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/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 #define MPAGE_DA_EXTENT_TAIL 0x01
46 static inline int ext4_begin_ordered_truncate(struct inode *inode,
49 return jbd2_journal_begin_ordered_truncate(
50 EXT4_SB(inode->i_sb)->s_journal,
51 &EXT4_I(inode)->jinode,
55 static void ext4_invalidatepage(struct page *page, unsigned long offset);
58 * Test whether an inode is a fast symlink.
60 static int ext4_inode_is_fast_symlink(struct inode *inode)
62 int ea_blocks = EXT4_I(inode)->i_file_acl ?
63 (inode->i_sb->s_blocksize >> 9) : 0;
65 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
69 * The ext4 forget function must perform a revoke if we are freeing data
70 * which has been journaled. Metadata (eg. indirect blocks) must be
71 * revoked in all cases.
73 * "bh" may be NULL: a metadata block may have been freed from memory
74 * but there may still be a record of it in the journal, and that record
75 * still needs to be revoked.
77 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
78 struct buffer_head *bh, ext4_fsblk_t blocknr)
84 BUFFER_TRACE(bh, "enter");
86 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
88 bh, is_metadata, inode->i_mode,
89 test_opt(inode->i_sb, DATA_FLAGS));
91 /* Never use the revoke function if we are doing full data
92 * journaling: there is no need to, and a V1 superblock won't
93 * support it. Otherwise, only skip the revoke on un-journaled
96 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
97 (!is_metadata && !ext4_should_journal_data(inode))) {
99 BUFFER_TRACE(bh, "call jbd2_journal_forget");
100 return ext4_journal_forget(handle, bh);
106 * data!=journal && (is_metadata || should_journal_data(inode))
108 BUFFER_TRACE(bh, "call ext4_journal_revoke");
109 err = ext4_journal_revoke(handle, blocknr, bh);
111 ext4_abort(inode->i_sb, __func__,
112 "error %d when attempting revoke", err);
113 BUFFER_TRACE(bh, "exit");
118 * Work out how many blocks we need to proceed with the next chunk of a
119 * truncate transaction.
121 static unsigned long blocks_for_truncate(struct inode *inode)
125 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
127 /* Give ourselves just enough room to cope with inodes in which
128 * i_blocks is corrupt: we've seen disk corruptions in the past
129 * which resulted in random data in an inode which looked enough
130 * like a regular file for ext4 to try to delete it. Things
131 * will go a bit crazy if that happens, but at least we should
132 * try not to panic the whole kernel. */
136 /* But we need to bound the transaction so we don't overflow the
138 if (needed > EXT4_MAX_TRANS_DATA)
139 needed = EXT4_MAX_TRANS_DATA;
141 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
145 * Truncate transactions can be complex and absolutely huge. So we need to
146 * be able to restart the transaction at a conventient checkpoint to make
147 * sure we don't overflow the journal.
149 * start_transaction gets us a new handle for a truncate transaction,
150 * and extend_transaction tries to extend the existing one a bit. If
151 * extend fails, we need to propagate the failure up and restart the
152 * transaction in the top-level truncate loop. --sct
154 static handle_t *start_transaction(struct inode *inode)
158 result = ext4_journal_start(inode, blocks_for_truncate(inode));
162 ext4_std_error(inode->i_sb, PTR_ERR(result));
167 * Try to extend this transaction for the purposes of truncation.
169 * Returns 0 if we managed to create more room. If we can't create more
170 * room, and the transaction must be restarted we return 1.
172 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
174 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
176 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
182 * Restart the transaction associated with *handle. This does a commit,
183 * so before we call here everything must be consistently dirtied against
186 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
188 jbd_debug(2, "restarting handle %p\n", handle);
189 return ext4_journal_restart(handle, blocks_for_truncate(inode));
193 * Called at the last iput() if i_nlink is zero.
195 void ext4_delete_inode(struct inode *inode)
200 if (ext4_should_order_data(inode))
201 ext4_begin_ordered_truncate(inode, 0);
202 truncate_inode_pages(&inode->i_data, 0);
204 if (is_bad_inode(inode))
207 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
208 if (IS_ERR(handle)) {
209 ext4_std_error(inode->i_sb, PTR_ERR(handle));
211 * If we're going to skip the normal cleanup, we still need to
212 * make sure that the in-core orphan linked list is properly
215 ext4_orphan_del(NULL, inode);
222 err = ext4_mark_inode_dirty(handle, inode);
224 ext4_warning(inode->i_sb, __func__,
225 "couldn't mark inode dirty (err %d)", err);
229 ext4_truncate(inode);
232 * ext4_ext_truncate() doesn't reserve any slop when it
233 * restarts journal transactions; therefore there may not be
234 * enough credits left in the handle to remove the inode from
235 * the orphan list and set the dtime field.
237 if (handle->h_buffer_credits < 3) {
238 err = ext4_journal_extend(handle, 3);
240 err = ext4_journal_restart(handle, 3);
242 ext4_warning(inode->i_sb, __func__,
243 "couldn't extend journal (err %d)", err);
245 ext4_journal_stop(handle);
251 * Kill off the orphan record which ext4_truncate created.
252 * AKPM: I think this can be inside the above `if'.
253 * Note that ext4_orphan_del() has to be able to cope with the
254 * deletion of a non-existent orphan - this is because we don't
255 * know if ext4_truncate() actually created an orphan record.
256 * (Well, we could do this if we need to, but heck - it works)
258 ext4_orphan_del(handle, inode);
259 EXT4_I(inode)->i_dtime = get_seconds();
262 * One subtle ordering requirement: if anything has gone wrong
263 * (transaction abort, IO errors, whatever), then we can still
264 * do these next steps (the fs will already have been marked as
265 * having errors), but we can't free the inode if the mark_dirty
268 if (ext4_mark_inode_dirty(handle, inode))
269 /* If that failed, just do the required in-core inode clear. */
272 ext4_free_inode(handle, inode);
273 ext4_journal_stop(handle);
276 clear_inode(inode); /* We must guarantee clearing of inode... */
282 struct buffer_head *bh;
285 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
287 p->key = *(p->p = v);
292 * ext4_block_to_path - parse the block number into array of offsets
293 * @inode: inode in question (we are only interested in its superblock)
294 * @i_block: block number to be parsed
295 * @offsets: array to store the offsets in
296 * @boundary: set this non-zero if the referred-to block is likely to be
297 * followed (on disk) by an indirect block.
299 * To store the locations of file's data ext4 uses a data structure common
300 * for UNIX filesystems - tree of pointers anchored in the inode, with
301 * data blocks at leaves and indirect blocks in intermediate nodes.
302 * This function translates the block number into path in that tree -
303 * return value is the path length and @offsets[n] is the offset of
304 * pointer to (n+1)th node in the nth one. If @block is out of range
305 * (negative or too large) warning is printed and zero returned.
307 * Note: function doesn't find node addresses, so no IO is needed. All
308 * we need to know is the capacity of indirect blocks (taken from the
313 * Portability note: the last comparison (check that we fit into triple
314 * indirect block) is spelled differently, because otherwise on an
315 * architecture with 32-bit longs and 8Kb pages we might get into trouble
316 * if our filesystem had 8Kb blocks. We might use long long, but that would
317 * kill us on x86. Oh, well, at least the sign propagation does not matter -
318 * i_block would have to be negative in the very beginning, so we would not
322 static int ext4_block_to_path(struct inode *inode,
324 ext4_lblk_t offsets[4], int *boundary)
326 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
327 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
328 const long direct_blocks = EXT4_NDIR_BLOCKS,
329 indirect_blocks = ptrs,
330 double_blocks = (1 << (ptrs_bits * 2));
335 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
336 } else if (i_block < direct_blocks) {
337 offsets[n++] = i_block;
338 final = direct_blocks;
339 } else if ((i_block -= direct_blocks) < indirect_blocks) {
340 offsets[n++] = EXT4_IND_BLOCK;
341 offsets[n++] = i_block;
343 } else if ((i_block -= indirect_blocks) < double_blocks) {
344 offsets[n++] = EXT4_DIND_BLOCK;
345 offsets[n++] = i_block >> ptrs_bits;
346 offsets[n++] = i_block & (ptrs - 1);
348 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
349 offsets[n++] = EXT4_TIND_BLOCK;
350 offsets[n++] = i_block >> (ptrs_bits * 2);
351 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
352 offsets[n++] = i_block & (ptrs - 1);
355 ext4_warning(inode->i_sb, "ext4_block_to_path",
356 "block %lu > max in inode %lu",
357 i_block + direct_blocks +
358 indirect_blocks + double_blocks, inode->i_ino);
361 *boundary = final - 1 - (i_block & (ptrs - 1));
366 * ext4_get_branch - read the chain of indirect blocks leading to data
367 * @inode: inode in question
368 * @depth: depth of the chain (1 - direct pointer, etc.)
369 * @offsets: offsets of pointers in inode/indirect blocks
370 * @chain: place to store the result
371 * @err: here we store the error value
373 * Function fills the array of triples <key, p, bh> and returns %NULL
374 * if everything went OK or the pointer to the last filled triple
375 * (incomplete one) otherwise. Upon the return chain[i].key contains
376 * the number of (i+1)-th block in the chain (as it is stored in memory,
377 * i.e. little-endian 32-bit), chain[i].p contains the address of that
378 * number (it points into struct inode for i==0 and into the bh->b_data
379 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
380 * block for i>0 and NULL for i==0. In other words, it holds the block
381 * numbers of the chain, addresses they were taken from (and where we can
382 * verify that chain did not change) and buffer_heads hosting these
385 * Function stops when it stumbles upon zero pointer (absent block)
386 * (pointer to last triple returned, *@err == 0)
387 * or when it gets an IO error reading an indirect block
388 * (ditto, *@err == -EIO)
389 * or when it reads all @depth-1 indirect blocks successfully and finds
390 * the whole chain, all way to the data (returns %NULL, *err == 0).
392 * Need to be called with
393 * down_read(&EXT4_I(inode)->i_data_sem)
395 static Indirect *ext4_get_branch(struct inode *inode, int depth,
396 ext4_lblk_t *offsets,
397 Indirect chain[4], int *err)
399 struct super_block *sb = inode->i_sb;
401 struct buffer_head *bh;
404 /* i_data is not going away, no lock needed */
405 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
409 bh = sb_bread(sb, le32_to_cpu(p->key));
412 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
426 * ext4_find_near - find a place for allocation with sufficient locality
428 * @ind: descriptor of indirect block.
430 * This function returns the preferred place for block allocation.
431 * It is used when heuristic for sequential allocation fails.
433 * + if there is a block to the left of our position - allocate near it.
434 * + if pointer will live in indirect block - allocate near that block.
435 * + if pointer will live in inode - allocate in the same
438 * In the latter case we colour the starting block by the callers PID to
439 * prevent it from clashing with concurrent allocations for a different inode
440 * in the same block group. The PID is used here so that functionally related
441 * files will be close-by on-disk.
443 * Caller must make sure that @ind is valid and will stay that way.
445 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
447 struct ext4_inode_info *ei = EXT4_I(inode);
448 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
450 ext4_fsblk_t bg_start;
451 ext4_fsblk_t last_block;
452 ext4_grpblk_t colour;
454 /* Try to find previous block */
455 for (p = ind->p - 1; p >= start; p--) {
457 return le32_to_cpu(*p);
460 /* No such thing, so let's try location of indirect block */
462 return ind->bh->b_blocknr;
465 * It is going to be referred to from the inode itself? OK, just put it
466 * into the same cylinder group then.
468 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
469 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
471 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
472 colour = (current->pid % 16) *
473 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
475 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
476 return bg_start + colour;
480 * ext4_find_goal - find a preferred place for allocation.
482 * @block: block we want
483 * @partial: pointer to the last triple within a chain
485 * Normally this function find the preferred place for block allocation,
488 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
492 * XXX need to get goal block from mballoc's data structures
495 return ext4_find_near(inode, partial);
499 * ext4_blks_to_allocate: Look up the block map and count the number
500 * of direct blocks need to be allocated for the given branch.
502 * @branch: chain of indirect blocks
503 * @k: number of blocks need for indirect blocks
504 * @blks: number of data blocks to be mapped.
505 * @blocks_to_boundary: the offset in the indirect block
507 * return the total number of blocks to be allocate, including the
508 * direct and indirect blocks.
510 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
511 int blocks_to_boundary)
513 unsigned long count = 0;
516 * Simple case, [t,d]Indirect block(s) has not allocated yet
517 * then it's clear blocks on that path have not allocated
520 /* right now we don't handle cross boundary allocation */
521 if (blks < blocks_to_boundary + 1)
524 count += blocks_to_boundary + 1;
529 while (count < blks && count <= blocks_to_boundary &&
530 le32_to_cpu(*(branch[0].p + count)) == 0) {
537 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
538 * @indirect_blks: the number of blocks need to allocate for indirect
541 * @new_blocks: on return it will store the new block numbers for
542 * the indirect blocks(if needed) and the first direct block,
543 * @blks: on return it will store the total number of allocated
546 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
547 ext4_lblk_t iblock, ext4_fsblk_t goal,
548 int indirect_blks, int blks,
549 ext4_fsblk_t new_blocks[4], int *err)
552 unsigned long count = 0, blk_allocated = 0;
554 ext4_fsblk_t current_block = 0;
558 * Here we try to allocate the requested multiple blocks at once,
559 * on a best-effort basis.
560 * To build a branch, we should allocate blocks for
561 * the indirect blocks(if not allocated yet), and at least
562 * the first direct block of this branch. That's the
563 * minimum number of blocks need to allocate(required)
565 /* first we try to allocate the indirect blocks */
566 target = indirect_blks;
569 /* allocating blocks for indirect blocks and direct blocks */
570 current_block = ext4_new_meta_blocks(handle, inode,
576 /* allocate blocks for indirect blocks */
577 while (index < indirect_blks && count) {
578 new_blocks[index++] = current_block++;
583 * save the new block number
584 * for the first direct block
586 new_blocks[index] = current_block;
587 printk(KERN_INFO "%s returned more blocks than "
588 "requested\n", __func__);
594 target = blks - count ;
595 blk_allocated = count;
598 /* Now allocate data blocks */
600 /* allocating blocks for data blocks */
601 current_block = ext4_new_blocks(handle, inode, iblock,
603 if (*err && (target == blks)) {
605 * if the allocation failed and we didn't allocate
611 if (target == blks) {
613 * save the new block number
614 * for the first direct block
616 new_blocks[index] = current_block;
618 blk_allocated += count;
621 /* total number of blocks allocated for direct blocks */
626 for (i = 0; i < index; i++)
627 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
632 * ext4_alloc_branch - allocate and set up a chain of blocks.
634 * @indirect_blks: number of allocated indirect blocks
635 * @blks: number of allocated direct blocks
636 * @offsets: offsets (in the blocks) to store the pointers to next.
637 * @branch: place to store the chain in.
639 * This function allocates blocks, zeroes out all but the last one,
640 * links them into chain and (if we are synchronous) writes them to disk.
641 * In other words, it prepares a branch that can be spliced onto the
642 * inode. It stores the information about that chain in the branch[], in
643 * the same format as ext4_get_branch() would do. We are calling it after
644 * we had read the existing part of chain and partial points to the last
645 * triple of that (one with zero ->key). Upon the exit we have the same
646 * picture as after the successful ext4_get_block(), except that in one
647 * place chain is disconnected - *branch->p is still zero (we did not
648 * set the last link), but branch->key contains the number that should
649 * be placed into *branch->p to fill that gap.
651 * If allocation fails we free all blocks we've allocated (and forget
652 * their buffer_heads) and return the error value the from failed
653 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
654 * as described above and return 0.
656 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
657 ext4_lblk_t iblock, int indirect_blks,
658 int *blks, ext4_fsblk_t goal,
659 ext4_lblk_t *offsets, Indirect *branch)
661 int blocksize = inode->i_sb->s_blocksize;
664 struct buffer_head *bh;
666 ext4_fsblk_t new_blocks[4];
667 ext4_fsblk_t current_block;
669 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
670 *blks, new_blocks, &err);
674 branch[0].key = cpu_to_le32(new_blocks[0]);
676 * metadata blocks and data blocks are allocated.
678 for (n = 1; n <= indirect_blks; n++) {
680 * Get buffer_head for parent block, zero it out
681 * and set the pointer to new one, then send
684 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
687 BUFFER_TRACE(bh, "call get_create_access");
688 err = ext4_journal_get_create_access(handle, bh);
695 memset(bh->b_data, 0, blocksize);
696 branch[n].p = (__le32 *) bh->b_data + offsets[n];
697 branch[n].key = cpu_to_le32(new_blocks[n]);
698 *branch[n].p = branch[n].key;
699 if (n == indirect_blks) {
700 current_block = new_blocks[n];
702 * End of chain, update the last new metablock of
703 * the chain to point to the new allocated
704 * data blocks numbers
706 for (i=1; i < num; i++)
707 *(branch[n].p + i) = cpu_to_le32(++current_block);
709 BUFFER_TRACE(bh, "marking uptodate");
710 set_buffer_uptodate(bh);
713 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
714 err = ext4_journal_dirty_metadata(handle, bh);
721 /* Allocation failed, free what we already allocated */
722 for (i = 1; i <= n ; i++) {
723 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
724 ext4_journal_forget(handle, branch[i].bh);
726 for (i = 0; i < indirect_blks; i++)
727 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
729 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
735 * ext4_splice_branch - splice the allocated branch onto inode.
737 * @block: (logical) number of block we are adding
738 * @chain: chain of indirect blocks (with a missing link - see
740 * @where: location of missing link
741 * @num: number of indirect blocks we are adding
742 * @blks: number of direct blocks we are adding
744 * This function fills the missing link and does all housekeeping needed in
745 * inode (->i_blocks, etc.). In case of success we end up with the full
746 * chain to new block and return 0.
748 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
749 ext4_lblk_t block, Indirect *where, int num, int blks)
753 ext4_fsblk_t current_block;
756 * If we're splicing into a [td]indirect block (as opposed to the
757 * inode) then we need to get write access to the [td]indirect block
761 BUFFER_TRACE(where->bh, "get_write_access");
762 err = ext4_journal_get_write_access(handle, where->bh);
768 *where->p = where->key;
771 * Update the host buffer_head or inode to point to more just allocated
772 * direct blocks blocks
774 if (num == 0 && blks > 1) {
775 current_block = le32_to_cpu(where->key) + 1;
776 for (i = 1; i < blks; i++)
777 *(where->p + i) = cpu_to_le32(current_block++);
780 /* We are done with atomic stuff, now do the rest of housekeeping */
782 inode->i_ctime = ext4_current_time(inode);
783 ext4_mark_inode_dirty(handle, inode);
785 /* had we spliced it onto indirect block? */
788 * If we spliced it onto an indirect block, we haven't
789 * altered the inode. Note however that if it is being spliced
790 * onto an indirect block at the very end of the file (the
791 * file is growing) then we *will* alter the inode to reflect
792 * the new i_size. But that is not done here - it is done in
793 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
795 jbd_debug(5, "splicing indirect only\n");
796 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
797 err = ext4_journal_dirty_metadata(handle, where->bh);
802 * OK, we spliced it into the inode itself on a direct block.
803 * Inode was dirtied above.
805 jbd_debug(5, "splicing direct\n");
810 for (i = 1; i <= num; i++) {
811 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
812 ext4_journal_forget(handle, where[i].bh);
813 ext4_free_blocks(handle, inode,
814 le32_to_cpu(where[i-1].key), 1, 0);
816 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
822 * Allocation strategy is simple: if we have to allocate something, we will
823 * have to go the whole way to leaf. So let's do it before attaching anything
824 * to tree, set linkage between the newborn blocks, write them if sync is
825 * required, recheck the path, free and repeat if check fails, otherwise
826 * set the last missing link (that will protect us from any truncate-generated
827 * removals - all blocks on the path are immune now) and possibly force the
828 * write on the parent block.
829 * That has a nice additional property: no special recovery from the failed
830 * allocations is needed - we simply release blocks and do not touch anything
831 * reachable from inode.
833 * `handle' can be NULL if create == 0.
835 * return > 0, # of blocks mapped or allocated.
836 * return = 0, if plain lookup failed.
837 * return < 0, error case.
840 * Need to be called with
841 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
842 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
844 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
845 ext4_lblk_t iblock, unsigned long maxblocks,
846 struct buffer_head *bh_result,
847 int create, int extend_disksize)
850 ext4_lblk_t offsets[4];
855 int blocks_to_boundary = 0;
857 struct ext4_inode_info *ei = EXT4_I(inode);
859 ext4_fsblk_t first_block = 0;
863 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
864 J_ASSERT(handle != NULL || create == 0);
865 depth = ext4_block_to_path(inode, iblock, offsets,
866 &blocks_to_boundary);
871 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
873 /* Simplest case - block found, no allocation needed */
875 first_block = le32_to_cpu(chain[depth - 1].key);
876 clear_buffer_new(bh_result);
879 while (count < maxblocks && count <= blocks_to_boundary) {
882 blk = le32_to_cpu(*(chain[depth-1].p + count));
884 if (blk == first_block + count)
892 /* Next simple case - plain lookup or failed read of indirect block */
893 if (!create || err == -EIO)
897 * Okay, we need to do block allocation.
899 goal = ext4_find_goal(inode, iblock, partial);
901 /* the number of blocks need to allocate for [d,t]indirect blocks */
902 indirect_blks = (chain + depth) - partial - 1;
905 * Next look up the indirect map to count the totoal number of
906 * direct blocks to allocate for this branch.
908 count = ext4_blks_to_allocate(partial, indirect_blks,
909 maxblocks, blocks_to_boundary);
911 * Block out ext4_truncate while we alter the tree
913 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
915 offsets + (partial - chain), partial);
918 * The ext4_splice_branch call will free and forget any buffers
919 * on the new chain if there is a failure, but that risks using
920 * up transaction credits, especially for bitmaps where the
921 * credits cannot be returned. Can we handle this somehow? We
922 * may need to return -EAGAIN upwards in the worst case. --sct
925 err = ext4_splice_branch(handle, inode, iblock,
926 partial, indirect_blks, count);
928 * i_disksize growing is protected by i_data_sem. Don't forget to
929 * protect it if you're about to implement concurrent
930 * ext4_get_block() -bzzz
932 if (!err && extend_disksize) {
933 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
934 if (disksize > i_size_read(inode))
935 disksize = i_size_read(inode);
936 if (disksize > ei->i_disksize)
937 ei->i_disksize = disksize;
942 set_buffer_new(bh_result);
944 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
945 if (count > blocks_to_boundary)
946 set_buffer_boundary(bh_result);
948 /* Clean up and exit */
949 partial = chain + depth - 1; /* the whole chain */
951 while (partial > chain) {
952 BUFFER_TRACE(partial->bh, "call brelse");
956 BUFFER_TRACE(bh_result, "returned");
962 * Calculate the number of metadata blocks need to reserve
963 * to allocate @blocks for non extent file based file
965 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
967 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
968 int ind_blks, dind_blks, tind_blks;
970 /* number of new indirect blocks needed */
971 ind_blks = (blocks + icap - 1) / icap;
973 dind_blks = (ind_blks + icap - 1) / icap;
977 return ind_blks + dind_blks + tind_blks;
981 * Calculate the number of metadata blocks need to reserve
982 * to allocate given number of blocks
984 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
989 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
990 return ext4_ext_calc_metadata_amount(inode, blocks);
992 return ext4_indirect_calc_metadata_amount(inode, blocks);
995 static void ext4_da_update_reserve_space(struct inode *inode, int used)
997 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
998 int total, mdb, mdb_free;
1000 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1001 /* recalculate the number of metablocks still need to be reserved */
1002 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1003 mdb = ext4_calc_metadata_amount(inode, total);
1005 /* figure out how many metablocks to release */
1006 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1007 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1010 /* Account for allocated meta_blocks */
1011 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1013 /* update fs dirty blocks counter */
1014 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1015 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1016 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1019 /* update per-inode reservations */
1020 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1021 EXT4_I(inode)->i_reserved_data_blocks -= used;
1023 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1027 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1028 * and returns if the blocks are already mapped.
1030 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1031 * and store the allocated blocks in the result buffer head and mark it
1034 * If file type is extents based, it will call ext4_ext_get_blocks(),
1035 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1038 * On success, it returns the number of blocks being mapped or allocate.
1039 * if create==0 and the blocks are pre-allocated and uninitialized block,
1040 * the result buffer head is unmapped. If the create ==1, it will make sure
1041 * the buffer head is mapped.
1043 * It returns 0 if plain look up failed (blocks have not been allocated), in
1044 * that casem, buffer head is unmapped
1046 * It returns the error in case of allocation failure.
1048 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1049 unsigned long max_blocks, struct buffer_head *bh,
1050 int create, int extend_disksize, int flag)
1054 clear_buffer_mapped(bh);
1057 * Try to see if we can get the block without requesting
1058 * for new file system block.
1060 down_read((&EXT4_I(inode)->i_data_sem));
1061 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1062 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1065 retval = ext4_get_blocks_handle(handle,
1066 inode, block, max_blocks, bh, 0, 0);
1068 up_read((&EXT4_I(inode)->i_data_sem));
1070 /* If it is only a block(s) look up */
1075 * Returns if the blocks have already allocated
1077 * Note that if blocks have been preallocated
1078 * ext4_ext_get_block() returns th create = 0
1079 * with buffer head unmapped.
1081 if (retval > 0 && buffer_mapped(bh))
1085 * New blocks allocate and/or writing to uninitialized extent
1086 * will possibly result in updating i_data, so we take
1087 * the write lock of i_data_sem, and call get_blocks()
1088 * with create == 1 flag.
1090 down_write((&EXT4_I(inode)->i_data_sem));
1093 * if the caller is from delayed allocation writeout path
1094 * we have already reserved fs blocks for allocation
1095 * let the underlying get_block() function know to
1096 * avoid double accounting
1099 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1101 * We need to check for EXT4 here because migrate
1102 * could have changed the inode type in between
1104 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1105 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1106 bh, create, extend_disksize);
1108 retval = ext4_get_blocks_handle(handle, inode, block,
1109 max_blocks, bh, create, extend_disksize);
1111 if (retval > 0 && buffer_new(bh)) {
1113 * We allocated new blocks which will result in
1114 * i_data's format changing. Force the migrate
1115 * to fail by clearing migrate flags
1117 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1123 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1125 * Update reserved blocks/metadata blocks
1126 * after successful block allocation
1127 * which were deferred till now
1129 if ((retval > 0) && buffer_delay(bh))
1130 ext4_da_update_reserve_space(inode, retval);
1133 up_write((&EXT4_I(inode)->i_data_sem));
1137 /* Maximum number of blocks we map for direct IO at once. */
1138 #define DIO_MAX_BLOCKS 4096
1140 int ext4_get_block(struct inode *inode, sector_t iblock,
1141 struct buffer_head *bh_result, int create)
1143 handle_t *handle = ext4_journal_current_handle();
1144 int ret = 0, started = 0;
1145 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1148 if (create && !handle) {
1149 /* Direct IO write... */
1150 if (max_blocks > DIO_MAX_BLOCKS)
1151 max_blocks = DIO_MAX_BLOCKS;
1152 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1153 handle = ext4_journal_start(inode, dio_credits);
1154 if (IS_ERR(handle)) {
1155 ret = PTR_ERR(handle);
1161 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1162 max_blocks, bh_result, create, 0, 0);
1164 bh_result->b_size = (ret << inode->i_blkbits);
1168 ext4_journal_stop(handle);
1174 * `handle' can be NULL if create is zero
1176 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1177 ext4_lblk_t block, int create, int *errp)
1179 struct buffer_head dummy;
1182 J_ASSERT(handle != NULL || create == 0);
1185 dummy.b_blocknr = -1000;
1186 buffer_trace_init(&dummy.b_history);
1187 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1188 &dummy, create, 1, 0);
1190 * ext4_get_blocks_handle() returns number of blocks
1191 * mapped. 0 in case of a HOLE.
1199 if (!err && buffer_mapped(&dummy)) {
1200 struct buffer_head *bh;
1201 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1206 if (buffer_new(&dummy)) {
1207 J_ASSERT(create != 0);
1208 J_ASSERT(handle != NULL);
1211 * Now that we do not always journal data, we should
1212 * keep in mind whether this should always journal the
1213 * new buffer as metadata. For now, regular file
1214 * writes use ext4_get_block instead, so it's not a
1218 BUFFER_TRACE(bh, "call get_create_access");
1219 fatal = ext4_journal_get_create_access(handle, bh);
1220 if (!fatal && !buffer_uptodate(bh)) {
1221 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1222 set_buffer_uptodate(bh);
1225 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1226 err = ext4_journal_dirty_metadata(handle, bh);
1230 BUFFER_TRACE(bh, "not a new buffer");
1243 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1244 ext4_lblk_t block, int create, int *err)
1246 struct buffer_head *bh;
1248 bh = ext4_getblk(handle, inode, block, create, err);
1251 if (buffer_uptodate(bh))
1253 ll_rw_block(READ_META, 1, &bh);
1255 if (buffer_uptodate(bh))
1262 static int walk_page_buffers(handle_t *handle,
1263 struct buffer_head *head,
1267 int (*fn)(handle_t *handle,
1268 struct buffer_head *bh))
1270 struct buffer_head *bh;
1271 unsigned block_start, block_end;
1272 unsigned blocksize = head->b_size;
1274 struct buffer_head *next;
1276 for (bh = head, block_start = 0;
1277 ret == 0 && (bh != head || !block_start);
1278 block_start = block_end, bh = next)
1280 next = bh->b_this_page;
1281 block_end = block_start + blocksize;
1282 if (block_end <= from || block_start >= to) {
1283 if (partial && !buffer_uptodate(bh))
1287 err = (*fn)(handle, bh);
1295 * To preserve ordering, it is essential that the hole instantiation and
1296 * the data write be encapsulated in a single transaction. We cannot
1297 * close off a transaction and start a new one between the ext4_get_block()
1298 * and the commit_write(). So doing the jbd2_journal_start at the start of
1299 * prepare_write() is the right place.
1301 * Also, this function can nest inside ext4_writepage() ->
1302 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1303 * has generated enough buffer credits to do the whole page. So we won't
1304 * block on the journal in that case, which is good, because the caller may
1307 * By accident, ext4 can be reentered when a transaction is open via
1308 * quota file writes. If we were to commit the transaction while thus
1309 * reentered, there can be a deadlock - we would be holding a quota
1310 * lock, and the commit would never complete if another thread had a
1311 * transaction open and was blocking on the quota lock - a ranking
1314 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1315 * will _not_ run commit under these circumstances because handle->h_ref
1316 * is elevated. We'll still have enough credits for the tiny quotafile
1319 static int do_journal_get_write_access(handle_t *handle,
1320 struct buffer_head *bh)
1322 if (!buffer_mapped(bh) || buffer_freed(bh))
1324 return ext4_journal_get_write_access(handle, bh);
1327 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1328 loff_t pos, unsigned len, unsigned flags,
1329 struct page **pagep, void **fsdata)
1331 struct inode *inode = mapping->host;
1332 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1339 index = pos >> PAGE_CACHE_SHIFT;
1340 from = pos & (PAGE_CACHE_SIZE - 1);
1344 handle = ext4_journal_start(inode, needed_blocks);
1345 if (IS_ERR(handle)) {
1346 ret = PTR_ERR(handle);
1350 page = grab_cache_page_write_begin(mapping, index, flags);
1352 ext4_journal_stop(handle);
1358 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1361 if (!ret && ext4_should_journal_data(inode)) {
1362 ret = walk_page_buffers(handle, page_buffers(page),
1363 from, to, NULL, do_journal_get_write_access);
1368 ext4_journal_stop(handle);
1369 page_cache_release(page);
1371 * block_write_begin may have instantiated a few blocks
1372 * outside i_size. Trim these off again. Don't need
1373 * i_size_read because we hold i_mutex.
1375 if (pos + len > inode->i_size)
1376 vmtruncate(inode, inode->i_size);
1379 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1385 /* For write_end() in data=journal mode */
1386 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1388 if (!buffer_mapped(bh) || buffer_freed(bh))
1390 set_buffer_uptodate(bh);
1391 return ext4_journal_dirty_metadata(handle, bh);
1395 * We need to pick up the new inode size which generic_commit_write gave us
1396 * `file' can be NULL - eg, when called from page_symlink().
1398 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1399 * buffers are managed internally.
1401 static int ext4_ordered_write_end(struct file *file,
1402 struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1406 handle_t *handle = ext4_journal_current_handle();
1407 struct inode *inode = mapping->host;
1410 ret = ext4_jbd2_file_inode(handle, inode);
1415 new_i_size = pos + copied;
1416 if (new_i_size > EXT4_I(inode)->i_disksize) {
1417 ext4_update_i_disksize(inode, new_i_size);
1418 /* We need to mark inode dirty even if
1419 * new_i_size is less that inode->i_size
1420 * bu greater than i_disksize.(hint delalloc)
1422 ext4_mark_inode_dirty(handle, inode);
1425 ret2 = generic_write_end(file, mapping, pos, len, copied,
1431 ret2 = ext4_journal_stop(handle);
1435 return ret ? ret : copied;
1438 static int ext4_writeback_write_end(struct file *file,
1439 struct address_space *mapping,
1440 loff_t pos, unsigned len, unsigned copied,
1441 struct page *page, void *fsdata)
1443 handle_t *handle = ext4_journal_current_handle();
1444 struct inode *inode = mapping->host;
1448 new_i_size = pos + copied;
1449 if (new_i_size > EXT4_I(inode)->i_disksize) {
1450 ext4_update_i_disksize(inode, new_i_size);
1451 /* We need to mark inode dirty even if
1452 * new_i_size is less that inode->i_size
1453 * bu greater than i_disksize.(hint delalloc)
1455 ext4_mark_inode_dirty(handle, inode);
1458 ret2 = generic_write_end(file, mapping, pos, len, copied,
1464 ret2 = ext4_journal_stop(handle);
1468 return ret ? ret : copied;
1471 static int ext4_journalled_write_end(struct file *file,
1472 struct address_space *mapping,
1473 loff_t pos, unsigned len, unsigned copied,
1474 struct page *page, void *fsdata)
1476 handle_t *handle = ext4_journal_current_handle();
1477 struct inode *inode = mapping->host;
1483 from = pos & (PAGE_CACHE_SIZE - 1);
1487 if (!PageUptodate(page))
1489 page_zero_new_buffers(page, from+copied, to);
1492 ret = walk_page_buffers(handle, page_buffers(page), from,
1493 to, &partial, write_end_fn);
1495 SetPageUptodate(page);
1496 new_i_size = pos + copied;
1497 if (new_i_size > inode->i_size)
1498 i_size_write(inode, pos+copied);
1499 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1500 if (new_i_size > EXT4_I(inode)->i_disksize) {
1501 ext4_update_i_disksize(inode, new_i_size);
1502 ret2 = ext4_mark_inode_dirty(handle, inode);
1508 ret2 = ext4_journal_stop(handle);
1511 page_cache_release(page);
1513 return ret ? ret : copied;
1516 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1519 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1520 unsigned long md_needed, mdblocks, total = 0;
1523 * recalculate the amount of metadata blocks to reserve
1524 * in order to allocate nrblocks
1525 * worse case is one extent per block
1528 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1529 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1530 mdblocks = ext4_calc_metadata_amount(inode, total);
1531 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1533 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1534 total = md_needed + nrblocks;
1536 if (ext4_claim_free_blocks(sbi, total)) {
1537 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1538 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1544 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1545 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1547 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1548 return 0; /* success */
1551 static void ext4_da_release_space(struct inode *inode, int to_free)
1553 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1554 int total, mdb, mdb_free, release;
1557 return; /* Nothing to release, exit */
1559 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1561 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1563 * if there is no reserved blocks, but we try to free some
1564 * then the counter is messed up somewhere.
1565 * but since this function is called from invalidate
1566 * page, it's harmless to return without any action
1568 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1569 "blocks for inode %lu, but there is no reserved "
1570 "data blocks\n", to_free, inode->i_ino);
1571 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1575 /* recalculate the number of metablocks still need to be reserved */
1576 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1577 mdb = ext4_calc_metadata_amount(inode, total);
1579 /* figure out how many metablocks to release */
1580 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1581 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1583 release = to_free + mdb_free;
1585 /* update fs dirty blocks counter for truncate case */
1586 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1588 /* update per-inode reservations */
1589 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1590 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1592 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1593 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1594 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1597 static void ext4_da_page_release_reservation(struct page *page,
1598 unsigned long offset)
1601 struct buffer_head *head, *bh;
1602 unsigned int curr_off = 0;
1604 head = page_buffers(page);
1607 unsigned int next_off = curr_off + bh->b_size;
1609 if ((offset <= curr_off) && (buffer_delay(bh))) {
1611 clear_buffer_delay(bh);
1613 curr_off = next_off;
1614 } while ((bh = bh->b_this_page) != head);
1615 ext4_da_release_space(page->mapping->host, to_release);
1619 * Delayed allocation stuff
1622 struct mpage_da_data {
1623 struct inode *inode;
1624 struct buffer_head lbh; /* extent of blocks */
1625 unsigned long first_page, next_page; /* extent of pages */
1626 get_block_t *get_block;
1627 struct writeback_control *wbc;
1634 * mpage_da_submit_io - walks through extent of pages and try to write
1635 * them with writepage() call back
1637 * @mpd->inode: inode
1638 * @mpd->first_page: first page of the extent
1639 * @mpd->next_page: page after the last page of the extent
1640 * @mpd->get_block: the filesystem's block mapper function
1642 * By the time mpage_da_submit_io() is called we expect all blocks
1643 * to be allocated. this may be wrong if allocation failed.
1645 * As pages are already locked by write_cache_pages(), we can't use it
1647 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1650 struct pagevec pvec;
1651 unsigned long index, end;
1652 int ret = 0, err, nr_pages, i;
1653 struct inode *inode = mpd->inode;
1654 struct address_space *mapping = inode->i_mapping;
1656 BUG_ON(mpd->next_page <= mpd->first_page);
1658 * We need to start from the first_page to the next_page - 1
1659 * to make sure we also write the mapped dirty buffer_heads.
1660 * If we look at mpd->lbh.b_blocknr we would only be looking
1661 * at the currently mapped buffer_heads.
1663 index = mpd->first_page;
1664 end = mpd->next_page - 1;
1666 pagevec_init(&pvec, 0);
1667 while (index <= end) {
1668 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1671 for (i = 0; i < nr_pages; i++) {
1672 struct page *page = pvec.pages[i];
1674 index = page->index;
1679 BUG_ON(!PageLocked(page));
1680 BUG_ON(PageWriteback(page));
1682 pages_skipped = mpd->wbc->pages_skipped;
1683 err = mapping->a_ops->writepage(page, mpd->wbc);
1684 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1686 * have successfully written the page
1687 * without skipping the same
1689 mpd->pages_written++;
1691 * In error case, we have to continue because
1692 * remaining pages are still locked
1693 * XXX: unlock and re-dirty them?
1698 pagevec_release(&pvec);
1704 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1706 * @mpd->inode - inode to walk through
1707 * @exbh->b_blocknr - first block on a disk
1708 * @exbh->b_size - amount of space in bytes
1709 * @logical - first logical block to start assignment with
1711 * the function goes through all passed space and put actual disk
1712 * block numbers into buffer heads, dropping BH_Delay
1714 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1715 struct buffer_head *exbh)
1717 struct inode *inode = mpd->inode;
1718 struct address_space *mapping = inode->i_mapping;
1719 int blocks = exbh->b_size >> inode->i_blkbits;
1720 sector_t pblock = exbh->b_blocknr, cur_logical;
1721 struct buffer_head *head, *bh;
1723 struct pagevec pvec;
1726 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1727 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1728 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1730 pagevec_init(&pvec, 0);
1732 while (index <= end) {
1733 /* XXX: optimize tail */
1734 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1737 for (i = 0; i < nr_pages; i++) {
1738 struct page *page = pvec.pages[i];
1740 index = page->index;
1745 BUG_ON(!PageLocked(page));
1746 BUG_ON(PageWriteback(page));
1747 BUG_ON(!page_has_buffers(page));
1749 bh = page_buffers(page);
1752 /* skip blocks out of the range */
1754 if (cur_logical >= logical)
1757 } while ((bh = bh->b_this_page) != head);
1760 if (cur_logical >= logical + blocks)
1762 if (buffer_delay(bh)) {
1763 bh->b_blocknr = pblock;
1764 clear_buffer_delay(bh);
1765 bh->b_bdev = inode->i_sb->s_bdev;
1766 } else if (buffer_unwritten(bh)) {
1767 bh->b_blocknr = pblock;
1768 clear_buffer_unwritten(bh);
1769 set_buffer_mapped(bh);
1771 bh->b_bdev = inode->i_sb->s_bdev;
1772 } else if (buffer_mapped(bh))
1773 BUG_ON(bh->b_blocknr != pblock);
1777 } while ((bh = bh->b_this_page) != head);
1779 pagevec_release(&pvec);
1785 * __unmap_underlying_blocks - just a helper function to unmap
1786 * set of blocks described by @bh
1788 static inline void __unmap_underlying_blocks(struct inode *inode,
1789 struct buffer_head *bh)
1791 struct block_device *bdev = inode->i_sb->s_bdev;
1794 blocks = bh->b_size >> inode->i_blkbits;
1795 for (i = 0; i < blocks; i++)
1796 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1799 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1800 sector_t logical, long blk_cnt)
1804 struct pagevec pvec;
1805 struct inode *inode = mpd->inode;
1806 struct address_space *mapping = inode->i_mapping;
1808 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1809 end = (logical + blk_cnt - 1) >>
1810 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1811 while (index <= end) {
1812 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1815 for (i = 0; i < nr_pages; i++) {
1816 struct page *page = pvec.pages[i];
1817 index = page->index;
1822 BUG_ON(!PageLocked(page));
1823 BUG_ON(PageWriteback(page));
1824 block_invalidatepage(page, 0);
1825 ClearPageUptodate(page);
1832 static void ext4_print_free_blocks(struct inode *inode)
1834 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1835 printk(KERN_EMERG "Total free blocks count %lld\n",
1836 ext4_count_free_blocks(inode->i_sb));
1837 printk(KERN_EMERG "Free/Dirty block details\n");
1838 printk(KERN_EMERG "free_blocks=%lld\n",
1839 percpu_counter_sum(&sbi->s_freeblocks_counter));
1840 printk(KERN_EMERG "dirty_blocks=%lld\n",
1841 percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1842 printk(KERN_EMERG "Block reservation details\n");
1843 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1844 EXT4_I(inode)->i_reserved_data_blocks);
1845 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1846 EXT4_I(inode)->i_reserved_meta_blocks);
1851 * mpage_da_map_blocks - go through given space
1853 * @mpd->lbh - bh describing space
1854 * @mpd->get_block - the filesystem's block mapper function
1856 * The function skips space we know is already mapped to disk blocks.
1859 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1862 struct buffer_head new;
1863 struct buffer_head *lbh = &mpd->lbh;
1867 * We consider only non-mapped and non-allocated blocks
1869 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1871 new.b_state = lbh->b_state;
1873 new.b_size = lbh->b_size;
1874 next = lbh->b_blocknr;
1876 * If we didn't accumulate anything
1877 * to write simply return
1881 err = mpd->get_block(mpd->inode, next, &new, 1);
1884 /* If get block returns with error
1885 * we simply return. Later writepage
1886 * will redirty the page and writepages
1887 * will find the dirty page again
1892 if (err == -ENOSPC &&
1893 ext4_count_free_blocks(mpd->inode->i_sb)) {
1899 * get block failure will cause us
1900 * to loop in writepages. Because
1901 * a_ops->writepage won't be able to
1902 * make progress. The page will be redirtied
1903 * by writepage and writepages will again
1904 * try to write the same.
1906 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1907 "at logical offset %llu with max blocks "
1908 "%zd with error %d\n",
1909 __func__, mpd->inode->i_ino,
1910 (unsigned long long)next,
1911 lbh->b_size >> mpd->inode->i_blkbits, err);
1912 printk(KERN_EMERG "This should not happen.!! "
1913 "Data will be lost\n");
1914 if (err == -ENOSPC) {
1915 ext4_print_free_blocks(mpd->inode);
1917 /* invlaidate all the pages */
1918 ext4_da_block_invalidatepages(mpd, next,
1919 lbh->b_size >> mpd->inode->i_blkbits);
1922 BUG_ON(new.b_size == 0);
1924 if (buffer_new(&new))
1925 __unmap_underlying_blocks(mpd->inode, &new);
1928 * If blocks are delayed marked, we need to
1929 * put actual blocknr and drop delayed bit
1931 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1932 mpage_put_bnr_to_bhs(mpd, next, &new);
1937 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1938 (1 << BH_Delay) | (1 << BH_Unwritten))
1941 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1943 * @mpd->lbh - extent of blocks
1944 * @logical - logical number of the block in the file
1945 * @bh - bh of the block (used to access block's state)
1947 * the function is used to collect contig. blocks in same state
1949 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1950 sector_t logical, struct buffer_head *bh)
1953 size_t b_size = bh->b_size;
1954 struct buffer_head *lbh = &mpd->lbh;
1955 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1957 /* check if thereserved journal credits might overflow */
1958 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1959 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1961 * With non-extent format we are limited by the journal
1962 * credit available. Total credit needed to insert
1963 * nrblocks contiguous blocks is dependent on the
1964 * nrblocks. So limit nrblocks.
1967 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1968 EXT4_MAX_TRANS_DATA) {
1970 * Adding the new buffer_head would make it cross the
1971 * allowed limit for which we have journal credit
1972 * reserved. So limit the new bh->b_size
1974 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1975 mpd->inode->i_blkbits;
1976 /* we will do mpage_da_submit_io in the next loop */
1980 * First block in the extent
1982 if (lbh->b_size == 0) {
1983 lbh->b_blocknr = logical;
1984 lbh->b_size = b_size;
1985 lbh->b_state = bh->b_state & BH_FLAGS;
1989 next = lbh->b_blocknr + nrblocks;
1991 * Can we merge the block to our big extent?
1993 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1994 lbh->b_size += b_size;
2000 * We couldn't merge the block to our extent, so we
2001 * need to flush current extent and start new one
2003 if (mpage_da_map_blocks(mpd) == 0)
2004 mpage_da_submit_io(mpd);
2010 * __mpage_da_writepage - finds extent of pages and blocks
2012 * @page: page to consider
2013 * @wbc: not used, we just follow rules
2016 * The function finds extents of pages and scan them for all blocks.
2018 static int __mpage_da_writepage(struct page *page,
2019 struct writeback_control *wbc, void *data)
2021 struct mpage_da_data *mpd = data;
2022 struct inode *inode = mpd->inode;
2023 struct buffer_head *bh, *head, fake;
2028 * Rest of the page in the page_vec
2029 * redirty then and skip then. We will
2030 * try to to write them again after
2031 * starting a new transaction
2033 redirty_page_for_writepage(wbc, page);
2035 return MPAGE_DA_EXTENT_TAIL;
2038 * Can we merge this page to current extent?
2040 if (mpd->next_page != page->index) {
2042 * Nope, we can't. So, we map non-allocated blocks
2043 * and start IO on them using writepage()
2045 if (mpd->next_page != mpd->first_page) {
2046 if (mpage_da_map_blocks(mpd) == 0)
2047 mpage_da_submit_io(mpd);
2049 * skip rest of the page in the page_vec
2052 redirty_page_for_writepage(wbc, page);
2054 return MPAGE_DA_EXTENT_TAIL;
2058 * Start next extent of pages ...
2060 mpd->first_page = page->index;
2065 mpd->lbh.b_size = 0;
2066 mpd->lbh.b_state = 0;
2067 mpd->lbh.b_blocknr = 0;
2070 mpd->next_page = page->index + 1;
2071 logical = (sector_t) page->index <<
2072 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2074 if (!page_has_buffers(page)) {
2076 * There is no attached buffer heads yet (mmap?)
2077 * we treat the page asfull of dirty blocks
2080 bh->b_size = PAGE_CACHE_SIZE;
2082 set_buffer_dirty(bh);
2083 set_buffer_uptodate(bh);
2084 mpage_add_bh_to_extent(mpd, logical, bh);
2086 return MPAGE_DA_EXTENT_TAIL;
2089 * Page with regular buffer heads, just add all dirty ones
2091 head = page_buffers(page);
2094 BUG_ON(buffer_locked(bh));
2096 * We need to try to allocate
2097 * unmapped blocks in the same page.
2098 * Otherwise we won't make progress
2099 * with the page in ext4_da_writepage
2101 if (buffer_dirty(bh) &&
2102 (!buffer_mapped(bh) || buffer_delay(bh))) {
2103 mpage_add_bh_to_extent(mpd, logical, bh);
2105 return MPAGE_DA_EXTENT_TAIL;
2106 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2108 * mapped dirty buffer. We need to update
2109 * the b_state because we look at
2110 * b_state in mpage_da_map_blocks. We don't
2111 * update b_size because if we find an
2112 * unmapped buffer_head later we need to
2113 * use the b_state flag of that buffer_head.
2115 if (mpd->lbh.b_size == 0)
2117 bh->b_state & BH_FLAGS;
2120 } while ((bh = bh->b_this_page) != head);
2127 * mpage_da_writepages - walk the list of dirty pages of the given
2128 * address space, allocates non-allocated blocks, maps newly-allocated
2129 * blocks to existing bhs and issue IO them
2131 * @mapping: address space structure to write
2132 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2133 * @get_block: the filesystem's block mapper function.
2135 * This is a library function, which implements the writepages()
2136 * address_space_operation.
2138 static int mpage_da_writepages(struct address_space *mapping,
2139 struct writeback_control *wbc,
2140 struct mpage_da_data *mpd)
2144 if (!mpd->get_block)
2145 return generic_writepages(mapping, wbc);
2147 mpd->lbh.b_size = 0;
2148 mpd->lbh.b_state = 0;
2149 mpd->lbh.b_blocknr = 0;
2150 mpd->first_page = 0;
2153 mpd->pages_written = 0;
2156 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2158 * Handle last extent of pages
2160 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2161 if (mpage_da_map_blocks(mpd) == 0)
2162 mpage_da_submit_io(mpd);
2165 ret = MPAGE_DA_EXTENT_TAIL;
2167 wbc->nr_to_write -= mpd->pages_written;
2172 * this is a special callback for ->write_begin() only
2173 * it's intention is to return mapped block or reserve space
2175 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2176 struct buffer_head *bh_result, int create)
2180 BUG_ON(create == 0);
2181 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2184 * first, we need to know whether the block is allocated already
2185 * preallocated blocks are unmapped but should treated
2186 * the same as allocated blocks.
2188 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2189 if ((ret == 0) && !buffer_delay(bh_result)) {
2190 /* the block isn't (pre)allocated yet, let's reserve space */
2192 * XXX: __block_prepare_write() unmaps passed block,
2195 ret = ext4_da_reserve_space(inode, 1);
2197 /* not enough space to reserve */
2200 map_bh(bh_result, inode->i_sb, 0);
2201 set_buffer_new(bh_result);
2202 set_buffer_delay(bh_result);
2203 } else if (ret > 0) {
2204 bh_result->b_size = (ret << inode->i_blkbits);
2210 #define EXT4_DELALLOC_RSVED 1
2211 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2212 struct buffer_head *bh_result, int create)
2215 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2216 loff_t disksize = EXT4_I(inode)->i_disksize;
2217 handle_t *handle = NULL;
2219 handle = ext4_journal_current_handle();
2221 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2222 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2225 bh_result->b_size = (ret << inode->i_blkbits);
2227 if (ext4_should_order_data(inode)) {
2229 retval = ext4_jbd2_file_inode(handle, inode);
2232 * Failed to add inode for ordered
2233 * mode. Don't update file size
2239 * Update on-disk size along with block allocation
2240 * we don't use 'extend_disksize' as size may change
2241 * within already allocated block -bzzz
2243 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2244 if (disksize > i_size_read(inode))
2245 disksize = i_size_read(inode);
2246 if (disksize > EXT4_I(inode)->i_disksize) {
2247 ext4_update_i_disksize(inode, disksize);
2248 ret = ext4_mark_inode_dirty(handle, inode);
2256 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2259 * unmapped buffer is possible for holes.
2260 * delay buffer is possible with delayed allocation
2262 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2265 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2266 struct buffer_head *bh_result, int create)
2269 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2272 * we don't want to do block allocation in writepage
2273 * so call get_block_wrap with create = 0
2275 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2276 bh_result, 0, 0, 0);
2278 bh_result->b_size = (ret << inode->i_blkbits);
2285 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2286 * get called via journal_submit_inode_data_buffers (no journal handle)
2287 * get called via shrink_page_list via pdflush (no journal handle)
2288 * or grab_page_cache when doing write_begin (have journal handle)
2290 static int ext4_da_writepage(struct page *page,
2291 struct writeback_control *wbc)
2296 struct buffer_head *page_bufs;
2297 struct inode *inode = page->mapping->host;
2299 size = i_size_read(inode);
2300 if (page->index == size >> PAGE_CACHE_SHIFT)
2301 len = size & ~PAGE_CACHE_MASK;
2303 len = PAGE_CACHE_SIZE;
2305 if (page_has_buffers(page)) {
2306 page_bufs = page_buffers(page);
2307 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2308 ext4_bh_unmapped_or_delay)) {
2310 * We don't want to do block allocation
2311 * So redirty the page and return
2312 * We may reach here when we do a journal commit
2313 * via journal_submit_inode_data_buffers.
2314 * If we don't have mapping block we just ignore
2315 * them. We can also reach here via shrink_page_list
2317 redirty_page_for_writepage(wbc, page);
2323 * The test for page_has_buffers() is subtle:
2324 * We know the page is dirty but it lost buffers. That means
2325 * that at some moment in time after write_begin()/write_end()
2326 * has been called all buffers have been clean and thus they
2327 * must have been written at least once. So they are all
2328 * mapped and we can happily proceed with mapping them
2329 * and writing the page.
2331 * Try to initialize the buffer_heads and check whether
2332 * all are mapped and non delay. We don't want to
2333 * do block allocation here.
2335 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2336 ext4_normal_get_block_write);
2338 page_bufs = page_buffers(page);
2339 /* check whether all are mapped and non delay */
2340 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2341 ext4_bh_unmapped_or_delay)) {
2342 redirty_page_for_writepage(wbc, page);
2348 * We can't do block allocation here
2349 * so just redity the page and unlock
2352 redirty_page_for_writepage(wbc, page);
2356 /* now mark the buffer_heads as dirty and uptodate */
2357 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2360 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2361 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2363 ret = block_write_full_page(page,
2364 ext4_normal_get_block_write,
2371 * This is called via ext4_da_writepages() to
2372 * calulate the total number of credits to reserve to fit
2373 * a single extent allocation into a single transaction,
2374 * ext4_da_writpeages() will loop calling this before
2375 * the block allocation.
2378 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2380 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2383 * With non-extent format the journal credit needed to
2384 * insert nrblocks contiguous block is dependent on
2385 * number of contiguous block. So we will limit
2386 * number of contiguous block to a sane value
2388 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2389 (max_blocks > EXT4_MAX_TRANS_DATA))
2390 max_blocks = EXT4_MAX_TRANS_DATA;
2392 return ext4_chunk_trans_blocks(inode, max_blocks);
2395 static int ext4_da_writepages(struct address_space *mapping,
2396 struct writeback_control *wbc)
2399 int range_whole = 0;
2400 handle_t *handle = NULL;
2401 struct mpage_da_data mpd;
2402 struct inode *inode = mapping->host;
2403 int no_nrwrite_index_update;
2404 long pages_written = 0, pages_skipped;
2405 int needed_blocks, ret = 0, nr_to_writebump = 0;
2406 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2409 * No pages to write? This is mainly a kludge to avoid starting
2410 * a transaction for special inodes like journal inode on last iput()
2411 * because that could violate lock ordering on umount
2413 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2417 * If the filesystem has aborted, it is read-only, so return
2418 * right away instead of dumping stack traces later on that
2419 * will obscure the real source of the problem. We test
2420 * EXT4_MOUNT_ABORT instead of sb->s_flag's MS_RDONLY because
2421 * the latter could be true if the filesystem is mounted
2422 * read-only, and in that case, ext4_da_writepages should
2423 * *never* be called, so if that ever happens, we would want
2426 if (unlikely(sbi->s_mount_opt & EXT4_MOUNT_ABORT))
2430 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2431 * This make sure small files blocks are allocated in
2432 * single attempt. This ensure that small files
2433 * get less fragmented.
2435 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2436 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2437 wbc->nr_to_write = sbi->s_mb_stream_request;
2439 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2442 if (wbc->range_cyclic)
2443 index = mapping->writeback_index;
2445 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2448 mpd.inode = mapping->host;
2451 * we don't want write_cache_pages to update
2452 * nr_to_write and writeback_index
2454 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2455 wbc->no_nrwrite_index_update = 1;
2456 pages_skipped = wbc->pages_skipped;
2458 while (!ret && wbc->nr_to_write > 0) {
2461 * we insert one extent at a time. So we need
2462 * credit needed for single extent allocation.
2463 * journalled mode is currently not supported
2466 BUG_ON(ext4_should_journal_data(inode));
2467 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2469 /* start a new transaction*/
2470 handle = ext4_journal_start(inode, needed_blocks);
2471 if (IS_ERR(handle)) {
2472 ret = PTR_ERR(handle);
2473 printk(KERN_CRIT "%s: jbd2_start: "
2474 "%ld pages, ino %lu; err %d\n", __func__,
2475 wbc->nr_to_write, inode->i_ino, ret);
2477 goto out_writepages;
2479 mpd.get_block = ext4_da_get_block_write;
2480 ret = mpage_da_writepages(mapping, wbc, &mpd);
2482 ext4_journal_stop(handle);
2484 if (mpd.retval == -ENOSPC) {
2485 /* commit the transaction which would
2486 * free blocks released in the transaction
2489 jbd2_journal_force_commit_nested(sbi->s_journal);
2490 wbc->pages_skipped = pages_skipped;
2492 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2494 * got one extent now try with
2497 pages_written += mpd.pages_written;
2498 wbc->pages_skipped = pages_skipped;
2500 } else if (wbc->nr_to_write)
2502 * There is no more writeout needed
2503 * or we requested for a noblocking writeout
2504 * and we found the device congested
2508 if (pages_skipped != wbc->pages_skipped)
2509 printk(KERN_EMERG "This should not happen leaving %s "
2510 "with nr_to_write = %ld ret = %d\n",
2511 __func__, wbc->nr_to_write, ret);
2514 index += pages_written;
2515 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2517 * set the writeback_index so that range_cyclic
2518 * mode will write it back later
2520 mapping->writeback_index = index;
2523 if (!no_nrwrite_index_update)
2524 wbc->no_nrwrite_index_update = 0;
2525 wbc->nr_to_write -= nr_to_writebump;
2529 #define FALL_BACK_TO_NONDELALLOC 1
2530 static int ext4_nonda_switch(struct super_block *sb)
2532 s64 free_blocks, dirty_blocks;
2533 struct ext4_sb_info *sbi = EXT4_SB(sb);
2536 * switch to non delalloc mode if we are running low
2537 * on free block. The free block accounting via percpu
2538 * counters can get slightly wrong with FBC_BATCH getting
2539 * accumulated on each CPU without updating global counters
2540 * Delalloc need an accurate free block accounting. So switch
2541 * to non delalloc when we are near to error range.
2543 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2544 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2545 if (2 * free_blocks < 3 * dirty_blocks ||
2546 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2548 * free block count is less that 150% of dirty blocks
2549 * or free blocks is less that watermark
2556 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2557 loff_t pos, unsigned len, unsigned flags,
2558 struct page **pagep, void **fsdata)
2560 int ret, retries = 0;
2564 struct inode *inode = mapping->host;
2567 index = pos >> PAGE_CACHE_SHIFT;
2568 from = pos & (PAGE_CACHE_SIZE - 1);
2571 if (ext4_nonda_switch(inode->i_sb)) {
2572 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2573 return ext4_write_begin(file, mapping, pos,
2574 len, flags, pagep, fsdata);
2576 *fsdata = (void *)0;
2579 * With delayed allocation, we don't log the i_disksize update
2580 * if there is delayed block allocation. But we still need
2581 * to journalling the i_disksize update if writes to the end
2582 * of file which has an already mapped buffer.
2584 handle = ext4_journal_start(inode, 1);
2585 if (IS_ERR(handle)) {
2586 ret = PTR_ERR(handle);
2590 page = grab_cache_page_write_begin(mapping, index, flags);
2592 ext4_journal_stop(handle);
2598 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2599 ext4_da_get_block_prep);
2602 ext4_journal_stop(handle);
2603 page_cache_release(page);
2605 * block_write_begin may have instantiated a few blocks
2606 * outside i_size. Trim these off again. Don't need
2607 * i_size_read because we hold i_mutex.
2609 if (pos + len > inode->i_size)
2610 vmtruncate(inode, inode->i_size);
2613 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2620 * Check if we should update i_disksize
2621 * when write to the end of file but not require block allocation
2623 static int ext4_da_should_update_i_disksize(struct page *page,
2624 unsigned long offset)
2626 struct buffer_head *bh;
2627 struct inode *inode = page->mapping->host;
2631 bh = page_buffers(page);
2632 idx = offset >> inode->i_blkbits;
2634 for (i = 0; i < idx; i++)
2635 bh = bh->b_this_page;
2637 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2642 static int ext4_da_write_end(struct file *file,
2643 struct address_space *mapping,
2644 loff_t pos, unsigned len, unsigned copied,
2645 struct page *page, void *fsdata)
2647 struct inode *inode = mapping->host;
2649 handle_t *handle = ext4_journal_current_handle();
2651 unsigned long start, end;
2652 int write_mode = (int)(unsigned long)fsdata;
2654 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2655 if (ext4_should_order_data(inode)) {
2656 return ext4_ordered_write_end(file, mapping, pos,
2657 len, copied, page, fsdata);
2658 } else if (ext4_should_writeback_data(inode)) {
2659 return ext4_writeback_write_end(file, mapping, pos,
2660 len, copied, page, fsdata);
2666 start = pos & (PAGE_CACHE_SIZE - 1);
2667 end = start + copied - 1;
2670 * generic_write_end() will run mark_inode_dirty() if i_size
2671 * changes. So let's piggyback the i_disksize mark_inode_dirty
2675 new_i_size = pos + copied;
2676 if (new_i_size > EXT4_I(inode)->i_disksize) {
2677 if (ext4_da_should_update_i_disksize(page, end)) {
2678 down_write(&EXT4_I(inode)->i_data_sem);
2679 if (new_i_size > EXT4_I(inode)->i_disksize) {
2681 * Updating i_disksize when extending file
2682 * without needing block allocation
2684 if (ext4_should_order_data(inode))
2685 ret = ext4_jbd2_file_inode(handle,
2688 EXT4_I(inode)->i_disksize = new_i_size;
2690 up_write(&EXT4_I(inode)->i_data_sem);
2691 /* We need to mark inode dirty even if
2692 * new_i_size is less that inode->i_size
2693 * bu greater than i_disksize.(hint delalloc)
2695 ext4_mark_inode_dirty(handle, inode);
2698 ret2 = generic_write_end(file, mapping, pos, len, copied,
2703 ret2 = ext4_journal_stop(handle);
2707 return ret ? ret : copied;
2710 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2713 * Drop reserved blocks
2715 BUG_ON(!PageLocked(page));
2716 if (!page_has_buffers(page))
2719 ext4_da_page_release_reservation(page, offset);
2722 ext4_invalidatepage(page, offset);
2729 * bmap() is special. It gets used by applications such as lilo and by
2730 * the swapper to find the on-disk block of a specific piece of data.
2732 * Naturally, this is dangerous if the block concerned is still in the
2733 * journal. If somebody makes a swapfile on an ext4 data-journaling
2734 * filesystem and enables swap, then they may get a nasty shock when the
2735 * data getting swapped to that swapfile suddenly gets overwritten by
2736 * the original zero's written out previously to the journal and
2737 * awaiting writeback in the kernel's buffer cache.
2739 * So, if we see any bmap calls here on a modified, data-journaled file,
2740 * take extra steps to flush any blocks which might be in the cache.
2742 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2744 struct inode *inode = mapping->host;
2748 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2749 test_opt(inode->i_sb, DELALLOC)) {
2751 * With delalloc we want to sync the file
2752 * so that we can make sure we allocate
2755 filemap_write_and_wait(mapping);
2758 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2760 * This is a REALLY heavyweight approach, but the use of
2761 * bmap on dirty files is expected to be extremely rare:
2762 * only if we run lilo or swapon on a freshly made file
2763 * do we expect this to happen.
2765 * (bmap requires CAP_SYS_RAWIO so this does not
2766 * represent an unprivileged user DOS attack --- we'd be
2767 * in trouble if mortal users could trigger this path at
2770 * NB. EXT4_STATE_JDATA is not set on files other than
2771 * regular files. If somebody wants to bmap a directory
2772 * or symlink and gets confused because the buffer
2773 * hasn't yet been flushed to disk, they deserve
2774 * everything they get.
2777 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2778 journal = EXT4_JOURNAL(inode);
2779 jbd2_journal_lock_updates(journal);
2780 err = jbd2_journal_flush(journal);
2781 jbd2_journal_unlock_updates(journal);
2787 return generic_block_bmap(mapping, block, ext4_get_block);
2790 static int bget_one(handle_t *handle, struct buffer_head *bh)
2796 static int bput_one(handle_t *handle, struct buffer_head *bh)
2803 * Note that we don't need to start a transaction unless we're journaling data
2804 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2805 * need to file the inode to the transaction's list in ordered mode because if
2806 * we are writing back data added by write(), the inode is already there and if
2807 * we are writing back data modified via mmap(), noone guarantees in which
2808 * transaction the data will hit the disk. In case we are journaling data, we
2809 * cannot start transaction directly because transaction start ranks above page
2810 * lock so we have to do some magic.
2812 * In all journaling modes block_write_full_page() will start the I/O.
2816 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2821 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2823 * Same applies to ext4_get_block(). We will deadlock on various things like
2824 * lock_journal and i_data_sem
2826 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2829 * 16May01: If we're reentered then journal_current_handle() will be
2830 * non-zero. We simply *return*.
2832 * 1 July 2001: @@@ FIXME:
2833 * In journalled data mode, a data buffer may be metadata against the
2834 * current transaction. But the same file is part of a shared mapping
2835 * and someone does a writepage() on it.
2837 * We will move the buffer onto the async_data list, but *after* it has
2838 * been dirtied. So there's a small window where we have dirty data on
2841 * Note that this only applies to the last partial page in the file. The
2842 * bit which block_write_full_page() uses prepare/commit for. (That's
2843 * broken code anyway: it's wrong for msync()).
2845 * It's a rare case: affects the final partial page, for journalled data
2846 * where the file is subject to bith write() and writepage() in the same
2847 * transction. To fix it we'll need a custom block_write_full_page().
2848 * We'll probably need that anyway for journalling writepage() output.
2850 * We don't honour synchronous mounts for writepage(). That would be
2851 * disastrous. Any write() or metadata operation will sync the fs for
2855 static int __ext4_normal_writepage(struct page *page,
2856 struct writeback_control *wbc)
2858 struct inode *inode = page->mapping->host;
2860 if (test_opt(inode->i_sb, NOBH))
2861 return nobh_writepage(page,
2862 ext4_normal_get_block_write, wbc);
2864 return block_write_full_page(page,
2865 ext4_normal_get_block_write,
2869 static int ext4_normal_writepage(struct page *page,
2870 struct writeback_control *wbc)
2872 struct inode *inode = page->mapping->host;
2873 loff_t size = i_size_read(inode);
2876 J_ASSERT(PageLocked(page));
2877 if (page->index == size >> PAGE_CACHE_SHIFT)
2878 len = size & ~PAGE_CACHE_MASK;
2880 len = PAGE_CACHE_SIZE;
2882 if (page_has_buffers(page)) {
2883 /* if page has buffers it should all be mapped
2884 * and allocated. If there are not buffers attached
2885 * to the page we know the page is dirty but it lost
2886 * buffers. That means that at some moment in time
2887 * after write_begin() / write_end() has been called
2888 * all buffers have been clean and thus they must have been
2889 * written at least once. So they are all mapped and we can
2890 * happily proceed with mapping them and writing the page.
2892 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2893 ext4_bh_unmapped_or_delay));
2896 if (!ext4_journal_current_handle())
2897 return __ext4_normal_writepage(page, wbc);
2899 redirty_page_for_writepage(wbc, page);
2904 static int __ext4_journalled_writepage(struct page *page,
2905 struct writeback_control *wbc)
2907 struct address_space *mapping = page->mapping;
2908 struct inode *inode = mapping->host;
2909 struct buffer_head *page_bufs;
2910 handle_t *handle = NULL;
2914 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2915 ext4_normal_get_block_write);
2919 page_bufs = page_buffers(page);
2920 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2922 /* As soon as we unlock the page, it can go away, but we have
2923 * references to buffers so we are safe */
2926 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2927 if (IS_ERR(handle)) {
2928 ret = PTR_ERR(handle);
2932 ret = walk_page_buffers(handle, page_bufs, 0,
2933 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2935 err = walk_page_buffers(handle, page_bufs, 0,
2936 PAGE_CACHE_SIZE, NULL, write_end_fn);
2939 err = ext4_journal_stop(handle);
2943 walk_page_buffers(handle, page_bufs, 0,
2944 PAGE_CACHE_SIZE, NULL, bput_one);
2945 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2954 static int ext4_journalled_writepage(struct page *page,
2955 struct writeback_control *wbc)
2957 struct inode *inode = page->mapping->host;
2958 loff_t size = i_size_read(inode);
2961 J_ASSERT(PageLocked(page));
2962 if (page->index == size >> PAGE_CACHE_SHIFT)
2963 len = size & ~PAGE_CACHE_MASK;
2965 len = PAGE_CACHE_SIZE;
2967 if (page_has_buffers(page)) {
2968 /* if page has buffers it should all be mapped
2969 * and allocated. If there are not buffers attached
2970 * to the page we know the page is dirty but it lost
2971 * buffers. That means that at some moment in time
2972 * after write_begin() / write_end() has been called
2973 * all buffers have been clean and thus they must have been
2974 * written at least once. So they are all mapped and we can
2975 * happily proceed with mapping them and writing the page.
2977 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2978 ext4_bh_unmapped_or_delay));
2981 if (ext4_journal_current_handle())
2984 if (PageChecked(page)) {
2986 * It's mmapped pagecache. Add buffers and journal it. There
2987 * doesn't seem much point in redirtying the page here.
2989 ClearPageChecked(page);
2990 return __ext4_journalled_writepage(page, wbc);
2993 * It may be a page full of checkpoint-mode buffers. We don't
2994 * really know unless we go poke around in the buffer_heads.
2995 * But block_write_full_page will do the right thing.
2997 return block_write_full_page(page,
2998 ext4_normal_get_block_write,
3002 redirty_page_for_writepage(wbc, page);
3007 static int ext4_readpage(struct file *file, struct page *page)
3009 return mpage_readpage(page, ext4_get_block);
3013 ext4_readpages(struct file *file, struct address_space *mapping,
3014 struct list_head *pages, unsigned nr_pages)
3016 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3019 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3021 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3024 * If it's a full truncate we just forget about the pending dirtying
3027 ClearPageChecked(page);
3029 jbd2_journal_invalidatepage(journal, page, offset);
3032 static int ext4_releasepage(struct page *page, gfp_t wait)
3034 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3036 WARN_ON(PageChecked(page));
3037 if (!page_has_buffers(page))
3039 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3043 * If the O_DIRECT write will extend the file then add this inode to the
3044 * orphan list. So recovery will truncate it back to the original size
3045 * if the machine crashes during the write.
3047 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3048 * crashes then stale disk data _may_ be exposed inside the file. But current
3049 * VFS code falls back into buffered path in that case so we are safe.
3051 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3052 const struct iovec *iov, loff_t offset,
3053 unsigned long nr_segs)
3055 struct file *file = iocb->ki_filp;
3056 struct inode *inode = file->f_mapping->host;
3057 struct ext4_inode_info *ei = EXT4_I(inode);
3061 size_t count = iov_length(iov, nr_segs);
3064 loff_t final_size = offset + count;
3066 if (final_size > inode->i_size) {
3067 /* Credits for sb + inode write */
3068 handle = ext4_journal_start(inode, 2);
3069 if (IS_ERR(handle)) {
3070 ret = PTR_ERR(handle);
3073 ret = ext4_orphan_add(handle, inode);
3075 ext4_journal_stop(handle);
3079 ei->i_disksize = inode->i_size;
3080 ext4_journal_stop(handle);
3084 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3086 ext4_get_block, NULL);
3091 /* Credits for sb + inode write */
3092 handle = ext4_journal_start(inode, 2);
3093 if (IS_ERR(handle)) {
3094 /* This is really bad luck. We've written the data
3095 * but cannot extend i_size. Bail out and pretend
3096 * the write failed... */
3097 ret = PTR_ERR(handle);
3101 ext4_orphan_del(handle, inode);
3103 loff_t end = offset + ret;
3104 if (end > inode->i_size) {
3105 ei->i_disksize = end;
3106 i_size_write(inode, end);
3108 * We're going to return a positive `ret'
3109 * here due to non-zero-length I/O, so there's
3110 * no way of reporting error returns from
3111 * ext4_mark_inode_dirty() to userspace. So
3114 ext4_mark_inode_dirty(handle, inode);
3117 err = ext4_journal_stop(handle);
3126 * Pages can be marked dirty completely asynchronously from ext4's journalling
3127 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3128 * much here because ->set_page_dirty is called under VFS locks. The page is
3129 * not necessarily locked.
3131 * We cannot just dirty the page and leave attached buffers clean, because the
3132 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3133 * or jbddirty because all the journalling code will explode.
3135 * So what we do is to mark the page "pending dirty" and next time writepage
3136 * is called, propagate that into the buffers appropriately.
3138 static int ext4_journalled_set_page_dirty(struct page *page)
3140 SetPageChecked(page);
3141 return __set_page_dirty_nobuffers(page);
3144 static const struct address_space_operations ext4_ordered_aops = {
3145 .readpage = ext4_readpage,
3146 .readpages = ext4_readpages,
3147 .writepage = ext4_normal_writepage,
3148 .sync_page = block_sync_page,
3149 .write_begin = ext4_write_begin,
3150 .write_end = ext4_ordered_write_end,
3152 .invalidatepage = ext4_invalidatepage,
3153 .releasepage = ext4_releasepage,
3154 .direct_IO = ext4_direct_IO,
3155 .migratepage = buffer_migrate_page,
3156 .is_partially_uptodate = block_is_partially_uptodate,
3159 static const struct address_space_operations ext4_writeback_aops = {
3160 .readpage = ext4_readpage,
3161 .readpages = ext4_readpages,
3162 .writepage = ext4_normal_writepage,
3163 .sync_page = block_sync_page,
3164 .write_begin = ext4_write_begin,
3165 .write_end = ext4_writeback_write_end,
3167 .invalidatepage = ext4_invalidatepage,
3168 .releasepage = ext4_releasepage,
3169 .direct_IO = ext4_direct_IO,
3170 .migratepage = buffer_migrate_page,
3171 .is_partially_uptodate = block_is_partially_uptodate,
3174 static const struct address_space_operations ext4_journalled_aops = {
3175 .readpage = ext4_readpage,
3176 .readpages = ext4_readpages,
3177 .writepage = ext4_journalled_writepage,
3178 .sync_page = block_sync_page,
3179 .write_begin = ext4_write_begin,
3180 .write_end = ext4_journalled_write_end,
3181 .set_page_dirty = ext4_journalled_set_page_dirty,
3183 .invalidatepage = ext4_invalidatepage,
3184 .releasepage = ext4_releasepage,
3185 .is_partially_uptodate = block_is_partially_uptodate,
3188 static const struct address_space_operations ext4_da_aops = {
3189 .readpage = ext4_readpage,
3190 .readpages = ext4_readpages,
3191 .writepage = ext4_da_writepage,
3192 .writepages = ext4_da_writepages,
3193 .sync_page = block_sync_page,
3194 .write_begin = ext4_da_write_begin,
3195 .write_end = ext4_da_write_end,
3197 .invalidatepage = ext4_da_invalidatepage,
3198 .releasepage = ext4_releasepage,
3199 .direct_IO = ext4_direct_IO,
3200 .migratepage = buffer_migrate_page,
3201 .is_partially_uptodate = block_is_partially_uptodate,
3204 void ext4_set_aops(struct inode *inode)
3206 if (ext4_should_order_data(inode) &&
3207 test_opt(inode->i_sb, DELALLOC))
3208 inode->i_mapping->a_ops = &ext4_da_aops;
3209 else if (ext4_should_order_data(inode))
3210 inode->i_mapping->a_ops = &ext4_ordered_aops;
3211 else if (ext4_should_writeback_data(inode) &&
3212 test_opt(inode->i_sb, DELALLOC))
3213 inode->i_mapping->a_ops = &ext4_da_aops;
3214 else if (ext4_should_writeback_data(inode))
3215 inode->i_mapping->a_ops = &ext4_writeback_aops;
3217 inode->i_mapping->a_ops = &ext4_journalled_aops;
3221 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3222 * up to the end of the block which corresponds to `from'.
3223 * This required during truncate. We need to physically zero the tail end
3224 * of that block so it doesn't yield old data if the file is later grown.
3226 int ext4_block_truncate_page(handle_t *handle,
3227 struct address_space *mapping, loff_t from)
3229 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3230 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3231 unsigned blocksize, length, pos;
3233 struct inode *inode = mapping->host;
3234 struct buffer_head *bh;
3238 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3242 blocksize = inode->i_sb->s_blocksize;
3243 length = blocksize - (offset & (blocksize - 1));
3244 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3247 * For "nobh" option, we can only work if we don't need to
3248 * read-in the page - otherwise we create buffers to do the IO.
3250 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3251 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3252 zero_user(page, offset, length);
3253 set_page_dirty(page);
3257 if (!page_has_buffers(page))
3258 create_empty_buffers(page, blocksize, 0);
3260 /* Find the buffer that contains "offset" */
3261 bh = page_buffers(page);
3263 while (offset >= pos) {
3264 bh = bh->b_this_page;
3270 if (buffer_freed(bh)) {
3271 BUFFER_TRACE(bh, "freed: skip");
3275 if (!buffer_mapped(bh)) {
3276 BUFFER_TRACE(bh, "unmapped");
3277 ext4_get_block(inode, iblock, bh, 0);
3278 /* unmapped? It's a hole - nothing to do */
3279 if (!buffer_mapped(bh)) {
3280 BUFFER_TRACE(bh, "still unmapped");
3285 /* Ok, it's mapped. Make sure it's up-to-date */
3286 if (PageUptodate(page))
3287 set_buffer_uptodate(bh);
3289 if (!buffer_uptodate(bh)) {
3291 ll_rw_block(READ, 1, &bh);
3293 /* Uhhuh. Read error. Complain and punt. */
3294 if (!buffer_uptodate(bh))
3298 if (ext4_should_journal_data(inode)) {
3299 BUFFER_TRACE(bh, "get write access");
3300 err = ext4_journal_get_write_access(handle, bh);
3305 zero_user(page, offset, length);
3307 BUFFER_TRACE(bh, "zeroed end of block");
3310 if (ext4_should_journal_data(inode)) {
3311 err = ext4_journal_dirty_metadata(handle, bh);
3313 if (ext4_should_order_data(inode))
3314 err = ext4_jbd2_file_inode(handle, inode);
3315 mark_buffer_dirty(bh);
3320 page_cache_release(page);
3325 * Probably it should be a library function... search for first non-zero word
3326 * or memcmp with zero_page, whatever is better for particular architecture.
3329 static inline int all_zeroes(__le32 *p, __le32 *q)
3338 * ext4_find_shared - find the indirect blocks for partial truncation.
3339 * @inode: inode in question
3340 * @depth: depth of the affected branch
3341 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3342 * @chain: place to store the pointers to partial indirect blocks
3343 * @top: place to the (detached) top of branch
3345 * This is a helper function used by ext4_truncate().
3347 * When we do truncate() we may have to clean the ends of several
3348 * indirect blocks but leave the blocks themselves alive. Block is
3349 * partially truncated if some data below the new i_size is refered
3350 * from it (and it is on the path to the first completely truncated
3351 * data block, indeed). We have to free the top of that path along
3352 * with everything to the right of the path. Since no allocation
3353 * past the truncation point is possible until ext4_truncate()
3354 * finishes, we may safely do the latter, but top of branch may
3355 * require special attention - pageout below the truncation point
3356 * might try to populate it.
3358 * We atomically detach the top of branch from the tree, store the
3359 * block number of its root in *@top, pointers to buffer_heads of
3360 * partially truncated blocks - in @chain[].bh and pointers to
3361 * their last elements that should not be removed - in
3362 * @chain[].p. Return value is the pointer to last filled element
3365 * The work left to caller to do the actual freeing of subtrees:
3366 * a) free the subtree starting from *@top
3367 * b) free the subtrees whose roots are stored in
3368 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3369 * c) free the subtrees growing from the inode past the @chain[0].
3370 * (no partially truncated stuff there). */
3372 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3373 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3375 Indirect *partial, *p;
3379 /* Make k index the deepest non-null offest + 1 */
3380 for (k = depth; k > 1 && !offsets[k-1]; k--)
3382 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3383 /* Writer: pointers */
3385 partial = chain + k-1;
3387 * If the branch acquired continuation since we've looked at it -
3388 * fine, it should all survive and (new) top doesn't belong to us.
3390 if (!partial->key && *partial->p)
3393 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3396 * OK, we've found the last block that must survive. The rest of our
3397 * branch should be detached before unlocking. However, if that rest
3398 * of branch is all ours and does not grow immediately from the inode
3399 * it's easier to cheat and just decrement partial->p.
3401 if (p == chain + k - 1 && p > chain) {
3405 /* Nope, don't do this in ext4. Must leave the tree intact */
3412 while (partial > p) {
3413 brelse(partial->bh);
3421 * Zero a number of block pointers in either an inode or an indirect block.
3422 * If we restart the transaction we must again get write access to the
3423 * indirect block for further modification.
3425 * We release `count' blocks on disk, but (last - first) may be greater
3426 * than `count' because there can be holes in there.
3428 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3429 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3430 unsigned long count, __le32 *first, __le32 *last)
3433 if (try_to_extend_transaction(handle, inode)) {
3435 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3436 ext4_journal_dirty_metadata(handle, bh);
3438 ext4_mark_inode_dirty(handle, inode);
3439 ext4_journal_test_restart(handle, inode);
3441 BUFFER_TRACE(bh, "retaking write access");
3442 ext4_journal_get_write_access(handle, bh);
3447 * Any buffers which are on the journal will be in memory. We find
3448 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3449 * on them. We've already detached each block from the file, so
3450 * bforget() in jbd2_journal_forget() should be safe.
3452 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3454 for (p = first; p < last; p++) {
3455 u32 nr = le32_to_cpu(*p);
3457 struct buffer_head *tbh;
3460 tbh = sb_find_get_block(inode->i_sb, nr);
3461 ext4_forget(handle, 0, inode, tbh, nr);
3465 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3469 * ext4_free_data - free a list of data blocks
3470 * @handle: handle for this transaction
3471 * @inode: inode we are dealing with
3472 * @this_bh: indirect buffer_head which contains *@first and *@last
3473 * @first: array of block numbers
3474 * @last: points immediately past the end of array
3476 * We are freeing all blocks refered from that array (numbers are stored as
3477 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3479 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3480 * blocks are contiguous then releasing them at one time will only affect one
3481 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3482 * actually use a lot of journal space.
3484 * @this_bh will be %NULL if @first and @last point into the inode's direct
3487 static void ext4_free_data(handle_t *handle, struct inode *inode,
3488 struct buffer_head *this_bh,
3489 __le32 *first, __le32 *last)
3491 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3492 unsigned long count = 0; /* Number of blocks in the run */
3493 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3496 ext4_fsblk_t nr; /* Current block # */
3497 __le32 *p; /* Pointer into inode/ind
3498 for current block */
3501 if (this_bh) { /* For indirect block */
3502 BUFFER_TRACE(this_bh, "get_write_access");
3503 err = ext4_journal_get_write_access(handle, this_bh);
3504 /* Important: if we can't update the indirect pointers
3505 * to the blocks, we can't free them. */
3510 for (p = first; p < last; p++) {
3511 nr = le32_to_cpu(*p);
3513 /* accumulate blocks to free if they're contiguous */
3516 block_to_free_p = p;
3518 } else if (nr == block_to_free + count) {
3521 ext4_clear_blocks(handle, inode, this_bh,
3523 count, block_to_free_p, p);
3525 block_to_free_p = p;
3532 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3533 count, block_to_free_p, p);
3536 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3539 * The buffer head should have an attached journal head at this
3540 * point. However, if the data is corrupted and an indirect
3541 * block pointed to itself, it would have been detached when
3542 * the block was cleared. Check for this instead of OOPSing.
3545 ext4_journal_dirty_metadata(handle, this_bh);
3547 ext4_error(inode->i_sb, __func__,
3548 "circular indirect block detected, "
3549 "inode=%lu, block=%llu",
3551 (unsigned long long) this_bh->b_blocknr);
3556 * ext4_free_branches - free an array of branches
3557 * @handle: JBD handle for this transaction
3558 * @inode: inode we are dealing with
3559 * @parent_bh: the buffer_head which contains *@first and *@last
3560 * @first: array of block numbers
3561 * @last: pointer immediately past the end of array
3562 * @depth: depth of the branches to free
3564 * We are freeing all blocks refered from these branches (numbers are
3565 * stored as little-endian 32-bit) and updating @inode->i_blocks
3568 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3569 struct buffer_head *parent_bh,
3570 __le32 *first, __le32 *last, int depth)
3575 if (is_handle_aborted(handle))
3579 struct buffer_head *bh;
3580 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3582 while (--p >= first) {
3583 nr = le32_to_cpu(*p);
3585 continue; /* A hole */
3587 /* Go read the buffer for the next level down */
3588 bh = sb_bread(inode->i_sb, nr);
3591 * A read failure? Report error and clear slot
3595 ext4_error(inode->i_sb, "ext4_free_branches",
3596 "Read failure, inode=%lu, block=%llu",
3601 /* This zaps the entire block. Bottom up. */
3602 BUFFER_TRACE(bh, "free child branches");
3603 ext4_free_branches(handle, inode, bh,
3604 (__le32 *) bh->b_data,
3605 (__le32 *) bh->b_data + addr_per_block,
3609 * We've probably journalled the indirect block several
3610 * times during the truncate. But it's no longer
3611 * needed and we now drop it from the transaction via
3612 * jbd2_journal_revoke().
3614 * That's easy if it's exclusively part of this
3615 * transaction. But if it's part of the committing
3616 * transaction then jbd2_journal_forget() will simply
3617 * brelse() it. That means that if the underlying
3618 * block is reallocated in ext4_get_block(),
3619 * unmap_underlying_metadata() will find this block
3620 * and will try to get rid of it. damn, damn.
3622 * If this block has already been committed to the
3623 * journal, a revoke record will be written. And
3624 * revoke records must be emitted *before* clearing
3625 * this block's bit in the bitmaps.
3627 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3630 * Everything below this this pointer has been
3631 * released. Now let this top-of-subtree go.
3633 * We want the freeing of this indirect block to be
3634 * atomic in the journal with the updating of the
3635 * bitmap block which owns it. So make some room in
3638 * We zero the parent pointer *after* freeing its
3639 * pointee in the bitmaps, so if extend_transaction()
3640 * for some reason fails to put the bitmap changes and
3641 * the release into the same transaction, recovery
3642 * will merely complain about releasing a free block,
3643 * rather than leaking blocks.
3645 if (is_handle_aborted(handle))
3647 if (try_to_extend_transaction(handle, inode)) {
3648 ext4_mark_inode_dirty(handle, inode);
3649 ext4_journal_test_restart(handle, inode);
3652 ext4_free_blocks(handle, inode, nr, 1, 1);
3656 * The block which we have just freed is
3657 * pointed to by an indirect block: journal it
3659 BUFFER_TRACE(parent_bh, "get_write_access");
3660 if (!ext4_journal_get_write_access(handle,
3663 BUFFER_TRACE(parent_bh,
3664 "call ext4_journal_dirty_metadata");
3665 ext4_journal_dirty_metadata(handle,
3671 /* We have reached the bottom of the tree. */
3672 BUFFER_TRACE(parent_bh, "free data blocks");
3673 ext4_free_data(handle, inode, parent_bh, first, last);
3677 int ext4_can_truncate(struct inode *inode)
3679 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3681 if (S_ISREG(inode->i_mode))
3683 if (S_ISDIR(inode->i_mode))
3685 if (S_ISLNK(inode->i_mode))
3686 return !ext4_inode_is_fast_symlink(inode);
3693 * We block out ext4_get_block() block instantiations across the entire
3694 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3695 * simultaneously on behalf of the same inode.
3697 * As we work through the truncate and commmit bits of it to the journal there
3698 * is one core, guiding principle: the file's tree must always be consistent on
3699 * disk. We must be able to restart the truncate after a crash.
3701 * The file's tree may be transiently inconsistent in memory (although it
3702 * probably isn't), but whenever we close off and commit a journal transaction,
3703 * the contents of (the filesystem + the journal) must be consistent and
3704 * restartable. It's pretty simple, really: bottom up, right to left (although
3705 * left-to-right works OK too).
3707 * Note that at recovery time, journal replay occurs *before* the restart of
3708 * truncate against the orphan inode list.
3710 * The committed inode has the new, desired i_size (which is the same as
3711 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3712 * that this inode's truncate did not complete and it will again call
3713 * ext4_truncate() to have another go. So there will be instantiated blocks
3714 * to the right of the truncation point in a crashed ext4 filesystem. But
3715 * that's fine - as long as they are linked from the inode, the post-crash
3716 * ext4_truncate() run will find them and release them.
3718 void ext4_truncate(struct inode *inode)
3721 struct ext4_inode_info *ei = EXT4_I(inode);
3722 __le32 *i_data = ei->i_data;
3723 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3724 struct address_space *mapping = inode->i_mapping;
3725 ext4_lblk_t offsets[4];
3730 ext4_lblk_t last_block;
3731 unsigned blocksize = inode->i_sb->s_blocksize;
3733 if (!ext4_can_truncate(inode))
3736 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3737 ext4_ext_truncate(inode);
3741 handle = start_transaction(inode);
3743 return; /* AKPM: return what? */
3745 last_block = (inode->i_size + blocksize-1)
3746 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3748 if (inode->i_size & (blocksize - 1))
3749 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3752 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3754 goto out_stop; /* error */
3757 * OK. This truncate is going to happen. We add the inode to the
3758 * orphan list, so that if this truncate spans multiple transactions,
3759 * and we crash, we will resume the truncate when the filesystem
3760 * recovers. It also marks the inode dirty, to catch the new size.
3762 * Implication: the file must always be in a sane, consistent
3763 * truncatable state while each transaction commits.
3765 if (ext4_orphan_add(handle, inode))
3769 * From here we block out all ext4_get_block() callers who want to
3770 * modify the block allocation tree.
3772 down_write(&ei->i_data_sem);
3774 ext4_discard_preallocations(inode);
3777 * The orphan list entry will now protect us from any crash which
3778 * occurs before the truncate completes, so it is now safe to propagate
3779 * the new, shorter inode size (held for now in i_size) into the
3780 * on-disk inode. We do this via i_disksize, which is the value which
3781 * ext4 *really* writes onto the disk inode.
3783 ei->i_disksize = inode->i_size;
3785 if (n == 1) { /* direct blocks */
3786 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3787 i_data + EXT4_NDIR_BLOCKS);
3791 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3792 /* Kill the top of shared branch (not detached) */
3794 if (partial == chain) {
3795 /* Shared branch grows from the inode */
3796 ext4_free_branches(handle, inode, NULL,
3797 &nr, &nr+1, (chain+n-1) - partial);
3800 * We mark the inode dirty prior to restart,
3801 * and prior to stop. No need for it here.
3804 /* Shared branch grows from an indirect block */
3805 BUFFER_TRACE(partial->bh, "get_write_access");
3806 ext4_free_branches(handle, inode, partial->bh,
3808 partial->p+1, (chain+n-1) - partial);
3811 /* Clear the ends of indirect blocks on the shared branch */
3812 while (partial > chain) {
3813 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3814 (__le32*)partial->bh->b_data+addr_per_block,
3815 (chain+n-1) - partial);
3816 BUFFER_TRACE(partial->bh, "call brelse");
3817 brelse (partial->bh);
3821 /* Kill the remaining (whole) subtrees */
3822 switch (offsets[0]) {
3824 nr = i_data[EXT4_IND_BLOCK];
3826 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3827 i_data[EXT4_IND_BLOCK] = 0;
3829 case EXT4_IND_BLOCK:
3830 nr = i_data[EXT4_DIND_BLOCK];
3832 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3833 i_data[EXT4_DIND_BLOCK] = 0;
3835 case EXT4_DIND_BLOCK:
3836 nr = i_data[EXT4_TIND_BLOCK];
3838 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3839 i_data[EXT4_TIND_BLOCK] = 0;
3841 case EXT4_TIND_BLOCK:
3845 up_write(&ei->i_data_sem);
3846 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3847 ext4_mark_inode_dirty(handle, inode);
3850 * In a multi-transaction truncate, we only make the final transaction
3857 * If this was a simple ftruncate(), and the file will remain alive
3858 * then we need to clear up the orphan record which we created above.
3859 * However, if this was a real unlink then we were called by
3860 * ext4_delete_inode(), and we allow that function to clean up the
3861 * orphan info for us.
3864 ext4_orphan_del(handle, inode);
3866 ext4_journal_stop(handle);
3870 * ext4_get_inode_loc returns with an extra refcount against the inode's
3871 * underlying buffer_head on success. If 'in_mem' is true, we have all
3872 * data in memory that is needed to recreate the on-disk version of this
3875 static int __ext4_get_inode_loc(struct inode *inode,
3876 struct ext4_iloc *iloc, int in_mem)
3878 struct ext4_group_desc *gdp;
3879 struct buffer_head *bh;
3880 struct super_block *sb = inode->i_sb;
3882 int inodes_per_block, inode_offset;
3885 if (!ext4_valid_inum(sb, inode->i_ino))
3888 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3889 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3894 * Figure out the offset within the block group inode table
3896 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3897 inode_offset = ((inode->i_ino - 1) %
3898 EXT4_INODES_PER_GROUP(sb));
3899 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3900 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3902 bh = sb_getblk(sb, block);
3904 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3905 "inode block - inode=%lu, block=%llu",
3906 inode->i_ino, block);
3909 if (!buffer_uptodate(bh)) {
3913 * If the buffer has the write error flag, we have failed
3914 * to write out another inode in the same block. In this
3915 * case, we don't have to read the block because we may
3916 * read the old inode data successfully.
3918 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3919 set_buffer_uptodate(bh);
3921 if (buffer_uptodate(bh)) {
3922 /* someone brought it uptodate while we waited */
3928 * If we have all information of the inode in memory and this
3929 * is the only valid inode in the block, we need not read the
3933 struct buffer_head *bitmap_bh;
3936 start = inode_offset & ~(inodes_per_block - 1);
3938 /* Is the inode bitmap in cache? */
3939 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3944 * If the inode bitmap isn't in cache then the
3945 * optimisation may end up performing two reads instead
3946 * of one, so skip it.
3948 if (!buffer_uptodate(bitmap_bh)) {
3952 for (i = start; i < start + inodes_per_block; i++) {
3953 if (i == inode_offset)
3955 if (ext4_test_bit(i, bitmap_bh->b_data))
3959 if (i == start + inodes_per_block) {
3960 /* all other inodes are free, so skip I/O */
3961 memset(bh->b_data, 0, bh->b_size);
3962 set_buffer_uptodate(bh);
3970 * If we need to do any I/O, try to pre-readahead extra
3971 * blocks from the inode table.
3973 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3974 ext4_fsblk_t b, end, table;
3977 table = ext4_inode_table(sb, gdp);
3978 /* Make sure s_inode_readahead_blks is a power of 2 */
3979 while (EXT4_SB(sb)->s_inode_readahead_blks &
3980 (EXT4_SB(sb)->s_inode_readahead_blks-1))
3981 EXT4_SB(sb)->s_inode_readahead_blks =
3982 (EXT4_SB(sb)->s_inode_readahead_blks &
3983 (EXT4_SB(sb)->s_inode_readahead_blks-1));
3984 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3987 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3988 num = EXT4_INODES_PER_GROUP(sb);
3989 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3990 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3991 num -= le16_to_cpu(gdp->bg_itable_unused);
3992 table += num / inodes_per_block;
3996 sb_breadahead(sb, b++);
4000 * There are other valid inodes in the buffer, this inode
4001 * has in-inode xattrs, or we don't have this inode in memory.
4002 * Read the block from disk.
4005 bh->b_end_io = end_buffer_read_sync;
4006 submit_bh(READ_META, bh);
4008 if (!buffer_uptodate(bh)) {
4009 ext4_error(sb, __func__,
4010 "unable to read inode block - inode=%lu, "
4011 "block=%llu", inode->i_ino, block);
4021 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4023 /* We have all inode data except xattrs in memory here. */
4024 return __ext4_get_inode_loc(inode, iloc,
4025 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
4028 void ext4_set_inode_flags(struct inode *inode)
4030 unsigned int flags = EXT4_I(inode)->i_flags;
4032 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4033 if (flags & EXT4_SYNC_FL)
4034 inode->i_flags |= S_SYNC;
4035 if (flags & EXT4_APPEND_FL)
4036 inode->i_flags |= S_APPEND;
4037 if (flags & EXT4_IMMUTABLE_FL)
4038 inode->i_flags |= S_IMMUTABLE;
4039 if (flags & EXT4_NOATIME_FL)
4040 inode->i_flags |= S_NOATIME;
4041 if (flags & EXT4_DIRSYNC_FL)
4042 inode->i_flags |= S_DIRSYNC;
4045 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4046 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4048 unsigned int flags = ei->vfs_inode.i_flags;
4050 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4051 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4053 ei->i_flags |= EXT4_SYNC_FL;
4054 if (flags & S_APPEND)
4055 ei->i_flags |= EXT4_APPEND_FL;
4056 if (flags & S_IMMUTABLE)
4057 ei->i_flags |= EXT4_IMMUTABLE_FL;
4058 if (flags & S_NOATIME)
4059 ei->i_flags |= EXT4_NOATIME_FL;
4060 if (flags & S_DIRSYNC)
4061 ei->i_flags |= EXT4_DIRSYNC_FL;
4063 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4064 struct ext4_inode_info *ei)
4067 struct inode *inode = &(ei->vfs_inode);
4068 struct super_block *sb = inode->i_sb;
4070 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4071 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4072 /* we are using combined 48 bit field */
4073 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4074 le32_to_cpu(raw_inode->i_blocks_lo);
4075 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4076 /* i_blocks represent file system block size */
4077 return i_blocks << (inode->i_blkbits - 9);
4082 return le32_to_cpu(raw_inode->i_blocks_lo);
4086 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4088 struct ext4_iloc iloc;
4089 struct ext4_inode *raw_inode;
4090 struct ext4_inode_info *ei;
4091 struct buffer_head *bh;
4092 struct inode *inode;
4096 inode = iget_locked(sb, ino);
4098 return ERR_PTR(-ENOMEM);
4099 if (!(inode->i_state & I_NEW))
4103 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4104 ei->i_acl = EXT4_ACL_NOT_CACHED;
4105 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4108 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4112 raw_inode = ext4_raw_inode(&iloc);
4113 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4114 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4115 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4116 if (!(test_opt(inode->i_sb, NO_UID32))) {
4117 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4118 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4120 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4123 ei->i_dir_start_lookup = 0;
4124 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4125 /* We now have enough fields to check if the inode was active or not.
4126 * This is needed because nfsd might try to access dead inodes
4127 * the test is that same one that e2fsck uses
4128 * NeilBrown 1999oct15
4130 if (inode->i_nlink == 0) {
4131 if (inode->i_mode == 0 ||
4132 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4133 /* this inode is deleted */
4138 /* The only unlinked inodes we let through here have
4139 * valid i_mode and are being read by the orphan
4140 * recovery code: that's fine, we're about to complete
4141 * the process of deleting those. */
4143 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4144 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4145 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4146 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4147 cpu_to_le32(EXT4_OS_HURD)) {
4149 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4151 inode->i_size = ext4_isize(raw_inode);
4152 ei->i_disksize = inode->i_size;
4153 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4154 ei->i_block_group = iloc.block_group;
4156 * NOTE! The in-memory inode i_data array is in little-endian order
4157 * even on big-endian machines: we do NOT byteswap the block numbers!
4159 for (block = 0; block < EXT4_N_BLOCKS; block++)
4160 ei->i_data[block] = raw_inode->i_block[block];
4161 INIT_LIST_HEAD(&ei->i_orphan);
4163 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4164 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4165 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4166 EXT4_INODE_SIZE(inode->i_sb)) {
4171 if (ei->i_extra_isize == 0) {
4172 /* The extra space is currently unused. Use it. */
4173 ei->i_extra_isize = sizeof(struct ext4_inode) -
4174 EXT4_GOOD_OLD_INODE_SIZE;
4176 __le32 *magic = (void *)raw_inode +
4177 EXT4_GOOD_OLD_INODE_SIZE +
4179 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4180 ei->i_state |= EXT4_STATE_XATTR;
4183 ei->i_extra_isize = 0;
4185 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4186 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4187 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4188 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4190 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4191 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4192 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4194 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4197 if (S_ISREG(inode->i_mode)) {
4198 inode->i_op = &ext4_file_inode_operations;
4199 inode->i_fop = &ext4_file_operations;
4200 ext4_set_aops(inode);
4201 } else if (S_ISDIR(inode->i_mode)) {
4202 inode->i_op = &ext4_dir_inode_operations;
4203 inode->i_fop = &ext4_dir_operations;
4204 } else if (S_ISLNK(inode->i_mode)) {
4205 if (ext4_inode_is_fast_symlink(inode))
4206 inode->i_op = &ext4_fast_symlink_inode_operations;
4208 inode->i_op = &ext4_symlink_inode_operations;
4209 ext4_set_aops(inode);
4212 inode->i_op = &ext4_special_inode_operations;
4213 if (raw_inode->i_block[0])
4214 init_special_inode(inode, inode->i_mode,
4215 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4217 init_special_inode(inode, inode->i_mode,
4218 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4221 ext4_set_inode_flags(inode);
4222 unlock_new_inode(inode);
4227 return ERR_PTR(ret);
4230 static int ext4_inode_blocks_set(handle_t *handle,
4231 struct ext4_inode *raw_inode,
4232 struct ext4_inode_info *ei)
4234 struct inode *inode = &(ei->vfs_inode);
4235 u64 i_blocks = inode->i_blocks;
4236 struct super_block *sb = inode->i_sb;
4238 if (i_blocks <= ~0U) {
4240 * i_blocks can be represnted in a 32 bit variable
4241 * as multiple of 512 bytes
4243 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4244 raw_inode->i_blocks_high = 0;
4245 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4248 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4251 if (i_blocks <= 0xffffffffffffULL) {
4253 * i_blocks can be represented in a 48 bit variable
4254 * as multiple of 512 bytes
4256 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4257 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4258 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4260 ei->i_flags |= EXT4_HUGE_FILE_FL;
4261 /* i_block is stored in file system block size */
4262 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4263 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4264 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4270 * Post the struct inode info into an on-disk inode location in the
4271 * buffer-cache. This gobbles the caller's reference to the
4272 * buffer_head in the inode location struct.
4274 * The caller must have write access to iloc->bh.
4276 static int ext4_do_update_inode(handle_t *handle,
4277 struct inode *inode,
4278 struct ext4_iloc *iloc)
4280 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4281 struct ext4_inode_info *ei = EXT4_I(inode);
4282 struct buffer_head *bh = iloc->bh;
4283 int err = 0, rc, block;
4285 /* For fields not not tracking in the in-memory inode,
4286 * initialise them to zero for new inodes. */
4287 if (ei->i_state & EXT4_STATE_NEW)
4288 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4290 ext4_get_inode_flags(ei);
4291 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4292 if (!(test_opt(inode->i_sb, NO_UID32))) {
4293 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4294 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4296 * Fix up interoperability with old kernels. Otherwise, old inodes get
4297 * re-used with the upper 16 bits of the uid/gid intact
4300 raw_inode->i_uid_high =
4301 cpu_to_le16(high_16_bits(inode->i_uid));
4302 raw_inode->i_gid_high =
4303 cpu_to_le16(high_16_bits(inode->i_gid));
4305 raw_inode->i_uid_high = 0;
4306 raw_inode->i_gid_high = 0;
4309 raw_inode->i_uid_low =
4310 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4311 raw_inode->i_gid_low =
4312 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4313 raw_inode->i_uid_high = 0;
4314 raw_inode->i_gid_high = 0;
4316 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4318 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4319 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4320 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4321 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4323 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4325 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4326 /* clear the migrate flag in the raw_inode */
4327 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4328 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4329 cpu_to_le32(EXT4_OS_HURD))
4330 raw_inode->i_file_acl_high =
4331 cpu_to_le16(ei->i_file_acl >> 32);
4332 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4333 ext4_isize_set(raw_inode, ei->i_disksize);
4334 if (ei->i_disksize > 0x7fffffffULL) {
4335 struct super_block *sb = inode->i_sb;
4336 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4337 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4338 EXT4_SB(sb)->s_es->s_rev_level ==
4339 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4340 /* If this is the first large file
4341 * created, add a flag to the superblock.
4343 err = ext4_journal_get_write_access(handle,
4344 EXT4_SB(sb)->s_sbh);
4347 ext4_update_dynamic_rev(sb);
4348 EXT4_SET_RO_COMPAT_FEATURE(sb,
4349 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4352 err = ext4_journal_dirty_metadata(handle,
4353 EXT4_SB(sb)->s_sbh);
4356 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4357 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4358 if (old_valid_dev(inode->i_rdev)) {
4359 raw_inode->i_block[0] =
4360 cpu_to_le32(old_encode_dev(inode->i_rdev));
4361 raw_inode->i_block[1] = 0;
4363 raw_inode->i_block[0] = 0;
4364 raw_inode->i_block[1] =
4365 cpu_to_le32(new_encode_dev(inode->i_rdev));
4366 raw_inode->i_block[2] = 0;
4368 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4369 raw_inode->i_block[block] = ei->i_data[block];
4371 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4372 if (ei->i_extra_isize) {
4373 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4374 raw_inode->i_version_hi =
4375 cpu_to_le32(inode->i_version >> 32);
4376 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4380 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4381 rc = ext4_journal_dirty_metadata(handle, bh);
4384 ei->i_state &= ~EXT4_STATE_NEW;
4388 ext4_std_error(inode->i_sb, err);
4393 * ext4_write_inode()
4395 * We are called from a few places:
4397 * - Within generic_file_write() for O_SYNC files.
4398 * Here, there will be no transaction running. We wait for any running
4399 * trasnaction to commit.
4401 * - Within sys_sync(), kupdate and such.
4402 * We wait on commit, if tol to.
4404 * - Within prune_icache() (PF_MEMALLOC == true)
4405 * Here we simply return. We can't afford to block kswapd on the
4408 * In all cases it is actually safe for us to return without doing anything,
4409 * because the inode has been copied into a raw inode buffer in
4410 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4413 * Note that we are absolutely dependent upon all inode dirtiers doing the
4414 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4415 * which we are interested.
4417 * It would be a bug for them to not do this. The code:
4419 * mark_inode_dirty(inode)
4421 * inode->i_size = expr;
4423 * is in error because a kswapd-driven write_inode() could occur while
4424 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4425 * will no longer be on the superblock's dirty inode list.
4427 int ext4_write_inode(struct inode *inode, int wait)
4429 if (current->flags & PF_MEMALLOC)
4432 if (ext4_journal_current_handle()) {
4433 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4441 return ext4_force_commit(inode->i_sb);
4447 * Called from notify_change.
4449 * We want to trap VFS attempts to truncate the file as soon as
4450 * possible. In particular, we want to make sure that when the VFS
4451 * shrinks i_size, we put the inode on the orphan list and modify
4452 * i_disksize immediately, so that during the subsequent flushing of
4453 * dirty pages and freeing of disk blocks, we can guarantee that any
4454 * commit will leave the blocks being flushed in an unused state on
4455 * disk. (On recovery, the inode will get truncated and the blocks will
4456 * be freed, so we have a strong guarantee that no future commit will
4457 * leave these blocks visible to the user.)
4459 * Another thing we have to assure is that if we are in ordered mode
4460 * and inode is still attached to the committing transaction, we must
4461 * we start writeout of all the dirty pages which are being truncated.
4462 * This way we are sure that all the data written in the previous
4463 * transaction are already on disk (truncate waits for pages under
4466 * Called with inode->i_mutex down.
4468 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4470 struct inode *inode = dentry->d_inode;
4472 const unsigned int ia_valid = attr->ia_valid;
4474 error = inode_change_ok(inode, attr);
4478 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4479 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4482 /* (user+group)*(old+new) structure, inode write (sb,
4483 * inode block, ? - but truncate inode update has it) */
4484 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4485 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4486 if (IS_ERR(handle)) {
4487 error = PTR_ERR(handle);
4490 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4492 ext4_journal_stop(handle);
4495 /* Update corresponding info in inode so that everything is in
4496 * one transaction */
4497 if (attr->ia_valid & ATTR_UID)
4498 inode->i_uid = attr->ia_uid;
4499 if (attr->ia_valid & ATTR_GID)
4500 inode->i_gid = attr->ia_gid;
4501 error = ext4_mark_inode_dirty(handle, inode);
4502 ext4_journal_stop(handle);
4505 if (attr->ia_valid & ATTR_SIZE) {
4506 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4507 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4509 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4516 if (S_ISREG(inode->i_mode) &&
4517 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4520 handle = ext4_journal_start(inode, 3);
4521 if (IS_ERR(handle)) {
4522 error = PTR_ERR(handle);
4526 error = ext4_orphan_add(handle, inode);
4527 EXT4_I(inode)->i_disksize = attr->ia_size;
4528 rc = ext4_mark_inode_dirty(handle, inode);
4531 ext4_journal_stop(handle);
4533 if (ext4_should_order_data(inode)) {
4534 error = ext4_begin_ordered_truncate(inode,
4537 /* Do as much error cleanup as possible */
4538 handle = ext4_journal_start(inode, 3);
4539 if (IS_ERR(handle)) {
4540 ext4_orphan_del(NULL, inode);
4543 ext4_orphan_del(handle, inode);
4544 ext4_journal_stop(handle);
4550 rc = inode_setattr(inode, attr);
4552 /* If inode_setattr's call to ext4_truncate failed to get a
4553 * transaction handle at all, we need to clean up the in-core
4554 * orphan list manually. */
4556 ext4_orphan_del(NULL, inode);
4558 if (!rc && (ia_valid & ATTR_MODE))
4559 rc = ext4_acl_chmod(inode);
4562 ext4_std_error(inode->i_sb, error);
4568 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4571 struct inode *inode;
4572 unsigned long delalloc_blocks;
4574 inode = dentry->d_inode;
4575 generic_fillattr(inode, stat);
4578 * We can't update i_blocks if the block allocation is delayed
4579 * otherwise in the case of system crash before the real block
4580 * allocation is done, we will have i_blocks inconsistent with
4581 * on-disk file blocks.
4582 * We always keep i_blocks updated together with real
4583 * allocation. But to not confuse with user, stat
4584 * will return the blocks that include the delayed allocation
4585 * blocks for this file.
4587 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4588 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4589 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4591 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4595 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4600 /* if nrblocks are contiguous */
4603 * With N contiguous data blocks, it need at most
4604 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4605 * 2 dindirect blocks
4608 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4609 return indirects + 3;
4612 * if nrblocks are not contiguous, worse case, each block touch
4613 * a indirect block, and each indirect block touch a double indirect
4614 * block, plus a triple indirect block
4616 indirects = nrblocks * 2 + 1;
4620 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4622 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4623 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4624 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4628 * Account for index blocks, block groups bitmaps and block group
4629 * descriptor blocks if modify datablocks and index blocks
4630 * worse case, the indexs blocks spread over different block groups
4632 * If datablocks are discontiguous, they are possible to spread over
4633 * different block groups too. If they are contiugous, with flexbg,
4634 * they could still across block group boundary.
4636 * Also account for superblock, inode, quota and xattr blocks
4638 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4640 int groups, gdpblocks;
4645 * How many index blocks need to touch to modify nrblocks?
4646 * The "Chunk" flag indicating whether the nrblocks is
4647 * physically contiguous on disk
4649 * For Direct IO and fallocate, they calls get_block to allocate
4650 * one single extent at a time, so they could set the "Chunk" flag
4652 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4657 * Now let's see how many group bitmaps and group descriptors need
4667 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4668 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4669 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4670 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4672 /* bitmaps and block group descriptor blocks */
4673 ret += groups + gdpblocks;
4675 /* Blocks for super block, inode, quota and xattr blocks */
4676 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4682 * Calulate the total number of credits to reserve to fit
4683 * the modification of a single pages into a single transaction,
4684 * which may include multiple chunks of block allocations.
4686 * This could be called via ext4_write_begin()
4688 * We need to consider the worse case, when
4689 * one new block per extent.
4691 int ext4_writepage_trans_blocks(struct inode *inode)
4693 int bpp = ext4_journal_blocks_per_page(inode);
4696 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4698 /* Account for data blocks for journalled mode */
4699 if (ext4_should_journal_data(inode))
4705 * Calculate the journal credits for a chunk of data modification.
4707 * This is called from DIO, fallocate or whoever calling
4708 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4710 * journal buffers for data blocks are not included here, as DIO
4711 * and fallocate do no need to journal data buffers.
4713 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4715 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4719 * The caller must have previously called ext4_reserve_inode_write().
4720 * Give this, we know that the caller already has write access to iloc->bh.
4722 int ext4_mark_iloc_dirty(handle_t *handle,
4723 struct inode *inode, struct ext4_iloc *iloc)
4727 if (test_opt(inode->i_sb, I_VERSION))
4728 inode_inc_iversion(inode);
4730 /* the do_update_inode consumes one bh->b_count */
4733 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4734 err = ext4_do_update_inode(handle, inode, iloc);
4740 * On success, We end up with an outstanding reference count against
4741 * iloc->bh. This _must_ be cleaned up later.
4745 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4746 struct ext4_iloc *iloc)
4750 err = ext4_get_inode_loc(inode, iloc);
4752 BUFFER_TRACE(iloc->bh, "get_write_access");
4753 err = ext4_journal_get_write_access(handle, iloc->bh);
4760 ext4_std_error(inode->i_sb, err);
4765 * Expand an inode by new_extra_isize bytes.
4766 * Returns 0 on success or negative error number on failure.
4768 static int ext4_expand_extra_isize(struct inode *inode,
4769 unsigned int new_extra_isize,
4770 struct ext4_iloc iloc,
4773 struct ext4_inode *raw_inode;
4774 struct ext4_xattr_ibody_header *header;
4775 struct ext4_xattr_entry *entry;
4777 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4780 raw_inode = ext4_raw_inode(&iloc);
4782 header = IHDR(inode, raw_inode);
4783 entry = IFIRST(header);
4785 /* No extended attributes present */
4786 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4787 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4788 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4790 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4794 /* try to expand with EAs present */
4795 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4800 * What we do here is to mark the in-core inode as clean with respect to inode
4801 * dirtiness (it may still be data-dirty).
4802 * This means that the in-core inode may be reaped by prune_icache
4803 * without having to perform any I/O. This is a very good thing,
4804 * because *any* task may call prune_icache - even ones which
4805 * have a transaction open against a different journal.
4807 * Is this cheating? Not really. Sure, we haven't written the
4808 * inode out, but prune_icache isn't a user-visible syncing function.
4809 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4810 * we start and wait on commits.
4812 * Is this efficient/effective? Well, we're being nice to the system
4813 * by cleaning up our inodes proactively so they can be reaped
4814 * without I/O. But we are potentially leaving up to five seconds'
4815 * worth of inodes floating about which prune_icache wants us to
4816 * write out. One way to fix that would be to get prune_icache()
4817 * to do a write_super() to free up some memory. It has the desired
4820 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4822 struct ext4_iloc iloc;
4823 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4824 static unsigned int mnt_count;
4828 err = ext4_reserve_inode_write(handle, inode, &iloc);
4829 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4830 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4832 * We need extra buffer credits since we may write into EA block
4833 * with this same handle. If journal_extend fails, then it will
4834 * only result in a minor loss of functionality for that inode.
4835 * If this is felt to be critical, then e2fsck should be run to
4836 * force a large enough s_min_extra_isize.
4838 if ((jbd2_journal_extend(handle,
4839 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4840 ret = ext4_expand_extra_isize(inode,
4841 sbi->s_want_extra_isize,
4844 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4846 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4847 ext4_warning(inode->i_sb, __func__,
4848 "Unable to expand inode %lu. Delete"
4849 " some EAs or run e2fsck.",
4852 le16_to_cpu(sbi->s_es->s_mnt_count);
4858 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4863 * ext4_dirty_inode() is called from __mark_inode_dirty()
4865 * We're really interested in the case where a file is being extended.
4866 * i_size has been changed by generic_commit_write() and we thus need
4867 * to include the updated inode in the current transaction.
4869 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4870 * are allocated to the file.
4872 * If the inode is marked synchronous, we don't honour that here - doing
4873 * so would cause a commit on atime updates, which we don't bother doing.
4874 * We handle synchronous inodes at the highest possible level.
4876 void ext4_dirty_inode(struct inode *inode)
4878 handle_t *current_handle = ext4_journal_current_handle();
4881 handle = ext4_journal_start(inode, 2);
4884 if (current_handle &&
4885 current_handle->h_transaction != handle->h_transaction) {
4886 /* This task has a transaction open against a different fs */
4887 printk(KERN_EMERG "%s: transactions do not match!\n",
4890 jbd_debug(5, "marking dirty. outer handle=%p\n",
4892 ext4_mark_inode_dirty(handle, inode);
4894 ext4_journal_stop(handle);
4901 * Bind an inode's backing buffer_head into this transaction, to prevent
4902 * it from being flushed to disk early. Unlike
4903 * ext4_reserve_inode_write, this leaves behind no bh reference and
4904 * returns no iloc structure, so the caller needs to repeat the iloc
4905 * lookup to mark the inode dirty later.
4907 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4909 struct ext4_iloc iloc;
4913 err = ext4_get_inode_loc(inode, &iloc);
4915 BUFFER_TRACE(iloc.bh, "get_write_access");
4916 err = jbd2_journal_get_write_access(handle, iloc.bh);
4918 err = ext4_journal_dirty_metadata(handle,
4923 ext4_std_error(inode->i_sb, err);
4928 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4935 * We have to be very careful here: changing a data block's
4936 * journaling status dynamically is dangerous. If we write a
4937 * data block to the journal, change the status and then delete
4938 * that block, we risk forgetting to revoke the old log record
4939 * from the journal and so a subsequent replay can corrupt data.
4940 * So, first we make sure that the journal is empty and that
4941 * nobody is changing anything.
4944 journal = EXT4_JOURNAL(inode);
4945 if (is_journal_aborted(journal))
4948 jbd2_journal_lock_updates(journal);
4949 jbd2_journal_flush(journal);
4952 * OK, there are no updates running now, and all cached data is
4953 * synced to disk. We are now in a completely consistent state
4954 * which doesn't have anything in the journal, and we know that
4955 * no filesystem updates are running, so it is safe to modify
4956 * the inode's in-core data-journaling state flag now.
4960 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4962 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4963 ext4_set_aops(inode);
4965 jbd2_journal_unlock_updates(journal);
4967 /* Finally we can mark the inode as dirty. */
4969 handle = ext4_journal_start(inode, 1);
4971 return PTR_ERR(handle);
4973 err = ext4_mark_inode_dirty(handle, inode);
4975 ext4_journal_stop(handle);
4976 ext4_std_error(inode->i_sb, err);
4981 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4983 return !buffer_mapped(bh);
4986 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4992 struct file *file = vma->vm_file;
4993 struct inode *inode = file->f_path.dentry->d_inode;
4994 struct address_space *mapping = inode->i_mapping;
4997 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4998 * get i_mutex because we are already holding mmap_sem.
5000 down_read(&inode->i_alloc_sem);
5001 size = i_size_read(inode);
5002 if (page->mapping != mapping || size <= page_offset(page)
5003 || !PageUptodate(page)) {
5004 /* page got truncated from under us? */
5008 if (PageMappedToDisk(page))
5011 if (page->index == size >> PAGE_CACHE_SHIFT)
5012 len = size & ~PAGE_CACHE_MASK;
5014 len = PAGE_CACHE_SIZE;
5016 if (page_has_buffers(page)) {
5017 /* return if we have all the buffers mapped */
5018 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5023 * OK, we need to fill the hole... Do write_begin write_end
5024 * to do block allocation/reservation.We are not holding
5025 * inode.i__mutex here. That allow * parallel write_begin,
5026 * write_end call. lock_page prevent this from happening
5027 * on the same page though
5029 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5030 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5033 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5034 len, len, page, fsdata);
5039 up_read(&inode->i_alloc_sem);