 * without skipping the same
1704 mpd->pages_written++;
1706 * In error case, we have to continue because
1707 * remaining pages are still locked
1708 * XXX: unlock and re-dirty them?
1713 pagevec_release(&pvec);
1719 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1721 * @mpd->inode - inode to walk through
1722 * @exbh->b_blocknr - first block on a disk
1723 * @exbh->b_size - amount of space in bytes
1724 * @logical - first logical block to start assignment with
1726 * the function goes through all passed space and put actual disk
1727 * block numbers into buffer heads, dropping BH_Delay
1729 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1730 struct buffer_head *exbh)
1732 struct inode *inode = mpd->inode;
1733 struct address_space *mapping = inode->i_mapping;
1734 int blocks = exbh->b_size >> inode->i_blkbits;
1735 sector_t pblock = exbh->b_blocknr, cur_logical;
1736 struct buffer_head *head, *bh;
1738 struct pagevec pvec;
1741 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1742 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1743 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1745 pagevec_init(&pvec, 0);
1747 while (index <= end) {
1748 /* XXX: optimize tail */
1749 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1752 for (i = 0; i < nr_pages; i++) {
1753 struct page *page = pvec.pages[i];
1755 index = page->index;
1760 BUG_ON(!PageLocked(page));
1761 BUG_ON(PageWriteback(page));
1762 BUG_ON(!page_has_buffers(page));
1764 bh = page_buffers(page);
1767 /* skip blocks out of the range */
1769 if (cur_logical >= logical)
1772 } while ((bh = bh->b_this_page) != head);
1775 if (cur_logical >= logical + blocks)
1777 if (buffer_delay(bh)) {
1778 bh->b_blocknr = pblock;
1779 clear_buffer_delay(bh);
1780 bh->b_bdev = inode->i_sb->s_bdev;
1781 } else if (buffer_unwritten(bh)) {
1782 bh->b_blocknr = pblock;
1783 clear_buffer_unwritten(bh);
1784 set_buffer_mapped(bh);
1786 bh->b_bdev = inode->i_sb->s_bdev;
1787 } else if (buffer_mapped(bh))
1788 BUG_ON(bh->b_blocknr != pblock);
1792 } while ((bh = bh->b_this_page) != head);
1794 pagevec_release(&pvec);
1800 * __unmap_underlying_blocks - just a helper function to unmap
1801 * set of blocks described by @bh
1803 static inline void __unmap_underlying_blocks(struct inode *inode,
1804 struct buffer_head *bh)
1806 struct block_device *bdev = inode->i_sb->s_bdev;
1809 blocks = bh->b_size >> inode->i_blkbits;
1810 for (i = 0; i < blocks; i++)
1811 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1814 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1815 sector_t logical, long blk_cnt)
1819 struct pagevec pvec;
1820 struct inode *inode = mpd->inode;
1821 struct address_space *mapping = inode->i_mapping;
1823 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1824 end = (logical + blk_cnt - 1) >>
1825 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1826 while (index <= end) {
1827 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1830 for (i = 0; i < nr_pages; i++) {
1831 struct page *page = pvec.pages[i];
1832 index = page->index;
1837 BUG_ON(!PageLocked(page));
1838 BUG_ON(PageWriteback(page));
1839 block_invalidatepage(page, 0);
1840 ClearPageUptodate(page);
1847 static void ext4_print_free_blocks(struct inode *inode)
1849 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1850 printk(KERN_EMERG "Total free blocks count %lld\n",
1851 ext4_count_free_blocks(inode->i_sb));
1852 printk(KERN_EMERG "Free/Dirty block details\n");
1853 printk(KERN_EMERG "free_blocks=%lld\n",
1854 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1855 printk(KERN_EMERG "dirty_blocks=%lld\n",
1856 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1857 printk(KERN_EMERG "Block reservation details\n");
1858 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1859 EXT4_I(inode)->i_reserved_data_blocks);
1860 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1861 EXT4_I(inode)->i_reserved_meta_blocks);
1866 * mpage_da_map_blocks - go through given space
1868 * @mpd->lbh - bh describing space
1869 * @mpd->get_block - the filesystem's block mapper function
1871 * The function skips space we know is already mapped to disk blocks.
1874 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1877 struct buffer_head new;
1878 struct buffer_head *lbh = &mpd->lbh;
1882 * We consider only non-mapped and non-allocated blocks
1884 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1886 new.b_state = lbh->b_state;
1888 new.b_size = lbh->b_size;
1889 next = lbh->b_blocknr;
1891 * If we didn't accumulate anything
1892 * to write simply return
1896 err = mpd->get_block(mpd->inode, next, &new, 1);
1899 /* If get block returns with error
1900 * we simply return. Later writepage
1901 * will redirty the page and writepages
1902 * will find the dirty page again
1907 if (err == -ENOSPC &&
1908 ext4_count_free_blocks(mpd->inode->i_sb)) {
1914 * get block failure will cause us
1915 * to loop in writepages. Because
1916 * a_ops->writepage won't be able to
1917 * make progress. The page will be redirtied
1918 * by writepage and writepages will again
1919 * try to write the same.
1921 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1922 "at logical offset %llu with max blocks "
1923 "%zd with error %d\n",
1924 __func__, mpd->inode->i_ino,
1925 (unsigned long long)next,
1926 lbh->b_size >> mpd->inode->i_blkbits, err);
1927 printk(KERN_EMERG "This should not happen.!! "
1928 "Data will be lost\n");
1929 if (err == -ENOSPC) {
1930 ext4_print_free_blocks(mpd->inode);
1932 /* invlaidate all the pages */
1933 ext4_da_block_invalidatepages(mpd, next,
1934 lbh->b_size >> mpd->inode->i_blkbits);
1937 BUG_ON(new.b_size == 0);
1939 if (buffer_new(&new))
1940 __unmap_underlying_blocks(mpd->inode, &new);
1943 * If blocks are delayed marked, we need to
1944 * put actual blocknr and drop delayed bit
1946 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1947 mpage_put_bnr_to_bhs(mpd, next, &new);
1952 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1953 (1 << BH_Delay) | (1 << BH_Unwritten))
1956 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1958 * @mpd->lbh - extent of blocks
1959 * @logical - logical number of the block in the file
1960 * @bh - bh of the block (used to access block's state)
1962 * the function is used to collect contig. blocks in same state
1964 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1965 sector_t logical, struct buffer_head *bh)
1968 size_t b_size = bh->b_size;
1969 struct buffer_head *lbh = &mpd->lbh;
1970 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1972 /* check if thereserved journal credits might overflow */
1973 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1974 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1976 * With non-extent format we are limited by the journal
1977 * credit available. Total credit needed to insert
1978 * nrblocks contiguous blocks is dependent on the
1979 * nrblocks. So limit nrblocks.
1982 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1983 EXT4_MAX_TRANS_DATA) {
1985 * Adding the new buffer_head would make it cross the
1986 * allowed limit for which we have journal credit
1987 * reserved. So limit the new bh->b_size
1989 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1990 mpd->inode->i_blkbits;
1991 /* we will do mpage_da_submit_io in the next loop */
1995 * First block in the extent
1997 if (lbh->b_size == 0) {
1998 lbh->b_blocknr = logical;
1999 lbh->b_size = b_size;
2000 lbh->b_state = bh->b_state & BH_FLAGS;
2004 next = lbh->b_blocknr + nrblocks;
2006 * Can we merge the block to our big extent?
2008 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
2009 lbh->b_size += b_size;
2015 * We couldn't merge the block to our extent, so we
2016 * need to flush current extent and start new one
2018 if (mpage_da_map_blocks(mpd) == 0)
2019 mpage_da_submit_io(mpd);
2025 * __mpage_da_writepage - finds extent of pages and blocks
2027 * @page: page to consider
2028 * @wbc: not used, we just follow rules
2031 * The function finds extents of pages and scan them for all blocks.
2033 static int __mpage_da_writepage(struct page *page,
2034 struct writeback_control *wbc, void *data)
2036 struct mpage_da_data *mpd = data;
2037 struct inode *inode = mpd->inode;
2038 struct buffer_head *bh, *head, fake;
2043 * Rest of the page in the page_vec
2044 * redirty then and skip then. We will
2045 * try to to write them again after
2046 * starting a new transaction
2048 redirty_page_for_writepage(wbc, page);
2050 return MPAGE_DA_EXTENT_TAIL;
2053 * Can we merge this page to current extent?
2055 if (mpd->next_page != page->index) {
2057 * Nope, we can't. So, we map non-allocated blocks
2058 * and start IO on them using writepage()
2060 if (mpd->next_page != mpd->first_page) {
2061 if (mpage_da_map_blocks(mpd) == 0)
2062 mpage_da_submit_io(mpd);
2064 * skip rest of the page in the page_vec
2067 redirty_page_for_writepage(wbc, page);
2069 return MPAGE_DA_EXTENT_TAIL;
2073 * Start next extent of pages ...
2075 mpd->first_page = page->index;
2080 mpd->lbh.b_size = 0;
2081 mpd->lbh.b_state = 0;
2082 mpd->lbh.b_blocknr = 0;
2085 mpd->next_page = page->index + 1;
2086 logical = (sector_t) page->index <<
2087 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2089 if (!page_has_buffers(page)) {
2091 * There is no attached buffer heads yet (mmap?)
2092 * we treat the page asfull of dirty blocks
2095 bh->b_size = PAGE_CACHE_SIZE;
2097 set_buffer_dirty(bh);
2098 set_buffer_uptodate(bh);
2099 mpage_add_bh_to_extent(mpd, logical, bh);
2101 return MPAGE_DA_EXTENT_TAIL;
2104 * Page with regular buffer heads, just add all dirty ones
2106 head = page_buffers(page);
2109 BUG_ON(buffer_locked(bh));
2111 * We need to try to allocate
2112 * unmapped blocks in the same page.
2113 * Otherwise we won't make progress
2114 * with the page in ext4_da_writepage
2116 if (buffer_dirty(bh) &&
2117 (!buffer_mapped(bh) || buffer_delay(bh))) {
2118 mpage_add_bh_to_extent(mpd, logical, bh);
2120 return MPAGE_DA_EXTENT_TAIL;
2121 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2123 * mapped dirty buffer. We need to update
2124 * the b_state because we look at
2125 * b_state in mpage_da_map_blocks. We don't
2126 * update b_size because if we find an
2127 * unmapped buffer_head later we need to
2128 * use the b_state flag of that buffer_head.
2130 if (mpd->lbh.b_size == 0)
2132 bh->b_state & BH_FLAGS;
2135 } while ((bh = bh->b_this_page) != head);
2142 * mpage_da_writepages - walk the list of dirty pages of the given
2143 * address space, allocates non-allocated blocks, maps newly-allocated
2144 * blocks to existing bhs and issue IO them
2146 * @mapping: address space structure to write
2147 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2148 * @get_block: the filesystem's block mapper function.
2150 * This is a library function, which implements the writepages()
2151 * address_space_operation.
2153 static int mpage_da_writepages(struct address_space *mapping,
2154 struct writeback_control *wbc,
2155 struct mpage_da_data *mpd)
2159 if (!mpd->get_block)
2160 return generic_writepages(mapping, wbc);
2162 mpd->lbh.b_size = 0;
2163 mpd->lbh.b_state = 0;
2164 mpd->lbh.b_blocknr = 0;
2165 mpd->first_page = 0;
2168 mpd->pages_written = 0;
2171 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2173 * Handle last extent of pages
2175 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2176 if (mpage_da_map_blocks(mpd) == 0)
2177 mpage_da_submit_io(mpd);
2180 ret = MPAGE_DA_EXTENT_TAIL;
2182 wbc->nr_to_write -= mpd->pages_written;
2187 * this is a special callback for ->write_begin() only
2188 * it's intention is to return mapped block or reserve space
2190 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2191 struct buffer_head *bh_result, int create)
2195 BUG_ON(create == 0);
2196 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2199 * first, we need to know whether the block is allocated already
2200 * preallocated blocks are unmapped but should treated
2201 * the same as allocated blocks.
2203 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2204 if ((ret == 0) && !buffer_delay(bh_result)) {
2205 /* the block isn't (pre)allocated yet, let's reserve space */
2207 * XXX: __block_prepare_write() unmaps passed block,
2210 ret = ext4_da_reserve_space(inode, 1);
2212 /* not enough space to reserve */
2215 map_bh(bh_result, inode->i_sb, 0);
2216 set_buffer_new(bh_result);
2217 set_buffer_delay(bh_result);
2218 } else if (ret > 0) {
2219 bh_result->b_size = (ret << inode->i_blkbits);
2225 #define EXT4_DELALLOC_RSVED 1
2226 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2227 struct buffer_head *bh_result, int create)
2230 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2231 loff_t disksize = EXT4_I(inode)->i_disksize;
2232 handle_t *handle = NULL;
2234 handle = ext4_journal_current_handle();
2236 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2237 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2240 bh_result->b_size = (ret << inode->i_blkbits);
2242 if (ext4_should_order_data(inode)) {
2244 retval = ext4_jbd2_file_inode(handle, inode);
2247 * Failed to add inode for ordered
2248 * mode. Don't update file size
2254 * Update on-disk size along with block allocation
2255 * we don't use 'extend_disksize' as size may change
2256 * within already allocated block -bzzz
2258 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2259 if (disksize > i_size_read(inode))
2260 disksize = i_size_read(inode);
2261 if (disksize > EXT4_I(inode)->i_disksize) {
2262 ext4_update_i_disksize(inode, disksize);
2263 ret = ext4_mark_inode_dirty(handle, inode);
2271 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2274 * unmapped buffer is possible for holes.
2275 * delay buffer is possible with delayed allocation
2277 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2280 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2281 struct buffer_head *bh_result, int create)
2284 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2287 * we don't want to do block allocation in writepage
2288 * so call get_block_wrap with create = 0
2290 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2291 bh_result, 0, 0, 0);
2293 bh_result->b_size = (ret << inode->i_blkbits);
2300 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2301 * get called via journal_submit_inode_data_buffers (no journal handle)
2302 * get called via shrink_page_list via pdflush (no journal handle)
2303 * or grab_page_cache when doing write_begin (have journal handle)
2305 static int ext4_da_writepage(struct page *page,
2306 struct writeback_control *wbc)
2311 struct buffer_head *page_bufs;
2312 struct inode *inode = page->mapping->host;
2314 size = i_size_read(inode);
2315 if (page->index == size >> PAGE_CACHE_SHIFT)
2316 len = size & ~PAGE_CACHE_MASK;
2318 len = PAGE_CACHE_SIZE;
2320 if (page_has_buffers(page)) {
2321 page_bufs = page_buffers(page);
2322 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2323 ext4_bh_unmapped_or_delay)) {
2325 * We don't want to do block allocation
2326 * So redirty the page and return
2327 * We may reach here when we do a journal commit
2328 * via journal_submit_inode_data_buffers.
2329 * If we don't have mapping block we just ignore
2330 * them. We can also reach here via shrink_page_list
2332 redirty_page_for_writepage(wbc, page);
2338 * The test for page_has_buffers() is subtle:
2339 * We know the page is dirty but it lost buffers. That means
2340 * that at some moment in time after write_begin()/write_end()
2341 * has been called all buffers have been clean and thus they
2342 * must have been written at least once. So they are all
2343 * mapped and we can happily proceed with mapping them
2344 * and writing the page.
2346 * Try to initialize the buffer_heads and check whether
2347 * all are mapped and non delay. We don't want to
2348 * do block allocation here.
2350 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2351 ext4_normal_get_block_write);
2353 page_bufs = page_buffers(page);
2354 /* check whether all are mapped and non delay */
2355 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2356 ext4_bh_unmapped_or_delay)) {
2357 redirty_page_for_writepage(wbc, page);
2363 * We can't do block allocation here
2364 * so just redity the page and unlock
2367 redirty_page_for_writepage(wbc, page);
2371 /* now mark the buffer_heads as dirty and uptodate */
2372 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2375 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2376 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2378 ret = block_write_full_page(page,
2379 ext4_normal_get_block_write,
2386 * This is called via ext4_da_writepages() to
2387 * calulate the total number of credits to reserve to fit
2388 * a single extent allocation into a single transaction,
2389 * ext4_da_writpeages() will loop calling this before
2390 * the block allocation.
2393 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2395 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2398 * With non-extent format the journal credit needed to
2399 * insert nrblocks contiguous block is dependent on
2400 * number of contiguous block. So we will limit
2401 * number of contiguous block to a sane value
2403 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2404 (max_blocks > EXT4_MAX_TRANS_DATA))
2405 max_blocks = EXT4_MAX_TRANS_DATA;
2407 return ext4_chunk_trans_blocks(inode, max_blocks);
2410 static int ext4_da_writepages(struct address_space *mapping,
2411 struct writeback_control *wbc)
2414 int range_whole = 0;
2415 handle_t *handle = NULL;
2416 struct mpage_da_data mpd;
2417 struct inode *inode = mapping->host;
2418 int no_nrwrite_index_update;
2419 long pages_written = 0, pages_skipped;
2420 int needed_blocks, ret = 0, nr_to_writebump = 0;
2421 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2424 * No pages to write? This is mainly a kludge to avoid starting
2425 * a transaction for special inodes like journal inode on last iput()
2426 * because that could violate lock ordering on umount
2428 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2432 * If the filesystem has aborted, it is read-only, so return
2433 * right away instead of dumping stack traces later on that
2434 * will obscure the real source of the problem. We test
2435 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2436 * the latter could be true if the filesystem is mounted
2437 * read-only, and in that case, ext4_da_writepages should
2438 * *never* be called, so if that ever happens, we would want
2441 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2445 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2446 * This make sure small files blocks are allocated in
2447 * single attempt. This ensure that small files
2448 * get less fragmented.
2450 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2451 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2452 wbc->nr_to_write = sbi->s_mb_stream_request;
2454 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2457 if (wbc->range_cyclic)
2458 index = mapping->writeback_index;
2460 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2463 mpd.inode = mapping->host;
2466 * we don't want write_cache_pages to update
2467 * nr_to_write and writeback_index
2469 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2470 wbc->no_nrwrite_index_update = 1;
2471 pages_skipped = wbc->pages_skipped;
2473 while (!ret && wbc->nr_to_write > 0) {
2476 * we insert one extent at a time. So we need
2477 * credit needed for single extent allocation.
2478 * journalled mode is currently not supported
2481 BUG_ON(ext4_should_journal_data(inode));
2482 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2484 /* start a new transaction*/
2485 handle = ext4_journal_start(inode, needed_blocks);
2486 if (IS_ERR(handle)) {
2487 ret = PTR_ERR(handle);
2488 printk(KERN_CRIT "%s: jbd2_start: "
2489 "%ld pages, ino %lu; err %d\n", __func__,
2490 wbc->nr_to_write, inode->i_ino, ret);
2492 goto out_writepages;
2494 mpd.get_block = ext4_da_get_block_write;
2495 ret = mpage_da_writepages(mapping, wbc, &mpd);
2497 ext4_journal_stop(handle);
2499 if (mpd.retval == -ENOSPC) {
2500 /* commit the transaction which would
2501 * free blocks released in the transaction
2504 jbd2_journal_force_commit_nested(sbi->s_journal);
2505 wbc->pages_skipped = pages_skipped;
2507 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2509 * got one extent now try with
2512 pages_written += mpd.pages_written;
2513 wbc->pages_skipped = pages_skipped;
2515 } else if (wbc->nr_to_write)
2517 * There is no more writeout needed
2518 * or we requested for a noblocking writeout
2519 * and we found the device congested
2523 if (pages_skipped != wbc->pages_skipped)
2524 printk(KERN_EMERG "This should not happen leaving %s "
2525 "with nr_to_write = %ld ret = %d\n",
2526 __func__, wbc->nr_to_write, ret);
2529 index += pages_written;
2530 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2532 * set the writeback_index so that range_cyclic
2533 * mode will write it back later
2535 mapping->writeback_index = index;
2538 if (!no_nrwrite_index_update)
2539 wbc->no_nrwrite_index_update = 0;
2540 wbc->nr_to_write -= nr_to_writebump;
2544 #define FALL_BACK_TO_NONDELALLOC 1
2545 static int ext4_nonda_switch(struct super_block *sb)
2547 s64 free_blocks, dirty_blocks;
2548 struct ext4_sb_info *sbi = EXT4_SB(sb);
2551 * switch to non delalloc mode if we are running low
2552 * on free block. The free block accounting via percpu
2553 * counters can get slightly wrong with FBC_BATCH getting
2554 * accumulated on each CPU without updating global counters
2555 * Delalloc need an accurate free block accounting. So switch
2556 * to non delalloc when we are near to error range.
2558 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2559 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2560 if (2 * free_blocks < 3 * dirty_blocks ||
2561 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2563 * free block count is less that 150% of dirty blocks
2564 * or free blocks is less that watermark
2571 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2572 loff_t pos, unsigned len, unsigned flags,
2573 struct page **pagep, void **fsdata)
2575 int ret, retries = 0;
2579 struct inode *inode = mapping->host;
2582 index = pos >> PAGE_CACHE_SHIFT;
2583 from = pos & (PAGE_CACHE_SIZE - 1);
2586 if (ext4_nonda_switch(inode->i_sb)) {
2587 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2588 return ext4_write_begin(file, mapping, pos,
2589 len, flags, pagep, fsdata);
2591 *fsdata = (void *)0;
2594 * With delayed allocation, we don't log the i_disksize update
2595 * if there is delayed block allocation. But we still need
2596 * to journalling the i_disksize update if writes to the end
2597 * of file which has an already mapped buffer.
2599 handle = ext4_journal_start(inode, 1);
2600 if (IS_ERR(handle)) {
2601 ret = PTR_ERR(handle);
2605 page = grab_cache_page_write_begin(mapping, index, flags);
2607 ext4_journal_stop(handle);
2613 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2614 ext4_da_get_block_prep);
2617 ext4_journal_stop(handle);
2618 page_cache_release(page);
2620 * block_write_begin may have instantiated a few blocks
2621 * outside i_size. Trim these off again. Don't need
2622 * i_size_read because we hold i_mutex.
2624 if (pos + len > inode->i_size)
2625 vmtruncate(inode, inode->i_size);
2628 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2635 * Check if we should update i_disksize
2636 * when write to the end of file but not require block allocation
2638 static int ext4_da_should_update_i_disksize(struct page *page,
2639 unsigned long offset)
2641 struct buffer_head *bh;
2642 struct inode *inode = page->mapping->host;
2646 bh = page_buffers(page);
2647 idx = offset >> inode->i_blkbits;
2649 for (i = 0; i < idx; i++)
2650 bh = bh->b_this_page;
2652 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2657 static int ext4_da_write_end(struct file *file,
2658 struct address_space *mapping,
2659 loff_t pos, unsigned len, unsigned copied,
2660 struct page *page, void *fsdata)
2662 struct inode *inode = mapping->host;
2664 handle_t *handle = ext4_journal_current_handle();
2666 unsigned long start, end;
2667 int write_mode = (int)(unsigned long)fsdata;
2669 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2670 if (ext4_should_order_data(inode)) {
2671 return ext4_ordered_write_end(file, mapping, pos,
2672 len, copied, page, fsdata);
2673 } else if (ext4_should_writeback_data(inode)) {
2674 return ext4_writeback_write_end(file, mapping, pos,
2675 len, copied, page, fsdata);
2681 start = pos & (PAGE_CACHE_SIZE - 1);
2682 end = start + copied - 1;
2685 * generic_write_end() will run mark_inode_dirty() if i_size
2686 * changes. So let's piggyback the i_disksize mark_inode_dirty
2690 new_i_size = pos + copied;
2691 if (new_i_size > EXT4_I(inode)->i_disksize) {
2692 if (ext4_da_should_update_i_disksize(page, end)) {
2693 down_write(&EXT4_I(inode)->i_data_sem);
2694 if (new_i_size > EXT4_I(inode)->i_disksize) {
2696 * Updating i_disksize when extending file
2697 * without needing block allocation
2699 if (ext4_should_order_data(inode))
2700 ret = ext4_jbd2_file_inode(handle,
2703 EXT4_I(inode)->i_disksize = new_i_size;
2705 up_write(&EXT4_I(inode)->i_data_sem);
2706 /* We need to mark inode dirty even if
2707 * new_i_size is less that inode->i_size
2708 * bu greater than i_disksize.(hint delalloc)
2710 ext4_mark_inode_dirty(handle, inode);
2713 ret2 = generic_write_end(file, mapping, pos, len, copied,
2718 ret2 = ext4_journal_stop(handle);
2722 return ret ? ret : copied;
2725 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2728 * Drop reserved blocks
2730 BUG_ON(!PageLocked(page));
2731 if (!page_has_buffers(page))
2734 ext4_da_page_release_reservation(page, offset);
2737 ext4_invalidatepage(page, offset);
2744 * bmap() is special. It gets used by applications such as lilo and by
2745 * the swapper to find the on-disk block of a specific piece of data.
2747 * Naturally, this is dangerous if the block concerned is still in the
2748 * journal. If somebody makes a swapfile on an ext4 data-journaling
2749 * filesystem and enables swap, then they may get a nasty shock when the
2750 * data getting swapped to that swapfile suddenly gets overwritten by
2751 * the original zero's written out previously to the journal and
2752 * awaiting writeback in the kernel's buffer cache.
2754 * So, if we see any bmap calls here on a modified, data-journaled file,
2755 * take extra steps to flush any blocks which might be in the cache.
2757 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2759 struct inode *inode = mapping->host;
2763 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2764 test_opt(inode->i_sb, DELALLOC)) {
2766 * With delalloc we want to sync the file
2767 * so that we can make sure we allocate
2770 filemap_write_and_wait(mapping);
2773 BUG_ON(!EXT4_JOURNAL(inode) &&
2774 EXT4_I(inode)->i_state & EXT4_STATE_JDATA);
2776 if (EXT4_JOURNAL(inode) && EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2778 * This is a REALLY heavyweight approach, but the use of
2779 * bmap on dirty files is expected to be extremely rare:
2780 * only if we run lilo or swapon on a freshly made file
2781 * do we expect this to happen.
2783 * (bmap requires CAP_SYS_RAWIO so this does not
2784 * represent an unprivileged user DOS attack --- we'd be
2785 * in trouble if mortal users could trigger this path at
2788 * NB. EXT4_STATE_JDATA is not set on files other than
2789 * regular files. If somebody wants to bmap a directory
2790 * or symlink and gets confused because the buffer
2791 * hasn't yet been flushed to disk, they deserve
2792 * everything they get.
2795 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2796 journal = EXT4_JOURNAL(inode);
2797 jbd2_journal_lock_updates(journal);
2798 err = jbd2_journal_flush(journal);
2799 jbd2_journal_unlock_updates(journal);
2805 return generic_block_bmap(mapping, block, ext4_get_block);
2808 static int bget_one(handle_t *handle, struct buffer_head *bh)
2814 static int bput_one(handle_t *handle, struct buffer_head *bh)
2821 * Note that we don't need to start a transaction unless we're journaling data
2822 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2823 * need to file the inode to the transaction's list in ordered mode because if
2824 * we are writing back data added by write(), the inode is already there and if
2825 * we are writing back data modified via mmap(), noone guarantees in which
2826 * transaction the data will hit the disk. In case we are journaling data, we
2827 * cannot start transaction directly because transaction start ranks above page
2828 * lock so we have to do some magic.
2830 * In all journaling modes block_write_full_page() will start the I/O.
2834 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2839 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2841 * Same applies to ext4_get_block(). We will deadlock on various things like
2842 * lock_journal and i_data_sem
2844 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2847 * 16May01: If we're reentered then journal_current_handle() will be
2848 * non-zero. We simply *return*.
2850 * 1 July 2001: @@@ FIXME:
2851 * In journalled data mode, a data buffer may be metadata against the
2852 * current transaction. But the same file is part of a shared mapping
2853 * and someone does a writepage() on it.
2855 * We will move the buffer onto the async_data list, but *after* it has
2856 * been dirtied. So there's a small window where we have dirty data on
2859 * Note that this only applies to the last partial page in the file. The
2860 * bit which block_write_full_page() uses prepare/commit for. (That's
2861 * broken code anyway: it's wrong for msync()).
2863 * It's a rare case: affects the final partial page, for journalled data
2864 * where the file is subject to bith write() and writepage() in the same
2865 * transction. To fix it we'll need a custom block_write_full_page().
2866 * We'll probably need that anyway for journalling writepage() output.
2868 * We don't honour synchronous mounts for writepage(). That would be
2869 * disastrous. Any write() or metadata operation will sync the fs for
2873 static int __ext4_normal_writepage(struct page *page,
2874 struct writeback_control *wbc)
2876 struct inode *inode = page->mapping->host;
2878 if (test_opt(inode->i_sb, NOBH))
2879 return nobh_writepage(page,
2880 ext4_normal_get_block_write, wbc);
2882 return block_write_full_page(page,
2883 ext4_normal_get_block_write,
2887 static int ext4_normal_writepage(struct page *page,
2888 struct writeback_control *wbc)
2890 struct inode *inode = page->mapping->host;
2891 loff_t size = i_size_read(inode);
2894 J_ASSERT(PageLocked(page));
2895 if (page->index == size >> PAGE_CACHE_SHIFT)
2896 len = size & ~PAGE_CACHE_MASK;
2898 len = PAGE_CACHE_SIZE;
2900 if (page_has_buffers(page)) {
2901 /* if page has buffers it should all be mapped
2902 * and allocated. If there are not buffers attached
2903 * to the page we know the page is dirty but it lost
2904 * buffers. That means that at some moment in time
2905 * after write_begin() / write_end() has been called
2906 * all buffers have been clean and thus they must have been
2907 * written at least once. So they are all mapped and we can
2908 * happily proceed with mapping them and writing the page.
2910 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2911 ext4_bh_unmapped_or_delay));
2914 if (!ext4_journal_current_handle())
2915 return __ext4_normal_writepage(page, wbc);
2917 redirty_page_for_writepage(wbc, page);
2922 static int __ext4_journalled_writepage(struct page *page,
2923 struct writeback_control *wbc)
2925 struct address_space *mapping = page->mapping;
2926 struct inode *inode = mapping->host;
2927 struct buffer_head *page_bufs;
2928 handle_t *handle = NULL;
2932 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2933 ext4_normal_get_block_write);
2937 page_bufs = page_buffers(page);
2938 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2940 /* As soon as we unlock the page, it can go away, but we have
2941 * references to buffers so we are safe */
2944 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2945 if (IS_ERR(handle)) {
2946 ret = PTR_ERR(handle);
2950 ret = walk_page_buffers(handle, page_bufs, 0,
2951 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2953 err = walk_page_buffers(handle, page_bufs, 0,
2954 PAGE_CACHE_SIZE, NULL, write_end_fn);
2957 err = ext4_journal_stop(handle);
2961 walk_page_buffers(handle, page_bufs, 0,
2962 PAGE_CACHE_SIZE, NULL, bput_one);
2963 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2972 static int ext4_journalled_writepage(struct page *page,
2973 struct writeback_control *wbc)
2975 struct inode *inode = page->mapping->host;
2976 loff_t size = i_size_read(inode);
2979 J_ASSERT(PageLocked(page));
2980 if (page->index == size >> PAGE_CACHE_SHIFT)
2981 len = size & ~PAGE_CACHE_MASK;
2983 len = PAGE_CACHE_SIZE;
2985 if (page_has_buffers(page)) {
2986 /* if page has buffers it should all be mapped
2987 * and allocated. If there are not buffers attached
2988 * to the page we know the page is dirty but it lost
2989 * buffers. That means that at some moment in time
2990 * after write_begin() / write_end() has been called
2991 * all buffers have been clean and thus they must have been
2992 * written at least once. So they are all mapped and we can
2993 * happily proceed with mapping them and writing the page.
2995 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2996 ext4_bh_unmapped_or_delay));
2999 if (ext4_journal_current_handle())
3002 if (PageChecked(page)) {
3004 * It's mmapped pagecache. Add buffers and journal it. There
3005 * doesn't seem much point in redirtying the page here.
3007 ClearPageChecked(page);
3008 return __ext4_journalled_writepage(page, wbc);
3011 * It may be a page full of checkpoint-mode buffers. We don't
3012 * really know unless we go poke around in the buffer_heads.
3013 * But block_write_full_page will do the right thing.
3015 return block_write_full_page(page,
3016 ext4_normal_get_block_write,
3020 redirty_page_for_writepage(wbc, page);
3025 static int ext4_readpage(struct file *file, struct page *page)
3027 return mpage_readpage(page, ext4_get_block);
3031 ext4_readpages(struct file *file, struct address_space *mapping,
3032 struct list_head *pages, unsigned nr_pages)
3034 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3037 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3039 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3042 * If it's a full truncate we just forget about the pending dirtying
3045 ClearPageChecked(page);
3048 jbd2_journal_invalidatepage(journal, page, offset);
3050 block_invalidatepage(page, offset);
3053 static int ext4_releasepage(struct page *page, gfp_t wait)
3055 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3057 WARN_ON(PageChecked(page));
3058 if (!page_has_buffers(page))
3061 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3063 return try_to_free_buffers(page);
3067 * If the O_DIRECT write will extend the file then add this inode to the
3068 * orphan list. So recovery will truncate it back to the original size
3069 * if the machine crashes during the write.
3071 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3072 * crashes then stale disk data _may_ be exposed inside the file. But current
3073 * VFS code falls back into buffered path in that case so we are safe.
3075 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3076 const struct iovec *iov, loff_t offset,
3077 unsigned long nr_segs)
3079 struct file *file = iocb->ki_filp;
3080 struct inode *inode = file->f_mapping->host;
3081 struct ext4_inode_info *ei = EXT4_I(inode);
3085 size_t count = iov_length(iov, nr_segs);
3088 loff_t final_size = offset + count;
3090 if (final_size > inode->i_size) {
3091 /* Credits for sb + inode write */
3092 handle = ext4_journal_start(inode, 2);
3093 if (IS_ERR(handle)) {
3094 ret = PTR_ERR(handle);
3097 ret = ext4_orphan_add(handle, inode);
3099 ext4_journal_stop(handle);
3103 ei->i_disksize = inode->i_size;
3104 ext4_journal_stop(handle);
3108 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3110 ext4_get_block, NULL);
3115 /* Credits for sb + inode write */
3116 handle = ext4_journal_start(inode, 2);
3117 if (IS_ERR(handle)) {
3118 /* This is really bad luck. We've written the data
3119 * but cannot extend i_size. Bail out and pretend
3120 * the write failed... */
3121 ret = PTR_ERR(handle);
3125 ext4_orphan_del(handle, inode);
3127 loff_t end = offset + ret;
3128 if (end > inode->i_size) {
3129 ei->i_disksize = end;
3130 i_size_write(inode, end);
3132 * We're going to return a positive `ret'
3133 * here due to non-zero-length I/O, so there's
3134 * no way of reporting error returns from
3135 * ext4_mark_inode_dirty() to userspace. So
3138 ext4_mark_inode_dirty(handle, inode);
3141 err = ext4_journal_stop(handle);
3150 * Pages can be marked dirty completely asynchronously from ext4's journalling
3151 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3152 * much here because ->set_page_dirty is called under VFS locks. The page is
3153 * not necessarily locked.
3155 * We cannot just dirty the page and leave attached buffers clean, because the
3156 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3157 * or jbddirty because all the journalling code will explode.
3159 * So what we do is to mark the page "pending dirty" and next time writepage
3160 * is called, propagate that into the buffers appropriately.
3162 static int ext4_journalled_set_page_dirty(struct page *page)
3164 SetPageChecked(page);
3165 return __set_page_dirty_nobuffers(page);
3168 static const struct address_space_operations ext4_ordered_aops = {
3169 .readpage = ext4_readpage,
3170 .readpages = ext4_readpages,
3171 .writepage = ext4_normal_writepage,
3172 .sync_page = block_sync_page,
3173 .write_begin = ext4_write_begin,
3174 .write_end = ext4_ordered_write_end,
3176 .invalidatepage = ext4_invalidatepage,
3177 .releasepage = ext4_releasepage,
3178 .direct_IO = ext4_direct_IO,
3179 .migratepage = buffer_migrate_page,
3180 .is_partially_uptodate = block_is_partially_uptodate,
3183 static const struct address_space_operations ext4_writeback_aops = {
3184 .readpage = ext4_readpage,
3185 .readpages = ext4_readpages,
3186 .writepage = ext4_normal_writepage,
3187 .sync_page = block_sync_page,
3188 .write_begin = ext4_write_begin,
3189 .write_end = ext4_writeback_write_end,
3191 .invalidatepage = ext4_invalidatepage,
3192 .releasepage = ext4_releasepage,
3193 .direct_IO = ext4_direct_IO,
3194 .migratepage = buffer_migrate_page,
3195 .is_partially_uptodate = block_is_partially_uptodate,
3198 static const struct address_space_operations ext4_journalled_aops = {
3199 .readpage = ext4_readpage,
3200 .readpages = ext4_readpages,
3201 .writepage = ext4_journalled_writepage,
3202 .sync_page = block_sync_page,
3203 .write_begin = ext4_write_begin,
3204 .write_end = ext4_journalled_write_end,
3205 .set_page_dirty = ext4_journalled_set_page_dirty,
3207 .invalidatepage = ext4_invalidatepage,
3208 .releasepage = ext4_releasepage,
3209 .is_partially_uptodate = block_is_partially_uptodate,
3212 static const struct address_space_operations ext4_da_aops = {
3213 .readpage = ext4_readpage,
3214 .readpages = ext4_readpages,
3215 .writepage = ext4_da_writepage,
3216 .writepages = ext4_da_writepages,
3217 .sync_page = block_sync_page,
3218 .write_begin = ext4_da_write_begin,
3219 .write_end = ext4_da_write_end,
3221 .invalidatepage = ext4_da_invalidatepage,
3222 .releasepage = ext4_releasepage,
3223 .direct_IO = ext4_direct_IO,
3224 .migratepage = buffer_migrate_page,
3225 .is_partially_uptodate = block_is_partially_uptodate,
3228 void ext4_set_aops(struct inode *inode)
3230 if (ext4_should_order_data(inode) &&
3231 test_opt(inode->i_sb, DELALLOC))
3232 inode->i_mapping->a_ops = &ext4_da_aops;
3233 else if (ext4_should_order_data(inode))
3234 inode->i_mapping->a_ops = &ext4_ordered_aops;
3235 else if (ext4_should_writeback_data(inode) &&
3236 test_opt(inode->i_sb, DELALLOC))
3237 inode->i_mapping->a_ops = &ext4_da_aops;
3238 else if (ext4_should_writeback_data(inode))
3239 inode->i_mapping->a_ops = &ext4_writeback_aops;
3241 inode->i_mapping->a_ops = &ext4_journalled_aops;
3245 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3246 * up to the end of the block which corresponds to `from'.
3247 * This required during truncate. We need to physically zero the tail end
3248 * of that block so it doesn't yield old data if the file is later grown.
3250 int ext4_block_truncate_page(handle_t *handle,
3251 struct address_space *mapping, loff_t from)
3253 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3254 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3255 unsigned blocksize, length, pos;
3257 struct inode *inode = mapping->host;
3258 struct buffer_head *bh;
3262 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3266 blocksize = inode->i_sb->s_blocksize;
3267 length = blocksize - (offset & (blocksize - 1));
3268 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3271 * For "nobh" option, we can only work if we don't need to
3272 * read-in the page - otherwise we create buffers to do the IO.
3274 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3275 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3276 zero_user(page, offset, length);
3277 set_page_dirty(page);
3281 if (!page_has_buffers(page))
3282 create_empty_buffers(page, blocksize, 0);
3284 /* Find the buffer that contains "offset" */
3285 bh = page_buffers(page);
3287 while (offset >= pos) {
3288 bh = bh->b_this_page;
3294 if (buffer_freed(bh)) {
3295 BUFFER_TRACE(bh, "freed: skip");
3299 if (!buffer_mapped(bh)) {
3300 BUFFER_TRACE(bh, "unmapped");
3301 ext4_get_block(inode, iblock, bh, 0);
3302 /* unmapped? It's a hole - nothing to do */
3303 if (!buffer_mapped(bh)) {
3304 BUFFER_TRACE(bh, "still unmapped");
3309 /* Ok, it's mapped. Make sure it's up-to-date */
3310 if (PageUptodate(page))
3311 set_buffer_uptodate(bh);
3313 if (!buffer_uptodate(bh)) {
3315 ll_rw_block(READ, 1, &bh);
3317 /* Uhhuh. Read error. Complain and punt. */
3318 if (!buffer_uptodate(bh))
3322 if (ext4_should_journal_data(inode)) {
3323 BUFFER_TRACE(bh, "get write access");
3324 err = ext4_journal_get_write_access(handle, bh);
3329 zero_user(page, offset, length);
3331 BUFFER_TRACE(bh, "zeroed end of block");
3334 if (ext4_should_journal_data(inode)) {
3335 err = ext4_handle_dirty_metadata(handle, inode, bh);
3337 if (ext4_should_order_data(inode))
3338 err = ext4_jbd2_file_inode(handle, inode);
3339 mark_buffer_dirty(bh);
3344 page_cache_release(page);
3349 * Probably it should be a library function... search for first non-zero word
3350 * or memcmp with zero_page, whatever is better for particular architecture.
3353 static inline int all_zeroes(__le32 *p, __le32 *q)
3362 * ext4_find_shared - find the indirect blocks for partial truncation.
3363 * @inode: inode in question
3364 * @depth: depth of the affected branch
3365 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3366 * @chain: place to store the pointers to partial indirect blocks
3367 * @top: place to the (detached) top of branch
3369 * This is a helper function used by ext4_truncate().
3371 * When we do truncate() we may have to clean the ends of several
3372 * indirect blocks but leave the blocks themselves alive. Block is
3373 * partially truncated if some data below the new i_size is refered
3374 * from it (and it is on the path to the first completely truncated
3375 * data block, indeed). We have to free the top of that path along
3376 * with everything to the right of the path. Since no allocation
3377 * past the truncation point is possible until ext4_truncate()
3378 * finishes, we may safely do the latter, but top of branch may
3379 * require special attention - pageout below the truncation point
3380 * might try to populate it.
3382 * We atomically detach the top of branch from the tree, store the
3383 * block number of its root in *@top, pointers to buffer_heads of
3384 * partially truncated blocks - in @chain[].bh and pointers to
3385 * their last elements that should not be removed - in
3386 * @chain[].p. Return value is the pointer to last filled element
3389 * The work left to caller to do the actual freeing of subtrees:
3390 * a) free the subtree starting from *@top
3391 * b) free the subtrees whose roots are stored in
3392 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3393 * c) free the subtrees growing from the inode past the @chain[0].
3394 * (no partially truncated stuff there). */
3396 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3397 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3399 Indirect *partial, *p;
3403 /* Make k index the deepest non-null offest + 1 */
3404 for (k = depth; k > 1 && !offsets[k-1]; k--)
3406 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3407 /* Writer: pointers */
3409 partial = chain + k-1;
3411 * If the branch acquired continuation since we've looked at it -
3412 * fine, it should all survive and (new) top doesn't belong to us.
3414 if (!partial->key && *partial->p)
3417 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3420 * OK, we've found the last block that must survive. The rest of our
3421 * branch should be detached before unlocking. However, if that rest
3422 * of branch is all ours and does not grow immediately from the inode
3423 * it's easier to cheat and just decrement partial->p.
3425 if (p == chain + k - 1 && p > chain) {
3429 /* Nope, don't do this in ext4. Must leave the tree intact */
3436 while (partial > p) {
3437 brelse(partial->bh);
3445 * Zero a number of block pointers in either an inode or an indirect block.
3446 * If we restart the transaction we must again get write access to the
3447 * indirect block for further modification.
3449 * We release `count' blocks on disk, but (last - first) may be greater
3450 * than `count' because there can be holes in there.
3452 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3453 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3454 unsigned long count, __le32 *first, __le32 *last)
3457 if (try_to_extend_transaction(handle, inode)) {
3459 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
3460 ext4_handle_dirty_metadata(handle, inode, bh);
3462 ext4_mark_inode_dirty(handle, inode);
3463 ext4_journal_test_restart(handle, inode);
3465 BUFFER_TRACE(bh, "retaking write access");
3466 ext4_journal_get_write_access(handle, bh);
3471 * Any buffers which are on the journal will be in memory. We find
3472 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3473 * on them. We've already detached each block from the file, so
3474 * bforget() in jbd2_journal_forget() should be safe.
3476 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3478 for (p = first; p < last; p++) {
3479 u32 nr = le32_to_cpu(*p);
3481 struct buffer_head *tbh;
3484 tbh = sb_find_get_block(inode->i_sb, nr);
3485 ext4_forget(handle, 0, inode, tbh, nr);
3489 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3493 * ext4_free_data - free a list of data blocks
3494 * @handle: handle for this transaction
3495 * @inode: inode we are dealing with
3496 * @this_bh: indirect buffer_head which contains *@first and *@last
3497 * @first: array of block numbers
3498 * @last: points immediately past the end of array
3500 * We are freeing all blocks refered from that array (numbers are stored as
3501 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3503 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3504 * blocks are contiguous then releasing them at one time will only affect one
3505 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3506 * actually use a lot of journal space.
3508 * @this_bh will be %NULL if @first and @last point into the inode's direct
3511 static void ext4_free_data(handle_t *handle, struct inode *inode,
3512 struct buffer_head *this_bh,
3513 __le32 *first, __le32 *last)
3515 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3516 unsigned long count = 0; /* Number of blocks in the run */
3517 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3520 ext4_fsblk_t nr; /* Current block # */
3521 __le32 *p; /* Pointer into inode/ind
3522 for current block */
3525 if (this_bh) { /* For indirect block */
3526 BUFFER_TRACE(this_bh, "get_write_access");
3527 err = ext4_journal_get_write_access(handle, this_bh);
3528 /* Important: if we can't update the indirect pointers
3529 * to the blocks, we can't free them. */
3534 for (p = first; p < last; p++) {
3535 nr = le32_to_cpu(*p);
3537 /* accumulate blocks to free if they're contiguous */
3540 block_to_free_p = p;
3542 } else if (nr == block_to_free + count) {
3545 ext4_clear_blocks(handle, inode, this_bh,
3547 count, block_to_free_p, p);
3549 block_to_free_p = p;
3556 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3557 count, block_to_free_p, p);
3560 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
3563 * The buffer head should have an attached journal head at this
3564 * point. However, if the data is corrupted and an indirect
3565 * block pointed to itself, it would have been detached when
3566 * the block was cleared. Check for this instead of OOPSing.
3569 ext4_handle_dirty_metadata(handle, inode, this_bh);
3571 ext4_error(inode->i_sb, __func__,
3572 "circular indirect block detected, "
3573 "inode=%lu, block=%llu",
3575 (unsigned long long) this_bh->b_blocknr);
3580 * ext4_free_branches - free an array of branches
3581 * @handle: JBD handle for this transaction
3582 * @inode: inode we are dealing with
3583 * @parent_bh: the buffer_head which contains *@first and *@last
3584 * @first: array of block numbers
3585 * @last: pointer immediately past the end of array
3586 * @depth: depth of the branches to free
3588 * We are freeing all blocks refered from these branches (numbers are
3589 * stored as little-endian 32-bit) and updating @inode->i_blocks
3592 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3593 struct buffer_head *parent_bh,
3594 __le32 *first, __le32 *last, int depth)
3599 if (ext4_handle_is_aborted(handle))
3603 struct buffer_head *bh;
3604 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3606 while (--p >= first) {
3607 nr = le32_to_cpu(*p);
3609 continue; /* A hole */
3611 /* Go read the buffer for the next level down */
3612 bh = sb_bread(inode->i_sb, nr);
3615 * A read failure? Report error and clear slot
3619 ext4_error(inode->i_sb, "ext4_free_branches",
3620 "Read failure, inode=%lu, block=%llu",
3625 /* This zaps the entire block. Bottom up. */
3626 BUFFER_TRACE(bh, "free child branches");
3627 ext4_free_branches(handle, inode, bh,
3628 (__le32 *) bh->b_data,
3629 (__le32 *) bh->b_data + addr_per_block,
3633 * We've probably journalled the indirect block several
3634 * times during the truncate. But it's no longer
3635 * needed and we now drop it from the transaction via
3636 * jbd2_journal_revoke().
3638 * That's easy if it's exclusively part of this
3639 * transaction. But if it's part of the committing
3640 * transaction then jbd2_journal_forget() will simply
3641 * brelse() it. That means that if the underlying
3642 * block is reallocated in ext4_get_block(),
3643 * unmap_underlying_metadata() will find this block
3644 * and will try to get rid of it. damn, damn.
3646 * If this block has already been committed to the
3647 * journal, a revoke record will be written. And
3648 * revoke records must be emitted *before* clearing
3649 * this block's bit in the bitmaps.
3651 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3654 * Everything below this this pointer has been
3655 * released. Now let this top-of-subtree go.
3657 * We want the freeing of this indirect block to be
3658 * atomic in the journal with the updating of the
3659 * bitmap block which owns it. So make some room in
3662 * We zero the parent pointer *after* freeing its
3663 * pointee in the bitmaps, so if extend_transaction()
3664 * for some reason fails to put the bitmap changes and
3665 * the release into the same transaction, recovery
3666 * will merely complain about releasing a free block,
3667 * rather than leaking blocks.
3669 if (ext4_handle_is_aborted(handle))
3671 if (try_to_extend_transaction(handle, inode)) {
3672 ext4_mark_inode_dirty(handle, inode);
3673 ext4_journal_test_restart(handle, inode);
3676 ext4_free_blocks(handle, inode, nr, 1, 1);
3680 * The block which we have just freed is
3681 * pointed to by an indirect block: journal it
3683 BUFFER_TRACE(parent_bh, "get_write_access");
3684 if (!ext4_journal_get_write_access(handle,
3687 BUFFER_TRACE(parent_bh,
3688 "call ext4_handle_dirty_metadata");
3689 ext4_handle_dirty_metadata(handle,
3696 /* We have reached the bottom of the tree. */
3697 BUFFER_TRACE(parent_bh, "free data blocks");
3698 ext4_free_data(handle, inode, parent_bh, first, last);
3702 int ext4_can_truncate(struct inode *inode)
3704 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3706 if (S_ISREG(inode->i_mode))
3708 if (S_ISDIR(inode->i_mode))
3710 if (S_ISLNK(inode->i_mode))
3711 return !ext4_inode_is_fast_symlink(inode);
3718 * We block out ext4_get_block() block instantiations across the entire
3719 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3720 * simultaneously on behalf of the same inode.
3722 * As we work through the truncate and commmit bits of it to the journal there
3723 * is one core, guiding principle: the file's tree must always be consistent on
3724 * disk. We must be able to restart the truncate after a crash.
3726 * The file's tree may be transiently inconsistent in memory (although it
3727 * probably isn't), but whenever we close off and commit a journal transaction,
3728 * the contents of (the filesystem + the journal) must be consistent and
3729 * restartable. It's pretty simple, really: bottom up, right to left (although
3730 * left-to-right works OK too).
3732 * Note that at recovery time, journal replay occurs *before* the restart of
3733 * truncate against the orphan inode list.
3735 * The committed inode has the new, desired i_size (which is the same as
3736 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3737 * that this inode's truncate did not complete and it will again call
3738 * ext4_truncate() to have another go. So there will be instantiated blocks
3739 * to the right of the truncation point in a crashed ext4 filesystem. But
3740 * that's fine - as long as they are linked from the inode, the post-crash
3741 * ext4_truncate() run will find them and release them.
3743 void ext4_truncate(struct inode *inode)
3746 struct ext4_inode_info *ei = EXT4_I(inode);
3747 __le32 *i_data = ei->i_data;
3748 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3749 struct address_space *mapping = inode->i_mapping;
3750 ext4_lblk_t offsets[4];
3755 ext4_lblk_t last_block;
3756 unsigned blocksize = inode->i_sb->s_blocksize;
3758 if (!ext4_can_truncate(inode))
3761 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3762 ext4_ext_truncate(inode);
3766 handle = start_transaction(inode);
3768 return; /* AKPM: return what? */
3770 last_block = (inode->i_size + blocksize-1)
3771 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3773 if (inode->i_size & (blocksize - 1))
3774 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3777 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3779 goto out_stop; /* error */
3782 * OK. This truncate is going to happen. We add the inode to the
3783 * orphan list, so that if this truncate spans multiple transactions,
3784 * and we crash, we will resume the truncate when the filesystem
3785 * recovers. It also marks the inode dirty, to catch the new size.
3787 * Implication: the file must always be in a sane, consistent
3788 * truncatable state while each transaction commits.
3790 if (ext4_orphan_add(handle, inode))
3794 * From here we block out all ext4_get_block() callers who want to
3795 * modify the block allocation tree.
3797 down_write(&ei->i_data_sem);
3799 ext4_discard_preallocations(inode);
3802 * The orphan list entry will now protect us from any crash which
3803 * occurs before the truncate completes, so it is now safe to propagate
3804 * the new, shorter inode size (held for now in i_size) into the
3805 * on-disk inode. We do this via i_disksize, which is the value which
3806 * ext4 *really* writes onto the disk inode.
3808 ei->i_disksize = inode->i_size;
3810 if (n == 1) { /* direct blocks */
3811 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3812 i_data + EXT4_NDIR_BLOCKS);
3816 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3817 /* Kill the top of shared branch (not detached) */
3819 if (partial == chain) {
3820 /* Shared branch grows from the inode */
3821 ext4_free_branches(handle, inode, NULL,
3822 &nr, &nr+1, (chain+n-1) - partial);
3825 * We mark the inode dirty prior to restart,
3826 * and prior to stop. No need for it here.
3829 /* Shared branch grows from an indirect block */
3830 BUFFER_TRACE(partial->bh, "get_write_access");
3831 ext4_free_branches(handle, inode, partial->bh,
3833 partial->p+1, (chain+n-1) - partial);
3836 /* Clear the ends of indirect blocks on the shared branch */
3837 while (partial > chain) {
3838 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3839 (__le32*)partial->bh->b_data+addr_per_block,
3840 (chain+n-1) - partial);
3841 BUFFER_TRACE(partial->bh, "call brelse");
3842 brelse (partial->bh);
3846 /* Kill the remaining (whole) subtrees */
3847 switch (offsets[0]) {
3849 nr = i_data[EXT4_IND_BLOCK];
3851 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3852 i_data[EXT4_IND_BLOCK] = 0;
3854 case EXT4_IND_BLOCK:
3855 nr = i_data[EXT4_DIND_BLOCK];
3857 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3858 i_data[EXT4_DIND_BLOCK] = 0;
3860 case EXT4_DIND_BLOCK:
3861 nr = i_data[EXT4_TIND_BLOCK];
3863 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3864 i_data[EXT4_TIND_BLOCK] = 0;
3866 case EXT4_TIND_BLOCK:
3870 up_write(&ei->i_data_sem);
3871 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3872 ext4_mark_inode_dirty(handle, inode);
3875 * In a multi-transaction truncate, we only make the final transaction
3879 ext4_handle_sync(handle);
3882 * If this was a simple ftruncate(), and the file will remain alive
3883 * then we need to clear up the orphan record which we created above.
3884 * However, if this was a real unlink then we were called by
3885 * ext4_delete_inode(), and we allow that function to clean up the
3886 * orphan info for us.
3889 ext4_orphan_del(handle, inode);
3891 ext4_journal_stop(handle);
3895 * ext4_get_inode_loc returns with an extra refcount against the inode's
3896 * underlying buffer_head on success. If 'in_mem' is true, we have all
3897 * data in memory that is needed to recreate the on-disk version of this
3900 static int __ext4_get_inode_loc(struct inode *inode,
3901 struct ext4_iloc *iloc, int in_mem)
3903 struct ext4_group_desc *gdp;
3904 struct buffer_head *bh;
3905 struct super_block *sb = inode->i_sb;
3907 int inodes_per_block, inode_offset;
3910 if (!ext4_valid_inum(sb, inode->i_ino))
3913 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3914 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3919 * Figure out the offset within the block group inode table
3921 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3922 inode_offset = ((inode->i_ino - 1) %
3923 EXT4_INODES_PER_GROUP(sb));
3924 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3925 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3927 bh = sb_getblk(sb, block);
3929 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3930 "inode block - inode=%lu, block=%llu",
3931 inode->i_ino, block);
3934 if (!buffer_uptodate(bh)) {
3938 * If the buffer has the write error flag, we have failed
3939 * to write out another inode in the same block. In this
3940 * case, we don't have to read the block because we may
3941 * read the old inode data successfully.
3943 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3944 set_buffer_uptodate(bh);
3946 if (buffer_uptodate(bh)) {
3947 /* someone brought it uptodate while we waited */
3953 * If we have all information of the inode in memory and this
3954 * is the only valid inode in the block, we need not read the
3958 struct buffer_head *bitmap_bh;
3961 start = inode_offset & ~(inodes_per_block - 1);
3963 /* Is the inode bitmap in cache? */
3964 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3969 * If the inode bitmap isn't in cache then the
3970 * optimisation may end up performing two reads instead
3971 * of one, so skip it.
3973 if (!buffer_uptodate(bitmap_bh)) {
3977 for (i = start; i < start + inodes_per_block; i++) {
3978 if (i == inode_offset)
3980 if (ext4_test_bit(i, bitmap_bh->b_data))
3984 if (i == start + inodes_per_block) {
3985 /* all other inodes are free, so skip I/O */
3986 memset(bh->b_data, 0, bh->b_size);
3987 set_buffer_uptodate(bh);
3995 * If we need to do any I/O, try to pre-readahead extra
3996 * blocks from the inode table.
3998 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3999 ext4_fsblk_t b, end, table;
4002 table = ext4_inode_table(sb, gdp);
4003 /* Make sure s_inode_readahead_blks is a power of 2 */
4004 while (EXT4_SB(sb)->s_inode_readahead_blks &
4005 (EXT4_SB(sb)->s_inode_readahead_blks-1))
4006 EXT4_SB(sb)->s_inode_readahead_blks =
4007 (EXT4_SB(sb)->s_inode_readahead_blks &
4008 (EXT4_SB(sb)->s_inode_readahead_blks-1));
4009 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4012 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4013 num = EXT4_INODES_PER_GROUP(sb);
4014 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4015 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4016 num -= le16_to_cpu(gdp->bg_itable_unused);
4017 table += num / inodes_per_block;
4021 sb_breadahead(sb, b++);
4025 * There are other valid inodes in the buffer, this inode
4026 * has in-inode xattrs, or we don't have this inode in memory.
4027 * Read the block from disk.
4030 bh->b_end_io = end_buffer_read_sync;
4031 submit_bh(READ_META, bh);
4033 if (!buffer_uptodate(bh)) {
4034 ext4_error(sb, __func__,
4035 "unable to read inode block - inode=%lu, "
4036 "block=%llu", inode->i_ino, block);
4046 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4048 /* We have all inode data except xattrs in memory here. */
4049 return __ext4_get_inode_loc(inode, iloc,
4050 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4053 void ext4_set_inode_flags(struct inode *inode)
4055 unsigned int flags = EXT4_I(inode)->i_flags;
4057 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4058 if (flags & EXT4_SYNC_FL)
4059 inode->i_flags |= S_SYNC;
4060 if (flags & EXT4_APPEND_FL)
4061 inode->i_flags |= S_APPEND;
4062 if (flags & EXT4_IMMUTABLE_FL)
4063 inode->i_flags |= S_IMMUTABLE;
4064 if (flags & EXT4_NOATIME_FL)
4065 inode->i_flags |= S_NOATIME;
4066 if (flags & EXT4_DIRSYNC_FL)
4067 inode->i_flags |= S_DIRSYNC;
4070 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4071 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4073 unsigned int flags = ei->vfs_inode.i_flags;
4075 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4076 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4078 ei->i_flags |= EXT4_SYNC_FL;
4079 if (flags & S_APPEND)
4080 ei->i_flags |= EXT4_APPEND_FL;
4081 if (flags & S_IMMUTABLE)
4082 ei->i_flags |= EXT4_IMMUTABLE_FL;
4083 if (flags & S_NOATIME)
4084 ei->i_flags |= EXT4_NOATIME_FL;
4085 if (flags & S_DIRSYNC)
4086 ei->i_flags |= EXT4_DIRSYNC_FL;
4088 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4089 struct ext4_inode_info *ei)
4092 struct inode *inode = &(ei->vfs_inode);
4093 struct super_block *sb = inode->i_sb;
4095 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4096 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4097 /* we are using combined 48 bit field */
4098 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4099 le32_to_cpu(raw_inode->i_blocks_lo);
4100 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4101 /* i_blocks represent file system block size */
4102 return i_blocks << (inode->i_blkbits - 9);
4107 return le32_to_cpu(raw_inode->i_blocks_lo);
4111 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4113 struct ext4_iloc iloc;
4114 struct ext4_inode *raw_inode;
4115 struct ext4_inode_info *ei;
4116 struct buffer_head *bh;
4117 struct inode *inode;
4121 inode = iget_locked(sb, ino);
4123 return ERR_PTR(-ENOMEM);
4124 if (!(inode->i_state & I_NEW))
4128 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4129 ei->i_acl = EXT4_ACL_NOT_CACHED;
4130 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4133 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4137 raw_inode = ext4_raw_inode(&iloc);
4138 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4139 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4140 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4141 if (!(test_opt(inode->i_sb, NO_UID32))) {
4142 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4143 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4145 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4148 ei->i_dir_start_lookup = 0;
4149 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4150 /* We now have enough fields to check if the inode was active or not.
4151 * This is needed because nfsd might try to access dead inodes
4152 * the test is that same one that e2fsck uses
4153 * NeilBrown 1999oct15
4155 if (inode->i_nlink == 0) {
4156 if (inode->i_mode == 0 ||
4157 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4158 /* this inode is deleted */
4163 /* The only unlinked inodes we let through here have
4164 * valid i_mode and are being read by the orphan
4165 * recovery code: that's fine, we're about to complete
4166 * the process of deleting those. */
4168 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4169 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4170 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4171 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4172 cpu_to_le32(EXT4_OS_HURD)) {
4174 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4176 inode->i_size = ext4_isize(raw_inode);
4177 ei->i_disksize = inode->i_size;
4178 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4179 ei->i_block_group = iloc.block_group;
4181 * NOTE! The in-memory inode i_data array is in little-endian order
4182 * even on big-endian machines: we do NOT byteswap the block numbers!
4184 for (block = 0; block < EXT4_N_BLOCKS; block++)
4185 ei->i_data[block] = raw_inode->i_block[block];
4186 INIT_LIST_HEAD(&ei->i_orphan);
4188 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4189 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4190 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4191 EXT4_INODE_SIZE(inode->i_sb)) {
4196 if (ei->i_extra_isize == 0) {
4197 /* The extra space is currently unused. Use it. */
4198 ei->i_extra_isize = sizeof(struct ext4_inode) -
4199 EXT4_GOOD_OLD_INODE_SIZE;
4201 __le32 *magic = (void *)raw_inode +
4202 EXT4_GOOD_OLD_INODE_SIZE +
4204 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4205 ei->i_state |= EXT4_STATE_XATTR;
4208 ei->i_extra_isize = 0;
4210 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4211 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4212 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4213 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4215 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4216 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4217 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4219 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4222 if (S_ISREG(inode->i_mode)) {
4223 inode->i_op = &ext4_file_inode_operations;
4224 inode->i_fop = &ext4_file_operations;
4225 ext4_set_aops(inode);
4226 } else if (S_ISDIR(inode->i_mode)) {
4227 inode->i_op = &ext4_dir_inode_operations;
4228 inode->i_fop = &ext4_dir_operations;
4229 } else if (S_ISLNK(inode->i_mode)) {
4230 if (ext4_inode_is_fast_symlink(inode)) {
4231 inode->i_op = &ext4_fast_symlink_inode_operations;
4232 nd_terminate_link(ei->i_data, inode->i_size,
4233 sizeof(ei->i_data) - 1);
4235 inode->i_op = &ext4_symlink_inode_operations;
4236 ext4_set_aops(inode);
4239 inode->i_op = &ext4_special_inode_operations;
4240 if (raw_inode->i_block[0])
4241 init_special_inode(inode, inode->i_mode,
4242 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4244 init_special_inode(inode, inode->i_mode,
4245 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4248 ext4_set_inode_flags(inode);
4249 unlock_new_inode(inode);
4254 return ERR_PTR(ret);
4257 static int ext4_inode_blocks_set(handle_t *handle,
4258 struct ext4_inode *raw_inode,
4259 struct ext4_inode_info *ei)
4261 struct inode *inode = &(ei->vfs_inode);
4262 u64 i_blocks = inode->i_blocks;
4263 struct super_block *sb = inode->i_sb;
4265 if (i_blocks <= ~0U) {
4267 * i_blocks can be represnted in a 32 bit variable
4268 * as multiple of 512 bytes
4270 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4271 raw_inode->i_blocks_high = 0;
4272 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4275 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4278 if (i_blocks <= 0xffffffffffffULL) {
4280 * i_blocks can be represented in a 48 bit variable
4281 * as multiple of 512 bytes
4283 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4284 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4285 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4287 ei->i_flags |= EXT4_HUGE_FILE_FL;
4288 /* i_block is stored in file system block size */
4289 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4290 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4291 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4297 * Post the struct inode info into an on-disk inode location in the
4298 * buffer-cache. This gobbles the caller's reference to the
4299 * buffer_head in the inode location struct.
4301 * The caller must have write access to iloc->bh.
4303 static int ext4_do_update_inode(handle_t *handle,
4304 struct inode *inode,
4305 struct ext4_iloc *iloc)
4307 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4308 struct ext4_inode_info *ei = EXT4_I(inode);
4309 struct buffer_head *bh = iloc->bh;
4310 int err = 0, rc, block;
4312 /* For fields not not tracking in the in-memory inode,
4313 * initialise them to zero for new inodes. */
4314 if (ei->i_state & EXT4_STATE_NEW)
4315 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4317 ext4_get_inode_flags(ei);
4318 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4319 if (!(test_opt(inode->i_sb, NO_UID32))) {
4320 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4321 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4323 * Fix up interoperability with old kernels. Otherwise, old inodes get
4324 * re-used with the upper 16 bits of the uid/gid intact
4327 raw_inode->i_uid_high =
4328 cpu_to_le16(high_16_bits(inode->i_uid));
4329 raw_inode->i_gid_high =
4330 cpu_to_le16(high_16_bits(inode->i_gid));
4332 raw_inode->i_uid_high = 0;
4333 raw_inode->i_gid_high = 0;
4336 raw_inode->i_uid_low =
4337 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4338 raw_inode->i_gid_low =
4339 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4340 raw_inode->i_uid_high = 0;
4341 raw_inode->i_gid_high = 0;
4343 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4345 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4346 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4347 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4348 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4350 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4352 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4353 /* clear the migrate flag in the raw_inode */
4354 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4355 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4356 cpu_to_le32(EXT4_OS_HURD))
4357 raw_inode->i_file_acl_high =
4358 cpu_to_le16(ei->i_file_acl >> 32);
4359 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4360 ext4_isize_set(raw_inode, ei->i_disksize);
4361 if (ei->i_disksize > 0x7fffffffULL) {
4362 struct super_block *sb = inode->i_sb;
4363 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4364 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4365 EXT4_SB(sb)->s_es->s_rev_level ==
4366 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4367 /* If this is the first large file
4368 * created, add a flag to the superblock.
4370 err = ext4_journal_get_write_access(handle,
4371 EXT4_SB(sb)->s_sbh);
4374 ext4_update_dynamic_rev(sb);
4375 EXT4_SET_RO_COMPAT_FEATURE(sb,
4376 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4378 ext4_handle_sync(handle);
4379 err = ext4_handle_dirty_metadata(handle, inode,
4380 EXT4_SB(sb)->s_sbh);
4383 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4384 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4385 if (old_valid_dev(inode->i_rdev)) {
4386 raw_inode->i_block[0] =
4387 cpu_to_le32(old_encode_dev(inode->i_rdev));
4388 raw_inode->i_block[1] = 0;
4390 raw_inode->i_block[0] = 0;
4391 raw_inode->i_block[1] =
4392 cpu_to_le32(new_encode_dev(inode->i_rdev));
4393 raw_inode->i_block[2] = 0;
4395 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4396 raw_inode->i_block[block] = ei->i_data[block];
4398 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4399 if (ei->i_extra_isize) {
4400 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4401 raw_inode->i_version_hi =
4402 cpu_to_le32(inode->i_version >> 32);
4403 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4406 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4407 rc = ext4_handle_dirty_metadata(handle, inode, bh);
4410 ei->i_state &= ~EXT4_STATE_NEW;
4414 ext4_std_error(inode->i_sb, err);
4419 * ext4_write_inode()
4421 * We are called from a few places:
4423 * - Within generic_file_write() for O_SYNC files.
4424 * Here, there will be no transaction running. We wait for any running
4425 * trasnaction to commit.
4427 * - Within sys_sync(), kupdate and such.
4428 * We wait on commit, if tol to.
4430 * - Within prune_icache() (PF_MEMALLOC == true)
4431 * Here we simply return. We can't afford to block kswapd on the
4434 * In all cases it is actually safe for us to return without doing anything,
4435 * because the inode has been copied into a raw inode buffer in
4436 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4439 * Note that we are absolutely dependent upon all inode dirtiers doing the
4440 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4441 * which we are interested.
4443 * It would be a bug for them to not do this. The code:
4445 * mark_inode_dirty(inode)
4447 * inode->i_size = expr;
4449 * is in error because a kswapd-driven write_inode() could occur while
4450 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4451 * will no longer be on the superblock's dirty inode list.
4453 int ext4_write_inode(struct inode *inode, int wait)
4455 if (current->flags & PF_MEMALLOC)
4458 if (ext4_journal_current_handle()) {
4459 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4467 return ext4_force_commit(inode->i_sb);
4470 int __ext4_write_dirty_metadata(struct inode *inode, struct buffer_head *bh)
4474 mark_buffer_dirty(bh);
4475 if (inode && inode_needs_sync(inode)) {
4476 sync_dirty_buffer(bh);
4477 if (buffer_req(bh) && !buffer_uptodate(bh)) {
4478 ext4_error(inode->i_sb, __func__,
4479 "IO error syncing inode, "
4480 "inode=%lu, block=%llu",
4482 (unsigned long long)bh->b_blocknr);
4492 * Called from notify_change.
4494 * We want to trap VFS attempts to truncate the file as soon as
4495 * possible. In particular, we want to make sure that when the VFS
4496 * shrinks i_size, we put the inode on the orphan list and modify
4497 * i_disksize immediately, so that during the subsequent flushing of
4498 * dirty pages and freeing of disk blocks, we can guarantee that any
4499 * commit will leave the blocks being flushed in an unused state on
4500 * disk. (On recovery, the inode will get truncated and the blocks will
4501 * be freed, so we have a strong guarantee that no future commit will
4502 * leave these blocks visible to the user.)
4504 * Another thing we have to assure is that if we are in ordered mode
4505 * and inode is still attached to the committing transaction, we must
4506 * we start writeout of all the dirty pages which are being truncated.
4507 * This way we are sure that all the data written in the previous
4508 * transaction are already on disk (truncate waits for pages under
4511 * Called with inode->i_mutex down.
4513 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4515 struct inode *inode = dentry->d_inode;
4517 const unsigned int ia_valid = attr->ia_valid;
4519 error = inode_change_ok(inode, attr);
4523 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4524 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4527 /* (user+group)*(old+new) structure, inode write (sb,
4528 * inode block, ? - but truncate inode update has it) */
4529 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4530 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4531 if (IS_ERR(handle)) {
4532 error = PTR_ERR(handle);
4535 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4537 ext4_journal_stop(handle);
4540 /* Update corresponding info in inode so that everything is in
4541 * one transaction */
4542 if (attr->ia_valid & ATTR_UID)
4543 inode->i_uid = attr->ia_uid;
4544 if (attr->ia_valid & ATTR_GID)
4545 inode->i_gid = attr->ia_gid;
4546 error = ext4_mark_inode_dirty(handle, inode);
4547 ext4_journal_stop(handle);
4550 if (attr->ia_valid & ATTR_SIZE) {
4551 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4552 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4554 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4561 if (S_ISREG(inode->i_mode) &&
4562 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4565 handle = ext4_journal_start(inode, 3);
4566 if (IS_ERR(handle)) {
4567 error = PTR_ERR(handle);
4571 error = ext4_orphan_add(handle, inode);
4572 EXT4_I(inode)->i_disksize = attr->ia_size;
4573 rc = ext4_mark_inode_dirty(handle, inode);
4576 ext4_journal_stop(handle);
4578 if (ext4_should_order_data(inode)) {
4579 error = ext4_begin_ordered_truncate(inode,
4582 /* Do as much error cleanup as possible */
4583 handle = ext4_journal_start(inode, 3);
4584 if (IS_ERR(handle)) {
4585 ext4_orphan_del(NULL, inode);
4588 ext4_orphan_del(handle, inode);
4589 ext4_journal_stop(handle);
4595 rc = inode_setattr(inode, attr);
4597 /* If inode_setattr's call to ext4_truncate failed to get a
4598 * transaction handle at all, we need to clean up the in-core
4599 * orphan list manually. */
4601 ext4_orphan_del(NULL, inode);
4603 if (!rc && (ia_valid & ATTR_MODE))
4604 rc = ext4_acl_chmod(inode);
4607 ext4_std_error(inode->i_sb, error);
4613 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4616 struct inode *inode;
4617 unsigned long delalloc_blocks;
4619 inode = dentry->d_inode;
4620 generic_fillattr(inode, stat);
4623 * We can't update i_blocks if the block allocation is delayed
4624 * otherwise in the case of system crash before the real block
4625 * allocation is done, we will have i_blocks inconsistent with
4626 * on-disk file blocks.
4627 * We always keep i_blocks updated together with real
4628 * allocation. But to not confuse with user, stat
4629 * will return the blocks that include the delayed allocation
4630 * blocks for this file.
4632 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4633 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4634 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4636 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4640 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4645 /* if nrblocks are contiguous */
4648 * With N contiguous data blocks, it need at most
4649 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4650 * 2 dindirect blocks
4653 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4654 return indirects + 3;
4657 * if nrblocks are not contiguous, worse case, each block touch
4658 * a indirect block, and each indirect block touch a double indirect
4659 * block, plus a triple indirect block
4661 indirects = nrblocks * 2 + 1;
4665 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4667 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4668 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4669 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4673 * Account for index blocks, block groups bitmaps and block group
4674 * descriptor blocks if modify datablocks and index blocks
4675 * worse case, the indexs blocks spread over different block groups
4677 * If datablocks are discontiguous, they are possible to spread over
4678 * different block groups too. If they are contiugous, with flexbg,
4679 * they could still across block group boundary.
4681 * Also account for superblock, inode, quota and xattr blocks
4683 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4685 int groups, gdpblocks;
4690 * How many index blocks need to touch to modify nrblocks?
4691 * The "Chunk" flag indicating whether the nrblocks is
4692 * physically contiguous on disk
4694 * For Direct IO and fallocate, they calls get_block to allocate
4695 * one single extent at a time, so they could set the "Chunk" flag
4697 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4702 * Now let's see how many group bitmaps and group descriptors need
4712 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4713 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4714 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4715 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4717 /* bitmaps and block group descriptor blocks */
4718 ret += groups + gdpblocks;
4720 /* Blocks for super block, inode, quota and xattr blocks */
4721 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4727 * Calulate the total number of credits to reserve to fit
4728 * the modification of a single pages into a single transaction,
4729 * which may include multiple chunks of block allocations.
4731 * This could be called via ext4_write_begin()
4733 * We need to consider the worse case, when
4734 * one new block per extent.
4736 int ext4_writepage_trans_blocks(struct inode *inode)
4738 int bpp = ext4_journal_blocks_per_page(inode);
4741 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4743 /* Account for data blocks for journalled mode */
4744 if (ext4_should_journal_data(inode))
4750 * Calculate the journal credits for a chunk of data modification.
4752 * This is called from DIO, fallocate or whoever calling
4753 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4755 * journal buffers for data blocks are not included here, as DIO
4756 * and fallocate do no need to journal data buffers.
4758 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4760 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4764 * The caller must have previously called ext4_reserve_inode_write().
4765 * Give this, we know that the caller already has write access to iloc->bh.
4767 int ext4_mark_iloc_dirty(handle_t *handle,
4768 struct inode *inode, struct ext4_iloc *iloc)
4772 if (test_opt(inode->i_sb, I_VERSION))
4773 inode_inc_iversion(inode);
4775 /* the do_update_inode consumes one bh->b_count */
4778 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4779 err = ext4_do_update_inode(handle, inode, iloc);
4785 * On success, We end up with an outstanding reference count against
4786 * iloc->bh. This _must_ be cleaned up later.
4790 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4791 struct ext4_iloc *iloc)
4795 err = ext4_get_inode_loc(inode, iloc);
4797 BUFFER_TRACE(iloc->bh, "get_write_access");
4798 err = ext4_journal_get_write_access(handle, iloc->bh);
4804 ext4_std_error(inode->i_sb, err);
4809 * Expand an inode by new_extra_isize bytes.
4810 * Returns 0 on success or negative error number on failure.
4812 static int ext4_expand_extra_isize(struct inode *inode,
4813 unsigned int new_extra_isize,
4814 struct ext4_iloc iloc,
4817 struct ext4_inode *raw_inode;
4818 struct ext4_xattr_ibody_header *header;
4819 struct ext4_xattr_entry *entry;
4821 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4824 raw_inode = ext4_raw_inode(&iloc);
4826 header = IHDR(inode, raw_inode);
4827 entry = IFIRST(header);
4829 /* No extended attributes present */
4830 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4831 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4832 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4834 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4838 /* try to expand with EAs present */
4839 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4844 * What we do here is to mark the in-core inode as clean with respect to inode
4845 * dirtiness (it may still be data-dirty).
4846 * This means that the in-core inode may be reaped by prune_icache
4847 * without having to perform any I/O. This is a very good thing,
4848 * because *any* task may call prune_icache - even ones which
4849 * have a transaction open against a different journal.
4851 * Is this cheating? Not really. Sure, we haven't written the
4852 * inode out, but prune_icache isn't a user-visible syncing function.
4853 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4854 * we start and wait on commits.
4856 * Is this efficient/effective? Well, we're being nice to the system
4857 * by cleaning up our inodes proactively so they can be reaped
4858 * without I/O. But we are potentially leaving up to five seconds'
4859 * worth of inodes floating about which prune_icache wants us to
4860 * write out. One way to fix that would be to get prune_icache()
4861 * to do a write_super() to free up some memory. It has the desired
4864 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4866 struct ext4_iloc iloc;
4867 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4868 static unsigned int mnt_count;
4872 err = ext4_reserve_inode_write(handle, inode, &iloc);
4873 if (ext4_handle_valid(handle) &&
4874 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4875 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4877 * We need extra buffer credits since we may write into EA block
4878 * with this same handle. If journal_extend fails, then it will
4879 * only result in a minor loss of functionality for that inode.
4880 * If this is felt to be critical, then e2fsck should be run to
4881 * force a large enough s_min_extra_isize.
4883 if ((jbd2_journal_extend(handle,
4884 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4885 ret = ext4_expand_extra_isize(inode,
4886 sbi->s_want_extra_isize,
4889 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4891 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4892 ext4_warning(inode->i_sb, __func__,
4893 "Unable to expand inode %lu. Delete"
4894 " some EAs or run e2fsck.",
4897 le16_to_cpu(sbi->s_es->s_mnt_count);
4903 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4908 * ext4_dirty_inode() is called from __mark_inode_dirty()
4910 * We're really interested in the case where a file is being extended.
4911 * i_size has been changed by generic_commit_write() and we thus need
4912 * to include the updated inode in the current transaction.
4914 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4915 * are allocated to the file.
4917 * If the inode is marked synchronous, we don't honour that here - doing
4918 * so would cause a commit on atime updates, which we don't bother doing.
4919 * We handle synchronous inodes at the highest possible level.
4921 void ext4_dirty_inode(struct inode *inode)
4923 handle_t *current_handle = ext4_journal_current_handle();
4926 if (!ext4_handle_valid(current_handle)) {
4927 ext4_mark_inode_dirty(current_handle, inode);
4931 handle = ext4_journal_start(inode, 2);
4934 if (current_handle &&
4935 current_handle->h_transaction != handle->h_transaction) {
4936 /* This task has a transaction open against a different fs */
4937 printk(KERN_EMERG "%s: transactions do not match!\n",
4940 jbd_debug(5, "marking dirty. outer handle=%p\n",
4942 ext4_mark_inode_dirty(handle, inode);
4944 ext4_journal_stop(handle);
4951 * Bind an inode's backing buffer_head into this transaction, to prevent
4952 * it from being flushed to disk early. Unlike
4953 * ext4_reserve_inode_write, this leaves behind no bh reference and
4954 * returns no iloc structure, so the caller needs to repeat the iloc
4955 * lookup to mark the inode dirty later.
4957 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4959 struct ext4_iloc iloc;
4963 err = ext4_get_inode_loc(inode, &iloc);
4965 BUFFER_TRACE(iloc.bh, "get_write_access");
4966 err = jbd2_journal_get_write_access(handle, iloc.bh);
4968 err = ext4_handle_dirty_metadata(handle,
4974 ext4_std_error(inode->i_sb, err);
4979 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4986 * We have to be very careful here: changing a data block's
4987 * journaling status dynamically is dangerous. If we write a
4988 * data block to the journal, change the status and then delete
4989 * that block, we risk forgetting to revoke the old log record
4990 * from the journal and so a subsequent replay can corrupt data.
4991 * So, first we make sure that the journal is empty and that
4992 * nobody is changing anything.
4995 journal = EXT4_JOURNAL(inode);
4998 if (is_journal_aborted(journal))
5001 jbd2_journal_lock_updates(journal);
5002 jbd2_journal_flush(journal);
5005 * OK, there are no updates running now, and all cached data is
5006 * synced to disk. We are now in a completely consistent state
5007 * which doesn't have anything in the journal, and we know that
5008 * no filesystem updates are running, so it is safe to modify
5009 * the inode's in-core data-journaling state flag now.
5013 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
5015 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
5016 ext4_set_aops(inode);
5018 jbd2_journal_unlock_updates(journal);
5020 /* Finally we can mark the inode as dirty. */
5022 handle = ext4_journal_start(inode, 1);
5024 return PTR_ERR(handle);
5026 err = ext4_mark_inode_dirty(handle, inode);
5027 ext4_handle_sync(handle);
5028 ext4_journal_stop(handle);
5029 ext4_std_error(inode->i_sb, err);
5034 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5036 return !buffer_mapped(bh);
5039 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
5045 struct file *file = vma->vm_file;
5046 struct inode *inode = file->f_path.dentry->d_inode;
5047 struct address_space *mapping = inode->i_mapping;
5050 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5051 * get i_mutex because we are already holding mmap_sem.
5053 down_read(&inode->i_alloc_sem);
5054 size = i_size_read(inode);
5055 if (page->mapping != mapping || size <= page_offset(page)
5056 || !PageUptodate(page)) {
5057 /* page got truncated from under us? */
5061 if (PageMappedToDisk(page))
5064 if (page->index == size >> PAGE_CACHE_SHIFT)
5065 len = size & ~PAGE_CACHE_MASK;
5067 len = PAGE_CACHE_SIZE;
5069 if (page_has_buffers(page)) {
5070 /* return if we have all the buffers mapped */
5071 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5076 * OK, we need to fill the hole... Do write_begin write_end
5077 * to do block allocation/reservation.We are not holding
5078 * inode.i__mutex here. That allow * parallel write_begin,
5079 * write_end call. lock_page prevent this from happening
5080 * on the same page though
5082 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5083 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5086 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5087 len, len, page, fsdata);
5092 up_read(&inode->i_alloc_sem);