2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include <linux/workqueue.h>
41 #include <linux/kernel.h>
42 #include <linux/slab.h>
44 #include "ext4_jbd2.h"
47 #include "ext4_extents.h"
49 #include <trace/events/ext4.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static inline int ext4_begin_ordered_truncate(struct inode *inode,
56 return jbd2_journal_begin_ordered_truncate(
57 EXT4_SB(inode->i_sb)->s_journal,
58 &EXT4_I(inode)->jinode,
62 static void ext4_invalidatepage(struct page *page, unsigned long offset);
63 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
64 struct buffer_head *bh_result, int create);
65 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
66 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
67 static int __ext4_journalled_writepage(struct page *page, unsigned int len);
68 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
71 * Test whether an inode is a fast symlink.
73 static int ext4_inode_is_fast_symlink(struct inode *inode)
75 int ea_blocks = EXT4_I(inode)->i_file_acl ?
76 (inode->i_sb->s_blocksize >> 9) : 0;
78 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
82 * Work out how many blocks we need to proceed with the next chunk of a
83 * truncate transaction.
85 static unsigned long blocks_for_truncate(struct inode *inode)
89 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
91 /* Give ourselves just enough room to cope with inodes in which
92 * i_blocks is corrupt: we've seen disk corruptions in the past
93 * which resulted in random data in an inode which looked enough
94 * like a regular file for ext4 to try to delete it. Things
95 * will go a bit crazy if that happens, but at least we should
96 * try not to panic the whole kernel. */
100 /* But we need to bound the transaction so we don't overflow the
102 if (needed > EXT4_MAX_TRANS_DATA)
103 needed = EXT4_MAX_TRANS_DATA;
105 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
109 * Truncate transactions can be complex and absolutely huge. So we need to
110 * be able to restart the transaction at a conventient checkpoint to make
111 * sure we don't overflow the journal.
113 * start_transaction gets us a new handle for a truncate transaction,
114 * and extend_transaction tries to extend the existing one a bit. If
115 * extend fails, we need to propagate the failure up and restart the
116 * transaction in the top-level truncate loop. --sct
118 static handle_t *start_transaction(struct inode *inode)
122 result = ext4_journal_start(inode, blocks_for_truncate(inode));
126 ext4_std_error(inode->i_sb, PTR_ERR(result));
131 * Try to extend this transaction for the purposes of truncation.
133 * Returns 0 if we managed to create more room. If we can't create more
134 * room, and the transaction must be restarted we return 1.
136 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
138 if (!ext4_handle_valid(handle))
140 if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
142 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
148 * Restart the transaction associated with *handle. This does a commit,
149 * so before we call here everything must be consistently dirtied against
152 int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
158 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
159 * moment, get_block can be called only for blocks inside i_size since
160 * page cache has been already dropped and writes are blocked by
161 * i_mutex. So we can safely drop the i_data_sem here.
163 BUG_ON(EXT4_JOURNAL(inode) == NULL);
164 jbd_debug(2, "restarting handle %p\n", handle);
165 up_write(&EXT4_I(inode)->i_data_sem);
166 ret = ext4_journal_restart(handle, blocks_for_truncate(inode));
167 down_write(&EXT4_I(inode)->i_data_sem);
168 ext4_discard_preallocations(inode);
174 * Called at the last iput() if i_nlink is zero.
176 void ext4_evict_inode(struct inode *inode)
181 if (inode->i_nlink) {
182 truncate_inode_pages(&inode->i_data, 0);
186 if (!is_bad_inode(inode))
187 dquot_initialize(inode);
189 if (ext4_should_order_data(inode))
190 ext4_begin_ordered_truncate(inode, 0);
191 truncate_inode_pages(&inode->i_data, 0);
193 if (is_bad_inode(inode))
196 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
197 if (IS_ERR(handle)) {
198 ext4_std_error(inode->i_sb, PTR_ERR(handle));
200 * If we're going to skip the normal cleanup, we still need to
201 * make sure that the in-core orphan linked list is properly
204 ext4_orphan_del(NULL, inode);
209 ext4_handle_sync(handle);
211 err = ext4_mark_inode_dirty(handle, inode);
213 ext4_warning(inode->i_sb,
214 "couldn't mark inode dirty (err %d)", err);
218 ext4_truncate(inode);
221 * ext4_ext_truncate() doesn't reserve any slop when it
222 * restarts journal transactions; therefore there may not be
223 * enough credits left in the handle to remove the inode from
224 * the orphan list and set the dtime field.
226 if (!ext4_handle_has_enough_credits(handle, 3)) {
227 err = ext4_journal_extend(handle, 3);
229 err = ext4_journal_restart(handle, 3);
231 ext4_warning(inode->i_sb,
232 "couldn't extend journal (err %d)", err);
234 ext4_journal_stop(handle);
235 ext4_orphan_del(NULL, inode);
241 * Kill off the orphan record which ext4_truncate created.
242 * AKPM: I think this can be inside the above `if'.
243 * Note that ext4_orphan_del() has to be able to cope with the
244 * deletion of a non-existent orphan - this is because we don't
245 * know if ext4_truncate() actually created an orphan record.
246 * (Well, we could do this if we need to, but heck - it works)
248 ext4_orphan_del(handle, inode);
249 EXT4_I(inode)->i_dtime = get_seconds();
252 * One subtle ordering requirement: if anything has gone wrong
253 * (transaction abort, IO errors, whatever), then we can still
254 * do these next steps (the fs will already have been marked as
255 * having errors), but we can't free the inode if the mark_dirty
258 if (ext4_mark_inode_dirty(handle, inode))
259 /* If that failed, just do the required in-core inode clear. */
260 ext4_clear_inode(inode);
262 ext4_free_inode(handle, inode);
263 ext4_journal_stop(handle);
266 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
272 struct buffer_head *bh;
275 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
277 p->key = *(p->p = v);
282 * ext4_block_to_path - parse the block number into array of offsets
283 * @inode: inode in question (we are only interested in its superblock)
284 * @i_block: block number to be parsed
285 * @offsets: array to store the offsets in
286 * @boundary: set this non-zero if the referred-to block is likely to be
287 * followed (on disk) by an indirect block.
289 * To store the locations of file's data ext4 uses a data structure common
290 * for UNIX filesystems - tree of pointers anchored in the inode, with
291 * data blocks at leaves and indirect blocks in intermediate nodes.
292 * This function translates the block number into path in that tree -
293 * return value is the path length and @offsets[n] is the offset of
294 * pointer to (n+1)th node in the nth one. If @block is out of range
295 * (negative or too large) warning is printed and zero returned.
297 * Note: function doesn't find node addresses, so no IO is needed. All
298 * we need to know is the capacity of indirect blocks (taken from the
303 * Portability note: the last comparison (check that we fit into triple
304 * indirect block) is spelled differently, because otherwise on an
305 * architecture with 32-bit longs and 8Kb pages we might get into trouble
306 * if our filesystem had 8Kb blocks. We might use long long, but that would
307 * kill us on x86. Oh, well, at least the sign propagation does not matter -
308 * i_block would have to be negative in the very beginning, so we would not
312 static int ext4_block_to_path(struct inode *inode,
314 ext4_lblk_t offsets[4], int *boundary)
316 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
317 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
318 const long direct_blocks = EXT4_NDIR_BLOCKS,
319 indirect_blocks = ptrs,
320 double_blocks = (1 << (ptrs_bits * 2));
324 if (i_block < direct_blocks) {
325 offsets[n++] = i_block;
326 final = direct_blocks;
327 } else if ((i_block -= direct_blocks) < indirect_blocks) {
328 offsets[n++] = EXT4_IND_BLOCK;
329 offsets[n++] = i_block;
331 } else if ((i_block -= indirect_blocks) < double_blocks) {
332 offsets[n++] = EXT4_DIND_BLOCK;
333 offsets[n++] = i_block >> ptrs_bits;
334 offsets[n++] = i_block & (ptrs - 1);
336 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
337 offsets[n++] = EXT4_TIND_BLOCK;
338 offsets[n++] = i_block >> (ptrs_bits * 2);
339 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
340 offsets[n++] = i_block & (ptrs - 1);
343 ext4_warning(inode->i_sb, "block %lu > max in inode %lu",
344 i_block + direct_blocks +
345 indirect_blocks + double_blocks, inode->i_ino);
348 *boundary = final - 1 - (i_block & (ptrs - 1));
352 static int __ext4_check_blockref(const char *function, unsigned int line,
354 __le32 *p, unsigned int max)
356 struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
360 while (bref < p+max) {
361 blk = le32_to_cpu(*bref++);
363 unlikely(!ext4_data_block_valid(EXT4_SB(inode->i_sb),
365 es->s_last_error_block = cpu_to_le64(blk);
366 ext4_error_inode(inode, function, line, blk,
375 #define ext4_check_indirect_blockref(inode, bh) \
376 __ext4_check_blockref(__func__, __LINE__, inode, \
377 (__le32 *)(bh)->b_data, \
378 EXT4_ADDR_PER_BLOCK((inode)->i_sb))
380 #define ext4_check_inode_blockref(inode) \
381 __ext4_check_blockref(__func__, __LINE__, inode, \
382 EXT4_I(inode)->i_data, \
386 * ext4_get_branch - read the chain of indirect blocks leading to data
387 * @inode: inode in question
388 * @depth: depth of the chain (1 - direct pointer, etc.)
389 * @offsets: offsets of pointers in inode/indirect blocks
390 * @chain: place to store the result
391 * @err: here we store the error value
393 * Function fills the array of triples <key, p, bh> and returns %NULL
394 * if everything went OK or the pointer to the last filled triple
395 * (incomplete one) otherwise. Upon the return chain[i].key contains
396 * the number of (i+1)-th block in the chain (as it is stored in memory,
397 * i.e. little-endian 32-bit), chain[i].p contains the address of that
398 * number (it points into struct inode for i==0 and into the bh->b_data
399 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
400 * block for i>0 and NULL for i==0. In other words, it holds the block
401 * numbers of the chain, addresses they were taken from (and where we can
402 * verify that chain did not change) and buffer_heads hosting these
405 * Function stops when it stumbles upon zero pointer (absent block)
406 * (pointer to last triple returned, *@err == 0)
407 * or when it gets an IO error reading an indirect block
408 * (ditto, *@err == -EIO)
409 * or when it reads all @depth-1 indirect blocks successfully and finds
410 * the whole chain, all way to the data (returns %NULL, *err == 0).
412 * Need to be called with
413 * down_read(&EXT4_I(inode)->i_data_sem)
415 static Indirect *ext4_get_branch(struct inode *inode, int depth,
416 ext4_lblk_t *offsets,
417 Indirect chain[4], int *err)
419 struct super_block *sb = inode->i_sb;
421 struct buffer_head *bh;
424 /* i_data is not going away, no lock needed */
425 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
429 bh = sb_getblk(sb, le32_to_cpu(p->key));
433 if (!bh_uptodate_or_lock(bh)) {
434 if (bh_submit_read(bh) < 0) {
438 /* validate block references */
439 if (ext4_check_indirect_blockref(inode, bh)) {
445 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
459 * ext4_find_near - find a place for allocation with sufficient locality
461 * @ind: descriptor of indirect block.
463 * This function returns the preferred place for block allocation.
464 * It is used when heuristic for sequential allocation fails.
466 * + if there is a block to the left of our position - allocate near it.
467 * + if pointer will live in indirect block - allocate near that block.
468 * + if pointer will live in inode - allocate in the same
471 * In the latter case we colour the starting block by the callers PID to
472 * prevent it from clashing with concurrent allocations for a different inode
473 * in the same block group. The PID is used here so that functionally related
474 * files will be close-by on-disk.
476 * Caller must make sure that @ind is valid and will stay that way.
478 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
480 struct ext4_inode_info *ei = EXT4_I(inode);
481 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
483 ext4_fsblk_t bg_start;
484 ext4_fsblk_t last_block;
485 ext4_grpblk_t colour;
486 ext4_group_t block_group;
487 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
489 /* Try to find previous block */
490 for (p = ind->p - 1; p >= start; p--) {
492 return le32_to_cpu(*p);
495 /* No such thing, so let's try location of indirect block */
497 return ind->bh->b_blocknr;
500 * It is going to be referred to from the inode itself? OK, just put it
501 * into the same cylinder group then.
503 block_group = ei->i_block_group;
504 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
505 block_group &= ~(flex_size-1);
506 if (S_ISREG(inode->i_mode))
509 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
510 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
513 * If we are doing delayed allocation, we don't need take
514 * colour into account.
516 if (test_opt(inode->i_sb, DELALLOC))
519 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
520 colour = (current->pid % 16) *
521 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
523 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
524 return bg_start + colour;
528 * ext4_find_goal - find a preferred place for allocation.
530 * @block: block we want
531 * @partial: pointer to the last triple within a chain
533 * Normally this function find the preferred place for block allocation,
535 * Because this is only used for non-extent files, we limit the block nr
538 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
544 * XXX need to get goal block from mballoc's data structures
547 goal = ext4_find_near(inode, partial);
548 goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
553 * ext4_blks_to_allocate: Look up the block map and count the number
554 * of direct blocks need to be allocated for the given branch.
556 * @branch: chain of indirect blocks
557 * @k: number of blocks need for indirect blocks
558 * @blks: number of data blocks to be mapped.
559 * @blocks_to_boundary: the offset in the indirect block
561 * return the total number of blocks to be allocate, including the
562 * direct and indirect blocks.
564 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned int blks,
565 int blocks_to_boundary)
567 unsigned int count = 0;
570 * Simple case, [t,d]Indirect block(s) has not allocated yet
571 * then it's clear blocks on that path have not allocated
574 /* right now we don't handle cross boundary allocation */
575 if (blks < blocks_to_boundary + 1)
578 count += blocks_to_boundary + 1;
583 while (count < blks && count <= blocks_to_boundary &&
584 le32_to_cpu(*(branch[0].p + count)) == 0) {
591 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
592 * @indirect_blks: the number of blocks need to allocate for indirect
595 * @new_blocks: on return it will store the new block numbers for
596 * the indirect blocks(if needed) and the first direct block,
597 * @blks: on return it will store the total number of allocated
600 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
601 ext4_lblk_t iblock, ext4_fsblk_t goal,
602 int indirect_blks, int blks,
603 ext4_fsblk_t new_blocks[4], int *err)
605 struct ext4_allocation_request ar;
607 unsigned long count = 0, blk_allocated = 0;
609 ext4_fsblk_t current_block = 0;
613 * Here we try to allocate the requested multiple blocks at once,
614 * on a best-effort basis.
615 * To build a branch, we should allocate blocks for
616 * the indirect blocks(if not allocated yet), and at least
617 * the first direct block of this branch. That's the
618 * minimum number of blocks need to allocate(required)
620 /* first we try to allocate the indirect blocks */
621 target = indirect_blks;
624 /* allocating blocks for indirect blocks and direct blocks */
625 current_block = ext4_new_meta_blocks(handle, inode,
630 if (unlikely(current_block + count > EXT4_MAX_BLOCK_FILE_PHYS)) {
631 EXT4_ERROR_INODE(inode,
632 "current_block %llu + count %lu > %d!",
633 current_block, count,
634 EXT4_MAX_BLOCK_FILE_PHYS);
640 /* allocate blocks for indirect blocks */
641 while (index < indirect_blks && count) {
642 new_blocks[index++] = current_block++;
647 * save the new block number
648 * for the first direct block
650 new_blocks[index] = current_block;
651 printk(KERN_INFO "%s returned more blocks than "
652 "requested\n", __func__);
658 target = blks - count ;
659 blk_allocated = count;
662 /* Now allocate data blocks */
663 memset(&ar, 0, sizeof(ar));
668 if (S_ISREG(inode->i_mode))
669 /* enable in-core preallocation only for regular files */
670 ar.flags = EXT4_MB_HINT_DATA;
672 current_block = ext4_mb_new_blocks(handle, &ar, err);
673 if (unlikely(current_block + ar.len > EXT4_MAX_BLOCK_FILE_PHYS)) {
674 EXT4_ERROR_INODE(inode,
675 "current_block %llu + ar.len %d > %d!",
676 current_block, ar.len,
677 EXT4_MAX_BLOCK_FILE_PHYS);
682 if (*err && (target == blks)) {
684 * if the allocation failed and we didn't allocate
690 if (target == blks) {
692 * save the new block number
693 * for the first direct block
695 new_blocks[index] = current_block;
697 blk_allocated += ar.len;
700 /* total number of blocks allocated for direct blocks */
705 for (i = 0; i < index; i++)
706 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
711 * ext4_alloc_branch - allocate and set up a chain of blocks.
713 * @indirect_blks: number of allocated indirect blocks
714 * @blks: number of allocated direct blocks
715 * @offsets: offsets (in the blocks) to store the pointers to next.
716 * @branch: place to store the chain in.
718 * This function allocates blocks, zeroes out all but the last one,
719 * links them into chain and (if we are synchronous) writes them to disk.
720 * In other words, it prepares a branch that can be spliced onto the
721 * inode. It stores the information about that chain in the branch[], in
722 * the same format as ext4_get_branch() would do. We are calling it after
723 * we had read the existing part of chain and partial points to the last
724 * triple of that (one with zero ->key). Upon the exit we have the same
725 * picture as after the successful ext4_get_block(), except that in one
726 * place chain is disconnected - *branch->p is still zero (we did not
727 * set the last link), but branch->key contains the number that should
728 * be placed into *branch->p to fill that gap.
730 * If allocation fails we free all blocks we've allocated (and forget
731 * their buffer_heads) and return the error value the from failed
732 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
733 * as described above and return 0.
735 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
736 ext4_lblk_t iblock, int indirect_blks,
737 int *blks, ext4_fsblk_t goal,
738 ext4_lblk_t *offsets, Indirect *branch)
740 int blocksize = inode->i_sb->s_blocksize;
743 struct buffer_head *bh;
745 ext4_fsblk_t new_blocks[4];
746 ext4_fsblk_t current_block;
748 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
749 *blks, new_blocks, &err);
753 branch[0].key = cpu_to_le32(new_blocks[0]);
755 * metadata blocks and data blocks are allocated.
757 for (n = 1; n <= indirect_blks; n++) {
759 * Get buffer_head for parent block, zero it out
760 * and set the pointer to new one, then send
763 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
771 BUFFER_TRACE(bh, "call get_create_access");
772 err = ext4_journal_get_create_access(handle, bh);
774 /* Don't brelse(bh) here; it's done in
775 * ext4_journal_forget() below */
780 memset(bh->b_data, 0, blocksize);
781 branch[n].p = (__le32 *) bh->b_data + offsets[n];
782 branch[n].key = cpu_to_le32(new_blocks[n]);
783 *branch[n].p = branch[n].key;
784 if (n == indirect_blks) {
785 current_block = new_blocks[n];
787 * End of chain, update the last new metablock of
788 * the chain to point to the new allocated
789 * data blocks numbers
791 for (i = 1; i < num; i++)
792 *(branch[n].p + i) = cpu_to_le32(++current_block);
794 BUFFER_TRACE(bh, "marking uptodate");
795 set_buffer_uptodate(bh);
798 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
799 err = ext4_handle_dirty_metadata(handle, inode, bh);
806 /* Allocation failed, free what we already allocated */
807 ext4_free_blocks(handle, inode, 0, new_blocks[0], 1, 0);
808 for (i = 1; i <= n ; i++) {
810 * branch[i].bh is newly allocated, so there is no
811 * need to revoke the block, which is why we don't
812 * need to set EXT4_FREE_BLOCKS_METADATA.
814 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1,
815 EXT4_FREE_BLOCKS_FORGET);
817 for (i = n+1; i < indirect_blks; i++)
818 ext4_free_blocks(handle, inode, 0, new_blocks[i], 1, 0);
820 ext4_free_blocks(handle, inode, 0, new_blocks[i], num, 0);
826 * ext4_splice_branch - splice the allocated branch onto inode.
828 * @block: (logical) number of block we are adding
829 * @chain: chain of indirect blocks (with a missing link - see
831 * @where: location of missing link
832 * @num: number of indirect blocks we are adding
833 * @blks: number of direct blocks we are adding
835 * This function fills the missing link and does all housekeeping needed in
836 * inode (->i_blocks, etc.). In case of success we end up with the full
837 * chain to new block and return 0.
839 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
840 ext4_lblk_t block, Indirect *where, int num,
845 ext4_fsblk_t current_block;
848 * If we're splicing into a [td]indirect block (as opposed to the
849 * inode) then we need to get write access to the [td]indirect block
853 BUFFER_TRACE(where->bh, "get_write_access");
854 err = ext4_journal_get_write_access(handle, where->bh);
860 *where->p = where->key;
863 * Update the host buffer_head or inode to point to more just allocated
864 * direct blocks blocks
866 if (num == 0 && blks > 1) {
867 current_block = le32_to_cpu(where->key) + 1;
868 for (i = 1; i < blks; i++)
869 *(where->p + i) = cpu_to_le32(current_block++);
872 /* We are done with atomic stuff, now do the rest of housekeeping */
873 /* had we spliced it onto indirect block? */
876 * If we spliced it onto an indirect block, we haven't
877 * altered the inode. Note however that if it is being spliced
878 * onto an indirect block at the very end of the file (the
879 * file is growing) then we *will* alter the inode to reflect
880 * the new i_size. But that is not done here - it is done in
881 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
883 jbd_debug(5, "splicing indirect only\n");
884 BUFFER_TRACE(where->bh, "call ext4_handle_dirty_metadata");
885 err = ext4_handle_dirty_metadata(handle, inode, where->bh);
890 * OK, we spliced it into the inode itself on a direct block.
892 ext4_mark_inode_dirty(handle, inode);
893 jbd_debug(5, "splicing direct\n");
898 for (i = 1; i <= num; i++) {
900 * branch[i].bh is newly allocated, so there is no
901 * need to revoke the block, which is why we don't
902 * need to set EXT4_FREE_BLOCKS_METADATA.
904 ext4_free_blocks(handle, inode, where[i].bh, 0, 1,
905 EXT4_FREE_BLOCKS_FORGET);
907 ext4_free_blocks(handle, inode, 0, le32_to_cpu(where[num].key),
914 * The ext4_ind_map_blocks() function handles non-extents inodes
915 * (i.e., using the traditional indirect/double-indirect i_blocks
916 * scheme) for ext4_map_blocks().
918 * Allocation strategy is simple: if we have to allocate something, we will
919 * have to go the whole way to leaf. So let's do it before attaching anything
920 * to tree, set linkage between the newborn blocks, write them if sync is
921 * required, recheck the path, free and repeat if check fails, otherwise
922 * set the last missing link (that will protect us from any truncate-generated
923 * removals - all blocks on the path are immune now) and possibly force the
924 * write on the parent block.
925 * That has a nice additional property: no special recovery from the failed
926 * allocations is needed - we simply release blocks and do not touch anything
927 * reachable from inode.
929 * `handle' can be NULL if create == 0.
931 * return > 0, # of blocks mapped or allocated.
932 * return = 0, if plain lookup failed.
933 * return < 0, error case.
935 * The ext4_ind_get_blocks() function should be called with
936 * down_write(&EXT4_I(inode)->i_data_sem) if allocating filesystem
937 * blocks (i.e., flags has EXT4_GET_BLOCKS_CREATE set) or
938 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system
941 static int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
942 struct ext4_map_blocks *map,
946 ext4_lblk_t offsets[4];
951 int blocks_to_boundary = 0;
954 ext4_fsblk_t first_block = 0;
956 J_ASSERT(!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)));
957 J_ASSERT(handle != NULL || (flags & EXT4_GET_BLOCKS_CREATE) == 0);
958 depth = ext4_block_to_path(inode, map->m_lblk, offsets,
959 &blocks_to_boundary);
964 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
966 /* Simplest case - block found, no allocation needed */
968 first_block = le32_to_cpu(chain[depth - 1].key);
971 while (count < map->m_len && count <= blocks_to_boundary) {
974 blk = le32_to_cpu(*(chain[depth-1].p + count));
976 if (blk == first_block + count)
984 /* Next simple case - plain lookup or failed read of indirect block */
985 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0 || err == -EIO)
989 * Okay, we need to do block allocation.
991 goal = ext4_find_goal(inode, map->m_lblk, partial);
993 /* the number of blocks need to allocate for [d,t]indirect blocks */
994 indirect_blks = (chain + depth) - partial - 1;
997 * Next look up the indirect map to count the totoal number of
998 * direct blocks to allocate for this branch.
1000 count = ext4_blks_to_allocate(partial, indirect_blks,
1001 map->m_len, blocks_to_boundary);
1003 * Block out ext4_truncate while we alter the tree
1005 err = ext4_alloc_branch(handle, inode, map->m_lblk, indirect_blks,
1007 offsets + (partial - chain), partial);
1010 * The ext4_splice_branch call will free and forget any buffers
1011 * on the new chain if there is a failure, but that risks using
1012 * up transaction credits, especially for bitmaps where the
1013 * credits cannot be returned. Can we handle this somehow? We
1014 * may need to return -EAGAIN upwards in the worst case. --sct
1017 err = ext4_splice_branch(handle, inode, map->m_lblk,
1018 partial, indirect_blks, count);
1022 map->m_flags |= EXT4_MAP_NEW;
1024 ext4_update_inode_fsync_trans(handle, inode, 1);
1026 map->m_flags |= EXT4_MAP_MAPPED;
1027 map->m_pblk = le32_to_cpu(chain[depth-1].key);
1029 if (count > blocks_to_boundary)
1030 map->m_flags |= EXT4_MAP_BOUNDARY;
1032 /* Clean up and exit */
1033 partial = chain + depth - 1; /* the whole chain */
1035 while (partial > chain) {
1036 BUFFER_TRACE(partial->bh, "call brelse");
1037 brelse(partial->bh);
1045 qsize_t *ext4_get_reserved_space(struct inode *inode)
1047 return &EXT4_I(inode)->i_reserved_quota;
1052 * Calculate the number of metadata blocks need to reserve
1053 * to allocate a new block at @lblocks for non extent file based file
1055 static int ext4_indirect_calc_metadata_amount(struct inode *inode,
1058 struct ext4_inode_info *ei = EXT4_I(inode);
1059 sector_t dind_mask = ~((sector_t)EXT4_ADDR_PER_BLOCK(inode->i_sb) - 1);
1062 if (lblock < EXT4_NDIR_BLOCKS)
1065 lblock -= EXT4_NDIR_BLOCKS;
1067 if (ei->i_da_metadata_calc_len &&
1068 (lblock & dind_mask) == ei->i_da_metadata_calc_last_lblock) {
1069 ei->i_da_metadata_calc_len++;
1072 ei->i_da_metadata_calc_last_lblock = lblock & dind_mask;
1073 ei->i_da_metadata_calc_len = 1;
1074 blk_bits = order_base_2(lblock);
1075 return (blk_bits / EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb)) + 1;
1079 * Calculate the number of metadata blocks need to reserve
1080 * to allocate a block located at @lblock
1082 static int ext4_calc_metadata_amount(struct inode *inode, sector_t lblock)
1084 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1085 return ext4_ext_calc_metadata_amount(inode, lblock);
1087 return ext4_indirect_calc_metadata_amount(inode, lblock);
1091 * Called with i_data_sem down, which is important since we can call
1092 * ext4_discard_preallocations() from here.
1094 void ext4_da_update_reserve_space(struct inode *inode,
1095 int used, int quota_claim)
1097 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1098 struct ext4_inode_info *ei = EXT4_I(inode);
1100 spin_lock(&ei->i_block_reservation_lock);
1101 trace_ext4_da_update_reserve_space(inode, used);
1102 if (unlikely(used > ei->i_reserved_data_blocks)) {
1103 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
1104 "with only %d reserved data blocks\n",
1105 __func__, inode->i_ino, used,
1106 ei->i_reserved_data_blocks);
1108 used = ei->i_reserved_data_blocks;
1111 /* Update per-inode reservations */
1112 ei->i_reserved_data_blocks -= used;
1113 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
1114 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1115 used + ei->i_allocated_meta_blocks);
1116 ei->i_allocated_meta_blocks = 0;
1118 if (ei->i_reserved_data_blocks == 0) {
1120 * We can release all of the reserved metadata blocks
1121 * only when we have written all of the delayed
1122 * allocation blocks.
1124 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1125 ei->i_reserved_meta_blocks);
1126 ei->i_reserved_meta_blocks = 0;
1127 ei->i_da_metadata_calc_len = 0;
1129 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1131 /* Update quota subsystem for data blocks */
1133 dquot_claim_block(inode, used);
1136 * We did fallocate with an offset that is already delayed
1137 * allocated. So on delayed allocated writeback we should
1138 * not re-claim the quota for fallocated blocks.
1140 dquot_release_reservation_block(inode, used);
1144 * If we have done all the pending block allocations and if
1145 * there aren't any writers on the inode, we can discard the
1146 * inode's preallocations.
1148 if ((ei->i_reserved_data_blocks == 0) &&
1149 (atomic_read(&inode->i_writecount) == 0))
1150 ext4_discard_preallocations(inode);
1153 static int __check_block_validity(struct inode *inode, const char *func,
1155 struct ext4_map_blocks *map)
1157 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
1159 ext4_error_inode(inode, func, line, map->m_pblk,
1160 "lblock %lu mapped to illegal pblock "
1161 "(length %d)", (unsigned long) map->m_lblk,
1168 #define check_block_validity(inode, map) \
1169 __check_block_validity((inode), __func__, __LINE__, (map))
1172 * Return the number of contiguous dirty pages in a given inode
1173 * starting at page frame idx.
1175 static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
1176 unsigned int max_pages)
1178 struct address_space *mapping = inode->i_mapping;
1180 struct pagevec pvec;
1182 int i, nr_pages, done = 0;
1186 pagevec_init(&pvec, 0);
1189 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1190 PAGECACHE_TAG_DIRTY,
1191 (pgoff_t)PAGEVEC_SIZE);
1194 for (i = 0; i < nr_pages; i++) {
1195 struct page *page = pvec.pages[i];
1196 struct buffer_head *bh, *head;
1199 if (unlikely(page->mapping != mapping) ||
1201 PageWriteback(page) ||
1202 page->index != idx) {
1207 if (page_has_buffers(page)) {
1208 bh = head = page_buffers(page);
1210 if (!buffer_delay(bh) &&
1211 !buffer_unwritten(bh))
1213 bh = bh->b_this_page;
1214 } while (!done && (bh != head));
1221 if (num >= max_pages) {
1226 pagevec_release(&pvec);
1232 * The ext4_map_blocks() function tries to look up the requested blocks,
1233 * and returns if the blocks are already mapped.
1235 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1236 * and store the allocated blocks in the result buffer head and mark it
1239 * If file type is extents based, it will call ext4_ext_map_blocks(),
1240 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
1243 * On success, it returns the number of blocks being mapped or allocate.
1244 * if create==0 and the blocks are pre-allocated and uninitialized block,
1245 * the result buffer head is unmapped. If the create ==1, it will make sure
1246 * the buffer head is mapped.
1248 * It returns 0 if plain look up failed (blocks have not been allocated), in
1249 * that casem, buffer head is unmapped
1251 * It returns the error in case of allocation failure.
1253 int ext4_map_blocks(handle_t *handle, struct inode *inode,
1254 struct ext4_map_blocks *map, int flags)
1259 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
1260 "logical block %lu\n", inode->i_ino, flags, map->m_len,
1261 (unsigned long) map->m_lblk);
1263 * Try to see if we can get the block without requesting a new
1264 * file system block.
1266 down_read((&EXT4_I(inode)->i_data_sem));
1267 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1268 retval = ext4_ext_map_blocks(handle, inode, map, 0);
1270 retval = ext4_ind_map_blocks(handle, inode, map, 0);
1272 up_read((&EXT4_I(inode)->i_data_sem));
1274 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1275 int ret = check_block_validity(inode, map);
1280 /* If it is only a block(s) look up */
1281 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
1285 * Returns if the blocks have already allocated
1287 * Note that if blocks have been preallocated
1288 * ext4_ext_get_block() returns th create = 0
1289 * with buffer head unmapped.
1291 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
1295 * When we call get_blocks without the create flag, the
1296 * BH_Unwritten flag could have gotten set if the blocks
1297 * requested were part of a uninitialized extent. We need to
1298 * clear this flag now that we are committed to convert all or
1299 * part of the uninitialized extent to be an initialized
1300 * extent. This is because we need to avoid the combination
1301 * of BH_Unwritten and BH_Mapped flags being simultaneously
1302 * set on the buffer_head.
1304 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
1307 * New blocks allocate and/or writing to uninitialized extent
1308 * will possibly result in updating i_data, so we take
1309 * the write lock of i_data_sem, and call get_blocks()
1310 * with create == 1 flag.
1312 down_write((&EXT4_I(inode)->i_data_sem));
1315 * if the caller is from delayed allocation writeout path
1316 * we have already reserved fs blocks for allocation
1317 * let the underlying get_block() function know to
1318 * avoid double accounting
1320 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1321 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1323 * We need to check for EXT4 here because migrate
1324 * could have changed the inode type in between
1326 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
1327 retval = ext4_ext_map_blocks(handle, inode, map, flags);
1329 retval = ext4_ind_map_blocks(handle, inode, map, flags);
1331 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
1333 * We allocated new blocks which will result in
1334 * i_data's format changing. Force the migrate
1335 * to fail by clearing migrate flags
1337 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
1341 * Update reserved blocks/metadata blocks after successful
1342 * block allocation which had been deferred till now. We don't
1343 * support fallocate for non extent files. So we can update
1344 * reserve space here.
1347 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
1348 ext4_da_update_reserve_space(inode, retval, 1);
1350 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1351 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1353 up_write((&EXT4_I(inode)->i_data_sem));
1354 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
1355 int ret = check_block_validity(inode, map);
1362 /* Maximum number of blocks we map for direct IO at once. */
1363 #define DIO_MAX_BLOCKS 4096
1365 static int _ext4_get_block(struct inode *inode, sector_t iblock,
1366 struct buffer_head *bh, int flags)
1368 handle_t *handle = ext4_journal_current_handle();
1369 struct ext4_map_blocks map;
1370 int ret = 0, started = 0;
1373 map.m_lblk = iblock;
1374 map.m_len = bh->b_size >> inode->i_blkbits;
1376 if (flags && !handle) {
1377 /* Direct IO write... */
1378 if (map.m_len > DIO_MAX_BLOCKS)
1379 map.m_len = DIO_MAX_BLOCKS;
1380 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
1381 handle = ext4_journal_start(inode, dio_credits);
1382 if (IS_ERR(handle)) {
1383 ret = PTR_ERR(handle);
1389 ret = ext4_map_blocks(handle, inode, &map, flags);
1391 map_bh(bh, inode->i_sb, map.m_pblk);
1392 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1393 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
1397 ext4_journal_stop(handle);
1401 int ext4_get_block(struct inode *inode, sector_t iblock,
1402 struct buffer_head *bh, int create)
1404 return _ext4_get_block(inode, iblock, bh,
1405 create ? EXT4_GET_BLOCKS_CREATE : 0);
1409 * `handle' can be NULL if create is zero
1411 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1412 ext4_lblk_t block, int create, int *errp)
1414 struct ext4_map_blocks map;
1415 struct buffer_head *bh;
1418 J_ASSERT(handle != NULL || create == 0);
1422 err = ext4_map_blocks(handle, inode, &map,
1423 create ? EXT4_GET_BLOCKS_CREATE : 0);
1431 bh = sb_getblk(inode->i_sb, map.m_pblk);
1436 if (map.m_flags & EXT4_MAP_NEW) {
1437 J_ASSERT(create != 0);
1438 J_ASSERT(handle != NULL);
1441 * Now that we do not always journal data, we should
1442 * keep in mind whether this should always journal the
1443 * new buffer as metadata. For now, regular file
1444 * writes use ext4_get_block instead, so it's not a
1448 BUFFER_TRACE(bh, "call get_create_access");
1449 fatal = ext4_journal_get_create_access(handle, bh);
1450 if (!fatal && !buffer_uptodate(bh)) {
1451 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1452 set_buffer_uptodate(bh);
1455 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1456 err = ext4_handle_dirty_metadata(handle, inode, bh);
1460 BUFFER_TRACE(bh, "not a new buffer");
1470 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1471 ext4_lblk_t block, int create, int *err)
1473 struct buffer_head *bh;
1475 bh = ext4_getblk(handle, inode, block, create, err);
1478 if (buffer_uptodate(bh))
1480 ll_rw_block(READ_META, 1, &bh);
1482 if (buffer_uptodate(bh))
1489 static int walk_page_buffers(handle_t *handle,
1490 struct buffer_head *head,
1494 int (*fn)(handle_t *handle,
1495 struct buffer_head *bh))
1497 struct buffer_head *bh;
1498 unsigned block_start, block_end;
1499 unsigned blocksize = head->b_size;
1501 struct buffer_head *next;
1503 for (bh = head, block_start = 0;
1504 ret == 0 && (bh != head || !block_start);
1505 block_start = block_end, bh = next) {
1506 next = bh->b_this_page;
1507 block_end = block_start + blocksize;
1508 if (block_end <= from || block_start >= to) {
1509 if (partial && !buffer_uptodate(bh))
1513 err = (*fn)(handle, bh);
1521 * To preserve ordering, it is essential that the hole instantiation and
1522 * the data write be encapsulated in a single transaction. We cannot
1523 * close off a transaction and start a new one between the ext4_get_block()
1524 * and the commit_write(). So doing the jbd2_journal_start at the start of
1525 * prepare_write() is the right place.
1527 * Also, this function can nest inside ext4_writepage() ->
1528 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1529 * has generated enough buffer credits to do the whole page. So we won't
1530 * block on the journal in that case, which is good, because the caller may
1533 * By accident, ext4 can be reentered when a transaction is open via
1534 * quota file writes. If we were to commit the transaction while thus
1535 * reentered, there can be a deadlock - we would be holding a quota
1536 * lock, and the commit would never complete if another thread had a
1537 * transaction open and was blocking on the quota lock - a ranking
1540 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1541 * will _not_ run commit under these circumstances because handle->h_ref
1542 * is elevated. We'll still have enough credits for the tiny quotafile
1545 static int do_journal_get_write_access(handle_t *handle,
1546 struct buffer_head *bh)
1548 int dirty = buffer_dirty(bh);
1551 if (!buffer_mapped(bh) || buffer_freed(bh))
1554 * __block_write_begin() could have dirtied some buffers. Clean
1555 * the dirty bit as jbd2_journal_get_write_access() could complain
1556 * otherwise about fs integrity issues. Setting of the dirty bit
1557 * by __block_write_begin() isn't a real problem here as we clear
1558 * the bit before releasing a page lock and thus writeback cannot
1559 * ever write the buffer.
1562 clear_buffer_dirty(bh);
1563 ret = ext4_journal_get_write_access(handle, bh);
1565 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1570 * Truncate blocks that were not used by write. We have to truncate the
1571 * pagecache as well so that corresponding buffers get properly unmapped.
1573 static void ext4_truncate_failed_write(struct inode *inode)
1575 truncate_inode_pages(inode->i_mapping, inode->i_size);
1576 ext4_truncate(inode);
1579 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
1580 struct buffer_head *bh_result, int create);
1581 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1582 loff_t pos, unsigned len, unsigned flags,
1583 struct page **pagep, void **fsdata)
1585 struct inode *inode = mapping->host;
1586 int ret, needed_blocks;
1593 trace_ext4_write_begin(inode, pos, len, flags);
1595 * Reserve one block more for addition to orphan list in case
1596 * we allocate blocks but write fails for some reason
1598 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1599 index = pos >> PAGE_CACHE_SHIFT;
1600 from = pos & (PAGE_CACHE_SIZE - 1);
1604 handle = ext4_journal_start(inode, needed_blocks);
1605 if (IS_ERR(handle)) {
1606 ret = PTR_ERR(handle);
1610 /* We cannot recurse into the filesystem as the transaction is already
1612 flags |= AOP_FLAG_NOFS;
1614 page = grab_cache_page_write_begin(mapping, index, flags);
1616 ext4_journal_stop(handle);
1622 if (ext4_should_dioread_nolock(inode))
1623 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
1625 ret = __block_write_begin(page, pos, len, ext4_get_block);
1627 if (!ret && ext4_should_journal_data(inode)) {
1628 ret = walk_page_buffers(handle, page_buffers(page),
1629 from, to, NULL, do_journal_get_write_access);
1634 page_cache_release(page);
1636 * __block_write_begin may have instantiated a few blocks
1637 * outside i_size. Trim these off again. Don't need
1638 * i_size_read because we hold i_mutex.
1640 * Add inode to orphan list in case we crash before
1643 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1644 ext4_orphan_add(handle, inode);
1646 ext4_journal_stop(handle);
1647 if (pos + len > inode->i_size) {
1648 ext4_truncate_failed_write(inode);
1650 * If truncate failed early the inode might
1651 * still be on the orphan list; we need to
1652 * make sure the inode is removed from the
1653 * orphan list in that case.
1656 ext4_orphan_del(NULL, inode);
1660 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1666 /* For write_end() in data=journal mode */
1667 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1669 if (!buffer_mapped(bh) || buffer_freed(bh))
1671 set_buffer_uptodate(bh);
1672 return ext4_handle_dirty_metadata(handle, NULL, bh);
1675 static int ext4_generic_write_end(struct file *file,
1676 struct address_space *mapping,
1677 loff_t pos, unsigned len, unsigned copied,
1678 struct page *page, void *fsdata)
1680 int i_size_changed = 0;
1681 struct inode *inode = mapping->host;
1682 handle_t *handle = ext4_journal_current_handle();
1684 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1687 * No need to use i_size_read() here, the i_size
1688 * cannot change under us because we hold i_mutex.
1690 * But it's important to update i_size while still holding page lock:
1691 * page writeout could otherwise come in and zero beyond i_size.
1693 if (pos + copied > inode->i_size) {
1694 i_size_write(inode, pos + copied);
1698 if (pos + copied > EXT4_I(inode)->i_disksize) {
1699 /* We need to mark inode dirty even if
1700 * new_i_size is less that inode->i_size
1701 * bu greater than i_disksize.(hint delalloc)
1703 ext4_update_i_disksize(inode, (pos + copied));
1707 page_cache_release(page);
1710 * Don't mark the inode dirty under page lock. First, it unnecessarily
1711 * makes the holding time of page lock longer. Second, it forces lock
1712 * ordering of page lock and transaction start for journaling
1716 ext4_mark_inode_dirty(handle, inode);
1722 * We need to pick up the new inode size which generic_commit_write gave us
1723 * `file' can be NULL - eg, when called from page_symlink().
1725 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1726 * buffers are managed internally.
1728 static int ext4_ordered_write_end(struct file *file,
1729 struct address_space *mapping,
1730 loff_t pos, unsigned len, unsigned copied,
1731 struct page *page, void *fsdata)
1733 handle_t *handle = ext4_journal_current_handle();
1734 struct inode *inode = mapping->host;
1737 trace_ext4_ordered_write_end(inode, pos, len, copied);
1738 ret = ext4_jbd2_file_inode(handle, inode);
1741 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1744 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1745 /* if we have allocated more blocks and copied
1746 * less. We will have blocks allocated outside
1747 * inode->i_size. So truncate them
1749 ext4_orphan_add(handle, inode);
1753 ret2 = ext4_journal_stop(handle);
1757 if (pos + len > inode->i_size) {
1758 ext4_truncate_failed_write(inode);
1760 * If truncate failed early the inode might still be
1761 * on the orphan list; we need to make sure the inode
1762 * is removed from the orphan list in that case.
1765 ext4_orphan_del(NULL, inode);
1769 return ret ? ret : copied;
1772 static int ext4_writeback_write_end(struct file *file,
1773 struct address_space *mapping,
1774 loff_t pos, unsigned len, unsigned copied,
1775 struct page *page, void *fsdata)
1777 handle_t *handle = ext4_journal_current_handle();
1778 struct inode *inode = mapping->host;
1781 trace_ext4_writeback_write_end(inode, pos, len, copied);
1782 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
1785 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1786 /* if we have allocated more blocks and copied
1787 * less. We will have blocks allocated outside
1788 * inode->i_size. So truncate them
1790 ext4_orphan_add(handle, inode);
1795 ret2 = ext4_journal_stop(handle);
1799 if (pos + len > inode->i_size) {
1800 ext4_truncate_failed_write(inode);
1802 * If truncate failed early the inode might still be
1803 * on the orphan list; we need to make sure the inode
1804 * is removed from the orphan list in that case.
1807 ext4_orphan_del(NULL, inode);
1810 return ret ? ret : copied;
1813 static int ext4_journalled_write_end(struct file *file,
1814 struct address_space *mapping,
1815 loff_t pos, unsigned len, unsigned copied,
1816 struct page *page, void *fsdata)
1818 handle_t *handle = ext4_journal_current_handle();
1819 struct inode *inode = mapping->host;
1825 trace_ext4_journalled_write_end(inode, pos, len, copied);
1826 from = pos & (PAGE_CACHE_SIZE - 1);
1830 if (!PageUptodate(page))
1832 page_zero_new_buffers(page, from+copied, to);
1835 ret = walk_page_buffers(handle, page_buffers(page), from,
1836 to, &partial, write_end_fn);
1838 SetPageUptodate(page);
1839 new_i_size = pos + copied;
1840 if (new_i_size > inode->i_size)
1841 i_size_write(inode, pos+copied);
1842 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1843 if (new_i_size > EXT4_I(inode)->i_disksize) {
1844 ext4_update_i_disksize(inode, new_i_size);
1845 ret2 = ext4_mark_inode_dirty(handle, inode);
1851 page_cache_release(page);
1852 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1853 /* if we have allocated more blocks and copied
1854 * less. We will have blocks allocated outside
1855 * inode->i_size. So truncate them
1857 ext4_orphan_add(handle, inode);
1859 ret2 = ext4_journal_stop(handle);
1862 if (pos + len > inode->i_size) {
1863 ext4_truncate_failed_write(inode);
1865 * If truncate failed early the inode might still be
1866 * on the orphan list; we need to make sure the inode
1867 * is removed from the orphan list in that case.
1870 ext4_orphan_del(NULL, inode);
1873 return ret ? ret : copied;
1877 * Reserve a single block located at lblock
1879 static int ext4_da_reserve_space(struct inode *inode, sector_t lblock)
1882 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1883 struct ext4_inode_info *ei = EXT4_I(inode);
1884 unsigned long md_needed;
1888 * recalculate the amount of metadata blocks to reserve
1889 * in order to allocate nrblocks
1890 * worse case is one extent per block
1893 spin_lock(&ei->i_block_reservation_lock);
1894 md_needed = ext4_calc_metadata_amount(inode, lblock);
1895 trace_ext4_da_reserve_space(inode, md_needed);
1896 spin_unlock(&ei->i_block_reservation_lock);
1899 * We will charge metadata quota at writeout time; this saves
1900 * us from metadata over-estimation, though we may go over by
1901 * a small amount in the end. Here we just reserve for data.
1903 ret = dquot_reserve_block(inode, 1);
1907 * We do still charge estimated metadata to the sb though;
1908 * we cannot afford to run out of free blocks.
1910 if (ext4_claim_free_blocks(sbi, md_needed + 1)) {
1911 dquot_release_reservation_block(inode, 1);
1912 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1918 spin_lock(&ei->i_block_reservation_lock);
1919 ei->i_reserved_data_blocks++;
1920 ei->i_reserved_meta_blocks += md_needed;
1921 spin_unlock(&ei->i_block_reservation_lock);
1923 return 0; /* success */
1926 static void ext4_da_release_space(struct inode *inode, int to_free)
1928 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1929 struct ext4_inode_info *ei = EXT4_I(inode);
1932 return; /* Nothing to release, exit */
1934 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1936 trace_ext4_da_release_space(inode, to_free);
1937 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1939 * if there aren't enough reserved blocks, then the
1940 * counter is messed up somewhere. Since this
1941 * function is called from invalidate page, it's
1942 * harmless to return without any action.
1944 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1945 "ino %lu, to_free %d with only %d reserved "
1946 "data blocks\n", inode->i_ino, to_free,
1947 ei->i_reserved_data_blocks);
1949 to_free = ei->i_reserved_data_blocks;
1951 ei->i_reserved_data_blocks -= to_free;
1953 if (ei->i_reserved_data_blocks == 0) {
1955 * We can release all of the reserved metadata blocks
1956 * only when we have written all of the delayed
1957 * allocation blocks.
1959 percpu_counter_sub(&sbi->s_dirtyblocks_counter,
1960 ei->i_reserved_meta_blocks);
1961 ei->i_reserved_meta_blocks = 0;
1962 ei->i_da_metadata_calc_len = 0;
1965 /* update fs dirty data blocks counter */
1966 percpu_counter_sub(&sbi->s_dirtyblocks_counter, to_free);
1968 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1970 dquot_release_reservation_block(inode, to_free);
1973 static void ext4_da_page_release_reservation(struct page *page,
1974 unsigned long offset)
1977 struct buffer_head *head, *bh;
1978 unsigned int curr_off = 0;
1980 head = page_buffers(page);
1983 unsigned int next_off = curr_off + bh->b_size;
1985 if ((offset <= curr_off) && (buffer_delay(bh))) {
1987 clear_buffer_delay(bh);
1989 curr_off = next_off;
1990 } while ((bh = bh->b_this_page) != head);
1991 ext4_da_release_space(page->mapping->host, to_release);
1995 * Delayed allocation stuff
1999 * mpage_da_submit_io - walks through extent of pages and try to write
2000 * them with writepage() call back
2002 * @mpd->inode: inode
2003 * @mpd->first_page: first page of the extent
2004 * @mpd->next_page: page after the last page of the extent
2006 * By the time mpage_da_submit_io() is called we expect all blocks
2007 * to be allocated. this may be wrong if allocation failed.
2009 * As pages are already locked by write_cache_pages(), we can't use it
2011 static int mpage_da_submit_io(struct mpage_da_data *mpd,
2012 struct ext4_map_blocks *map)
2014 struct pagevec pvec;
2015 unsigned long index, end;
2016 int ret = 0, err, nr_pages, i;
2017 struct inode *inode = mpd->inode;
2018 struct address_space *mapping = inode->i_mapping;
2019 loff_t size = i_size_read(inode);
2020 unsigned int len, block_start;
2021 struct buffer_head *bh, *page_bufs = NULL;
2022 int journal_data = ext4_should_journal_data(inode);
2023 sector_t pblock = 0, cur_logical = 0;
2024 struct ext4_io_submit io_submit;
2026 BUG_ON(mpd->next_page <= mpd->first_page);
2027 memset(&io_submit, 0, sizeof(io_submit));
2029 * We need to start from the first_page to the next_page - 1
2030 * to make sure we also write the mapped dirty buffer_heads.
2031 * If we look at mpd->b_blocknr we would only be looking
2032 * at the currently mapped buffer_heads.
2034 index = mpd->first_page;
2035 end = mpd->next_page - 1;
2037 pagevec_init(&pvec, 0);
2038 while (index <= end) {
2039 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2042 for (i = 0; i < nr_pages; i++) {
2043 int commit_write = 0, redirty_page = 0;
2044 struct page *page = pvec.pages[i];
2046 index = page->index;
2050 if (index == size >> PAGE_CACHE_SHIFT)
2051 len = size & ~PAGE_CACHE_MASK;
2053 len = PAGE_CACHE_SIZE;
2055 cur_logical = index << (PAGE_CACHE_SHIFT -
2057 pblock = map->m_pblk + (cur_logical -
2062 BUG_ON(!PageLocked(page));
2063 BUG_ON(PageWriteback(page));
2066 * If the page does not have buffers (for
2067 * whatever reason), try to create them using
2068 * __block_write_begin. If this fails,
2069 * redirty the page and move on.
2071 if (!page_has_buffers(page)) {
2072 if (__block_write_begin(page, 0, len,
2073 noalloc_get_block_write)) {
2075 redirty_page_for_writepage(mpd->wbc,
2083 bh = page_bufs = page_buffers(page);
2088 if (map && (cur_logical >= map->m_lblk) &&
2089 (cur_logical <= (map->m_lblk +
2090 (map->m_len - 1)))) {
2091 if (buffer_delay(bh)) {
2092 clear_buffer_delay(bh);
2093 bh->b_blocknr = pblock;
2095 if (buffer_unwritten(bh) ||
2097 BUG_ON(bh->b_blocknr != pblock);
2098 if (map->m_flags & EXT4_MAP_UNINIT)
2099 set_buffer_uninit(bh);
2100 clear_buffer_unwritten(bh);
2103 /* redirty page if block allocation undone */
2104 if (buffer_delay(bh) || buffer_unwritten(bh))
2106 bh = bh->b_this_page;
2107 block_start += bh->b_size;
2110 } while (bh != page_bufs);
2116 /* mark the buffer_heads as dirty & uptodate */
2117 block_commit_write(page, 0, len);
2120 * Delalloc doesn't support data journalling,
2121 * but eventually maybe we'll lift this
2124 if (unlikely(journal_data && PageChecked(page)))
2125 err = __ext4_journalled_writepage(page, len);
2127 err = ext4_bio_write_page(&io_submit, page,
2131 mpd->pages_written++;
2133 * In error case, we have to continue because
2134 * remaining pages are still locked
2139 pagevec_release(&pvec);
2141 ext4_io_submit(&io_submit);
2145 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
2146 sector_t logical, long blk_cnt)
2150 struct pagevec pvec;
2151 struct inode *inode = mpd->inode;
2152 struct address_space *mapping = inode->i_mapping;
2154 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
2155 end = (logical + blk_cnt - 1) >>
2156 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2157 while (index <= end) {
2158 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
2161 for (i = 0; i < nr_pages; i++) {
2162 struct page *page = pvec.pages[i];
2163 if (page->index > end)
2165 BUG_ON(!PageLocked(page));
2166 BUG_ON(PageWriteback(page));
2167 block_invalidatepage(page, 0);
2168 ClearPageUptodate(page);
2171 index = pvec.pages[nr_pages - 1]->index + 1;
2172 pagevec_release(&pvec);
2177 static void ext4_print_free_blocks(struct inode *inode)
2179 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2180 printk(KERN_CRIT "Total free blocks count %lld\n",
2181 ext4_count_free_blocks(inode->i_sb));
2182 printk(KERN_CRIT "Free/Dirty block details\n");
2183 printk(KERN_CRIT "free_blocks=%lld\n",
2184 (long long) percpu_counter_sum(&sbi->s_freeblocks_counter));
2185 printk(KERN_CRIT "dirty_blocks=%lld\n",
2186 (long long) percpu_counter_sum(&sbi->s_dirtyblocks_counter));
2187 printk(KERN_CRIT "Block reservation details\n");
2188 printk(KERN_CRIT "i_reserved_data_blocks=%u\n",
2189 EXT4_I(inode)->i_reserved_data_blocks);
2190 printk(KERN_CRIT "i_reserved_meta_blocks=%u\n",
2191 EXT4_I(inode)->i_reserved_meta_blocks);
2196 * mpage_da_map_and_submit - go through given space, map them
2197 * if necessary, and then submit them for I/O
2199 * @mpd - bh describing space
2201 * The function skips space we know is already mapped to disk blocks.
2204 static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
2206 int err, blks, get_blocks_flags;
2207 struct ext4_map_blocks map, *mapp = NULL;
2208 sector_t next = mpd->b_blocknr;
2209 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
2210 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
2211 handle_t *handle = NULL;
2214 * If the blocks are mapped already, or we couldn't accumulate
2215 * any blocks, then proceed immediately to the submission stage.
2217 if ((mpd->b_size == 0) ||
2218 ((mpd->b_state & (1 << BH_Mapped)) &&
2219 !(mpd->b_state & (1 << BH_Delay)) &&
2220 !(mpd->b_state & (1 << BH_Unwritten))))
2223 handle = ext4_journal_current_handle();
2227 * Call ext4_map_blocks() to allocate any delayed allocation
2228 * blocks, or to convert an uninitialized extent to be
2229 * initialized (in the case where we have written into
2230 * one or more preallocated blocks).
2232 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
2233 * indicate that we are on the delayed allocation path. This
2234 * affects functions in many different parts of the allocation
2235 * call path. This flag exists primarily because we don't
2236 * want to change *many* call functions, so ext4_map_blocks()
2237 * will set the magic i_delalloc_reserved_flag once the
2238 * inode's allocation semaphore is taken.
2240 * If the blocks in questions were delalloc blocks, set
2241 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
2242 * variables are updated after the blocks have been allocated.
2245 map.m_len = max_blocks;
2246 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
2247 if (ext4_should_dioread_nolock(mpd->inode))
2248 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2249 if (mpd->b_state & (1 << BH_Delay))
2250 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2252 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
2254 struct super_block *sb = mpd->inode->i_sb;
2258 * If get block returns EAGAIN or ENOSPC and there
2259 * appears to be free blocks we will call
2260 * ext4_writepage() for all of the pages which will
2261 * just redirty the pages.
2266 if (err == -ENOSPC &&
2267 ext4_count_free_blocks(sb)) {
2273 * get block failure will cause us to loop in
2274 * writepages, because a_ops->writepage won't be able
2275 * to make progress. The page will be redirtied by
2276 * writepage and writepages will again try to write
2279 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2280 ext4_msg(sb, KERN_CRIT,
2281 "delayed block allocation failed for inode %lu "
2282 "at logical offset %llu with max blocks %zd "
2283 "with error %d", mpd->inode->i_ino,
2284 (unsigned long long) next,
2285 mpd->b_size >> mpd->inode->i_blkbits, err);
2286 ext4_msg(sb, KERN_CRIT,
2287 "This should not happen!! Data will be lost\n");
2289 ext4_print_free_blocks(mpd->inode);
2291 /* invalidate all the pages */
2292 ext4_da_block_invalidatepages(mpd, next,
2293 mpd->b_size >> mpd->inode->i_blkbits);
2299 if (map.m_flags & EXT4_MAP_NEW) {
2300 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
2303 for (i = 0; i < map.m_len; i++)
2304 unmap_underlying_metadata(bdev, map.m_pblk + i);
2307 if (ext4_should_order_data(mpd->inode)) {
2308 err = ext4_jbd2_file_inode(handle, mpd->inode);
2310 /* This only happens if the journal is aborted */
2315 * Update on-disk size along with block allocation.
2317 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
2318 if (disksize > i_size_read(mpd->inode))
2319 disksize = i_size_read(mpd->inode);
2320 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
2321 ext4_update_i_disksize(mpd->inode, disksize);
2322 err = ext4_mark_inode_dirty(handle, mpd->inode);
2324 ext4_error(mpd->inode->i_sb,
2325 "Failed to mark inode %lu dirty",
2330 mpage_da_submit_io(mpd, mapp);
2334 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
2335 (1 << BH_Delay) | (1 << BH_Unwritten))
2338 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
2340 * @mpd->lbh - extent of blocks
2341 * @logical - logical number of the block in the file
2342 * @bh - bh of the block (used to access block's state)
2344 * the function is used to collect contig. blocks in same state
2346 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
2347 sector_t logical, size_t b_size,
2348 unsigned long b_state)
2351 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
2354 * XXX Don't go larger than mballoc is willing to allocate
2355 * This is a stopgap solution. We eventually need to fold
2356 * mpage_da_submit_io() into this function and then call
2357 * ext4_map_blocks() multiple times in a loop
2359 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
2362 /* check if thereserved journal credits might overflow */
2363 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
2364 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
2366 * With non-extent format we are limited by the journal
2367 * credit available. Total credit needed to insert
2368 * nrblocks contiguous blocks is dependent on the
2369 * nrblocks. So limit nrblocks.
2372 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
2373 EXT4_MAX_TRANS_DATA) {
2375 * Adding the new buffer_head would make it cross the
2376 * allowed limit for which we have journal credit
2377 * reserved. So limit the new bh->b_size
2379 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
2380 mpd->inode->i_blkbits;
2381 /* we will do mpage_da_submit_io in the next loop */
2385 * First block in the extent
2387 if (mpd->b_size == 0) {
2388 mpd->b_blocknr = logical;
2389 mpd->b_size = b_size;
2390 mpd->b_state = b_state & BH_FLAGS;
2394 next = mpd->b_blocknr + nrblocks;
2396 * Can we merge the block to our big extent?
2398 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
2399 mpd->b_size += b_size;
2405 * We couldn't merge the block to our extent, so we
2406 * need to flush current extent and start new one
2408 mpage_da_map_and_submit(mpd);
2412 static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
2414 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
2418 * __mpage_da_writepage - finds extent of pages and blocks
2420 * @page: page to consider
2421 * @wbc: not used, we just follow rules
2424 * The function finds extents of pages and scan them for all blocks.
2426 static int __mpage_da_writepage(struct page *page,
2427 struct writeback_control *wbc,
2428 struct mpage_da_data *mpd)
2430 struct inode *inode = mpd->inode;
2431 struct buffer_head *bh, *head;
2435 * Can we merge this page to current extent?
2437 if (mpd->next_page != page->index) {
2439 * Nope, we can't. So, we map non-allocated blocks
2440 * and start IO on them
2442 if (mpd->next_page != mpd->first_page) {
2443 mpage_da_map_and_submit(mpd);
2445 * skip rest of the page in the page_vec
2447 redirty_page_for_writepage(wbc, page);
2449 return MPAGE_DA_EXTENT_TAIL;
2453 * Start next extent of pages ...
2455 mpd->first_page = page->index;
2465 mpd->next_page = page->index + 1;
2466 logical = (sector_t) page->index <<
2467 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2469 if (!page_has_buffers(page)) {
2470 mpage_add_bh_to_extent(mpd, logical, PAGE_CACHE_SIZE,
2471 (1 << BH_Dirty) | (1 << BH_Uptodate));
2473 return MPAGE_DA_EXTENT_TAIL;
2476 * Page with regular buffer heads, just add all dirty ones
2478 head = page_buffers(page);
2481 BUG_ON(buffer_locked(bh));
2483 * We need to try to allocate
2484 * unmapped blocks in the same page.
2485 * Otherwise we won't make progress
2486 * with the page in ext4_writepage
2488 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2489 mpage_add_bh_to_extent(mpd, logical,
2493 return MPAGE_DA_EXTENT_TAIL;
2494 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2496 * mapped dirty buffer. We need to update
2497 * the b_state because we look at
2498 * b_state in mpage_da_map_blocks. We don't
2499 * update b_size because if we find an
2500 * unmapped buffer_head later we need to
2501 * use the b_state flag of that buffer_head.
2503 if (mpd->b_size == 0)
2504 mpd->b_state = bh->b_state & BH_FLAGS;
2507 } while ((bh = bh->b_this_page) != head);
2514 * This is a special get_blocks_t callback which is used by
2515 * ext4_da_write_begin(). It will either return mapped block or
2516 * reserve space for a single block.
2518 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2519 * We also have b_blocknr = -1 and b_bdev initialized properly
2521 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2522 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2523 * initialized properly.
2525 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2526 struct buffer_head *bh, int create)
2528 struct ext4_map_blocks map;
2530 sector_t invalid_block = ~((sector_t) 0xffff);
2532 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
2535 BUG_ON(create == 0);
2536 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
2538 map.m_lblk = iblock;
2542 * first, we need to know whether the block is allocated already
2543 * preallocated blocks are unmapped but should treated
2544 * the same as allocated blocks.
2546 ret = ext4_map_blocks(NULL, inode, &map, 0);
2550 if (buffer_delay(bh))
2551 return 0; /* Not sure this could or should happen */
2553 * XXX: __block_write_begin() unmaps passed block, is it OK?
2555 ret = ext4_da_reserve_space(inode, iblock);
2557 /* not enough space to reserve */
2560 map_bh(bh, inode->i_sb, invalid_block);
2562 set_buffer_delay(bh);
2566 map_bh(bh, inode->i_sb, map.m_pblk);
2567 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
2569 if (buffer_unwritten(bh)) {
2570 /* A delayed write to unwritten bh should be marked
2571 * new and mapped. Mapped ensures that we don't do
2572 * get_block multiple times when we write to the same
2573 * offset and new ensures that we do proper zero out
2574 * for partial write.
2577 set_buffer_mapped(bh);
2583 * This function is used as a standard get_block_t calback function
2584 * when there is no desire to allocate any blocks. It is used as a
2585 * callback function for block_write_begin() and block_write_full_page().
2586 * These functions should only try to map a single block at a time.
2588 * Since this function doesn't do block allocations even if the caller
2589 * requests it by passing in create=1, it is critically important that
2590 * any caller checks to make sure that any buffer heads are returned
2591 * by this function are either all already mapped or marked for
2592 * delayed allocation before calling block_write_full_page(). Otherwise,
2593 * b_blocknr could be left unitialized, and the page write functions will
2594 * be taken by surprise.
2596 static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
2597 struct buffer_head *bh_result, int create)
2599 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2600 return _ext4_get_block(inode, iblock, bh_result, 0);
2603 static int bget_one(handle_t *handle, struct buffer_head *bh)
2609 static int bput_one(handle_t *handle, struct buffer_head *bh)
2615 static int __ext4_journalled_writepage(struct page *page,
2618 struct address_space *mapping = page->mapping;
2619 struct inode *inode = mapping->host;
2620 struct buffer_head *page_bufs;
2621 handle_t *handle = NULL;
2625 ClearPageChecked(page);
2626 page_bufs = page_buffers(page);
2628 walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one);
2629 /* As soon as we unlock the page, it can go away, but we have
2630 * references to buffers so we are safe */
2633 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2634 if (IS_ERR(handle)) {
2635 ret = PTR_ERR(handle);
2639 ret = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2640 do_journal_get_write_access);
2642 err = walk_page_buffers(handle, page_bufs, 0, len, NULL,
2646 err = ext4_journal_stop(handle);
2650 walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one);
2651 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2656 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
2657 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
2660 * Note that we don't need to start a transaction unless we're journaling data
2661 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2662 * need to file the inode to the transaction's list in ordered mode because if
2663 * we are writing back data added by write(), the inode is already there and if
2664 * we are writing back data modified via mmap(), noone guarantees in which
2665 * transaction the data will hit the disk. In case we are journaling data, we
2666 * cannot start transaction directly because transaction start ranks above page
2667 * lock so we have to do some magic.
2669 * This function can get called via...
2670 * - ext4_da_writepages after taking page lock (have journal handle)
2671 * - journal_submit_inode_data_buffers (no journal handle)
2672 * - shrink_page_list via pdflush (no journal handle)
2673 * - grab_page_cache when doing write_begin (have journal handle)
2675 * We don't do any block allocation in this function. If we have page with
2676 * multiple blocks we need to write those buffer_heads that are mapped. This
2677 * is important for mmaped based write. So if we do with blocksize 1K
2678 * truncate(f, 1024);
2679 * a = mmap(f, 0, 4096);
2681 * truncate(f, 4096);
2682 * we have in the page first buffer_head mapped via page_mkwrite call back
2683 * but other bufer_heads would be unmapped but dirty(dirty done via the
2684 * do_wp_page). So writepage should write the first block. If we modify
2685 * the mmap area beyond 1024 we will again get a page_fault and the
2686 * page_mkwrite callback will do the block allocation and mark the
2687 * buffer_heads mapped.
2689 * We redirty the page if we have any buffer_heads that is either delay or
2690 * unwritten in the page.
2692 * We can get recursively called as show below.
2694 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2697 * But since we don't do any block allocation we should not deadlock.
2698 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2700 static int ext4_writepage(struct page *page,
2701 struct writeback_control *wbc)
2703 int ret = 0, commit_write = 0;
2706 struct buffer_head *page_bufs = NULL;
2707 struct inode *inode = page->mapping->host;
2709 trace_ext4_writepage(inode, page);
2710 size = i_size_read(inode);
2711 if (page->index == size >> PAGE_CACHE_SHIFT)
2712 len = size & ~PAGE_CACHE_MASK;
2714 len = PAGE_CACHE_SIZE;
2717 * If the page does not have buffers (for whatever reason),
2718 * try to create them using __block_write_begin. If this
2719 * fails, redirty the page and move on.
2721 if (!page_has_buffers(page)) {
2722 if (__block_write_begin(page, 0, len,
2723 noalloc_get_block_write)) {
2725 redirty_page_for_writepage(wbc, page);
2731 page_bufs = page_buffers(page);
2732 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2733 ext4_bh_delay_or_unwritten)) {
2735 * We don't want to do block allocation, so redirty
2736 * the page and return. We may reach here when we do
2737 * a journal commit via journal_submit_inode_data_buffers.
2738 * We can also reach here via shrink_page_list
2743 /* now mark the buffer_heads as dirty and uptodate */
2744 block_commit_write(page, 0, len);
2746 if (PageChecked(page) && ext4_should_journal_data(inode))
2748 * It's mmapped pagecache. Add buffers and journal it. There
2749 * doesn't seem much point in redirtying the page here.
2751 return __ext4_journalled_writepage(page, len);
2753 if (buffer_uninit(page_bufs)) {
2754 ext4_set_bh_endio(page_bufs, inode);
2755 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2756 wbc, ext4_end_io_buffer_write);
2758 ret = block_write_full_page(page, noalloc_get_block_write,
2765 * This is called via ext4_da_writepages() to
2766 * calulate the total number of credits to reserve to fit
2767 * a single extent allocation into a single transaction,
2768 * ext4_da_writpeages() will loop calling this before
2769 * the block allocation.
2772 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2774 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2777 * With non-extent format the journal credit needed to
2778 * insert nrblocks contiguous block is dependent on
2779 * number of contiguous block. So we will limit
2780 * number of contiguous block to a sane value
2782 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
2783 (max_blocks > EXT4_MAX_TRANS_DATA))
2784 max_blocks = EXT4_MAX_TRANS_DATA;
2786 return ext4_chunk_trans_blocks(inode, max_blocks);
2790 * write_cache_pages_da - walk the list of dirty pages of the given
2791 * address space and call the callback function (which usually writes
2794 * This is a forked version of write_cache_pages(). Differences:
2795 * Range cyclic is ignored.
2796 * no_nrwrite_index_update is always presumed true
2798 static int write_cache_pages_da(struct address_space *mapping,
2799 struct writeback_control *wbc,
2800 struct mpage_da_data *mpd,
2801 pgoff_t *done_index)
2805 struct pagevec pvec;
2808 pgoff_t end; /* Inclusive */
2809 long nr_to_write = wbc->nr_to_write;
2812 pagevec_init(&pvec, 0);
2813 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2814 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2816 if (wbc->sync_mode == WB_SYNC_ALL)
2817 tag = PAGECACHE_TAG_TOWRITE;
2819 tag = PAGECACHE_TAG_DIRTY;
2821 *done_index = index;
2822 while (!done && (index <= end)) {
2825 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
2826 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2830 for (i = 0; i < nr_pages; i++) {
2831 struct page *page = pvec.pages[i];
2834 * At this point, the page may be truncated or
2835 * invalidated (changing page->mapping to NULL), or
2836 * even swizzled back from swapper_space to tmpfs file
2837 * mapping. However, page->index will not change
2838 * because we have a reference on the page.
2840 if (page->index > end) {
2845 *done_index = page->index + 1;
2850 * Page truncated or invalidated. We can freely skip it
2851 * then, even for data integrity operations: the page
2852 * has disappeared concurrently, so there could be no
2853 * real expectation of this data interity operation
2854 * even if there is now a new, dirty page at the same
2855 * pagecache address.
2857 if (unlikely(page->mapping != mapping)) {
2863 if (!PageDirty(page)) {
2864 /* someone wrote it for us */
2865 goto continue_unlock;
2868 if (PageWriteback(page)) {
2869 if (wbc->sync_mode != WB_SYNC_NONE)
2870 wait_on_page_writeback(page);
2872 goto continue_unlock;
2875 BUG_ON(PageWriteback(page));
2876 if (!clear_page_dirty_for_io(page))
2877 goto continue_unlock;
2879 ret = __mpage_da_writepage(page, wbc, mpd);
2880 if (unlikely(ret)) {
2881 if (ret == AOP_WRITEPAGE_ACTIVATE) {
2890 if (nr_to_write > 0) {
2892 if (nr_to_write == 0 &&
2893 wbc->sync_mode == WB_SYNC_NONE) {
2895 * We stop writing back only if we are
2896 * not doing integrity sync. In case of
2897 * integrity sync we have to keep going
2898 * because someone may be concurrently
2899 * dirtying pages, and we might have
2900 * synced a lot of newly appeared dirty
2901 * pages, but have not synced all of the
2909 pagevec_release(&pvec);
2916 static int ext4_da_writepages(struct address_space *mapping,
2917 struct writeback_control *wbc)
2920 int range_whole = 0;
2921 handle_t *handle = NULL;
2922 struct mpage_da_data mpd;
2923 struct inode *inode = mapping->host;
2924 int pages_written = 0;
2926 unsigned int max_pages;
2927 int range_cyclic, cycled = 1, io_done = 0;
2928 int needed_blocks, ret = 0;
2929 long desired_nr_to_write, nr_to_writebump = 0;
2930 loff_t range_start = wbc->range_start;
2931 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2932 pgoff_t done_index = 0;
2935 trace_ext4_da_writepages(inode, wbc);
2938 * No pages to write? This is mainly a kludge to avoid starting
2939 * a transaction for special inodes like journal inode on last iput()
2940 * because that could violate lock ordering on umount
2942 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2946 * If the filesystem has aborted, it is read-only, so return
2947 * right away instead of dumping stack traces later on that
2948 * will obscure the real source of the problem. We test
2949 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2950 * the latter could be true if the filesystem is mounted
2951 * read-only, and in that case, ext4_da_writepages should
2952 * *never* be called, so if that ever happens, we would want
2955 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
2958 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2961 range_cyclic = wbc->range_cyclic;
2962 if (wbc->range_cyclic) {
2963 index = mapping->writeback_index;
2966 wbc->range_start = index << PAGE_CACHE_SHIFT;
2967 wbc->range_end = LLONG_MAX;
2968 wbc->range_cyclic = 0;
2971 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2972 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2976 * This works around two forms of stupidity. The first is in
2977 * the writeback code, which caps the maximum number of pages
2978 * written to be 1024 pages. This is wrong on multiple
2979 * levels; different architectues have a different page size,
2980 * which changes the maximum amount of data which gets
2981 * written. Secondly, 4 megabytes is way too small. XFS
2982 * forces this value to be 16 megabytes by multiplying
2983 * nr_to_write parameter by four, and then relies on its
2984 * allocator to allocate larger extents to make them
2985 * contiguous. Unfortunately this brings us to the second
2986 * stupidity, which is that ext4's mballoc code only allocates
2987 * at most 2048 blocks. So we force contiguous writes up to
2988 * the number of dirty blocks in the inode, or
2989 * sbi->max_writeback_mb_bump whichever is smaller.
2991 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
2992 if (!range_cyclic && range_whole) {
2993 if (wbc->nr_to_write == LONG_MAX)
2994 desired_nr_to_write = wbc->nr_to_write;
2996 desired_nr_to_write = wbc->nr_to_write * 8;
2998 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
3000 if (desired_nr_to_write > max_pages)
3001 desired_nr_to_write = max_pages;
3003 if (wbc->nr_to_write < desired_nr_to_write) {
3004 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
3005 wbc->nr_to_write = desired_nr_to_write;
3009 mpd.inode = mapping->host;
3011 pages_skipped = wbc->pages_skipped;
3014 if (wbc->sync_mode == WB_SYNC_ALL)
3015 tag_pages_for_writeback(mapping, index, end);
3017 while (!ret && wbc->nr_to_write > 0) {
3020 * we insert one extent at a time. So we need
3021 * credit needed for single extent allocation.
3022 * journalled mode is currently not supported
3025 BUG_ON(ext4_should_journal_data(inode));
3026 needed_blocks = ext4_da_writepages_trans_blocks(inode);
3028 /* start a new transaction*/
3029 handle = ext4_journal_start(inode, needed_blocks);
3030 if (IS_ERR(handle)) {
3031 ret = PTR_ERR(handle);
3032 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
3033 "%ld pages, ino %lu; err %d", __func__,
3034 wbc->nr_to_write, inode->i_ino, ret);
3035 goto out_writepages;
3039 * Now call __mpage_da_writepage to find the next
3040 * contiguous region of logical blocks that need
3041 * blocks to be allocated by ext4. We don't actually
3042 * submit the blocks for I/O here, even though
3043 * write_cache_pages thinks it will, and will set the
3044 * pages as clean for write before calling
3045 * __mpage_da_writepage().
3053 mpd.pages_written = 0;
3055 ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index);
3057 * If we have a contiguous extent of pages and we
3058 * haven't done the I/O yet, map the blocks and submit
3061 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
3062 mpage_da_map_and_submit(&mpd);
3063 ret = MPAGE_DA_EXTENT_TAIL;
3065 trace_ext4_da_write_pages(inode, &mpd);
3066 wbc->nr_to_write -= mpd.pages_written;
3068 ext4_journal_stop(handle);
3070 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
3071 /* commit the transaction which would
3072 * free blocks released in the transaction
3075 jbd2_journal_force_commit_nested(sbi->s_journal);
3076 wbc->pages_skipped = pages_skipped;
3078 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
3080 * got one extent now try with
3083 pages_written += mpd.pages_written;
3084 wbc->pages_skipped = pages_skipped;
3087 } else if (wbc->nr_to_write)
3089 * There is no more writeout needed
3090 * or we requested for a noblocking writeout
3091 * and we found the device congested
3095 if (!io_done && !cycled) {
3098 wbc->range_start = index << PAGE_CACHE_SHIFT;
3099 wbc->range_end = mapping->writeback_index - 1;
3102 if (pages_skipped != wbc->pages_skipped)
3103 ext4_msg(inode->i_sb, KERN_CRIT,
3104 "This should not happen leaving %s "
3105 "with nr_to_write = %ld ret = %d",
3106 __func__, wbc->nr_to_write, ret);
3109 wbc->range_cyclic = range_cyclic;
3110 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3112 * set the writeback_index so that range_cyclic
3113 * mode will write it back later
3115 mapping->writeback_index = done_index;
3118 wbc->nr_to_write -= nr_to_writebump;
3119 wbc->range_start = range_start;
3120 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
3124 #define FALL_BACK_TO_NONDELALLOC 1
3125 static int ext4_nonda_switch(struct super_block *sb)
3127 s64 free_blocks, dirty_blocks;
3128 struct ext4_sb_info *sbi = EXT4_SB(sb);
3131 * switch to non delalloc mode if we are running low
3132 * on free block. The free block accounting via percpu
3133 * counters can get slightly wrong with percpu_counter_batch getting
3134 * accumulated on each CPU without updating global counters
3135 * Delalloc need an accurate free block accounting. So switch
3136 * to non delalloc when we are near to error range.
3138 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
3139 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
3140 if (2 * free_blocks < 3 * dirty_blocks ||
3141 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
3143 * free block count is less than 150% of dirty blocks
3144 * or free blocks is less than watermark
3149 * Even if we don't switch but are nearing capacity,
3150 * start pushing delalloc when 1/2 of free blocks are dirty.
3152 if (free_blocks < 2 * dirty_blocks)
3153 writeback_inodes_sb_if_idle(sb);
3158 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3159 loff_t pos, unsigned len, unsigned flags,
3160 struct page **pagep, void **fsdata)
3162 int ret, retries = 0;
3165 struct inode *inode = mapping->host;
3168 index = pos >> PAGE_CACHE_SHIFT;
3170 if (ext4_nonda_switch(inode->i_sb)) {
3171 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3172 return ext4_write_begin(file, mapping, pos,
3173 len, flags, pagep, fsdata);
3175 *fsdata = (void *)0;
3176 trace_ext4_da_write_begin(inode, pos, len, flags);
3179 * With delayed allocation, we don't log the i_disksize update
3180 * if there is delayed block allocation. But we still need
3181 * to journalling the i_disksize update if writes to the end
3182 * of file which has an already mapped buffer.
3184 handle = ext4_journal_start(inode, 1);
3185 if (IS_ERR(handle)) {
3186 ret = PTR_ERR(handle);
3189 /* We cannot recurse into the filesystem as the transaction is already
3191 flags |= AOP_FLAG_NOFS;
3193 page = grab_cache_page_write_begin(mapping, index, flags);
3195 ext4_journal_stop(handle);
3201 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3204 ext4_journal_stop(handle);
3205 page_cache_release(page);
3207 * block_write_begin may have instantiated a few blocks
3208 * outside i_size. Trim these off again. Don't need
3209 * i_size_read because we hold i_mutex.
3211 if (pos + len > inode->i_size)
3212 ext4_truncate_failed_write(inode);
3215 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3222 * Check if we should update i_disksize
3223 * when write to the end of file but not require block allocation
3225 static int ext4_da_should_update_i_disksize(struct page *page,
3226 unsigned long offset)
3228 struct buffer_head *bh;
3229 struct inode *inode = page->mapping->host;
3233 bh = page_buffers(page);
3234 idx = offset >> inode->i_blkbits;
3236 for (i = 0; i < idx; i++)
3237 bh = bh->b_this_page;
3239 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3244 static int ext4_da_write_end(struct file *file,
3245 struct address_space *mapping,
3246 loff_t pos, unsigned len, unsigned copied,
3247 struct page *page, void *fsdata)
3249 struct inode *inode = mapping->host;
3251 handle_t *handle = ext4_journal_current_handle();
3253 unsigned long start, end;
3254 int write_mode = (int)(unsigned long)fsdata;
3256 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
3257 if (ext4_should_order_data(inode)) {
3258 return ext4_ordered_write_end(file, mapping, pos,
3259 len, copied, page, fsdata);
3260 } else if (ext4_should_writeback_data(inode)) {
3261 return ext4_writeback_write_end(file, mapping, pos,
3262 len, copied, page, fsdata);
3268 trace_ext4_da_write_end(inode, pos, len, copied);
3269 start = pos & (PAGE_CACHE_SIZE - 1);
3270 end = start + copied - 1;
3273 * generic_write_end() will run mark_inode_dirty() if i_size
3274 * changes. So let's piggyback the i_disksize mark_inode_dirty
3278 new_i_size = pos + copied;
3279 if (new_i_size > EXT4_I(inode)->i_disksize) {
3280 if (ext4_da_should_update_i_disksize(page, end)) {
3281 down_write(&EXT4_I(inode)->i_data_sem);
3282 if (new_i_size > EXT4_I(inode)->i_disksize) {
3284 * Updating i_disksize when extending file
3285 * without needing block allocation
3287 if (ext4_should_order_data(inode))
3288 ret = ext4_jbd2_file_inode(handle,
3291 EXT4_I(inode)->i_disksize = new_i_size;
3293 up_write(&EXT4_I(inode)->i_data_sem);
3294 /* We need to mark inode dirty even if
3295 * new_i_size is less that inode->i_size
3296 * bu greater than i_disksize.(hint delalloc)
3298 ext4_mark_inode_dirty(handle, inode);
3301 ret2 = generic_write_end(file, mapping, pos, len, copied,
3306 ret2 = ext4_journal_stop(handle);
3310 return ret ? ret : copied;
3313 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
3316 * Drop reserved blocks
3318 BUG_ON(!PageLocked(page));
3319 if (!page_has_buffers(page))
3322 ext4_da_page_release_reservation(page, offset);
3325 ext4_invalidatepage(page, offset);
3331 * Force all delayed allocation blocks to be allocated for a given inode.
3333 int ext4_alloc_da_blocks(struct inode *inode)
3335 trace_ext4_alloc_da_blocks(inode);
3337 if (!EXT4_I(inode)->i_reserved_data_blocks &&
3338 !EXT4_I(inode)->i_reserved_meta_blocks)
3342 * We do something simple for now. The filemap_flush() will
3343 * also start triggering a write of the data blocks, which is
3344 * not strictly speaking necessary (and for users of
3345 * laptop_mode, not even desirable). However, to do otherwise
3346 * would require replicating code paths in:
3348 * ext4_da_writepages() ->
3349 * write_cache_pages() ---> (via passed in callback function)
3350 * __mpage_da_writepage() -->
3351 * mpage_add_bh_to_extent()
3352 * mpage_da_map_blocks()
3354 * The problem is that write_cache_pages(), located in
3355 * mm/page-writeback.c, marks pages clean in preparation for
3356 * doing I/O, which is not desirable if we're not planning on
3359 * We could call write_cache_pages(), and then redirty all of
3360 * the pages by calling redirty_page_for_writeback() but that
3361 * would be ugly in the extreme. So instead we would need to
3362 * replicate parts of the code in the above functions,
3363 * simplifying them becuase we wouldn't actually intend to
3364 * write out the pages, but rather only collect contiguous
3365 * logical block extents, call the multi-block allocator, and
3366 * then update the buffer heads with the block allocations.
3368 * For now, though, we'll cheat by calling filemap_flush(),
3369 * which will map the blocks, and start the I/O, but not
3370 * actually wait for the I/O to complete.
3372 return filemap_flush(inode->i_mapping);
3376 * bmap() is special. It gets used by applications such as lilo and by
3377 * the swapper to find the on-disk block of a specific piece of data.
3379 * Naturally, this is dangerous if the block concerned is still in the
3380 * journal. If somebody makes a swapfile on an ext4 data-journaling
3381 * filesystem and enables swap, then they may get a nasty shock when the
3382 * data getting swapped to that swapfile suddenly gets overwritten by
3383 * the original zero's written out previously to the journal and
3384 * awaiting writeback in the kernel's buffer cache.
3386 * So, if we see any bmap calls here on a modified, data-journaled file,
3387 * take extra steps to flush any blocks which might be in the cache.
3389 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3391 struct inode *inode = mapping->host;
3395 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3396 test_opt(inode->i_sb, DELALLOC)) {
3398 * With delalloc we want to sync the file
3399 * so that we can make sure we allocate
3402 filemap_write_and_wait(mapping);
3405 if (EXT4_JOURNAL(inode) &&
3406 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3408 * This is a REALLY heavyweight approach, but the use of
3409 * bmap on dirty files is expected to be extremely rare:
3410 * only if we run lilo or swapon on a freshly made file
3411 * do we expect this to happen.
3413 * (bmap requires CAP_SYS_RAWIO so this does not
3414 * represent an unprivileged user DOS attack --- we'd be
3415 * in trouble if mortal users could trigger this path at
3418 * NB. EXT4_STATE_JDATA is not set on files other than
3419 * regular files. If somebody wants to bmap a directory
3420 * or symlink and gets confused because the buffer
3421 * hasn't yet been flushed to disk, they deserve
3422 * everything they get.
3425 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3426 journal = EXT4_JOURNAL(inode);
3427 jbd2_journal_lock_updates(journal);
3428 err = jbd2_journal_flush(journal);
3429 jbd2_journal_unlock_updates(journal);
3435 return generic_block_bmap(mapping, block, ext4_get_block);
3438 static int ext4_readpage(struct file *file, struct page *page)
3440 return mpage_readpage(page, ext4_get_block);
3444 ext4_readpages(struct file *file, struct address_space *mapping,
3445 struct list_head *pages, unsigned nr_pages)
3447 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
3450 static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
3452 struct buffer_head *head, *bh;
3453 unsigned int curr_off = 0;
3455 if (!page_has_buffers(page))
3457 head = bh = page_buffers(page);
3459 if (offset <= curr_off && test_clear_buffer_uninit(bh)
3461 ext4_free_io_end(bh->b_private);
3462 bh->b_private = NULL;
3463 bh->b_end_io = NULL;
3465 curr_off = curr_off + bh->b_size;
3466 bh = bh->b_this_page;
3467 } while (bh != head);
3470 static void ext4_invalidatepage(struct page *page, unsigned long offset)
3472 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3475 * free any io_end structure allocated for buffers to be discarded
3477 if (ext4_should_dioread_nolock(page->mapping->host))
3478 ext4_invalidatepage_free_endio(page, offset);
3480 * If it's a full truncate we just forget about the pending dirtying
3483 ClearPageChecked(page);
3486 jbd2_journal_invalidatepage(journal, page, offset);
3488 block_invalidatepage(page, offset);
3491 static int ext4_releasepage(struct page *page, gfp_t wait)
3493 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3495 WARN_ON(PageChecked(page));
3496 if (!page_has_buffers(page))
3499 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3501 return try_to_free_buffers(page);
3505 * O_DIRECT for ext3 (or indirect map) based files
3507 * If the O_DIRECT write will extend the file then add this inode to the
3508 * orphan list. So recovery will truncate it back to the original size
3509 * if the machine crashes during the write.
3511 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3512 * crashes then stale disk data _may_ be exposed inside the file. But current
3513 * VFS code falls back into buffered path in that case so we are safe.
3515 static ssize_t ext4_ind_direct_IO(int rw, struct kiocb *iocb,
3516 const struct iovec *iov, loff_t offset,
3517 unsigned long nr_segs)
3519 struct file *file = iocb->ki_filp;
3520 struct inode *inode = file->f_mapping->host;
3521 struct ext4_inode_info *ei = EXT4_I(inode);
3525 size_t count = iov_length(iov, nr_segs);
3529 loff_t final_size = offset + count;
3531 if (final_size > inode->i_size) {
3532 /* Credits for sb + inode write */
3533 handle = ext4_journal_start(inode, 2);
3534 if (IS_ERR(handle)) {
3535 ret = PTR_ERR(handle);
3538 ret = ext4_orphan_add(handle, inode);
3540 ext4_journal_stop(handle);
3544 ei->i_disksize = inode->i_size;
3545 ext4_journal_stop(handle);
3550 if (rw == READ && ext4_should_dioread_nolock(inode))
3551 ret = __blockdev_direct_IO(rw, iocb, inode,
3552 inode->i_sb->s_bdev, iov,
3554 ext4_get_block, NULL, NULL, 0);
3556 ret = blockdev_direct_IO(rw, iocb, inode,
3557 inode->i_sb->s_bdev, iov,
3559 ext4_get_block, NULL);
3561 if (unlikely((rw & WRITE) && ret < 0)) {
3562 loff_t isize = i_size_read(inode);
3563 loff_t end = offset + iov_length(iov, nr_segs);
3566 vmtruncate(inode, isize);
3569 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3575 /* Credits for sb + inode write */
3576 handle = ext4_journal_start(inode, 2);
3577 if (IS_ERR(handle)) {
3578 /* This is really bad luck. We've written the data
3579 * but cannot extend i_size. Bail out and pretend
3580 * the write failed... */
3581 ret = PTR_ERR(handle);
3583 ext4_orphan_del(NULL, inode);
3588 ext4_orphan_del(handle, inode);
3590 loff_t end = offset + ret;
3591 if (end > inode->i_size) {
3592 ei->i_disksize = end;
3593 i_size_write(inode, end);
3595 * We're going to return a positive `ret'
3596 * here due to non-zero-length I/O, so there's
3597 * no way of reporting error returns from
3598 * ext4_mark_inode_dirty() to userspace. So
3601 ext4_mark_inode_dirty(handle, inode);
3604 err = ext4_journal_stop(handle);
3613 * ext4_get_block used when preparing for a DIO write or buffer write.
3614 * We allocate an uinitialized extent if blocks haven't been allocated.
3615 * The extent will be converted to initialized after the IO is complete.
3617 static int ext4_get_block_write(struct inode *inode, sector_t iblock,
3618 struct buffer_head *bh_result, int create)
3620 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3621 inode->i_ino, create);
3622 return _ext4_get_block(inode, iblock, bh_result,
3623 EXT4_GET_BLOCKS_IO_CREATE_EXT);
3626 static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3627 ssize_t size, void *private, int ret,
3630 ext4_io_end_t *io_end = iocb->private;
3631 struct workqueue_struct *wq;
3632 unsigned long flags;
3633 struct ext4_inode_info *ei;
3635 /* if not async direct IO or dio with 0 bytes write, just return */
3636 if (!io_end || !size)
3639 ext_debug("ext4_end_io_dio(): io_end 0x%p"
3640 "for inode %lu, iocb 0x%p, offset %llu, size %llu\n",
3641 iocb->private, io_end->inode->i_ino, iocb, offset,
3644 /* if not aio dio with unwritten extents, just free io and return */
3645 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
3646 ext4_free_io_end(io_end);
3647 iocb->private = NULL;
3650 aio_complete(iocb, ret, 0);
3654 io_end->offset = offset;
3655 io_end->size = size;
3657 io_end->iocb = iocb;
3658 io_end->result = ret;
3660 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
3662 /* Add the io_end to per-inode completed aio dio list*/
3663 ei = EXT4_I(io_end->inode);
3664 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
3665 list_add_tail(&io_end->list, &ei->i_completed_io_list);
3666 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
3668 /* queue the work to convert unwritten extents to written */
3669 queue_work(wq, &io_end->work);
3670 iocb->private = NULL;
3673 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
3675 ext4_io_end_t *io_end = bh->b_private;
3676 struct workqueue_struct *wq;
3677 struct inode *inode;
3678 unsigned long flags;
3680 if (!test_clear_buffer_uninit(bh) || !io_end)
3683 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
3684 printk("sb umounted, discard end_io request for inode %lu\n",
3685 io_end->inode->i_ino);
3686 ext4_free_io_end(io_end);
3690 io_end->flag = EXT4_IO_END_UNWRITTEN;
3691 inode = io_end->inode;
3693 /* Add the io_end to per-inode completed io list*/
3694 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
3695 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
3696 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
3698 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
3699 /* queue the work to convert unwritten extents to written */
3700 queue_work(wq, &io_end->work);
3702 bh->b_private = NULL;
3703 bh->b_end_io = NULL;
3704 clear_buffer_uninit(bh);
3705 end_buffer_async_write(bh, uptodate);
3708 static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3710 ext4_io_end_t *io_end;
3711 struct page *page = bh->b_page;
3712 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3713 size_t size = bh->b_size;
3716 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3718 if (printk_ratelimit())
3719 printk(KERN_WARNING "%s: allocation fail\n", __func__);
3723 io_end->offset = offset;
3724 io_end->size = size;
3726 * We need to hold a reference to the page to make sure it
3727 * doesn't get evicted before ext4_end_io_work() has a chance
3728 * to convert the extent from written to unwritten.
3730 io_end->page = page;
3731 get_page(io_end->page);
3733 bh->b_private = io_end;
3734 bh->b_end_io = ext4_end_io_buffer_write;
3739 * For ext4 extent files, ext4 will do direct-io write to holes,
3740 * preallocated extents, and those write extend the file, no need to
3741 * fall back to buffered IO.
3743 * For holes, we fallocate those blocks, mark them as unintialized
3744 * If those blocks were preallocated, we mark sure they are splited, but
3745 * still keep the range to write as unintialized.
3747 * The unwrritten extents will be converted to written when DIO is completed.
3748 * For async direct IO, since the IO may still pending when return, we
3749 * set up an end_io call back function, which will do the convertion
3750 * when async direct IO completed.
3752 * If the O_DIRECT write will extend the file then add this inode to the
3753 * orphan list. So recovery will truncate it back to the original size
3754 * if the machine crashes during the write.
3757 static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3758 const struct iovec *iov, loff_t offset,
3759 unsigned long nr_segs)
3761 struct file *file = iocb->ki_filp;
3762 struct inode *inode = file->f_mapping->host;
3764 size_t count = iov_length(iov, nr_segs);
3766 loff_t final_size = offset + count;
3767 if (rw == WRITE && final_size <= inode->i_size) {
3769 * We could direct write to holes and fallocate.
3771 * Allocated blocks to fill the hole are marked as uninitialized
3772 * to prevent paralel buffered read to expose the stale data
3773 * before DIO complete the data IO.
3775 * As to previously fallocated extents, ext4 get_block
3776 * will just simply mark the buffer mapped but still
3777 * keep the extents uninitialized.
3779 * for non AIO case, we will convert those unwritten extents
3780 * to written after return back from blockdev_direct_IO.
3782 * for async DIO, the conversion needs to be defered when
3783 * the IO is completed. The ext4 end_io callback function
3784 * will be called to take care of the conversion work.
3785 * Here for async case, we allocate an io_end structure to
3788 iocb->private = NULL;
3789 EXT4_I(inode)->cur_aio_dio = NULL;
3790 if (!is_sync_kiocb(iocb)) {
3791 iocb->private = ext4_init_io_end(inode, GFP_NOFS);
3795 * we save the io structure for current async
3796 * direct IO, so that later ext4_map_blocks()
3797 * could flag the io structure whether there
3798 * is a unwritten extents needs to be converted
3799 * when IO is completed.
3801 EXT4_I(inode)->cur_aio_dio = iocb->private;
3804 ret = blockdev_direct_IO(rw, iocb, inode,
3805 inode->i_sb->s_bdev, iov,
3807 ext4_get_block_write,
3810 EXT4_I(inode)->cur_aio_dio = NULL;
3812 * The io_end structure takes a reference to the inode,
3813 * that structure needs to be destroyed and the
3814 * reference to the inode need to be dropped, when IO is
3815 * complete, even with 0 byte write, or failed.
3817 * In the successful AIO DIO case, the io_end structure will be
3818 * desctroyed and the reference to the inode will be dropped
3819 * after the end_io call back function is called.
3821 * In the case there is 0 byte write, or error case, since
3822 * VFS direct IO won't invoke the end_io call back function,
3823 * we need to free the end_io structure here.
3825 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3826 ext4_free_io_end(iocb->private);
3827 iocb->private = NULL;
3828 } else if (ret > 0 && ext4_test_inode_state(inode,
3829 EXT4_STATE_DIO_UNWRITTEN)) {
3832 * for non AIO case, since the IO is already
3833 * completed, we could do the convertion right here
3835 err = ext4_convert_unwritten_extents(inode,
3839 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3844 /* for write the the end of file case, we fall back to old way */
3845 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3848 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3849 const struct iovec *iov, loff_t offset,
3850 unsigned long nr_segs)
3852 struct file *file = iocb->ki_filp;
3853 struct inode *inode = file->f_mapping->host;
3855 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3856 return ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3858 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3862 * Pages can be marked dirty completely asynchronously from ext4's journalling
3863 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3864 * much here because ->set_page_dirty is called under VFS locks. The page is
3865 * not necessarily locked.
3867 * We cannot just dirty the page and leave attached buffers clean, because the
3868 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3869 * or jbddirty because all the journalling code will explode.
3871 * So what we do is to mark the page "pending dirty" and next time writepage
3872 * is called, propagate that into the buffers appropriately.
3874 static int ext4_journalled_set_page_dirty(struct page *page)
3876 SetPageChecked(page);
3877 return __set_page_dirty_nobuffers(page);
3880 static const struct address_space_operations ext4_ordered_aops = {
3881 .readpage = ext4_readpage,
3882 .readpages = ext4_readpages,
3883 .writepage = ext4_writepage,
3884 .sync_page = block_sync_page,
3885 .write_begin = ext4_write_begin,
3886 .write_end = ext4_ordered_write_end,
3888 .invalidatepage = ext4_invalidatepage,
3889 .releasepage = ext4_releasepage,
3890 .direct_IO = ext4_direct_IO,
3891 .migratepage = buffer_migrate_page,
3892 .is_partially_uptodate = block_is_partially_uptodate,
3893 .error_remove_page = generic_error_remove_page,
3896 static const struct address_space_operations ext4_writeback_aops = {
3897 .readpage = ext4_readpage,
3898 .readpages = ext4_readpages,
3899 .writepage = ext4_writepage,
3900 .sync_page = block_sync_page,
3901 .write_begin = ext4_write_begin,
3902 .write_end = ext4_writeback_write_end,
3904 .invalidatepage = ext4_invalidatepage,
3905 .releasepage = ext4_releasepage,
3906 .direct_IO = ext4_direct_IO,
3907 .migratepage = buffer_migrate_page,
3908 .is_partially_uptodate = block_is_partially_uptodate,
3909 .error_remove_page = generic_error_remove_page,
3912 static const struct address_space_operations ext4_journalled_aops = {
3913 .readpage = ext4_readpage,
3914 .readpages = ext4_readpages,
3915 .writepage = ext4_writepage,
3916 .sync_page = block_sync_page,
3917 .write_begin = ext4_write_begin,
3918 .write_end = ext4_journalled_write_end,
3919 .set_page_dirty = ext4_journalled_set_page_dirty,
3921 .invalidatepage = ext4_invalidatepage,
3922 .releasepage = ext4_releasepage,
3923 .is_partially_uptodate = block_is_partially_uptodate,
3924 .error_remove_page = generic_error_remove_page,
3927 static const struct address_space_operations ext4_da_aops = {
3928 .readpage = ext4_readpage,
3929 .readpages = ext4_readpages,
3930 .writepage = ext4_writepage,
3931 .writepages = ext4_da_writepages,
3932 .sync_page = block_sync_page,
3933 .write_begin = ext4_da_write_begin,
3934 .write_end = ext4_da_write_end,
3936 .invalidatepage = ext4_da_invalidatepage,
3937 .releasepage = ext4_releasepage,
3938 .direct_IO = ext4_direct_IO,
3939 .migratepage = buffer_migrate_page,
3940 .is_partially_uptodate = block_is_partially_uptodate,
3941 .error_remove_page = generic_error_remove_page,
3944 void ext4_set_aops(struct inode *inode)
3946 if (ext4_should_order_data(inode) &&
3947 test_opt(inode->i_sb, DELALLOC))
3948 inode->i_mapping->a_ops = &ext4_da_aops;
3949 else if (ext4_should_order_data(inode))
3950 inode->i_mapping->a_ops = &ext4_ordered_aops;
3951 else if (ext4_should_writeback_data(inode) &&
3952 test_opt(inode->i_sb, DELALLOC))
3953 inode->i_mapping->a_ops = &ext4_da_aops;
3954 else if (ext4_should_writeback_data(inode))
3955 inode->i_mapping->a_ops = &ext4_writeback_aops;
3957 inode->i_mapping->a_ops = &ext4_journalled_aops;
3961 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3962 * up to the end of the block which corresponds to `from'.
3963 * This required during truncate. We need to physically zero the tail end
3964 * of that block so it doesn't yield old data if the file is later grown.
3966 int ext4_block_truncate_page(handle_t *handle,
3967 struct address_space *mapping, loff_t from)
3969 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3970 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3971 unsigned blocksize, length, pos;
3973 struct inode *inode = mapping->host;
3974 struct buffer_head *bh;
3978 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3979 mapping_gfp_mask(mapping) & ~__GFP_FS);
3983 blocksize = inode->i_sb->s_blocksize;
3984 length = blocksize - (offset & (blocksize - 1));
3985 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3987 if (!page_has_buffers(page))
3988 create_empty_buffers(page, blocksize, 0);
3990 /* Find the buffer that contains "offset" */
3991 bh = page_buffers(page);
3993 while (offset >= pos) {
3994 bh = bh->b_this_page;
4000 if (buffer_freed(bh)) {
4001 BUFFER_TRACE(bh, "freed: skip");
4005 if (!buffer_mapped(bh)) {
4006 BUFFER_TRACE(bh, "unmapped");
4007 ext4_get_block(inode, iblock, bh, 0);
4008 /* unmapped? It's a hole - nothing to do */
4009 if (!buffer_mapped(bh)) {
4010 BUFFER_TRACE(bh, "still unmapped");
4015 /* Ok, it's mapped. Make sure it's up-to-date */
4016 if (PageUptodate(page))
4017 set_buffer_uptodate(bh);
4019 if (!buffer_uptodate(bh)) {
4021 ll_rw_block(READ, 1, &bh);
4023 /* Uhhuh. Read error. Complain and punt. */
4024 if (!buffer_uptodate(bh))
4028 if (ext4_should_journal_data(inode)) {
4029 BUFFER_TRACE(bh, "get write access");
4030 err = ext4_journal_get_write_access(handle, bh);
4035 zero_user(page, offset, length);
4037 BUFFER_TRACE(bh, "zeroed end of block");
4040 if (ext4_should_journal_data(inode)) {
4041 err = ext4_handle_dirty_metadata(handle, inode, bh);
4043 if (ext4_should_order_data(inode))
4044 err = ext4_jbd2_file_inode(handle, inode);
4045 mark_buffer_dirty(bh);
4050 page_cache_release(page);
4055 * Probably it should be a library function... search for first non-zero word
4056 * or memcmp with zero_page, whatever is better for particular architecture.
4059 static inline int all_zeroes(__le32 *p, __le32 *q)
4068 * ext4_find_shared - find the indirect blocks for partial truncation.
4069 * @inode: inode in question
4070 * @depth: depth of the affected branch
4071 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
4072 * @chain: place to store the pointers to partial indirect blocks
4073 * @top: place to the (detached) top of branch
4075 * This is a helper function used by ext4_truncate().
4077 * When we do truncate() we may have to clean the ends of several
4078 * indirect blocks but leave the blocks themselves alive. Block is
4079 * partially truncated if some data below the new i_size is refered
4080 * from it (and it is on the path to the first completely truncated
4081 * data block, indeed). We have to free the top of that path along
4082 * with everything to the right of the path. Since no allocation
4083 * past the truncation point is possible until ext4_truncate()
4084 * finishes, we may safely do the latter, but top of branch may
4085 * require special attention - pageout below the truncation point
4086 * might try to populate it.
4088 * We atomically detach the top of branch from the tree, store the
4089 * block number of its root in *@top, pointers to buffer_heads of
4090 * partially truncated blocks - in @chain[].bh and pointers to
4091 * their last elements that should not be removed - in
4092 * @chain[].p. Return value is the pointer to last filled element
4095 * The work left to caller to do the actual freeing of subtrees:
4096 * a) free the subtree starting from *@top
4097 * b) free the subtrees whose roots are stored in
4098 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
4099 * c) free the subtrees growing from the inode past the @chain[0].
4100 * (no partially truncated stuff there). */
4102 static Indirect *ext4_find_shared(struct inode *inode, int depth,
4103 ext4_lblk_t offsets[4], Indirect chain[4],
4106 Indirect *partial, *p;
4110 /* Make k index the deepest non-null offset + 1 */
4111 for (k = depth; k > 1 && !offsets[k-1]; k--)
4113 partial = ext4_get_branch(inode, k, offsets, chain, &err);
4114 /* Writer: pointers */
4116 partial = chain + k-1;
4118 * If the branch acquired continuation since we've looked at it -
4119 * fine, it should all survive and (new) top doesn't belong to us.
4121 if (!partial->key && *partial->p)
4124 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
4127 * OK, we've found the last block that must survive. The rest of our
4128 * branch should be detached before unlocking. However, if that rest
4129 * of branch is all ours and does not grow immediately from the inode
4130 * it's easier to cheat and just decrement partial->p.
4132 if (p == chain + k - 1 && p > chain) {
4136 /* Nope, don't do this in ext4. Must leave the tree intact */
4143 while (partial > p) {
4144 brelse(partial->bh);
4152 * Zero a number of block pointers in either an inode or an indirect block.
4153 * If we restart the transaction we must again get write access to the
4154 * indirect block for further modification.
4156 * We release `count' blocks on disk, but (last - first) may be greater
4157 * than `count' because there can be holes in there.
4159 static int ext4_clear_blocks(handle_t *handle, struct inode *inode,
4160 struct buffer_head *bh,
4161 ext4_fsblk_t block_to_free,
4162 unsigned long count, __le32 *first,
4166 int flags = EXT4_FREE_BLOCKS_FORGET | EXT4_FREE_BLOCKS_VALIDATED;
4168 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
4169 flags |= EXT4_FREE_BLOCKS_METADATA;
4171 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), block_to_free,
4173 EXT4_ERROR_INODE(inode, "attempt to clear invalid "
4174 "blocks %llu len %lu",
4175 (unsigned long long) block_to_free, count);
4179 if (try_to_extend_transaction(handle, inode)) {
4181 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
4182 ext4_handle_dirty_metadata(handle, inode, bh);
4184 ext4_mark_inode_dirty(handle, inode);
4185 ext4_truncate_restart_trans(handle, inode,
4186 blocks_for_truncate(inode));
4188 BUFFER_TRACE(bh, "retaking write access");
4189 ext4_journal_get_write_access(handle, bh);
4193 for (p = first; p < last; p++)
4196 ext4_free_blocks(handle, inode, 0, block_to_free, count, flags);
4201 * ext4_free_data - free a list of data blocks
4202 * @handle: handle for this transaction
4203 * @inode: inode we are dealing with
4204 * @this_bh: indirect buffer_head which contains *@first and *@last
4205 * @first: array of block numbers
4206 * @last: points immediately past the end of array
4208 * We are freeing all blocks refered from that array (numbers are stored as
4209 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
4211 * We accumulate contiguous runs of blocks to free. Conveniently, if these
4212 * blocks are contiguous then releasing them at one time will only affect one
4213 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
4214 * actually use a lot of journal space.
4216 * @this_bh will be %NULL if @first and @last point into the inode's direct
4219 static void ext4_free_data(handle_t *handle, struct inode *inode,
4220 struct buffer_head *this_bh,
4221 __le32 *first, __le32 *last)
4223 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
4224 unsigned long count = 0; /* Number of blocks in the run */
4225 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
4228 ext4_fsblk_t nr; /* Current block # */
4229 __le32 *p; /* Pointer into inode/ind
4230 for current block */
4233 if (this_bh) { /* For indirect block */
4234 BUFFER_TRACE(this_bh, "get_write_access");
4235 err = ext4_journal_get_write_access(handle, this_bh);
4236 /* Important: if we can't update the indirect pointers
4237 * to the blocks, we can't free them. */
4242 for (p = first; p < last; p++) {
4243 nr = le32_to_cpu(*p);
4245 /* accumulate blocks to free if they're contiguous */
4248 block_to_free_p = p;
4250 } else if (nr == block_to_free + count) {
4253 if (ext4_clear_blocks(handle, inode, this_bh,
4254 block_to_free, count,
4255 block_to_free_p, p))
4258 block_to_free_p = p;
4265 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
4266 count, block_to_free_p, p);
4269 BUFFER_TRACE(this_bh, "call ext4_handle_dirty_metadata");
4272 * The buffer head should have an attached journal head at this
4273 * point. However, if the data is corrupted and an indirect
4274 * block pointed to itself, it would have been detached when
4275 * the block was cleared. Check for this instead of OOPSing.
4277 if ((EXT4_JOURNAL(inode) == NULL) || bh2jh(this_bh))
4278 ext4_handle_dirty_metadata(handle, inode, this_bh);
4280 EXT4_ERROR_INODE(inode,
4281 "circular indirect block detected at "
4283 (unsigned long long) this_bh->b_blocknr);
4288 * ext4_free_branches - free an array of branches
4289 * @handle: JBD handle for this transaction
4290 * @inode: inode we are dealing with
4291 * @parent_bh: the buffer_head which contains *@first and *@last
4292 * @first: array of block numbers
4293 * @last: pointer immediately past the end of array
4294 * @depth: depth of the branches to free
4296 * We are freeing all blocks refered from these branches (numbers are
4297 * stored as little-endian 32-bit) and updating @inode->i_blocks
4300 static void ext4_free_branches(handle_t *handle, struct inode *inode,
4301 struct buffer_head *parent_bh,
4302 __le32 *first, __le32 *last, int depth)
4307 if (ext4_handle_is_aborted(handle))
4311 struct buffer_head *bh;
4312 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4314 while (--p >= first) {
4315 nr = le32_to_cpu(*p);
4317 continue; /* A hole */
4319 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb),
4321 EXT4_ERROR_INODE(inode,
4322 "invalid indirect mapped "
4323 "block %lu (level %d)",
4324 (unsigned long) nr, depth);
4328 /* Go read the buffer for the next level down */
4329 bh = sb_bread(inode->i_sb, nr);
4332 * A read failure? Report error and clear slot
4336 EXT4_ERROR_INODE_BLOCK(inode, nr,
4341 /* This zaps the entire block. Bottom up. */
4342 BUFFER_TRACE(bh, "free child branches");
4343 ext4_free_branches(handle, inode, bh,
4344 (__le32 *) bh->b_data,
4345 (__le32 *) bh->b_data + addr_per_block,
4349 * Everything below this this pointer has been
4350 * released. Now let this top-of-subtree go.
4352 * We want the freeing of this indirect block to be
4353 * atomic in the journal with the updating of the
4354 * bitmap block which owns it. So make some room in
4357 * We zero the parent pointer *after* freeing its
4358 * pointee in the bitmaps, so if extend_transaction()
4359 * for some reason fails to put the bitmap changes and
4360 * the release into the same transaction, recovery
4361 * will merely complain about releasing a free block,
4362 * rather than leaking blocks.
4364 if (ext4_handle_is_aborted(handle))
4366 if (try_to_extend_transaction(handle, inode)) {
4367 ext4_mark_inode_dirty(handle, inode);
4368 ext4_truncate_restart_trans(handle, inode,
4369 blocks_for_truncate(inode));
4373 * The forget flag here is critical because if
4374 * we are journaling (and not doing data
4375 * journaling), we have to make sure a revoke
4376 * record is written to prevent the journal
4377 * replay from overwriting the (former)
4378 * indirect block if it gets reallocated as a
4379 * data block. This must happen in the same
4380 * transaction where the data blocks are
4383 ext4_free_blocks(handle, inode, 0, nr, 1,
4384 EXT4_FREE_BLOCKS_METADATA|
4385 EXT4_FREE_BLOCKS_FORGET);
4389 * The block which we have just freed is
4390 * pointed to by an indirect block: journal it
4392 BUFFER_TRACE(parent_bh, "get_write_access");
4393 if (!ext4_journal_get_write_access(handle,
4396 BUFFER_TRACE(parent_bh,
4397 "call ext4_handle_dirty_metadata");
4398 ext4_handle_dirty_metadata(handle,
4405 /* We have reached the bottom of the tree. */
4406 BUFFER_TRACE(parent_bh, "free data blocks");
4407 ext4_free_data(handle, inode, parent_bh, first, last);
4411 int ext4_can_truncate(struct inode *inode)
4413 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4415 if (S_ISREG(inode->i_mode))
4417 if (S_ISDIR(inode->i_mode))
4419 if (S_ISLNK(inode->i_mode))
4420 return !ext4_inode_is_fast_symlink(inode);
4427 * We block out ext4_get_block() block instantiations across the entire
4428 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4429 * simultaneously on behalf of the same inode.
4431 * As we work through the truncate and commmit bits of it to the journal there
4432 * is one core, guiding principle: the file's tree must always be consistent on
4433 * disk. We must be able to restart the truncate after a crash.
4435 * The file's tree may be transiently inconsistent in memory (although it
4436 * probably isn't), but whenever we close off and commit a journal transaction,
4437 * the contents of (the filesystem + the journal) must be consistent and
4438 * restartable. It's pretty simple, really: bottom up, right to left (although
4439 * left-to-right works OK too).
4441 * Note that at recovery time, journal replay occurs *before* the restart of
4442 * truncate against the orphan inode list.
4444 * The committed inode has the new, desired i_size (which is the same as
4445 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4446 * that this inode's truncate did not complete and it will again call
4447 * ext4_truncate() to have another go. So there will be instantiated blocks
4448 * to the right of the truncation point in a crashed ext4 filesystem. But
4449 * that's fine - as long as they are linked from the inode, the post-crash
4450 * ext4_truncate() run will find them and release them.
4452 void ext4_truncate(struct inode *inode)
4455 struct ext4_inode_info *ei = EXT4_I(inode);
4456 __le32 *i_data = ei->i_data;
4457 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
4458 struct address_space *mapping = inode->i_mapping;
4459 ext4_lblk_t offsets[4];
4464 ext4_lblk_t last_block;
4465 unsigned blocksize = inode->i_sb->s_blocksize;
4467 if (!ext4_can_truncate(inode))
4470 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4472 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4473 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4475 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4476 ext4_ext_truncate(inode);
4480 handle = start_transaction(inode);
4482 return; /* AKPM: return what? */
4484 last_block = (inode->i_size + blocksize-1)
4485 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
4487 if (inode->i_size & (blocksize - 1))
4488 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
4491 n = ext4_block_to_path(inode, last_block, offsets, NULL);
4493 goto out_stop; /* error */
4496 * OK. This truncate is going to happen. We add the inode to the
4497 * orphan list, so that if this truncate spans multiple transactions,
4498 * and we crash, we will resume the truncate when the filesystem
4499 * recovers. It also marks the inode dirty, to catch the new size.
4501 * Implication: the file must always be in a sane, consistent
4502 * truncatable state while each transaction commits.
4504 if (ext4_orphan_add(handle, inode))
4508 * From here we block out all ext4_get_block() callers who want to
4509 * modify the block allocation tree.
4511 down_write(&ei->i_data_sem);
4513 ext4_discard_preallocations(inode);
4516 * The orphan list entry will now protect us from any crash which
4517 * occurs before the truncate completes, so it is now safe to propagate
4518 * the new, shorter inode size (held for now in i_size) into the
4519 * on-disk inode. We do this via i_disksize, which is the value which
4520 * ext4 *really* writes onto the disk inode.
4522 ei->i_disksize = inode->i_size;
4524 if (n == 1) { /* direct blocks */
4525 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
4526 i_data + EXT4_NDIR_BLOCKS);
4530 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
4531 /* Kill the top of shared branch (not detached) */
4533 if (partial == chain) {
4534 /* Shared branch grows from the inode */
4535 ext4_free_branches(handle, inode, NULL,
4536 &nr, &nr+1, (chain+n-1) - partial);
4539 * We mark the inode dirty prior to restart,
4540 * and prior to stop. No need for it here.
4543 /* Shared branch grows from an indirect block */
4544 BUFFER_TRACE(partial->bh, "get_write_access");
4545 ext4_free_branches(handle, inode, partial->bh,
4547 partial->p+1, (chain+n-1) - partial);
4550 /* Clear the ends of indirect blocks on the shared branch */
4551 while (partial > chain) {
4552 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
4553 (__le32*)partial->bh->b_data+addr_per_block,
4554 (chain+n-1) - partial);
4555 BUFFER_TRACE(partial->bh, "call brelse");
4556 brelse(partial->bh);
4560 /* Kill the remaining (whole) subtrees */
4561 switch (offsets[0]) {
4563 nr = i_data[EXT4_IND_BLOCK];
4565 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
4566 i_data[EXT4_IND_BLOCK] = 0;
4568 case EXT4_IND_BLOCK:
4569 nr = i_data[EXT4_DIND_BLOCK];
4571 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
4572 i_data[EXT4_DIND_BLOCK] = 0;
4574 case EXT4_DIND_BLOCK:
4575 nr = i_data[EXT4_TIND_BLOCK];
4577 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
4578 i_data[EXT4_TIND_BLOCK] = 0;
4580 case EXT4_TIND_BLOCK:
4584 up_write(&ei->i_data_sem);
4585 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
4586 ext4_mark_inode_dirty(handle, inode);
4589 * In a multi-transaction truncate, we only make the final transaction
4593 ext4_handle_sync(handle);
4596 * If this was a simple ftruncate(), and the file will remain alive
4597 * then we need to clear up the orphan record which we created above.
4598 * However, if this was a real unlink then we were called by
4599 * ext4_delete_inode(), and we allow that function to clean up the
4600 * orphan info for us.
4603 ext4_orphan_del(handle, inode);
4605 ext4_journal_stop(handle);
4609 * ext4_get_inode_loc returns with an extra refcount against the inode's
4610 * underlying buffer_head on success. If 'in_mem' is true, we have all
4611 * data in memory that is needed to recreate the on-disk version of this
4614 static int __ext4_get_inode_loc(struct inode *inode,
4615 struct ext4_iloc *iloc, int in_mem)
4617 struct ext4_group_desc *gdp;
4618 struct buffer_head *bh;
4619 struct super_block *sb = inode->i_sb;
4621 int inodes_per_block, inode_offset;
4624 if (!ext4_valid_inum(sb, inode->i_ino))
4627 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4628 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4633 * Figure out the offset within the block group inode table
4635 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
4636 inode_offset = ((inode->i_ino - 1) %
4637 EXT4_INODES_PER_GROUP(sb));
4638 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4639 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4641 bh = sb_getblk(sb, block);
4643 EXT4_ERROR_INODE_BLOCK(inode, block,
4644 "unable to read itable block");
4647 if (!buffer_uptodate(bh)) {
4651 * If the buffer has the write error flag, we have failed
4652 * to write out another inode in the same block. In this
4653 * case, we don't have to read the block because we may
4654 * read the old inode data successfully.
4656 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4657 set_buffer_uptodate(bh);
4659 if (buffer_uptodate(bh)) {
4660 /* someone brought it uptodate while we waited */
4666 * If we have all information of the inode in memory and this
4667 * is the only valid inode in the block, we need not read the
4671 struct buffer_head *bitmap_bh;
4674 start = inode_offset & ~(inodes_per_block - 1);
4676 /* Is the inode bitmap in cache? */
4677 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4682 * If the inode bitmap isn't in cache then the
4683 * optimisation may end up performing two reads instead
4684 * of one, so skip it.
4686 if (!buffer_uptodate(bitmap_bh)) {
4690 for (i = start; i < start + inodes_per_block; i++) {
4691 if (i == inode_offset)
4693 if (ext4_test_bit(i, bitmap_bh->b_data))
4697 if (i == start + inodes_per_block) {
4698 /* all other inodes are free, so skip I/O */
4699 memset(bh->b_data, 0, bh->b_size);
4700 set_buffer_uptodate(bh);
4708 * If we need to do any I/O, try to pre-readahead extra
4709 * blocks from the inode table.
4711 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4712 ext4_fsblk_t b, end, table;
4715 table = ext4_inode_table(sb, gdp);
4716 /* s_inode_readahead_blks is always a power of 2 */
4717 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
4720 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
4721 num = EXT4_INODES_PER_GROUP(sb);
4722 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4723 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
4724 num -= ext4_itable_unused_count(sb, gdp);
4725 table += num / inodes_per_block;
4729 sb_breadahead(sb, b++);
4733 * There are other valid inodes in the buffer, this inode
4734 * has in-inode xattrs, or we don't have this inode in memory.
4735 * Read the block from disk.
4738 bh->b_end_io = end_buffer_read_sync;
4739 submit_bh(READ_META, bh);
4741 if (!buffer_uptodate(bh)) {
4742 EXT4_ERROR_INODE_BLOCK(inode, block,
4743 "unable to read itable block");
4753 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4755 /* We have all inode data except xattrs in memory here. */
4756 return __ext4_get_inode_loc(inode, iloc,
4757 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4760 void ext4_set_inode_flags(struct inode *inode)
4762 unsigned int flags = EXT4_I(inode)->i_flags;
4764 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4765 if (flags & EXT4_SYNC_FL)
4766 inode->i_flags |= S_SYNC;
4767 if (flags & EXT4_APPEND_FL)
4768 inode->i_flags |= S_APPEND;
4769 if (flags & EXT4_IMMUTABLE_FL)
4770 inode->i_flags |= S_IMMUTABLE;
4771 if (flags & EXT4_NOATIME_FL)
4772 inode->i_flags |= S_NOATIME;
4773 if (flags & EXT4_DIRSYNC_FL)
4774 inode->i_flags |= S_DIRSYNC;
4777 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4778 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4780 unsigned int vfs_fl;
4781 unsigned long old_fl, new_fl;
4784 vfs_fl = ei->vfs_inode.i_flags;
4785 old_fl = ei->i_flags;
4786 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4787 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
4789 if (vfs_fl & S_SYNC)
4790 new_fl |= EXT4_SYNC_FL;
4791 if (vfs_fl & S_APPEND)
4792 new_fl |= EXT4_APPEND_FL;
4793 if (vfs_fl & S_IMMUTABLE)
4794 new_fl |= EXT4_IMMUTABLE_FL;
4795 if (vfs_fl & S_NOATIME)
4796 new_fl |= EXT4_NOATIME_FL;
4797 if (vfs_fl & S_DIRSYNC)
4798 new_fl |= EXT4_DIRSYNC_FL;
4799 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
4802 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4803 struct ext4_inode_info *ei)
4806 struct inode *inode = &(ei->vfs_inode);
4807 struct super_block *sb = inode->i_sb;
4809 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4810 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4811 /* we are using combined 48 bit field */
4812 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4813 le32_to_cpu(raw_inode->i_blocks_lo);
4814 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4815 /* i_blocks represent file system block size */
4816 return i_blocks << (inode->i_blkbits - 9);
4821 return le32_to_cpu(raw_inode->i_blocks_lo);
4825 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4827 struct ext4_iloc iloc;
4828 struct ext4_inode *raw_inode;
4829 struct ext4_inode_info *ei;
4830 struct inode *inode;
4831 journal_t *journal = EXT4_SB(sb)->s_journal;
4835 inode = iget_locked(sb, ino);
4837 return ERR_PTR(-ENOMEM);
4838 if (!(inode->i_state & I_NEW))
4844 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4847 raw_inode = ext4_raw_inode(&iloc);
4848 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4849 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4850 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4851 if (!(test_opt(inode->i_sb, NO_UID32))) {
4852 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4853 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4855 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4857 ei->i_state_flags = 0;
4858 ei->i_dir_start_lookup = 0;
4859 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4860 /* We now have enough fields to check if the inode was active or not.
4861 * This is needed because nfsd might try to access dead inodes
4862 * the test is that same one that e2fsck uses
4863 * NeilBrown 1999oct15
4865 if (inode->i_nlink == 0) {
4866 if (inode->i_mode == 0 ||
4867 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4868 /* this inode is deleted */
4872 /* The only unlinked inodes we let through here have
4873 * valid i_mode and are being read by the orphan
4874 * recovery code: that's fine, we're about to complete
4875 * the process of deleting those. */
4877 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4878 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4879 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4880 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
4882 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4883 inode->i_size = ext4_isize(raw_inode);
4884 ei->i_disksize = inode->i_size;
4886 ei->i_reserved_quota = 0;
4888 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4889 ei->i_block_group = iloc.block_group;
4890 ei->i_last_alloc_group = ~0;
4892 * NOTE! The in-memory inode i_data array is in little-endian order
4893 * even on big-endian machines: we do NOT byteswap the block numbers!
4895 for (block = 0; block < EXT4_N_BLOCKS; block++)
4896 ei->i_data[block] = raw_inode->i_block[block];
4897 INIT_LIST_HEAD(&ei->i_orphan);
4900 * Set transaction id's of transactions that have to be committed
4901 * to finish f[data]sync. We set them to currently running transaction
4902 * as we cannot be sure that the inode or some of its metadata isn't
4903 * part of the transaction - the inode could have been reclaimed and
4904 * now it is reread from disk.
4907 transaction_t *transaction;
4910 read_lock(&journal->j_state_lock);
4911 if (journal->j_running_transaction)
4912 transaction = journal->j_running_transaction;
4914 transaction = journal->j_committing_transaction;
4916 tid = transaction->t_tid;
4918 tid = journal->j_commit_sequence;
4919 read_unlock(&journal->j_state_lock);
4920 ei->i_sync_tid = tid;
4921 ei->i_datasync_tid = tid;
4924 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4925 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4926 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4927 EXT4_INODE_SIZE(inode->i_sb)) {
4931 if (ei->i_extra_isize == 0) {
4932 /* The extra space is currently unused. Use it. */
4933 ei->i_extra_isize = sizeof(struct ext4_inode) -
4934 EXT4_GOOD_OLD_INODE_SIZE;
4936 __le32 *magic = (void *)raw_inode +
4937 EXT4_GOOD_OLD_INODE_SIZE +
4939 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4940 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4943 ei->i_extra_isize = 0;
4945 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4946 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4947 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4948 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4950 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4951 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4952 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4954 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4958 if (ei->i_file_acl &&
4959 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4960 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4964 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4965 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4966 (S_ISLNK(inode->i_mode) &&
4967 !ext4_inode_is_fast_symlink(inode)))
4968 /* Validate extent which is part of inode */
4969 ret = ext4_ext_check_inode(inode);
4970 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4971 (S_ISLNK(inode->i_mode) &&
4972 !ext4_inode_is_fast_symlink(inode))) {
4973 /* Validate block references which are part of inode */
4974 ret = ext4_check_inode_blockref(inode);
4979 if (S_ISREG(inode->i_mode)) {
4980 inode->i_op = &ext4_file_inode_operations;
4981 inode->i_fop = &ext4_file_operations;
4982 ext4_set_aops(inode);
4983 } else if (S_ISDIR(inode->i_mode)) {
4984 inode->i_op = &ext4_dir_inode_operations;
4985 inode->i_fop = &ext4_dir_operations;
4986 } else if (S_ISLNK(inode->i_mode)) {
4987 if (ext4_inode_is_fast_symlink(inode)) {
4988 inode->i_op = &ext4_fast_symlink_inode_operations;
4989 nd_terminate_link(ei->i_data, inode->i_size,
4990 sizeof(ei->i_data) - 1);
4992 inode->i_op = &ext4_symlink_inode_operations;
4993 ext4_set_aops(inode);
4995 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4996 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4997 inode->i_op = &ext4_special_inode_operations;
4998 if (raw_inode->i_block[0])
4999 init_special_inode(inode, inode->i_mode,
5000 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5002 init_special_inode(inode, inode->i_mode,
5003 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5006 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
5010 ext4_set_inode_flags(inode);
5011 unlock_new_inode(inode);
5017 return ERR_PTR(ret);
5020 static int ext4_inode_blocks_set(handle_t *handle,
5021 struct ext4_inode *raw_inode,
5022 struct ext4_inode_info *ei)
5024 struct inode *inode = &(ei->vfs_inode);
5025 u64 i_blocks = inode->i_blocks;
5026 struct super_block *sb = inode->i_sb;
5028 if (i_blocks <= ~0U) {
5030 * i_blocks can be represnted in a 32 bit variable
5031 * as multiple of 512 bytes
5033 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5034 raw_inode->i_blocks_high = 0;
5035 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5038 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
5041 if (i_blocks <= 0xffffffffffffULL) {
5043 * i_blocks can be represented in a 48 bit variable
5044 * as multiple of 512 bytes
5046 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5047 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5048 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5050 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5051 /* i_block is stored in file system block size */
5052 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5053 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5054 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5060 * Post the struct inode info into an on-disk inode location in the
5061 * buffer-cache. This gobbles the caller's reference to the
5062 * buffer_head in the inode location struct.
5064 * The caller must have write access to iloc->bh.
5066 static int ext4_do_update_inode(handle_t *handle,
5067 struct inode *inode,
5068 struct ext4_iloc *iloc)
5070 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5071 struct ext4_inode_info *ei = EXT4_I(inode);
5072 struct buffer_head *bh = iloc->bh;
5073 int err = 0, rc, block;
5075 /* For fields not not tracking in the in-memory inode,
5076 * initialise them to zero for new inodes. */
5077 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5078 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5080 ext4_get_inode_flags(ei);
5081 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5082 if (!(test_opt(inode->i_sb, NO_UID32))) {
5083 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
5084 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
5086 * Fix up interoperability with old kernels. Otherwise, old inodes get
5087 * re-used with the upper 16 bits of the uid/gid intact
5090 raw_inode->i_uid_high =
5091 cpu_to_le16(high_16_bits(inode->i_uid));
5092 raw_inode->i_gid_high =
5093 cpu_to_le16(high_16_bits(inode->i_gid));
5095 raw_inode->i_uid_high = 0;
5096 raw_inode->i_gid_high = 0;
5099 raw_inode->i_uid_low =
5100 cpu_to_le16(fs_high2lowuid(inode->i_uid));
5101 raw_inode->i_gid_low =
5102 cpu_to_le16(fs_high2lowgid(inode->i_gid));
5103 raw_inode->i_uid_high = 0;
5104 raw_inode->i_gid_high = 0;
5106 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5108 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5109 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5110 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5111 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5113 if (ext4_inode_blocks_set(handle, raw_inode, ei))
5115 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5116 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
5117 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
5118 cpu_to_le32(EXT4_OS_HURD))
5119 raw_inode->i_file_acl_high =
5120 cpu_to_le16(ei->i_file_acl >> 32);
5121 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5122 ext4_isize_set(raw_inode, ei->i_disksize);
5123 if (ei->i_disksize > 0x7fffffffULL) {
5124 struct super_block *sb = inode->i_sb;
5125 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
5126 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
5127 EXT4_SB(sb)->s_es->s_rev_level ==
5128 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
5129 /* If this is the first large file
5130 * created, add a flag to the superblock.
5132 err = ext4_journal_get_write_access(handle,
5133 EXT4_SB(sb)->s_sbh);
5136 ext4_update_dynamic_rev(sb);
5137 EXT4_SET_RO_COMPAT_FEATURE(sb,
5138 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
5140 ext4_handle_sync(handle);
5141 err = ext4_handle_dirty_metadata(handle, NULL,
5142 EXT4_SB(sb)->s_sbh);
5145 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5146 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5147 if (old_valid_dev(inode->i_rdev)) {
5148 raw_inode->i_block[0] =
5149 cpu_to_le32(old_encode_dev(inode->i_rdev));
5150 raw_inode->i_block[1] = 0;
5152 raw_inode->i_block[0] = 0;
5153 raw_inode->i_block[1] =
5154 cpu_to_le32(new_encode_dev(inode->i_rdev));
5155 raw_inode->i_block[2] = 0;
5158 for (block = 0; block < EXT4_N_BLOCKS; block++)
5159 raw_inode->i_block[block] = ei->i_data[block];
5161 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
5162 if (ei->i_extra_isize) {
5163 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5164 raw_inode->i_version_hi =
5165 cpu_to_le32(inode->i_version >> 32);
5166 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
5169 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5170 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5173 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5175 ext4_update_inode_fsync_trans(handle, inode, 0);
5178 ext4_std_error(inode->i_sb, err);
5183 * ext4_write_inode()
5185 * We are called from a few places:
5187 * - Within generic_file_write() for O_SYNC files.
5188 * Here, there will be no transaction running. We wait for any running
5189 * trasnaction to commit.
5191 * - Within sys_sync(), kupdate and such.
5192 * We wait on commit, if tol to.
5194 * - Within prune_icache() (PF_MEMALLOC == true)
5195 * Here we simply return. We can't afford to block kswapd on the
5198 * In all cases it is actually safe for us to return without doing anything,
5199 * because the inode has been copied into a raw inode buffer in
5200 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
5203 * Note that we are absolutely dependent upon all inode dirtiers doing the
5204 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5205 * which we are interested.
5207 * It would be a bug for them to not do this. The code:
5209 * mark_inode_dirty(inode)
5211 * inode->i_size = expr;
5213 * is in error because a kswapd-driven write_inode() could occur while
5214 * `stuff()' is running, and the new i_size will be lost. Plus the inode
5215 * will no longer be on the superblock's dirty inode list.
5217 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5221 if (current->flags & PF_MEMALLOC)
5224 if (EXT4_SB(inode->i_sb)->s_journal) {
5225 if (ext4_journal_current_handle()) {
5226 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5231 if (wbc->sync_mode != WB_SYNC_ALL)
5234 err = ext4_force_commit(inode->i_sb);
5236 struct ext4_iloc iloc;
5238 err = __ext4_get_inode_loc(inode, &iloc, 0);
5241 if (wbc->sync_mode == WB_SYNC_ALL)
5242 sync_dirty_buffer(iloc.bh);
5243 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5244 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5245 "IO error syncing inode");
5256 * Called from notify_change.
5258 * We want to trap VFS attempts to truncate the file as soon as
5259 * possible. In particular, we want to make sure that when the VFS
5260 * shrinks i_size, we put the inode on the orphan list and modify
5261 * i_disksize immediately, so that during the subsequent flushing of
5262 * dirty pages and freeing of disk blocks, we can guarantee that any
5263 * commit will leave the blocks being flushed in an unused state on
5264 * disk. (On recovery, the inode will get truncated and the blocks will
5265 * be freed, so we have a strong guarantee that no future commit will
5266 * leave these blocks visible to the user.)
5268 * Another thing we have to assure is that if we are in ordered mode
5269 * and inode is still attached to the committing transaction, we must
5270 * we start writeout of all the dirty pages which are being truncated.
5271 * This way we are sure that all the data written in the previous
5272 * transaction are already on disk (truncate waits for pages under
5275 * Called with inode->i_mutex down.
5277 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5279 struct inode *inode = dentry->d_inode;
5282 const unsigned int ia_valid = attr->ia_valid;
5284 error = inode_change_ok(inode, attr);
5288 if (is_quota_modification(inode, attr))
5289 dquot_initialize(inode);
5290 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
5291 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
5294 /* (user+group)*(old+new) structure, inode write (sb,
5295 * inode block, ? - but truncate inode update has it) */
5296 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
5297 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
5298 if (IS_ERR(handle)) {
5299 error = PTR_ERR(handle);
5302 error = dquot_transfer(inode, attr);
5304 ext4_journal_stop(handle);
5307 /* Update corresponding info in inode so that everything is in
5308 * one transaction */
5309 if (attr->ia_valid & ATTR_UID)
5310 inode->i_uid = attr->ia_uid;
5311 if (attr->ia_valid & ATTR_GID)
5312 inode->i_gid = attr->ia_gid;
5313 error = ext4_mark_inode_dirty(handle, inode);
5314 ext4_journal_stop(handle);
5317 if (attr->ia_valid & ATTR_SIZE) {
5318 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5319 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5321 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5326 if (S_ISREG(inode->i_mode) &&
5327 attr->ia_valid & ATTR_SIZE &&
5328 (attr->ia_size < inode->i_size ||
5329 (ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))) {
5332 handle = ext4_journal_start(inode, 3);
5333 if (IS_ERR(handle)) {
5334 error = PTR_ERR(handle);
5337 if (ext4_handle_valid(handle)) {
5338 error = ext4_orphan_add(handle, inode);
5341 EXT4_I(inode)->i_disksize = attr->ia_size;
5342 rc = ext4_mark_inode_dirty(handle, inode);
5345 ext4_journal_stop(handle);
5347 if (ext4_should_order_data(inode)) {
5348 error = ext4_begin_ordered_truncate(inode,
5351 /* Do as much error cleanup as possible */
5352 handle = ext4_journal_start(inode, 3);
5353 if (IS_ERR(handle)) {
5354 ext4_orphan_del(NULL, inode);
5357 ext4_orphan_del(handle, inode);
5359 ext4_journal_stop(handle);
5363 /* ext4_truncate will clear the flag */
5364 if ((ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS)))
5365 ext4_truncate(inode);
5368 if ((attr->ia_valid & ATTR_SIZE) &&
5369 attr->ia_size != i_size_read(inode))
5370 rc = vmtruncate(inode, attr->ia_size);
5373 setattr_copy(inode, attr);
5374 mark_inode_dirty(inode);
5378 * If the call to ext4_truncate failed to get a transaction handle at
5379 * all, we need to clean up the in-core orphan list manually.
5381 if (orphan && inode->i_nlink)
5382 ext4_orphan_del(NULL, inode);
5384 if (!rc && (ia_valid & ATTR_MODE))
5385 rc = ext4_acl_chmod(inode);
5388 ext4_std_error(inode->i_sb, error);
5394 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
5397 struct inode *inode;
5398 unsigned long delalloc_blocks;
5400 inode = dentry->d_inode;
5401 generic_fillattr(inode, stat);
5404 * We can't update i_blocks if the block allocation is delayed
5405 * otherwise in the case of system crash before the real block
5406 * allocation is done, we will have i_blocks inconsistent with
5407 * on-disk file blocks.
5408 * We always keep i_blocks updated together with real
5409 * allocation. But to not confuse with user, stat
5410 * will return the blocks that include the delayed allocation
5411 * blocks for this file.
5413 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
5414 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
5415 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
5417 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
5421 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
5426 /* if nrblocks are contiguous */
5429 * With N contiguous data blocks, it need at most
5430 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
5431 * 2 dindirect blocks
5434 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
5435 return indirects + 3;
5438 * if nrblocks are not contiguous, worse case, each block touch
5439 * a indirect block, and each indirect block touch a double indirect
5440 * block, plus a triple indirect block
5442 indirects = nrblocks * 2 + 1;
5446 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5448 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5449 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
5450 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
5454 * Account for index blocks, block groups bitmaps and block group
5455 * descriptor blocks if modify datablocks and index blocks
5456 * worse case, the indexs blocks spread over different block groups
5458 * If datablocks are discontiguous, they are possible to spread over
5459 * different block groups too. If they are contiuguous, with flexbg,
5460 * they could still across block group boundary.
5462 * Also account for superblock, inode, quota and xattr blocks
5464 static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
5466 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5472 * How many index blocks need to touch to modify nrblocks?
5473 * The "Chunk" flag indicating whether the nrblocks is
5474 * physically contiguous on disk
5476 * For Direct IO and fallocate, they calls get_block to allocate
5477 * one single extent at a time, so they could set the "Chunk" flag
5479 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
5484 * Now let's see how many group bitmaps and group descriptors need
5494 if (groups > ngroups)
5496 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5497 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5499 /* bitmaps and block group descriptor blocks */
5500 ret += groups + gdpblocks;
5502 /* Blocks for super block, inode, quota and xattr blocks */
5503 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5509 * Calulate the total number of credits to reserve to fit
5510 * the modification of a single pages into a single transaction,
5511 * which may include multiple chunks of block allocations.
5513 * This could be called via ext4_write_begin()
5515 * We need to consider the worse case, when
5516 * one new block per extent.
5518 int ext4_writepage_trans_blocks(struct inode *inode)
5520 int bpp = ext4_journal_blocks_per_page(inode);
5523 ret = ext4_meta_trans_blocks(inode, bpp, 0);
5525 /* Account for data blocks for journalled mode */
5526 if (ext4_should_journal_data(inode))
5532 * Calculate the journal credits for a chunk of data modification.
5534 * This is called from DIO, fallocate or whoever calling
5535 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5537 * journal buffers for data blocks are not included here, as DIO
5538 * and fallocate do no need to journal data buffers.
5540 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5542 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5546 * The caller must have previously called ext4_reserve_inode_write().
5547 * Give this, we know that the caller already has write access to iloc->bh.
5549 int ext4_mark_iloc_dirty(handle_t *handle,
5550 struct inode *inode, struct ext4_iloc *iloc)
5554 if (test_opt(inode->i_sb, I_VERSION))
5555 inode_inc_iversion(inode);
5557 /* the do_update_inode consumes one bh->b_count */
5560 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5561 err = ext4_do_update_inode(handle, inode, iloc);
5567 * On success, We end up with an outstanding reference count against
5568 * iloc->bh. This _must_ be cleaned up later.
5572 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5573 struct ext4_iloc *iloc)
5577 err = ext4_get_inode_loc(inode, iloc);
5579 BUFFER_TRACE(iloc->bh, "get_write_access");
5580 err = ext4_journal_get_write_access(handle, iloc->bh);
5586 ext4_std_error(inode->i_sb, err);
5591 * Expand an inode by new_extra_isize bytes.
5592 * Returns 0 on success or negative error number on failure.
5594 static int ext4_expand_extra_isize(struct inode *inode,
5595 unsigned int new_extra_isize,
5596 struct ext4_iloc iloc,
5599 struct ext4_inode *raw_inode;
5600 struct ext4_xattr_ibody_header *header;
5602 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
5605 raw_inode = ext4_raw_inode(&iloc);
5607 header = IHDR(inode, raw_inode);
5609 /* No extended attributes present */
5610 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5611 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5612 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
5614 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5618 /* try to expand with EAs present */
5619 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
5624 * What we do here is to mark the in-core inode as clean with respect to inode
5625 * dirtiness (it may still be data-dirty).
5626 * This means that the in-core inode may be reaped by prune_icache
5627 * without having to perform any I/O. This is a very good thing,
5628 * because *any* task may call prune_icache - even ones which
5629 * have a transaction open against a different journal.
5631 * Is this cheating? Not really. Sure, we haven't written the
5632 * inode out, but prune_icache isn't a user-visible syncing function.
5633 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5634 * we start and wait on commits.
5636 * Is this efficient/effective? Well, we're being nice to the system
5637 * by cleaning up our inodes proactively so they can be reaped
5638 * without I/O. But we are potentially leaving up to five seconds'
5639 * worth of inodes floating about which prune_icache wants us to
5640 * write out. One way to fix that would be to get prune_icache()
5641 * to do a write_super() to free up some memory. It has the desired
5644 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5646 struct ext4_iloc iloc;
5647 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5648 static unsigned int mnt_count;
5652 err = ext4_reserve_inode_write(handle, inode, &iloc);
5653 if (ext4_handle_valid(handle) &&
5654 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
5655 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5657 * We need extra buffer credits since we may write into EA block
5658 * with this same handle. If journal_extend fails, then it will
5659 * only result in a minor loss of functionality for that inode.
5660 * If this is felt to be critical, then e2fsck should be run to
5661 * force a large enough s_min_extra_isize.
5663 if ((jbd2_journal_extend(handle,
5664 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5665 ret = ext4_expand_extra_isize(inode,
5666 sbi->s_want_extra_isize,
5669 ext4_set_inode_state(inode,
5670 EXT4_STATE_NO_EXPAND);
5672 le16_to_cpu(sbi->s_es->s_mnt_count)) {
5673 ext4_warning(inode->i_sb,
5674 "Unable to expand inode %lu. Delete"
5675 " some EAs or run e2fsck.",
5678 le16_to_cpu(sbi->s_es->s_mnt_count);
5684 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5689 * ext4_dirty_inode() is called from __mark_inode_dirty()
5691 * We're really interested in the case where a file is being extended.
5692 * i_size has been changed by generic_commit_write() and we thus need
5693 * to include the updated inode in the current transaction.
5695 * Also, dquot_alloc_block() will always dirty the inode when blocks
5696 * are allocated to the file.
5698 * If the inode is marked synchronous, we don't honour that here - doing
5699 * so would cause a commit on atime updates, which we don't bother doing.
5700 * We handle synchronous inodes at the highest possible level.
5702 void ext4_dirty_inode(struct inode *inode)
5706 handle = ext4_journal_start(inode, 2);
5710 ext4_mark_inode_dirty(handle, inode);
5712 ext4_journal_stop(handle);
5719 * Bind an inode's backing buffer_head into this transaction, to prevent
5720 * it from being flushed to disk early. Unlike
5721 * ext4_reserve_inode_write, this leaves behind no bh reference and
5722 * returns no iloc structure, so the caller needs to repeat the iloc
5723 * lookup to mark the inode dirty later.
5725 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
5727 struct ext4_iloc iloc;
5731 err = ext4_get_inode_loc(inode, &iloc);
5733 BUFFER_TRACE(iloc.bh, "get_write_access");
5734 err = jbd2_journal_get_write_access(handle, iloc.bh);
5736 err = ext4_handle_dirty_metadata(handle,
5742 ext4_std_error(inode->i_sb, err);
5747 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5754 * We have to be very careful here: changing a data block's
5755 * journaling status dynamically is dangerous. If we write a
5756 * data block to the journal, change the status and then delete
5757 * that block, we risk forgetting to revoke the old log record
5758 * from the journal and so a subsequent replay can corrupt data.
5759 * So, first we make sure that the journal is empty and that
5760 * nobody is changing anything.
5763 journal = EXT4_JOURNAL(inode);
5766 if (is_journal_aborted(journal))
5769 jbd2_journal_lock_updates(journal);
5770 jbd2_journal_flush(journal);
5773 * OK, there are no updates running now, and all cached data is
5774 * synced to disk. We are now in a completely consistent state
5775 * which doesn't have anything in the journal, and we know that
5776 * no filesystem updates are running, so it is safe to modify
5777 * the inode's in-core data-journaling state flag now.
5781 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5783 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5784 ext4_set_aops(inode);
5786 jbd2_journal_unlock_updates(journal);
5788 /* Finally we can mark the inode as dirty. */
5790 handle = ext4_journal_start(inode, 1);
5792 return PTR_ERR(handle);
5794 err = ext4_mark_inode_dirty(handle, inode);
5795 ext4_handle_sync(handle);
5796 ext4_journal_stop(handle);
5797 ext4_std_error(inode->i_sb, err);
5802 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5804 return !buffer_mapped(bh);
5807 int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5809 struct page *page = vmf->page;
5814 struct file *file = vma->vm_file;
5815 struct inode *inode = file->f_path.dentry->d_inode;
5816 struct address_space *mapping = inode->i_mapping;
5819 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
5820 * get i_mutex because we are already holding mmap_sem.
5822 down_read(&inode->i_alloc_sem);
5823 size = i_size_read(inode);
5824 if (page->mapping != mapping || size <= page_offset(page)
5825 || !PageUptodate(page)) {
5826 /* page got truncated from under us? */
5830 if (PageMappedToDisk(page))
5833 if (page->index == size >> PAGE_CACHE_SHIFT)
5834 len = size & ~PAGE_CACHE_MASK;
5836 len = PAGE_CACHE_SIZE;
5840 * return if we have all the buffers mapped. This avoid
5841 * the need to call write_begin/write_end which does a
5842 * journal_start/journal_stop which can block and take
5845 if (page_has_buffers(page)) {
5846 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
5847 ext4_bh_unmapped)) {
5854 * OK, we need to fill the hole... Do write_begin write_end
5855 * to do block allocation/reservation.We are not holding
5856 * inode.i__mutex here. That allow * parallel write_begin,
5857 * write_end call. lock_page prevent this from happening
5858 * on the same page though
5860 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5861 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5864 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5865 len, len, page, fsdata);
5871 ret = VM_FAULT_SIGBUS;
5872 up_read(&inode->i_alloc_sem);