2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/namei.h>
38 #include <linux/uio.h>
39 #include <linux/bio.h>
40 #include "ext4_jbd2.h"
43 #include "ext4_extents.h"
45 #define MPAGE_DA_EXTENT_TAIL 0x01
47 static inline int ext4_begin_ordered_truncate(struct inode *inode,
50 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
54 static void ext4_invalidatepage(struct page *page, unsigned long offset);
57 * Test whether an inode is a fast symlink.
59 static int ext4_inode_is_fast_symlink(struct inode *inode)
61 int ea_blocks = EXT4_I(inode)->i_file_acl ?
62 (inode->i_sb->s_blocksize >> 9) : 0;
64 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
68 * The ext4 forget function must perform a revoke if we are freeing data
69 * which has been journaled. Metadata (eg. indirect blocks) must be
70 * revoked in all cases.
72 * "bh" may be NULL: a metadata block may have been freed from memory
73 * but there may still be a record of it in the journal, and that record
74 * still needs to be revoked.
76 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
77 struct buffer_head *bh, ext4_fsblk_t blocknr)
83 BUFFER_TRACE(bh, "enter");
85 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
87 bh, is_metadata, inode->i_mode,
88 test_opt(inode->i_sb, DATA_FLAGS));
90 /* Never use the revoke function if we are doing full data
91 * journaling: there is no need to, and a V1 superblock won't
92 * support it. Otherwise, only skip the revoke on un-journaled
95 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
96 (!is_metadata && !ext4_should_journal_data(inode))) {
98 BUFFER_TRACE(bh, "call jbd2_journal_forget");
99 return ext4_journal_forget(handle, bh);
105 * data!=journal && (is_metadata || should_journal_data(inode))
107 BUFFER_TRACE(bh, "call ext4_journal_revoke");
108 err = ext4_journal_revoke(handle, blocknr, bh);
110 ext4_abort(inode->i_sb, __func__,
111 "error %d when attempting revoke", err);
112 BUFFER_TRACE(bh, "exit");
117 * Work out how many blocks we need to proceed with the next chunk of a
118 * truncate transaction.
120 static unsigned long blocks_for_truncate(struct inode *inode)
124 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
126 /* Give ourselves just enough room to cope with inodes in which
127 * i_blocks is corrupt: we've seen disk corruptions in the past
128 * which resulted in random data in an inode which looked enough
129 * like a regular file for ext4 to try to delete it. Things
130 * will go a bit crazy if that happens, but at least we should
131 * try not to panic the whole kernel. */
135 /* But we need to bound the transaction so we don't overflow the
137 if (needed > EXT4_MAX_TRANS_DATA)
138 needed = EXT4_MAX_TRANS_DATA;
140 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
144 * Truncate transactions can be complex and absolutely huge. So we need to
145 * be able to restart the transaction at a conventient checkpoint to make
146 * sure we don't overflow the journal.
148 * start_transaction gets us a new handle for a truncate transaction,
149 * and extend_transaction tries to extend the existing one a bit. If
150 * extend fails, we need to propagate the failure up and restart the
151 * transaction in the top-level truncate loop. --sct
153 static handle_t *start_transaction(struct inode *inode)
157 result = ext4_journal_start(inode, blocks_for_truncate(inode));
161 ext4_std_error(inode->i_sb, PTR_ERR(result));
166 * Try to extend this transaction for the purposes of truncation.
168 * Returns 0 if we managed to create more room. If we can't create more
169 * room, and the transaction must be restarted we return 1.
171 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
173 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
175 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
181 * Restart the transaction associated with *handle. This does a commit,
182 * so before we call here everything must be consistently dirtied against
185 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
187 jbd_debug(2, "restarting handle %p\n", handle);
188 return ext4_journal_restart(handle, blocks_for_truncate(inode));
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_delete_inode(struct inode *inode)
199 if (ext4_should_order_data(inode))
200 ext4_begin_ordered_truncate(inode, 0);
201 truncate_inode_pages(&inode->i_data, 0);
203 if (is_bad_inode(inode))
206 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
207 if (IS_ERR(handle)) {
208 ext4_std_error(inode->i_sb, PTR_ERR(handle));
210 * If we're going to skip the normal cleanup, we still need to
211 * make sure that the in-core orphan linked list is properly
214 ext4_orphan_del(NULL, inode);
221 err = ext4_mark_inode_dirty(handle, inode);
223 ext4_warning(inode->i_sb, __func__,
224 "couldn't mark inode dirty (err %d)", err);
228 ext4_truncate(inode);
231 * ext4_ext_truncate() doesn't reserve any slop when it
232 * restarts journal transactions; therefore there may not be
233 * enough credits left in the handle to remove the inode from
234 * the orphan list and set the dtime field.
236 if (handle->h_buffer_credits < 3) {
237 err = ext4_journal_extend(handle, 3);
239 err = ext4_journal_restart(handle, 3);
241 ext4_warning(inode->i_sb, __func__,
242 "couldn't extend journal (err %d)", err);
244 ext4_journal_stop(handle);
250 * Kill off the orphan record which ext4_truncate created.
251 * AKPM: I think this can be inside the above `if'.
252 * Note that ext4_orphan_del() has to be able to cope with the
253 * deletion of a non-existent orphan - this is because we don't
254 * know if ext4_truncate() actually created an orphan record.
255 * (Well, we could do this if we need to, but heck - it works)
257 ext4_orphan_del(handle, inode);
258 EXT4_I(inode)->i_dtime = get_seconds();
261 * One subtle ordering requirement: if anything has gone wrong
262 * (transaction abort, IO errors, whatever), then we can still
263 * do these next steps (the fs will already have been marked as
264 * having errors), but we can't free the inode if the mark_dirty
267 if (ext4_mark_inode_dirty(handle, inode))
268 /* If that failed, just do the required in-core inode clear. */
271 ext4_free_inode(handle, inode);
272 ext4_journal_stop(handle);
275 clear_inode(inode); /* We must guarantee clearing of inode... */
281 struct buffer_head *bh;
284 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
286 p->key = *(p->p = v);
291 * ext4_block_to_path - parse the block number into array of offsets
292 * @inode: inode in question (we are only interested in its superblock)
293 * @i_block: block number to be parsed
294 * @offsets: array to store the offsets in
295 * @boundary: set this non-zero if the referred-to block is likely to be
296 * followed (on disk) by an indirect block.
298 * To store the locations of file's data ext4 uses a data structure common
299 * for UNIX filesystems - tree of pointers anchored in the inode, with
300 * data blocks at leaves and indirect blocks in intermediate nodes.
301 * This function translates the block number into path in that tree -
302 * return value is the path length and @offsets[n] is the offset of
303 * pointer to (n+1)th node in the nth one. If @block is out of range
304 * (negative or too large) warning is printed and zero returned.
306 * Note: function doesn't find node addresses, so no IO is needed. All
307 * we need to know is the capacity of indirect blocks (taken from the
312 * Portability note: the last comparison (check that we fit into triple
313 * indirect block) is spelled differently, because otherwise on an
314 * architecture with 32-bit longs and 8Kb pages we might get into trouble
315 * if our filesystem had 8Kb blocks. We might use long long, but that would
316 * kill us on x86. Oh, well, at least the sign propagation does not matter -
317 * i_block would have to be negative in the very beginning, so we would not
321 static int ext4_block_to_path(struct inode *inode,
323 ext4_lblk_t offsets[4], int *boundary)
325 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
326 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
327 const long direct_blocks = EXT4_NDIR_BLOCKS,
328 indirect_blocks = ptrs,
329 double_blocks = (1 << (ptrs_bits * 2));
334 ext4_warning(inode->i_sb, "ext4_block_to_path", "block < 0");
335 } else if (i_block < direct_blocks) {
336 offsets[n++] = i_block;
337 final = direct_blocks;
338 } else if ((i_block -= direct_blocks) < indirect_blocks) {
339 offsets[n++] = EXT4_IND_BLOCK;
340 offsets[n++] = i_block;
342 } else if ((i_block -= indirect_blocks) < double_blocks) {
343 offsets[n++] = EXT4_DIND_BLOCK;
344 offsets[n++] = i_block >> ptrs_bits;
345 offsets[n++] = i_block & (ptrs - 1);
347 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
348 offsets[n++] = EXT4_TIND_BLOCK;
349 offsets[n++] = i_block >> (ptrs_bits * 2);
350 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
351 offsets[n++] = i_block & (ptrs - 1);
354 ext4_warning(inode->i_sb, "ext4_block_to_path",
356 i_block + direct_blocks +
357 indirect_blocks + double_blocks);
360 *boundary = final - 1 - (i_block & (ptrs - 1));
365 * ext4_get_branch - read the chain of indirect blocks leading to data
366 * @inode: inode in question
367 * @depth: depth of the chain (1 - direct pointer, etc.)
368 * @offsets: offsets of pointers in inode/indirect blocks
369 * @chain: place to store the result
370 * @err: here we store the error value
372 * Function fills the array of triples <key, p, bh> and returns %NULL
373 * if everything went OK or the pointer to the last filled triple
374 * (incomplete one) otherwise. Upon the return chain[i].key contains
375 * the number of (i+1)-th block in the chain (as it is stored in memory,
376 * i.e. little-endian 32-bit), chain[i].p contains the address of that
377 * number (it points into struct inode for i==0 and into the bh->b_data
378 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
379 * block for i>0 and NULL for i==0. In other words, it holds the block
380 * numbers of the chain, addresses they were taken from (and where we can
381 * verify that chain did not change) and buffer_heads hosting these
384 * Function stops when it stumbles upon zero pointer (absent block)
385 * (pointer to last triple returned, *@err == 0)
386 * or when it gets an IO error reading an indirect block
387 * (ditto, *@err == -EIO)
388 * or when it reads all @depth-1 indirect blocks successfully and finds
389 * the whole chain, all way to the data (returns %NULL, *err == 0).
391 * Need to be called with
392 * down_read(&EXT4_I(inode)->i_data_sem)
394 static Indirect *ext4_get_branch(struct inode *inode, int depth,
395 ext4_lblk_t *offsets,
396 Indirect chain[4], int *err)
398 struct super_block *sb = inode->i_sb;
400 struct buffer_head *bh;
403 /* i_data is not going away, no lock needed */
404 add_chain(chain, NULL, EXT4_I(inode)->i_data + *offsets);
408 bh = sb_bread(sb, le32_to_cpu(p->key));
411 add_chain(++p, bh, (__le32 *)bh->b_data + *++offsets);
425 * ext4_find_near - find a place for allocation with sufficient locality
427 * @ind: descriptor of indirect block.
429 * This function returns the preferred place for block allocation.
430 * It is used when heuristic for sequential allocation fails.
432 * + if there is a block to the left of our position - allocate near it.
433 * + if pointer will live in indirect block - allocate near that block.
434 * + if pointer will live in inode - allocate in the same
437 * In the latter case we colour the starting block by the callers PID to
438 * prevent it from clashing with concurrent allocations for a different inode
439 * in the same block group. The PID is used here so that functionally related
440 * files will be close-by on-disk.
442 * Caller must make sure that @ind is valid and will stay that way.
444 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
446 struct ext4_inode_info *ei = EXT4_I(inode);
447 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
449 ext4_fsblk_t bg_start;
450 ext4_fsblk_t last_block;
451 ext4_grpblk_t colour;
453 /* Try to find previous block */
454 for (p = ind->p - 1; p >= start; p--) {
456 return le32_to_cpu(*p);
459 /* No such thing, so let's try location of indirect block */
461 return ind->bh->b_blocknr;
464 * It is going to be referred to from the inode itself? OK, just put it
465 * into the same cylinder group then.
467 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
468 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
470 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
471 colour = (current->pid % 16) *
472 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
474 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
475 return bg_start + colour;
479 * ext4_find_goal - find a preferred place for allocation.
481 * @block: block we want
482 * @partial: pointer to the last triple within a chain
484 * Normally this function find the preferred place for block allocation,
487 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
491 * XXX need to get goal block from mballoc's data structures
494 return ext4_find_near(inode, partial);
498 * ext4_blks_to_allocate: Look up the block map and count the number
499 * of direct blocks need to be allocated for the given branch.
501 * @branch: chain of indirect blocks
502 * @k: number of blocks need for indirect blocks
503 * @blks: number of data blocks to be mapped.
504 * @blocks_to_boundary: the offset in the indirect block
506 * return the total number of blocks to be allocate, including the
507 * direct and indirect blocks.
509 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
510 int blocks_to_boundary)
512 unsigned long count = 0;
515 * Simple case, [t,d]Indirect block(s) has not allocated yet
516 * then it's clear blocks on that path have not allocated
519 /* right now we don't handle cross boundary allocation */
520 if (blks < blocks_to_boundary + 1)
523 count += blocks_to_boundary + 1;
528 while (count < blks && count <= blocks_to_boundary &&
529 le32_to_cpu(*(branch[0].p + count)) == 0) {
536 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
537 * @indirect_blks: the number of blocks need to allocate for indirect
540 * @new_blocks: on return it will store the new block numbers for
541 * the indirect blocks(if needed) and the first direct block,
542 * @blks: on return it will store the total number of allocated
545 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
546 ext4_lblk_t iblock, ext4_fsblk_t goal,
547 int indirect_blks, int blks,
548 ext4_fsblk_t new_blocks[4], int *err)
550 struct ext4_allocation_request ar;
552 unsigned long count = 0, blk_allocated = 0;
554 ext4_fsblk_t current_block = 0;
558 * Here we try to allocate the requested multiple blocks at once,
559 * on a best-effort basis.
560 * To build a branch, we should allocate blocks for
561 * the indirect blocks(if not allocated yet), and at least
562 * the first direct block of this branch. That's the
563 * minimum number of blocks need to allocate(required)
565 /* first we try to allocate the indirect blocks */
566 target = indirect_blks;
569 /* allocating blocks for indirect blocks and direct blocks */
570 current_block = ext4_new_meta_blocks(handle, inode,
576 /* allocate blocks for indirect blocks */
577 while (index < indirect_blks && count) {
578 new_blocks[index++] = current_block++;
583 * save the new block number
584 * for the first direct block
586 new_blocks[index] = current_block;
587 printk(KERN_INFO "%s returned more blocks than "
588 "requested\n", __func__);
594 target = blks - count ;
595 blk_allocated = count;
598 /* Now allocate data blocks */
599 memset(&ar, 0, sizeof(ar));
604 if (S_ISREG(inode->i_mode))
605 /* enable in-core preallocation only for regular files */
606 ar.flags = EXT4_MB_HINT_DATA;
608 current_block = ext4_mb_new_blocks(handle, &ar, err);
610 if (*err && (target == blks)) {
612 * if the allocation failed and we didn't allocate
618 if (target == blks) {
620 * save the new block number
621 * for the first direct block
623 new_blocks[index] = current_block;
625 blk_allocated += ar.len;
628 /* total number of blocks allocated for direct blocks */
633 for (i = 0; i < index; i++)
634 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
639 * ext4_alloc_branch - allocate and set up a chain of blocks.
641 * @indirect_blks: number of allocated indirect blocks
642 * @blks: number of allocated direct blocks
643 * @offsets: offsets (in the blocks) to store the pointers to next.
644 * @branch: place to store the chain in.
646 * This function allocates blocks, zeroes out all but the last one,
647 * links them into chain and (if we are synchronous) writes them to disk.
648 * In other words, it prepares a branch that can be spliced onto the
649 * inode. It stores the information about that chain in the branch[], in
650 * the same format as ext4_get_branch() would do. We are calling it after
651 * we had read the existing part of chain and partial points to the last
652 * triple of that (one with zero ->key). Upon the exit we have the same
653 * picture as after the successful ext4_get_block(), except that in one
654 * place chain is disconnected - *branch->p is still zero (we did not
655 * set the last link), but branch->key contains the number that should
656 * be placed into *branch->p to fill that gap.
658 * If allocation fails we free all blocks we've allocated (and forget
659 * their buffer_heads) and return the error value the from failed
660 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
661 * as described above and return 0.
663 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
664 ext4_lblk_t iblock, int indirect_blks,
665 int *blks, ext4_fsblk_t goal,
666 ext4_lblk_t *offsets, Indirect *branch)
668 int blocksize = inode->i_sb->s_blocksize;
671 struct buffer_head *bh;
673 ext4_fsblk_t new_blocks[4];
674 ext4_fsblk_t current_block;
676 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
677 *blks, new_blocks, &err);
681 branch[0].key = cpu_to_le32(new_blocks[0]);
683 * metadata blocks and data blocks are allocated.
685 for (n = 1; n <= indirect_blks; n++) {
687 * Get buffer_head for parent block, zero it out
688 * and set the pointer to new one, then send
691 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
694 BUFFER_TRACE(bh, "call get_create_access");
695 err = ext4_journal_get_create_access(handle, bh);
702 memset(bh->b_data, 0, blocksize);
703 branch[n].p = (__le32 *) bh->b_data + offsets[n];
704 branch[n].key = cpu_to_le32(new_blocks[n]);
705 *branch[n].p = branch[n].key;
706 if (n == indirect_blks) {
707 current_block = new_blocks[n];
709 * End of chain, update the last new metablock of
710 * the chain to point to the new allocated
711 * data blocks numbers
713 for (i=1; i < num; i++)
714 *(branch[n].p + i) = cpu_to_le32(++current_block);
716 BUFFER_TRACE(bh, "marking uptodate");
717 set_buffer_uptodate(bh);
720 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
721 err = ext4_journal_dirty_metadata(handle, bh);
728 /* Allocation failed, free what we already allocated */
729 for (i = 1; i <= n ; i++) {
730 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
731 ext4_journal_forget(handle, branch[i].bh);
733 for (i = 0; i < indirect_blks; i++)
734 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
736 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
742 * ext4_splice_branch - splice the allocated branch onto inode.
744 * @block: (logical) number of block we are adding
745 * @chain: chain of indirect blocks (with a missing link - see
747 * @where: location of missing link
748 * @num: number of indirect blocks we are adding
749 * @blks: number of direct blocks we are adding
751 * This function fills the missing link and does all housekeeping needed in
752 * inode (->i_blocks, etc.). In case of success we end up with the full
753 * chain to new block and return 0.
755 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
756 ext4_lblk_t block, Indirect *where, int num, int blks)
760 ext4_fsblk_t current_block;
763 * If we're splicing into a [td]indirect block (as opposed to the
764 * inode) then we need to get write access to the [td]indirect block
768 BUFFER_TRACE(where->bh, "get_write_access");
769 err = ext4_journal_get_write_access(handle, where->bh);
775 *where->p = where->key;
778 * Update the host buffer_head or inode to point to more just allocated
779 * direct blocks blocks
781 if (num == 0 && blks > 1) {
782 current_block = le32_to_cpu(where->key) + 1;
783 for (i = 1; i < blks; i++)
784 *(where->p + i) = cpu_to_le32(current_block++);
787 /* We are done with atomic stuff, now do the rest of housekeeping */
789 inode->i_ctime = ext4_current_time(inode);
790 ext4_mark_inode_dirty(handle, inode);
792 /* had we spliced it onto indirect block? */
795 * If we spliced it onto an indirect block, we haven't
796 * altered the inode. Note however that if it is being spliced
797 * onto an indirect block at the very end of the file (the
798 * file is growing) then we *will* alter the inode to reflect
799 * the new i_size. But that is not done here - it is done in
800 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
802 jbd_debug(5, "splicing indirect only\n");
803 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
804 err = ext4_journal_dirty_metadata(handle, where->bh);
809 * OK, we spliced it into the inode itself on a direct block.
810 * Inode was dirtied above.
812 jbd_debug(5, "splicing direct\n");
817 for (i = 1; i <= num; i++) {
818 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
819 ext4_journal_forget(handle, where[i].bh);
820 ext4_free_blocks(handle, inode,
821 le32_to_cpu(where[i-1].key), 1, 0);
823 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
829 * Allocation strategy is simple: if we have to allocate something, we will
830 * have to go the whole way to leaf. So let's do it before attaching anything
831 * to tree, set linkage between the newborn blocks, write them if sync is
832 * required, recheck the path, free and repeat if check fails, otherwise
833 * set the last missing link (that will protect us from any truncate-generated
834 * removals - all blocks on the path are immune now) and possibly force the
835 * write on the parent block.
836 * That has a nice additional property: no special recovery from the failed
837 * allocations is needed - we simply release blocks and do not touch anything
838 * reachable from inode.
840 * `handle' can be NULL if create == 0.
842 * return > 0, # of blocks mapped or allocated.
843 * return = 0, if plain lookup failed.
844 * return < 0, error case.
847 * Need to be called with
848 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
849 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
851 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
852 ext4_lblk_t iblock, unsigned long maxblocks,
853 struct buffer_head *bh_result,
854 int create, int extend_disksize)
857 ext4_lblk_t offsets[4];
862 int blocks_to_boundary = 0;
864 struct ext4_inode_info *ei = EXT4_I(inode);
866 ext4_fsblk_t first_block = 0;
870 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
871 J_ASSERT(handle != NULL || create == 0);
872 depth = ext4_block_to_path(inode, iblock, offsets,
873 &blocks_to_boundary);
878 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
880 /* Simplest case - block found, no allocation needed */
882 first_block = le32_to_cpu(chain[depth - 1].key);
883 clear_buffer_new(bh_result);
886 while (count < maxblocks && count <= blocks_to_boundary) {
889 blk = le32_to_cpu(*(chain[depth-1].p + count));
891 if (blk == first_block + count)
899 /* Next simple case - plain lookup or failed read of indirect block */
900 if (!create || err == -EIO)
904 * Okay, we need to do block allocation.
906 goal = ext4_find_goal(inode, iblock, partial);
908 /* the number of blocks need to allocate for [d,t]indirect blocks */
909 indirect_blks = (chain + depth) - partial - 1;
912 * Next look up the indirect map to count the totoal number of
913 * direct blocks to allocate for this branch.
915 count = ext4_blks_to_allocate(partial, indirect_blks,
916 maxblocks, blocks_to_boundary);
918 * Block out ext4_truncate while we alter the tree
920 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
922 offsets + (partial - chain), partial);
925 * The ext4_splice_branch call will free and forget any buffers
926 * on the new chain if there is a failure, but that risks using
927 * up transaction credits, especially for bitmaps where the
928 * credits cannot be returned. Can we handle this somehow? We
929 * may need to return -EAGAIN upwards in the worst case. --sct
932 err = ext4_splice_branch(handle, inode, iblock,
933 partial, indirect_blks, count);
935 * i_disksize growing is protected by i_data_sem. Don't forget to
936 * protect it if you're about to implement concurrent
937 * ext4_get_block() -bzzz
939 if (!err && extend_disksize) {
940 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
941 if (disksize > i_size_read(inode))
942 disksize = i_size_read(inode);
943 if (disksize > ei->i_disksize)
944 ei->i_disksize = disksize;
949 set_buffer_new(bh_result);
951 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
952 if (count > blocks_to_boundary)
953 set_buffer_boundary(bh_result);
955 /* Clean up and exit */
956 partial = chain + depth - 1; /* the whole chain */
958 while (partial > chain) {
959 BUFFER_TRACE(partial->bh, "call brelse");
963 BUFFER_TRACE(bh_result, "returned");
969 * Calculate the number of metadata blocks need to reserve
970 * to allocate @blocks for non extent file based file
972 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
974 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
975 int ind_blks, dind_blks, tind_blks;
977 /* number of new indirect blocks needed */
978 ind_blks = (blocks + icap - 1) / icap;
980 dind_blks = (ind_blks + icap - 1) / icap;
984 return ind_blks + dind_blks + tind_blks;
988 * Calculate the number of metadata blocks need to reserve
989 * to allocate given number of blocks
991 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
996 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
997 return ext4_ext_calc_metadata_amount(inode, blocks);
999 return ext4_indirect_calc_metadata_amount(inode, blocks);
1002 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1004 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1005 int total, mdb, mdb_free;
1007 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1008 /* recalculate the number of metablocks still need to be reserved */
1009 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1010 mdb = ext4_calc_metadata_amount(inode, total);
1012 /* figure out how many metablocks to release */
1013 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1014 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1017 /* Account for allocated meta_blocks */
1018 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1020 /* update fs dirty blocks counter */
1021 percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
1022 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1023 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1026 /* update per-inode reservations */
1027 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1028 EXT4_I(inode)->i_reserved_data_blocks -= used;
1030 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1034 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1035 * and returns if the blocks are already mapped.
1037 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1038 * and store the allocated blocks in the result buffer head and mark it
1041 * If file type is extents based, it will call ext4_ext_get_blocks(),
1042 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1045 * On success, it returns the number of blocks being mapped or allocate.
1046 * if create==0 and the blocks are pre-allocated and uninitialized block,
1047 * the result buffer head is unmapped. If the create ==1, it will make sure
1048 * the buffer head is mapped.
1050 * It returns 0 if plain look up failed (blocks have not been allocated), in
1051 * that casem, buffer head is unmapped
1053 * It returns the error in case of allocation failure.
1055 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1056 unsigned long max_blocks, struct buffer_head *bh,
1057 int create, int extend_disksize, int flag)
1061 clear_buffer_mapped(bh);
1064 * Try to see if we can get the block without requesting
1065 * for new file system block.
1067 down_read((&EXT4_I(inode)->i_data_sem));
1068 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1069 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1072 retval = ext4_get_blocks_handle(handle,
1073 inode, block, max_blocks, bh, 0, 0);
1075 up_read((&EXT4_I(inode)->i_data_sem));
1077 /* If it is only a block(s) look up */
1082 * Returns if the blocks have already allocated
1084 * Note that if blocks have been preallocated
1085 * ext4_ext_get_block() returns th create = 0
1086 * with buffer head unmapped.
1088 if (retval > 0 && buffer_mapped(bh))
1092 * New blocks allocate and/or writing to uninitialized extent
1093 * will possibly result in updating i_data, so we take
1094 * the write lock of i_data_sem, and call get_blocks()
1095 * with create == 1 flag.
1097 down_write((&EXT4_I(inode)->i_data_sem));
1100 * if the caller is from delayed allocation writeout path
1101 * we have already reserved fs blocks for allocation
1102 * let the underlying get_block() function know to
1103 * avoid double accounting
1106 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1108 * We need to check for EXT4 here because migrate
1109 * could have changed the inode type in between
1111 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1112 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1113 bh, create, extend_disksize);
1115 retval = ext4_get_blocks_handle(handle, inode, block,
1116 max_blocks, bh, create, extend_disksize);
1118 if (retval > 0 && buffer_new(bh)) {
1120 * We allocated new blocks which will result in
1121 * i_data's format changing. Force the migrate
1122 * to fail by clearing migrate flags
1124 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1130 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1132 * Update reserved blocks/metadata blocks
1133 * after successful block allocation
1134 * which were deferred till now
1136 if ((retval > 0) && buffer_delay(bh))
1137 ext4_da_update_reserve_space(inode, retval);
1140 up_write((&EXT4_I(inode)->i_data_sem));
1144 /* Maximum number of blocks we map for direct IO at once. */
1145 #define DIO_MAX_BLOCKS 4096
1147 int ext4_get_block(struct inode *inode, sector_t iblock,
1148 struct buffer_head *bh_result, int create)
1150 handle_t *handle = ext4_journal_current_handle();
1151 int ret = 0, started = 0;
1152 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1155 if (create && !handle) {
1156 /* Direct IO write... */
1157 if (max_blocks > DIO_MAX_BLOCKS)
1158 max_blocks = DIO_MAX_BLOCKS;
1159 dio_credits = ext4_chunk_trans_blocks(inode, max_blocks);
1160 handle = ext4_journal_start(inode, dio_credits);
1161 if (IS_ERR(handle)) {
1162 ret = PTR_ERR(handle);
1168 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1169 max_blocks, bh_result, create, 0, 0);
1171 bh_result->b_size = (ret << inode->i_blkbits);
1175 ext4_journal_stop(handle);
1181 * `handle' can be NULL if create is zero
1183 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1184 ext4_lblk_t block, int create, int *errp)
1186 struct buffer_head dummy;
1189 J_ASSERT(handle != NULL || create == 0);
1192 dummy.b_blocknr = -1000;
1193 buffer_trace_init(&dummy.b_history);
1194 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1195 &dummy, create, 1, 0);
1197 * ext4_get_blocks_handle() returns number of blocks
1198 * mapped. 0 in case of a HOLE.
1206 if (!err && buffer_mapped(&dummy)) {
1207 struct buffer_head *bh;
1208 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1213 if (buffer_new(&dummy)) {
1214 J_ASSERT(create != 0);
1215 J_ASSERT(handle != NULL);
1218 * Now that we do not always journal data, we should
1219 * keep in mind whether this should always journal the
1220 * new buffer as metadata. For now, regular file
1221 * writes use ext4_get_block instead, so it's not a
1225 BUFFER_TRACE(bh, "call get_create_access");
1226 fatal = ext4_journal_get_create_access(handle, bh);
1227 if (!fatal && !buffer_uptodate(bh)) {
1228 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1229 set_buffer_uptodate(bh);
1232 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1233 err = ext4_journal_dirty_metadata(handle, bh);
1237 BUFFER_TRACE(bh, "not a new buffer");
1250 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1251 ext4_lblk_t block, int create, int *err)
1253 struct buffer_head *bh;
1255 bh = ext4_getblk(handle, inode, block, create, err);
1258 if (buffer_uptodate(bh))
1260 ll_rw_block(READ_META, 1, &bh);
1262 if (buffer_uptodate(bh))
1269 static int walk_page_buffers(handle_t *handle,
1270 struct buffer_head *head,
1274 int (*fn)(handle_t *handle,
1275 struct buffer_head *bh))
1277 struct buffer_head *bh;
1278 unsigned block_start, block_end;
1279 unsigned blocksize = head->b_size;
1281 struct buffer_head *next;
1283 for (bh = head, block_start = 0;
1284 ret == 0 && (bh != head || !block_start);
1285 block_start = block_end, bh = next)
1287 next = bh->b_this_page;
1288 block_end = block_start + blocksize;
1289 if (block_end <= from || block_start >= to) {
1290 if (partial && !buffer_uptodate(bh))
1294 err = (*fn)(handle, bh);
1302 * To preserve ordering, it is essential that the hole instantiation and
1303 * the data write be encapsulated in a single transaction. We cannot
1304 * close off a transaction and start a new one between the ext4_get_block()
1305 * and the commit_write(). So doing the jbd2_journal_start at the start of
1306 * prepare_write() is the right place.
1308 * Also, this function can nest inside ext4_writepage() ->
1309 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1310 * has generated enough buffer credits to do the whole page. So we won't
1311 * block on the journal in that case, which is good, because the caller may
1314 * By accident, ext4 can be reentered when a transaction is open via
1315 * quota file writes. If we were to commit the transaction while thus
1316 * reentered, there can be a deadlock - we would be holding a quota
1317 * lock, and the commit would never complete if another thread had a
1318 * transaction open and was blocking on the quota lock - a ranking
1321 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1322 * will _not_ run commit under these circumstances because handle->h_ref
1323 * is elevated. We'll still have enough credits for the tiny quotafile
1326 static int do_journal_get_write_access(handle_t *handle,
1327 struct buffer_head *bh)
1329 if (!buffer_mapped(bh) || buffer_freed(bh))
1331 return ext4_journal_get_write_access(handle, bh);
1334 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1335 loff_t pos, unsigned len, unsigned flags,
1336 struct page **pagep, void **fsdata)
1338 struct inode *inode = mapping->host;
1339 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1346 index = pos >> PAGE_CACHE_SHIFT;
1347 from = pos & (PAGE_CACHE_SIZE - 1);
1351 handle = ext4_journal_start(inode, needed_blocks);
1352 if (IS_ERR(handle)) {
1353 ret = PTR_ERR(handle);
1357 page = grab_cache_page_write_begin(mapping, index, flags);
1359 ext4_journal_stop(handle);
1365 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1368 if (!ret && ext4_should_journal_data(inode)) {
1369 ret = walk_page_buffers(handle, page_buffers(page),
1370 from, to, NULL, do_journal_get_write_access);
1375 ext4_journal_stop(handle);
1376 page_cache_release(page);
1378 * block_write_begin may have instantiated a few blocks
1379 * outside i_size. Trim these off again. Don't need
1380 * i_size_read because we hold i_mutex.
1382 if (pos + len > inode->i_size)
1383 vmtruncate(inode, inode->i_size);
1386 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1392 /* For write_end() in data=journal mode */
1393 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1395 if (!buffer_mapped(bh) || buffer_freed(bh))
1397 set_buffer_uptodate(bh);
1398 return ext4_journal_dirty_metadata(handle, bh);
1402 * We need to pick up the new inode size which generic_commit_write gave us
1403 * `file' can be NULL - eg, when called from page_symlink().
1405 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1406 * buffers are managed internally.
1408 static int ext4_ordered_write_end(struct file *file,
1409 struct address_space *mapping,
1410 loff_t pos, unsigned len, unsigned copied,
1411 struct page *page, void *fsdata)
1413 handle_t *handle = ext4_journal_current_handle();
1414 struct inode *inode = mapping->host;
1417 ret = ext4_jbd2_file_inode(handle, inode);
1422 new_i_size = pos + copied;
1423 if (new_i_size > EXT4_I(inode)->i_disksize) {
1424 ext4_update_i_disksize(inode, new_i_size);
1425 /* We need to mark inode dirty even if
1426 * new_i_size is less that inode->i_size
1427 * bu greater than i_disksize.(hint delalloc)
1429 ext4_mark_inode_dirty(handle, inode);
1432 ret2 = generic_write_end(file, mapping, pos, len, copied,
1438 ret2 = ext4_journal_stop(handle);
1442 return ret ? ret : copied;
1445 static int ext4_writeback_write_end(struct file *file,
1446 struct address_space *mapping,
1447 loff_t pos, unsigned len, unsigned copied,
1448 struct page *page, void *fsdata)
1450 handle_t *handle = ext4_journal_current_handle();
1451 struct inode *inode = mapping->host;
1455 new_i_size = pos + copied;
1456 if (new_i_size > EXT4_I(inode)->i_disksize) {
1457 ext4_update_i_disksize(inode, new_i_size);
1458 /* We need to mark inode dirty even if
1459 * new_i_size is less that inode->i_size
1460 * bu greater than i_disksize.(hint delalloc)
1462 ext4_mark_inode_dirty(handle, inode);
1465 ret2 = generic_write_end(file, mapping, pos, len, copied,
1471 ret2 = ext4_journal_stop(handle);
1475 return ret ? ret : copied;
1478 static int ext4_journalled_write_end(struct file *file,
1479 struct address_space *mapping,
1480 loff_t pos, unsigned len, unsigned copied,
1481 struct page *page, void *fsdata)
1483 handle_t *handle = ext4_journal_current_handle();
1484 struct inode *inode = mapping->host;
1490 from = pos & (PAGE_CACHE_SIZE - 1);
1494 if (!PageUptodate(page))
1496 page_zero_new_buffers(page, from+copied, to);
1499 ret = walk_page_buffers(handle, page_buffers(page), from,
1500 to, &partial, write_end_fn);
1502 SetPageUptodate(page);
1503 new_i_size = pos + copied;
1504 if (new_i_size > inode->i_size)
1505 i_size_write(inode, pos+copied);
1506 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1507 if (new_i_size > EXT4_I(inode)->i_disksize) {
1508 ext4_update_i_disksize(inode, new_i_size);
1509 ret2 = ext4_mark_inode_dirty(handle, inode);
1515 ret2 = ext4_journal_stop(handle);
1518 page_cache_release(page);
1520 return ret ? ret : copied;
1523 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1526 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1527 unsigned long md_needed, mdblocks, total = 0;
1530 * recalculate the amount of metadata blocks to reserve
1531 * in order to allocate nrblocks
1532 * worse case is one extent per block
1535 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1536 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1537 mdblocks = ext4_calc_metadata_amount(inode, total);
1538 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1540 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1541 total = md_needed + nrblocks;
1543 if (ext4_claim_free_blocks(sbi, total)) {
1544 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1545 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1551 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1552 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1554 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1555 return 0; /* success */
1558 static void ext4_da_release_space(struct inode *inode, int to_free)
1560 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1561 int total, mdb, mdb_free, release;
1564 return; /* Nothing to release, exit */
1566 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1568 if (!EXT4_I(inode)->i_reserved_data_blocks) {
1570 * if there is no reserved blocks, but we try to free some
1571 * then the counter is messed up somewhere.
1572 * but since this function is called from invalidate
1573 * page, it's harmless to return without any action
1575 printk(KERN_INFO "ext4 delalloc try to release %d reserved "
1576 "blocks for inode %lu, but there is no reserved "
1577 "data blocks\n", to_free, inode->i_ino);
1578 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1582 /* recalculate the number of metablocks still need to be reserved */
1583 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1584 mdb = ext4_calc_metadata_amount(inode, total);
1586 /* figure out how many metablocks to release */
1587 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1588 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1590 release = to_free + mdb_free;
1592 /* update fs dirty blocks counter for truncate case */
1593 percpu_counter_sub(&sbi->s_dirtyblocks_counter, release);
1595 /* update per-inode reservations */
1596 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1597 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1599 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1600 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1601 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1604 static void ext4_da_page_release_reservation(struct page *page,
1605 unsigned long offset)
1608 struct buffer_head *head, *bh;
1609 unsigned int curr_off = 0;
1611 head = page_buffers(page);
1614 unsigned int next_off = curr_off + bh->b_size;
1616 if ((offset <= curr_off) && (buffer_delay(bh))) {
1618 clear_buffer_delay(bh);
1620 curr_off = next_off;
1621 } while ((bh = bh->b_this_page) != head);
1622 ext4_da_release_space(page->mapping->host, to_release);
1626 * Delayed allocation stuff
1629 struct mpage_da_data {
1630 struct inode *inode;
1631 struct buffer_head lbh; /* extent of blocks */
1632 unsigned long first_page, next_page; /* extent of pages */
1633 get_block_t *get_block;
1634 struct writeback_control *wbc;
1641 * mpage_da_submit_io - walks through extent of pages and try to write
1642 * them with writepage() call back
1644 * @mpd->inode: inode
1645 * @mpd->first_page: first page of the extent
1646 * @mpd->next_page: page after the last page of the extent
1647 * @mpd->get_block: the filesystem's block mapper function
1649 * By the time mpage_da_submit_io() is called we expect all blocks
1650 * to be allocated. this may be wrong if allocation failed.
1652 * As pages are already locked by write_cache_pages(), we can't use it
1654 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1656 struct address_space *mapping = mpd->inode->i_mapping;
1657 int ret = 0, err, nr_pages, i;
1658 unsigned long index, end;
1659 struct pagevec pvec;
1662 BUG_ON(mpd->next_page <= mpd->first_page);
1663 pagevec_init(&pvec, 0);
1664 index = mpd->first_page;
1665 end = mpd->next_page - 1;
1667 while (index <= end) {
1669 * We can use PAGECACHE_TAG_DIRTY lookup here because
1670 * even though we have cleared the dirty flag on the page
1671 * We still keep the page in the radix tree with tag
1672 * PAGECACHE_TAG_DIRTY. See clear_page_dirty_for_io.
1673 * The PAGECACHE_TAG_DIRTY is cleared in set_page_writeback
1674 * which is called via the below writepage callback.
1676 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1677 PAGECACHE_TAG_DIRTY,
1679 (pgoff_t)PAGEVEC_SIZE-1) + 1);
1682 for (i = 0; i < nr_pages; i++) {
1683 struct page *page = pvec.pages[i];
1685 pages_skipped = mpd->wbc->pages_skipped;
1686 err = mapping->a_ops->writepage(page, mpd->wbc);
1687 if (!err && (pages_skipped == mpd->wbc->pages_skipped))
1689 * have successfully written the page
1690 * without skipping the same
1692 mpd->pages_written++;
1694 * In error case, we have to continue because
1695 * remaining pages are still locked
1696 * XXX: unlock and re-dirty them?
1701 pagevec_release(&pvec);
1707 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1709 * @mpd->inode - inode to walk through
1710 * @exbh->b_blocknr - first block on a disk
1711 * @exbh->b_size - amount of space in bytes
1712 * @logical - first logical block to start assignment with
1714 * the function goes through all passed space and put actual disk
1715 * block numbers into buffer heads, dropping BH_Delay
1717 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1718 struct buffer_head *exbh)
1720 struct inode *inode = mpd->inode;
1721 struct address_space *mapping = inode->i_mapping;
1722 int blocks = exbh->b_size >> inode->i_blkbits;
1723 sector_t pblock = exbh->b_blocknr, cur_logical;
1724 struct buffer_head *head, *bh;
1726 struct pagevec pvec;
1729 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1730 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1731 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1733 pagevec_init(&pvec, 0);
1735 while (index <= end) {
1736 /* XXX: optimize tail */
1737 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1740 for (i = 0; i < nr_pages; i++) {
1741 struct page *page = pvec.pages[i];
1743 index = page->index;
1748 BUG_ON(!PageLocked(page));
1749 BUG_ON(PageWriteback(page));
1750 BUG_ON(!page_has_buffers(page));
1752 bh = page_buffers(page);
1755 /* skip blocks out of the range */
1757 if (cur_logical >= logical)
1760 } while ((bh = bh->b_this_page) != head);
1763 if (cur_logical >= logical + blocks)
1765 if (buffer_delay(bh)) {
1766 bh->b_blocknr = pblock;
1767 clear_buffer_delay(bh);
1768 bh->b_bdev = inode->i_sb->s_bdev;
1769 } else if (buffer_unwritten(bh)) {
1770 bh->b_blocknr = pblock;
1771 clear_buffer_unwritten(bh);
1772 set_buffer_mapped(bh);
1774 bh->b_bdev = inode->i_sb->s_bdev;
1775 } else if (buffer_mapped(bh))
1776 BUG_ON(bh->b_blocknr != pblock);
1780 } while ((bh = bh->b_this_page) != head);
1782 pagevec_release(&pvec);
1788 * __unmap_underlying_blocks - just a helper function to unmap
1789 * set of blocks described by @bh
1791 static inline void __unmap_underlying_blocks(struct inode *inode,
1792 struct buffer_head *bh)
1794 struct block_device *bdev = inode->i_sb->s_bdev;
1797 blocks = bh->b_size >> inode->i_blkbits;
1798 for (i = 0; i < blocks; i++)
1799 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1802 static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd,
1803 sector_t logical, long blk_cnt)
1807 struct pagevec pvec;
1808 struct inode *inode = mpd->inode;
1809 struct address_space *mapping = inode->i_mapping;
1811 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1812 end = (logical + blk_cnt - 1) >>
1813 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1814 while (index <= end) {
1815 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1818 for (i = 0; i < nr_pages; i++) {
1819 struct page *page = pvec.pages[i];
1820 index = page->index;
1825 BUG_ON(!PageLocked(page));
1826 BUG_ON(PageWriteback(page));
1827 block_invalidatepage(page, 0);
1828 ClearPageUptodate(page);
1835 static void ext4_print_free_blocks(struct inode *inode)
1837 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1838 printk(KERN_EMERG "Total free blocks count %lld\n",
1839 ext4_count_free_blocks(inode->i_sb));
1840 printk(KERN_EMERG "Free/Dirty block details\n");
1841 printk(KERN_EMERG "free_blocks=%lld\n",
1842 (long long)percpu_counter_sum(&sbi->s_freeblocks_counter));
1843 printk(KERN_EMERG "dirty_blocks=%lld\n",
1844 (long long)percpu_counter_sum(&sbi->s_dirtyblocks_counter));
1845 printk(KERN_EMERG "Block reservation details\n");
1846 printk(KERN_EMERG "i_reserved_data_blocks=%lu\n",
1847 EXT4_I(inode)->i_reserved_data_blocks);
1848 printk(KERN_EMERG "i_reserved_meta_blocks=%lu\n",
1849 EXT4_I(inode)->i_reserved_meta_blocks);
1854 * mpage_da_map_blocks - go through given space
1856 * @mpd->lbh - bh describing space
1857 * @mpd->get_block - the filesystem's block mapper function
1859 * The function skips space we know is already mapped to disk blocks.
1862 static int mpage_da_map_blocks(struct mpage_da_data *mpd)
1865 struct buffer_head new;
1866 struct buffer_head *lbh = &mpd->lbh;
1870 * We consider only non-mapped and non-allocated blocks
1872 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1874 new.b_state = lbh->b_state;
1876 new.b_size = lbh->b_size;
1877 next = lbh->b_blocknr;
1879 * If we didn't accumulate anything
1880 * to write simply return
1884 err = mpd->get_block(mpd->inode, next, &new, 1);
1887 /* If get block returns with error
1888 * we simply return. Later writepage
1889 * will redirty the page and writepages
1890 * will find the dirty page again
1895 if (err == -ENOSPC &&
1896 ext4_count_free_blocks(mpd->inode->i_sb)) {
1902 * get block failure will cause us
1903 * to loop in writepages. Because
1904 * a_ops->writepage won't be able to
1905 * make progress. The page will be redirtied
1906 * by writepage and writepages will again
1907 * try to write the same.
1909 printk(KERN_EMERG "%s block allocation failed for inode %lu "
1910 "at logical offset %llu with max blocks "
1911 "%zd with error %d\n",
1912 __func__, mpd->inode->i_ino,
1913 (unsigned long long)next,
1914 lbh->b_size >> mpd->inode->i_blkbits, err);
1915 printk(KERN_EMERG "This should not happen.!! "
1916 "Data will be lost\n");
1917 if (err == -ENOSPC) {
1918 ext4_print_free_blocks(mpd->inode);
1920 /* invlaidate all the pages */
1921 ext4_da_block_invalidatepages(mpd, next,
1922 lbh->b_size >> mpd->inode->i_blkbits);
1925 BUG_ON(new.b_size == 0);
1927 if (buffer_new(&new))
1928 __unmap_underlying_blocks(mpd->inode, &new);
1931 * If blocks are delayed marked, we need to
1932 * put actual blocknr and drop delayed bit
1934 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1935 mpage_put_bnr_to_bhs(mpd, next, &new);
1940 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1941 (1 << BH_Delay) | (1 << BH_Unwritten))
1944 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1946 * @mpd->lbh - extent of blocks
1947 * @logical - logical number of the block in the file
1948 * @bh - bh of the block (used to access block's state)
1950 * the function is used to collect contig. blocks in same state
1952 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1953 sector_t logical, struct buffer_head *bh)
1956 size_t b_size = bh->b_size;
1957 struct buffer_head *lbh = &mpd->lbh;
1958 int nrblocks = lbh->b_size >> mpd->inode->i_blkbits;
1960 /* check if thereserved journal credits might overflow */
1961 if (!(EXT4_I(mpd->inode)->i_flags & EXT4_EXTENTS_FL)) {
1962 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1964 * With non-extent format we are limited by the journal
1965 * credit available. Total credit needed to insert
1966 * nrblocks contiguous blocks is dependent on the
1967 * nrblocks. So limit nrblocks.
1970 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1971 EXT4_MAX_TRANS_DATA) {
1973 * Adding the new buffer_head would make it cross the
1974 * allowed limit for which we have journal credit
1975 * reserved. So limit the new bh->b_size
1977 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1978 mpd->inode->i_blkbits;
1979 /* we will do mpage_da_submit_io in the next loop */
1983 * First block in the extent
1985 if (lbh->b_size == 0) {
1986 lbh->b_blocknr = logical;
1987 lbh->b_size = b_size;
1988 lbh->b_state = bh->b_state & BH_FLAGS;
1992 next = lbh->b_blocknr + nrblocks;
1994 * Can we merge the block to our big extent?
1996 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1997 lbh->b_size += b_size;
2003 * We couldn't merge the block to our extent, so we
2004 * need to flush current extent and start new one
2006 if (mpage_da_map_blocks(mpd) == 0)
2007 mpage_da_submit_io(mpd);
2013 * __mpage_da_writepage - finds extent of pages and blocks
2015 * @page: page to consider
2016 * @wbc: not used, we just follow rules
2019 * The function finds extents of pages and scan them for all blocks.
2021 static int __mpage_da_writepage(struct page *page,
2022 struct writeback_control *wbc, void *data)
2024 struct mpage_da_data *mpd = data;
2025 struct inode *inode = mpd->inode;
2026 struct buffer_head *bh, *head, fake;
2031 * Rest of the page in the page_vec
2032 * redirty then and skip then. We will
2033 * try to to write them again after
2034 * starting a new transaction
2036 redirty_page_for_writepage(wbc, page);
2038 return MPAGE_DA_EXTENT_TAIL;
2041 * Can we merge this page to current extent?
2043 if (mpd->next_page != page->index) {
2045 * Nope, we can't. So, we map non-allocated blocks
2046 * and start IO on them using writepage()
2048 if (mpd->next_page != mpd->first_page) {
2049 if (mpage_da_map_blocks(mpd) == 0)
2050 mpage_da_submit_io(mpd);
2052 * skip rest of the page in the page_vec
2055 redirty_page_for_writepage(wbc, page);
2057 return MPAGE_DA_EXTENT_TAIL;
2061 * Start next extent of pages ...
2063 mpd->first_page = page->index;
2068 mpd->lbh.b_size = 0;
2069 mpd->lbh.b_state = 0;
2070 mpd->lbh.b_blocknr = 0;
2073 mpd->next_page = page->index + 1;
2074 logical = (sector_t) page->index <<
2075 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2077 if (!page_has_buffers(page)) {
2079 * There is no attached buffer heads yet (mmap?)
2080 * we treat the page asfull of dirty blocks
2083 bh->b_size = PAGE_CACHE_SIZE;
2085 set_buffer_dirty(bh);
2086 set_buffer_uptodate(bh);
2087 mpage_add_bh_to_extent(mpd, logical, bh);
2089 return MPAGE_DA_EXTENT_TAIL;
2092 * Page with regular buffer heads, just add all dirty ones
2094 head = page_buffers(page);
2097 BUG_ON(buffer_locked(bh));
2098 if (buffer_dirty(bh) &&
2099 (!buffer_mapped(bh) || buffer_delay(bh))) {
2100 mpage_add_bh_to_extent(mpd, logical, bh);
2102 return MPAGE_DA_EXTENT_TAIL;
2105 } while ((bh = bh->b_this_page) != head);
2112 * mpage_da_writepages - walk the list of dirty pages of the given
2113 * address space, allocates non-allocated blocks, maps newly-allocated
2114 * blocks to existing bhs and issue IO them
2116 * @mapping: address space structure to write
2117 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
2118 * @get_block: the filesystem's block mapper function.
2120 * This is a library function, which implements the writepages()
2121 * address_space_operation.
2123 static int mpage_da_writepages(struct address_space *mapping,
2124 struct writeback_control *wbc,
2125 struct mpage_da_data *mpd)
2129 if (!mpd->get_block)
2130 return generic_writepages(mapping, wbc);
2132 mpd->lbh.b_size = 0;
2133 mpd->lbh.b_state = 0;
2134 mpd->lbh.b_blocknr = 0;
2135 mpd->first_page = 0;
2138 mpd->pages_written = 0;
2141 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, mpd);
2143 * Handle last extent of pages
2145 if (!mpd->io_done && mpd->next_page != mpd->first_page) {
2146 if (mpage_da_map_blocks(mpd) == 0)
2147 mpage_da_submit_io(mpd);
2150 ret = MPAGE_DA_EXTENT_TAIL;
2152 wbc->nr_to_write -= mpd->pages_written;
2157 * this is a special callback for ->write_begin() only
2158 * it's intention is to return mapped block or reserve space
2160 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2161 struct buffer_head *bh_result, int create)
2165 BUG_ON(create == 0);
2166 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2169 * first, we need to know whether the block is allocated already
2170 * preallocated blocks are unmapped but should treated
2171 * the same as allocated blocks.
2173 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2174 if ((ret == 0) && !buffer_delay(bh_result)) {
2175 /* the block isn't (pre)allocated yet, let's reserve space */
2177 * XXX: __block_prepare_write() unmaps passed block,
2180 ret = ext4_da_reserve_space(inode, 1);
2182 /* not enough space to reserve */
2185 map_bh(bh_result, inode->i_sb, 0);
2186 set_buffer_new(bh_result);
2187 set_buffer_delay(bh_result);
2188 } else if (ret > 0) {
2189 bh_result->b_size = (ret << inode->i_blkbits);
2195 #define EXT4_DELALLOC_RSVED 1
2196 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2197 struct buffer_head *bh_result, int create)
2200 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2201 loff_t disksize = EXT4_I(inode)->i_disksize;
2202 handle_t *handle = NULL;
2204 handle = ext4_journal_current_handle();
2206 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2207 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2210 bh_result->b_size = (ret << inode->i_blkbits);
2212 if (ext4_should_order_data(inode)) {
2214 retval = ext4_jbd2_file_inode(handle, inode);
2217 * Failed to add inode for ordered
2218 * mode. Don't update file size
2224 * Update on-disk size along with block allocation
2225 * we don't use 'extend_disksize' as size may change
2226 * within already allocated block -bzzz
2228 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2229 if (disksize > i_size_read(inode))
2230 disksize = i_size_read(inode);
2231 if (disksize > EXT4_I(inode)->i_disksize) {
2232 ext4_update_i_disksize(inode, disksize);
2233 ret = ext4_mark_inode_dirty(handle, inode);
2241 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2244 * unmapped buffer is possible for holes.
2245 * delay buffer is possible with delayed allocation
2247 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2250 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2251 struct buffer_head *bh_result, int create)
2254 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2257 * we don't want to do block allocation in writepage
2258 * so call get_block_wrap with create = 0
2260 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2261 bh_result, 0, 0, 0);
2263 bh_result->b_size = (ret << inode->i_blkbits);
2270 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2271 * get called via journal_submit_inode_data_buffers (no journal handle)
2272 * get called via shrink_page_list via pdflush (no journal handle)
2273 * or grab_page_cache when doing write_begin (have journal handle)
2275 static int ext4_da_writepage(struct page *page,
2276 struct writeback_control *wbc)
2281 struct buffer_head *page_bufs;
2282 struct inode *inode = page->mapping->host;
2284 size = i_size_read(inode);
2285 if (page->index == size >> PAGE_CACHE_SHIFT)
2286 len = size & ~PAGE_CACHE_MASK;
2288 len = PAGE_CACHE_SIZE;
2290 if (page_has_buffers(page)) {
2291 page_bufs = page_buffers(page);
2292 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2293 ext4_bh_unmapped_or_delay)) {
2295 * We don't want to do block allocation
2296 * So redirty the page and return
2297 * We may reach here when we do a journal commit
2298 * via journal_submit_inode_data_buffers.
2299 * If we don't have mapping block we just ignore
2300 * them. We can also reach here via shrink_page_list
2302 redirty_page_for_writepage(wbc, page);
2308 * The test for page_has_buffers() is subtle:
2309 * We know the page is dirty but it lost buffers. That means
2310 * that at some moment in time after write_begin()/write_end()
2311 * has been called all buffers have been clean and thus they
2312 * must have been written at least once. So they are all
2313 * mapped and we can happily proceed with mapping them
2314 * and writing the page.
2316 * Try to initialize the buffer_heads and check whether
2317 * all are mapped and non delay. We don't want to
2318 * do block allocation here.
2320 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2321 ext4_normal_get_block_write);
2323 page_bufs = page_buffers(page);
2324 /* check whether all are mapped and non delay */
2325 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2326 ext4_bh_unmapped_or_delay)) {
2327 redirty_page_for_writepage(wbc, page);
2333 * We can't do block allocation here
2334 * so just redity the page and unlock
2337 redirty_page_for_writepage(wbc, page);
2341 /* now mark the buffer_heads as dirty and uptodate */
2342 block_commit_write(page, 0, PAGE_CACHE_SIZE);
2345 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2346 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2348 ret = block_write_full_page(page,
2349 ext4_normal_get_block_write,
2356 * This is called via ext4_da_writepages() to
2357 * calulate the total number of credits to reserve to fit
2358 * a single extent allocation into a single transaction,
2359 * ext4_da_writpeages() will loop calling this before
2360 * the block allocation.
2363 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2365 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2368 * With non-extent format the journal credit needed to
2369 * insert nrblocks contiguous block is dependent on
2370 * number of contiguous block. So we will limit
2371 * number of contiguous block to a sane value
2373 if (!(inode->i_flags & EXT4_EXTENTS_FL) &&
2374 (max_blocks > EXT4_MAX_TRANS_DATA))
2375 max_blocks = EXT4_MAX_TRANS_DATA;
2377 return ext4_chunk_trans_blocks(inode, max_blocks);
2380 static int ext4_da_writepages(struct address_space *mapping,
2381 struct writeback_control *wbc)
2384 int range_whole = 0;
2385 handle_t *handle = NULL;
2386 struct mpage_da_data mpd;
2387 struct inode *inode = mapping->host;
2388 int no_nrwrite_index_update;
2389 long pages_written = 0, pages_skipped;
2390 int needed_blocks, ret = 0, nr_to_writebump = 0;
2391 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2394 * No pages to write? This is mainly a kludge to avoid starting
2395 * a transaction for special inodes like journal inode on last iput()
2396 * because that could violate lock ordering on umount
2398 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2401 * Make sure nr_to_write is >= sbi->s_mb_stream_request
2402 * This make sure small files blocks are allocated in
2403 * single attempt. This ensure that small files
2404 * get less fragmented.
2406 if (wbc->nr_to_write < sbi->s_mb_stream_request) {
2407 nr_to_writebump = sbi->s_mb_stream_request - wbc->nr_to_write;
2408 wbc->nr_to_write = sbi->s_mb_stream_request;
2410 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2413 if (wbc->range_cyclic)
2414 index = mapping->writeback_index;
2416 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2419 mpd.inode = mapping->host;
2422 * we don't want write_cache_pages to update
2423 * nr_to_write and writeback_index
2425 no_nrwrite_index_update = wbc->no_nrwrite_index_update;
2426 wbc->no_nrwrite_index_update = 1;
2427 pages_skipped = wbc->pages_skipped;
2429 while (!ret && wbc->nr_to_write > 0) {
2432 * we insert one extent at a time. So we need
2433 * credit needed for single extent allocation.
2434 * journalled mode is currently not supported
2437 BUG_ON(ext4_should_journal_data(inode));
2438 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2440 /* start a new transaction*/
2441 handle = ext4_journal_start(inode, needed_blocks);
2442 if (IS_ERR(handle)) {
2443 ret = PTR_ERR(handle);
2444 printk(KERN_EMERG "%s: jbd2_start: "
2445 "%ld pages, ino %lu; err %d\n", __func__,
2446 wbc->nr_to_write, inode->i_ino, ret);
2448 goto out_writepages;
2450 mpd.get_block = ext4_da_get_block_write;
2451 ret = mpage_da_writepages(mapping, wbc, &mpd);
2453 ext4_journal_stop(handle);
2455 if (mpd.retval == -ENOSPC) {
2456 /* commit the transaction which would
2457 * free blocks released in the transaction
2460 jbd2_journal_force_commit_nested(sbi->s_journal);
2461 wbc->pages_skipped = pages_skipped;
2463 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
2465 * got one extent now try with
2468 pages_written += mpd.pages_written;
2469 wbc->pages_skipped = pages_skipped;
2471 } else if (wbc->nr_to_write)
2473 * There is no more writeout needed
2474 * or we requested for a noblocking writeout
2475 * and we found the device congested
2479 if (pages_skipped != wbc->pages_skipped)
2480 printk(KERN_EMERG "This should not happen leaving %s "
2481 "with nr_to_write = %ld ret = %d\n",
2482 __func__, wbc->nr_to_write, ret);
2485 index += pages_written;
2486 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2488 * set the writeback_index so that range_cyclic
2489 * mode will write it back later
2491 mapping->writeback_index = index;
2494 if (!no_nrwrite_index_update)
2495 wbc->no_nrwrite_index_update = 0;
2496 wbc->nr_to_write -= nr_to_writebump;
2500 #define FALL_BACK_TO_NONDELALLOC 1
2501 static int ext4_nonda_switch(struct super_block *sb)
2503 s64 free_blocks, dirty_blocks;
2504 struct ext4_sb_info *sbi = EXT4_SB(sb);
2507 * switch to non delalloc mode if we are running low
2508 * on free block. The free block accounting via percpu
2509 * counters can get slightly wrong with FBC_BATCH getting
2510 * accumulated on each CPU without updating global counters
2511 * Delalloc need an accurate free block accounting. So switch
2512 * to non delalloc when we are near to error range.
2514 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
2515 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyblocks_counter);
2516 if (2 * free_blocks < 3 * dirty_blocks ||
2517 free_blocks < (dirty_blocks + EXT4_FREEBLOCKS_WATERMARK)) {
2519 * free block count is less that 150% of dirty blocks
2520 * or free blocks is less that watermark
2527 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2528 loff_t pos, unsigned len, unsigned flags,
2529 struct page **pagep, void **fsdata)
2531 int ret, retries = 0;
2535 struct inode *inode = mapping->host;
2538 index = pos >> PAGE_CACHE_SHIFT;
2539 from = pos & (PAGE_CACHE_SIZE - 1);
2542 if (ext4_nonda_switch(inode->i_sb)) {
2543 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2544 return ext4_write_begin(file, mapping, pos,
2545 len, flags, pagep, fsdata);
2547 *fsdata = (void *)0;
2550 * With delayed allocation, we don't log the i_disksize update
2551 * if there is delayed block allocation. But we still need
2552 * to journalling the i_disksize update if writes to the end
2553 * of file which has an already mapped buffer.
2555 handle = ext4_journal_start(inode, 1);
2556 if (IS_ERR(handle)) {
2557 ret = PTR_ERR(handle);
2561 page = grab_cache_page_write_begin(mapping, index, flags);
2563 ext4_journal_stop(handle);
2569 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2570 ext4_da_get_block_prep);
2573 ext4_journal_stop(handle);
2574 page_cache_release(page);
2576 * block_write_begin may have instantiated a few blocks
2577 * outside i_size. Trim these off again. Don't need
2578 * i_size_read because we hold i_mutex.
2580 if (pos + len > inode->i_size)
2581 vmtruncate(inode, inode->i_size);
2584 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2591 * Check if we should update i_disksize
2592 * when write to the end of file but not require block allocation
2594 static int ext4_da_should_update_i_disksize(struct page *page,
2595 unsigned long offset)
2597 struct buffer_head *bh;
2598 struct inode *inode = page->mapping->host;
2602 bh = page_buffers(page);
2603 idx = offset >> inode->i_blkbits;
2605 for (i = 0; i < idx; i++)
2606 bh = bh->b_this_page;
2608 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2613 static int ext4_da_write_end(struct file *file,
2614 struct address_space *mapping,
2615 loff_t pos, unsigned len, unsigned copied,
2616 struct page *page, void *fsdata)
2618 struct inode *inode = mapping->host;
2620 handle_t *handle = ext4_journal_current_handle();
2622 unsigned long start, end;
2623 int write_mode = (int)(unsigned long)fsdata;
2625 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
2626 if (ext4_should_order_data(inode)) {
2627 return ext4_ordered_write_end(file, mapping, pos,
2628 len, copied, page, fsdata);
2629 } else if (ext4_should_writeback_data(inode)) {
2630 return ext4_writeback_write_end(file, mapping, pos,
2631 len, copied, page, fsdata);
2637 start = pos & (PAGE_CACHE_SIZE - 1);
2638 end = start + copied - 1;
2641 * generic_write_end() will run mark_inode_dirty() if i_size
2642 * changes. So let's piggyback the i_disksize mark_inode_dirty
2646 new_i_size = pos + copied;
2647 if (new_i_size > EXT4_I(inode)->i_disksize) {
2648 if (ext4_da_should_update_i_disksize(page, end)) {
2649 down_write(&EXT4_I(inode)->i_data_sem);
2650 if (new_i_size > EXT4_I(inode)->i_disksize) {
2652 * Updating i_disksize when extending file
2653 * without needing block allocation
2655 if (ext4_should_order_data(inode))
2656 ret = ext4_jbd2_file_inode(handle,
2659 EXT4_I(inode)->i_disksize = new_i_size;
2661 up_write(&EXT4_I(inode)->i_data_sem);
2662 /* We need to mark inode dirty even if
2663 * new_i_size is less that inode->i_size
2664 * bu greater than i_disksize.(hint delalloc)
2666 ext4_mark_inode_dirty(handle, inode);
2669 ret2 = generic_write_end(file, mapping, pos, len, copied,
2674 ret2 = ext4_journal_stop(handle);
2678 return ret ? ret : copied;
2681 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2684 * Drop reserved blocks
2686 BUG_ON(!PageLocked(page));
2687 if (!page_has_buffers(page))
2690 ext4_da_page_release_reservation(page, offset);
2693 ext4_invalidatepage(page, offset);
2700 * bmap() is special. It gets used by applications such as lilo and by
2701 * the swapper to find the on-disk block of a specific piece of data.
2703 * Naturally, this is dangerous if the block concerned is still in the
2704 * journal. If somebody makes a swapfile on an ext4 data-journaling
2705 * filesystem and enables swap, then they may get a nasty shock when the
2706 * data getting swapped to that swapfile suddenly gets overwritten by
2707 * the original zero's written out previously to the journal and
2708 * awaiting writeback in the kernel's buffer cache.
2710 * So, if we see any bmap calls here on a modified, data-journaled file,
2711 * take extra steps to flush any blocks which might be in the cache.
2713 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2715 struct inode *inode = mapping->host;
2719 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2720 test_opt(inode->i_sb, DELALLOC)) {
2722 * With delalloc we want to sync the file
2723 * so that we can make sure we allocate
2726 filemap_write_and_wait(mapping);
2729 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2731 * This is a REALLY heavyweight approach, but the use of
2732 * bmap on dirty files is expected to be extremely rare:
2733 * only if we run lilo or swapon on a freshly made file
2734 * do we expect this to happen.
2736 * (bmap requires CAP_SYS_RAWIO so this does not
2737 * represent an unprivileged user DOS attack --- we'd be
2738 * in trouble if mortal users could trigger this path at
2741 * NB. EXT4_STATE_JDATA is not set on files other than
2742 * regular files. If somebody wants to bmap a directory
2743 * or symlink and gets confused because the buffer
2744 * hasn't yet been flushed to disk, they deserve
2745 * everything they get.
2748 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2749 journal = EXT4_JOURNAL(inode);
2750 jbd2_journal_lock_updates(journal);
2751 err = jbd2_journal_flush(journal);
2752 jbd2_journal_unlock_updates(journal);
2758 return generic_block_bmap(mapping, block, ext4_get_block);
2761 static int bget_one(handle_t *handle, struct buffer_head *bh)
2767 static int bput_one(handle_t *handle, struct buffer_head *bh)
2774 * Note that we don't need to start a transaction unless we're journaling data
2775 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2776 * need to file the inode to the transaction's list in ordered mode because if
2777 * we are writing back data added by write(), the inode is already there and if
2778 * we are writing back data modified via mmap(), noone guarantees in which
2779 * transaction the data will hit the disk. In case we are journaling data, we
2780 * cannot start transaction directly because transaction start ranks above page
2781 * lock so we have to do some magic.
2783 * In all journaling modes block_write_full_page() will start the I/O.
2787 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2792 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2794 * Same applies to ext4_get_block(). We will deadlock on various things like
2795 * lock_journal and i_data_sem
2797 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2800 * 16May01: If we're reentered then journal_current_handle() will be
2801 * non-zero. We simply *return*.
2803 * 1 July 2001: @@@ FIXME:
2804 * In journalled data mode, a data buffer may be metadata against the
2805 * current transaction. But the same file is part of a shared mapping
2806 * and someone does a writepage() on it.
2808 * We will move the buffer onto the async_data list, but *after* it has
2809 * been dirtied. So there's a small window where we have dirty data on
2812 * Note that this only applies to the last partial page in the file. The
2813 * bit which block_write_full_page() uses prepare/commit for. (That's
2814 * broken code anyway: it's wrong for msync()).
2816 * It's a rare case: affects the final partial page, for journalled data
2817 * where the file is subject to bith write() and writepage() in the same
2818 * transction. To fix it we'll need a custom block_write_full_page().
2819 * We'll probably need that anyway for journalling writepage() output.
2821 * We don't honour synchronous mounts for writepage(). That would be
2822 * disastrous. Any write() or metadata operation will sync the fs for
2826 static int __ext4_normal_writepage(struct page *page,
2827 struct writeback_control *wbc)
2829 struct inode *inode = page->mapping->host;
2831 if (test_opt(inode->i_sb, NOBH))
2832 return nobh_writepage(page,
2833 ext4_normal_get_block_write, wbc);
2835 return block_write_full_page(page,
2836 ext4_normal_get_block_write,
2840 static int ext4_normal_writepage(struct page *page,
2841 struct writeback_control *wbc)
2843 struct inode *inode = page->mapping->host;
2844 loff_t size = i_size_read(inode);
2847 J_ASSERT(PageLocked(page));
2848 if (page->index == size >> PAGE_CACHE_SHIFT)
2849 len = size & ~PAGE_CACHE_MASK;
2851 len = PAGE_CACHE_SIZE;
2853 if (page_has_buffers(page)) {
2854 /* if page has buffers it should all be mapped
2855 * and allocated. If there are not buffers attached
2856 * to the page we know the page is dirty but it lost
2857 * buffers. That means that at some moment in time
2858 * after write_begin() / write_end() has been called
2859 * all buffers have been clean and thus they must have been
2860 * written at least once. So they are all mapped and we can
2861 * happily proceed with mapping them and writing the page.
2863 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2864 ext4_bh_unmapped_or_delay));
2867 if (!ext4_journal_current_handle())
2868 return __ext4_normal_writepage(page, wbc);
2870 redirty_page_for_writepage(wbc, page);
2875 static int __ext4_journalled_writepage(struct page *page,
2876 struct writeback_control *wbc)
2878 struct address_space *mapping = page->mapping;
2879 struct inode *inode = mapping->host;
2880 struct buffer_head *page_bufs;
2881 handle_t *handle = NULL;
2885 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2886 ext4_normal_get_block_write);
2890 page_bufs = page_buffers(page);
2891 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2893 /* As soon as we unlock the page, it can go away, but we have
2894 * references to buffers so we are safe */
2897 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2898 if (IS_ERR(handle)) {
2899 ret = PTR_ERR(handle);
2903 ret = walk_page_buffers(handle, page_bufs, 0,
2904 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2906 err = walk_page_buffers(handle, page_bufs, 0,
2907 PAGE_CACHE_SIZE, NULL, write_end_fn);
2910 err = ext4_journal_stop(handle);
2914 walk_page_buffers(handle, page_bufs, 0,
2915 PAGE_CACHE_SIZE, NULL, bput_one);
2916 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2925 static int ext4_journalled_writepage(struct page *page,
2926 struct writeback_control *wbc)
2928 struct inode *inode = page->mapping->host;
2929 loff_t size = i_size_read(inode);
2932 J_ASSERT(PageLocked(page));
2933 if (page->index == size >> PAGE_CACHE_SHIFT)
2934 len = size & ~PAGE_CACHE_MASK;
2936 len = PAGE_CACHE_SIZE;
2938 if (page_has_buffers(page)) {
2939 /* if page has buffers it should all be mapped
2940 * and allocated. If there are not buffers attached
2941 * to the page we know the page is dirty but it lost
2942 * buffers. That means that at some moment in time
2943 * after write_begin() / write_end() has been called
2944 * all buffers have been clean and thus they must have been
2945 * written at least once. So they are all mapped and we can
2946 * happily proceed with mapping them and writing the page.
2948 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2949 ext4_bh_unmapped_or_delay));
2952 if (ext4_journal_current_handle())
2955 if (PageChecked(page)) {
2957 * It's mmapped pagecache. Add buffers and journal it. There
2958 * doesn't seem much point in redirtying the page here.
2960 ClearPageChecked(page);
2961 return __ext4_journalled_writepage(page, wbc);
2964 * It may be a page full of checkpoint-mode buffers. We don't
2965 * really know unless we go poke around in the buffer_heads.
2966 * But block_write_full_page will do the right thing.
2968 return block_write_full_page(page,
2969 ext4_normal_get_block_write,
2973 redirty_page_for_writepage(wbc, page);
2978 static int ext4_readpage(struct file *file, struct page *page)
2980 return mpage_readpage(page, ext4_get_block);
2984 ext4_readpages(struct file *file, struct address_space *mapping,
2985 struct list_head *pages, unsigned nr_pages)
2987 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2990 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2992 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2995 * If it's a full truncate we just forget about the pending dirtying
2998 ClearPageChecked(page);
3000 jbd2_journal_invalidatepage(journal, page, offset);
3003 static int ext4_releasepage(struct page *page, gfp_t wait)
3005 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3007 WARN_ON(PageChecked(page));
3008 if (!page_has_buffers(page))
3010 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3014 * If the O_DIRECT write will extend the file then add this inode to the
3015 * orphan list. So recovery will truncate it back to the original size
3016 * if the machine crashes during the write.
3018 * If the O_DIRECT write is intantiating holes inside i_size and the machine
3019 * crashes then stale disk data _may_ be exposed inside the file. But current
3020 * VFS code falls back into buffered path in that case so we are safe.
3022 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3023 const struct iovec *iov, loff_t offset,
3024 unsigned long nr_segs)
3026 struct file *file = iocb->ki_filp;
3027 struct inode *inode = file->f_mapping->host;
3028 struct ext4_inode_info *ei = EXT4_I(inode);
3032 size_t count = iov_length(iov, nr_segs);
3035 loff_t final_size = offset + count;
3037 if (final_size > inode->i_size) {
3038 /* Credits for sb + inode write */
3039 handle = ext4_journal_start(inode, 2);
3040 if (IS_ERR(handle)) {
3041 ret = PTR_ERR(handle);
3044 ret = ext4_orphan_add(handle, inode);
3046 ext4_journal_stop(handle);
3050 ei->i_disksize = inode->i_size;
3051 ext4_journal_stop(handle);
3055 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3057 ext4_get_block, NULL);
3062 /* Credits for sb + inode write */
3063 handle = ext4_journal_start(inode, 2);
3064 if (IS_ERR(handle)) {
3065 /* This is really bad luck. We've written the data
3066 * but cannot extend i_size. Bail out and pretend
3067 * the write failed... */
3068 ret = PTR_ERR(handle);
3072 ext4_orphan_del(handle, inode);
3074 loff_t end = offset + ret;
3075 if (end > inode->i_size) {
3076 ei->i_disksize = end;
3077 i_size_write(inode, end);
3079 * We're going to return a positive `ret'
3080 * here due to non-zero-length I/O, so there's
3081 * no way of reporting error returns from
3082 * ext4_mark_inode_dirty() to userspace. So
3085 ext4_mark_inode_dirty(handle, inode);
3088 err = ext4_journal_stop(handle);
3097 * Pages can be marked dirty completely asynchronously from ext4's journalling
3098 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3099 * much here because ->set_page_dirty is called under VFS locks. The page is
3100 * not necessarily locked.
3102 * We cannot just dirty the page and leave attached buffers clean, because the
3103 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3104 * or jbddirty because all the journalling code will explode.
3106 * So what we do is to mark the page "pending dirty" and next time writepage
3107 * is called, propagate that into the buffers appropriately.
3109 static int ext4_journalled_set_page_dirty(struct page *page)
3111 SetPageChecked(page);
3112 return __set_page_dirty_nobuffers(page);
3115 static const struct address_space_operations ext4_ordered_aops = {
3116 .readpage = ext4_readpage,
3117 .readpages = ext4_readpages,
3118 .writepage = ext4_normal_writepage,
3119 .sync_page = block_sync_page,
3120 .write_begin = ext4_write_begin,
3121 .write_end = ext4_ordered_write_end,
3123 .invalidatepage = ext4_invalidatepage,
3124 .releasepage = ext4_releasepage,
3125 .direct_IO = ext4_direct_IO,
3126 .migratepage = buffer_migrate_page,
3127 .is_partially_uptodate = block_is_partially_uptodate,
3130 static const struct address_space_operations ext4_writeback_aops = {
3131 .readpage = ext4_readpage,
3132 .readpages = ext4_readpages,
3133 .writepage = ext4_normal_writepage,
3134 .sync_page = block_sync_page,
3135 .write_begin = ext4_write_begin,
3136 .write_end = ext4_writeback_write_end,
3138 .invalidatepage = ext4_invalidatepage,
3139 .releasepage = ext4_releasepage,
3140 .direct_IO = ext4_direct_IO,
3141 .migratepage = buffer_migrate_page,
3142 .is_partially_uptodate = block_is_partially_uptodate,
3145 static const struct address_space_operations ext4_journalled_aops = {
3146 .readpage = ext4_readpage,
3147 .readpages = ext4_readpages,
3148 .writepage = ext4_journalled_writepage,
3149 .sync_page = block_sync_page,
3150 .write_begin = ext4_write_begin,
3151 .write_end = ext4_journalled_write_end,
3152 .set_page_dirty = ext4_journalled_set_page_dirty,
3154 .invalidatepage = ext4_invalidatepage,
3155 .releasepage = ext4_releasepage,
3156 .is_partially_uptodate = block_is_partially_uptodate,
3159 static const struct address_space_operations ext4_da_aops = {
3160 .readpage = ext4_readpage,
3161 .readpages = ext4_readpages,
3162 .writepage = ext4_da_writepage,
3163 .writepages = ext4_da_writepages,
3164 .sync_page = block_sync_page,
3165 .write_begin = ext4_da_write_begin,
3166 .write_end = ext4_da_write_end,
3168 .invalidatepage = ext4_da_invalidatepage,
3169 .releasepage = ext4_releasepage,
3170 .direct_IO = ext4_direct_IO,
3171 .migratepage = buffer_migrate_page,
3172 .is_partially_uptodate = block_is_partially_uptodate,
3175 void ext4_set_aops(struct inode *inode)
3177 if (ext4_should_order_data(inode) &&
3178 test_opt(inode->i_sb, DELALLOC))
3179 inode->i_mapping->a_ops = &ext4_da_aops;
3180 else if (ext4_should_order_data(inode))
3181 inode->i_mapping->a_ops = &ext4_ordered_aops;
3182 else if (ext4_should_writeback_data(inode) &&
3183 test_opt(inode->i_sb, DELALLOC))
3184 inode->i_mapping->a_ops = &ext4_da_aops;
3185 else if (ext4_should_writeback_data(inode))
3186 inode->i_mapping->a_ops = &ext4_writeback_aops;
3188 inode->i_mapping->a_ops = &ext4_journalled_aops;
3192 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3193 * up to the end of the block which corresponds to `from'.
3194 * This required during truncate. We need to physically zero the tail end
3195 * of that block so it doesn't yield old data if the file is later grown.
3197 int ext4_block_truncate_page(handle_t *handle,
3198 struct address_space *mapping, loff_t from)
3200 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3201 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3202 unsigned blocksize, length, pos;
3204 struct inode *inode = mapping->host;
3205 struct buffer_head *bh;
3209 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
3213 blocksize = inode->i_sb->s_blocksize;
3214 length = blocksize - (offset & (blocksize - 1));
3215 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3218 * For "nobh" option, we can only work if we don't need to
3219 * read-in the page - otherwise we create buffers to do the IO.
3221 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
3222 ext4_should_writeback_data(inode) && PageUptodate(page)) {
3223 zero_user(page, offset, length);
3224 set_page_dirty(page);
3228 if (!page_has_buffers(page))
3229 create_empty_buffers(page, blocksize, 0);
3231 /* Find the buffer that contains "offset" */
3232 bh = page_buffers(page);
3234 while (offset >= pos) {
3235 bh = bh->b_this_page;
3241 if (buffer_freed(bh)) {
3242 BUFFER_TRACE(bh, "freed: skip");
3246 if (!buffer_mapped(bh)) {
3247 BUFFER_TRACE(bh, "unmapped");
3248 ext4_get_block(inode, iblock, bh, 0);
3249 /* unmapped? It's a hole - nothing to do */
3250 if (!buffer_mapped(bh)) {
3251 BUFFER_TRACE(bh, "still unmapped");
3256 /* Ok, it's mapped. Make sure it's up-to-date */
3257 if (PageUptodate(page))
3258 set_buffer_uptodate(bh);
3260 if (!buffer_uptodate(bh)) {
3262 ll_rw_block(READ, 1, &bh);
3264 /* Uhhuh. Read error. Complain and punt. */
3265 if (!buffer_uptodate(bh))
3269 if (ext4_should_journal_data(inode)) {
3270 BUFFER_TRACE(bh, "get write access");
3271 err = ext4_journal_get_write_access(handle, bh);
3276 zero_user(page, offset, length);
3278 BUFFER_TRACE(bh, "zeroed end of block");
3281 if (ext4_should_journal_data(inode)) {
3282 err = ext4_journal_dirty_metadata(handle, bh);
3284 if (ext4_should_order_data(inode))
3285 err = ext4_jbd2_file_inode(handle, inode);
3286 mark_buffer_dirty(bh);
3291 page_cache_release(page);
3296 * Probably it should be a library function... search for first non-zero word
3297 * or memcmp with zero_page, whatever is better for particular architecture.
3300 static inline int all_zeroes(__le32 *p, __le32 *q)
3309 * ext4_find_shared - find the indirect blocks for partial truncation.
3310 * @inode: inode in question
3311 * @depth: depth of the affected branch
3312 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3313 * @chain: place to store the pointers to partial indirect blocks
3314 * @top: place to the (detached) top of branch
3316 * This is a helper function used by ext4_truncate().
3318 * When we do truncate() we may have to clean the ends of several
3319 * indirect blocks but leave the blocks themselves alive. Block is
3320 * partially truncated if some data below the new i_size is refered
3321 * from it (and it is on the path to the first completely truncated
3322 * data block, indeed). We have to free the top of that path along
3323 * with everything to the right of the path. Since no allocation
3324 * past the truncation point is possible until ext4_truncate()
3325 * finishes, we may safely do the latter, but top of branch may
3326 * require special attention - pageout below the truncation point
3327 * might try to populate it.
3329 * We atomically detach the top of branch from the tree, store the
3330 * block number of its root in *@top, pointers to buffer_heads of
3331 * partially truncated blocks - in @chain[].bh and pointers to
3332 * their last elements that should not be removed - in
3333 * @chain[].p. Return value is the pointer to last filled element
3336 * The work left to caller to do the actual freeing of subtrees:
3337 * a) free the subtree starting from *@top
3338 * b) free the subtrees whose roots are stored in
3339 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3340 * c) free the subtrees growing from the inode past the @chain[0].
3341 * (no partially truncated stuff there). */
3343 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3344 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3346 Indirect *partial, *p;
3350 /* Make k index the deepest non-null offest + 1 */
3351 for (k = depth; k > 1 && !offsets[k-1]; k--)
3353 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3354 /* Writer: pointers */
3356 partial = chain + k-1;
3358 * If the branch acquired continuation since we've looked at it -
3359 * fine, it should all survive and (new) top doesn't belong to us.
3361 if (!partial->key && *partial->p)
3364 for (p = partial; (p > chain) && all_zeroes((__le32 *) p->bh->b_data, p->p); p--)
3367 * OK, we've found the last block that must survive. The rest of our
3368 * branch should be detached before unlocking. However, if that rest
3369 * of branch is all ours and does not grow immediately from the inode
3370 * it's easier to cheat and just decrement partial->p.
3372 if (p == chain + k - 1 && p > chain) {
3376 /* Nope, don't do this in ext4. Must leave the tree intact */
3383 while (partial > p) {
3384 brelse(partial->bh);
3392 * Zero a number of block pointers in either an inode or an indirect block.
3393 * If we restart the transaction we must again get write access to the
3394 * indirect block for further modification.
3396 * We release `count' blocks on disk, but (last - first) may be greater
3397 * than `count' because there can be holes in there.
3399 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3400 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3401 unsigned long count, __le32 *first, __le32 *last)
3404 if (try_to_extend_transaction(handle, inode)) {
3406 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3407 ext4_journal_dirty_metadata(handle, bh);
3409 ext4_mark_inode_dirty(handle, inode);
3410 ext4_journal_test_restart(handle, inode);
3412 BUFFER_TRACE(bh, "retaking write access");
3413 ext4_journal_get_write_access(handle, bh);
3418 * Any buffers which are on the journal will be in memory. We find
3419 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3420 * on them. We've already detached each block from the file, so
3421 * bforget() in jbd2_journal_forget() should be safe.
3423 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3425 for (p = first; p < last; p++) {
3426 u32 nr = le32_to_cpu(*p);
3428 struct buffer_head *tbh;
3431 tbh = sb_find_get_block(inode->i_sb, nr);
3432 ext4_forget(handle, 0, inode, tbh, nr);
3436 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3440 * ext4_free_data - free a list of data blocks
3441 * @handle: handle for this transaction
3442 * @inode: inode we are dealing with
3443 * @this_bh: indirect buffer_head which contains *@first and *@last
3444 * @first: array of block numbers
3445 * @last: points immediately past the end of array
3447 * We are freeing all blocks refered from that array (numbers are stored as
3448 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3450 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3451 * blocks are contiguous then releasing them at one time will only affect one
3452 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3453 * actually use a lot of journal space.
3455 * @this_bh will be %NULL if @first and @last point into the inode's direct
3458 static void ext4_free_data(handle_t *handle, struct inode *inode,
3459 struct buffer_head *this_bh,
3460 __le32 *first, __le32 *last)
3462 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3463 unsigned long count = 0; /* Number of blocks in the run */
3464 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3467 ext4_fsblk_t nr; /* Current block # */
3468 __le32 *p; /* Pointer into inode/ind
3469 for current block */
3472 if (this_bh) { /* For indirect block */
3473 BUFFER_TRACE(this_bh, "get_write_access");
3474 err = ext4_journal_get_write_access(handle, this_bh);
3475 /* Important: if we can't update the indirect pointers
3476 * to the blocks, we can't free them. */
3481 for (p = first; p < last; p++) {
3482 nr = le32_to_cpu(*p);
3484 /* accumulate blocks to free if they're contiguous */
3487 block_to_free_p = p;
3489 } else if (nr == block_to_free + count) {
3492 ext4_clear_blocks(handle, inode, this_bh,
3494 count, block_to_free_p, p);
3496 block_to_free_p = p;
3503 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3504 count, block_to_free_p, p);
3507 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3510 * The buffer head should have an attached journal head at this
3511 * point. However, if the data is corrupted and an indirect
3512 * block pointed to itself, it would have been detached when
3513 * the block was cleared. Check for this instead of OOPSing.
3516 ext4_journal_dirty_metadata(handle, this_bh);
3518 ext4_error(inode->i_sb, __func__,
3519 "circular indirect block detected, "
3520 "inode=%lu, block=%llu",
3522 (unsigned long long) this_bh->b_blocknr);
3527 * ext4_free_branches - free an array of branches
3528 * @handle: JBD handle for this transaction
3529 * @inode: inode we are dealing with
3530 * @parent_bh: the buffer_head which contains *@first and *@last
3531 * @first: array of block numbers
3532 * @last: pointer immediately past the end of array
3533 * @depth: depth of the branches to free
3535 * We are freeing all blocks refered from these branches (numbers are
3536 * stored as little-endian 32-bit) and updating @inode->i_blocks
3539 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3540 struct buffer_head *parent_bh,
3541 __le32 *first, __le32 *last, int depth)
3546 if (is_handle_aborted(handle))
3550 struct buffer_head *bh;
3551 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3553 while (--p >= first) {
3554 nr = le32_to_cpu(*p);
3556 continue; /* A hole */
3558 /* Go read the buffer for the next level down */
3559 bh = sb_bread(inode->i_sb, nr);
3562 * A read failure? Report error and clear slot
3566 ext4_error(inode->i_sb, "ext4_free_branches",
3567 "Read failure, inode=%lu, block=%llu",
3572 /* This zaps the entire block. Bottom up. */
3573 BUFFER_TRACE(bh, "free child branches");
3574 ext4_free_branches(handle, inode, bh,
3575 (__le32 *) bh->b_data,
3576 (__le32 *) bh->b_data + addr_per_block,
3580 * We've probably journalled the indirect block several
3581 * times during the truncate. But it's no longer
3582 * needed and we now drop it from the transaction via
3583 * jbd2_journal_revoke().
3585 * That's easy if it's exclusively part of this
3586 * transaction. But if it's part of the committing
3587 * transaction then jbd2_journal_forget() will simply
3588 * brelse() it. That means that if the underlying
3589 * block is reallocated in ext4_get_block(),
3590 * unmap_underlying_metadata() will find this block
3591 * and will try to get rid of it. damn, damn.
3593 * If this block has already been committed to the
3594 * journal, a revoke record will be written. And
3595 * revoke records must be emitted *before* clearing
3596 * this block's bit in the bitmaps.
3598 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3601 * Everything below this this pointer has been
3602 * released. Now let this top-of-subtree go.
3604 * We want the freeing of this indirect block to be
3605 * atomic in the journal with the updating of the
3606 * bitmap block which owns it. So make some room in
3609 * We zero the parent pointer *after* freeing its
3610 * pointee in the bitmaps, so if extend_transaction()
3611 * for some reason fails to put the bitmap changes and
3612 * the release into the same transaction, recovery
3613 * will merely complain about releasing a free block,
3614 * rather than leaking blocks.
3616 if (is_handle_aborted(handle))
3618 if (try_to_extend_transaction(handle, inode)) {
3619 ext4_mark_inode_dirty(handle, inode);
3620 ext4_journal_test_restart(handle, inode);
3623 ext4_free_blocks(handle, inode, nr, 1, 1);
3627 * The block which we have just freed is
3628 * pointed to by an indirect block: journal it
3630 BUFFER_TRACE(parent_bh, "get_write_access");
3631 if (!ext4_journal_get_write_access(handle,
3634 BUFFER_TRACE(parent_bh,
3635 "call ext4_journal_dirty_metadata");
3636 ext4_journal_dirty_metadata(handle,
3642 /* We have reached the bottom of the tree. */
3643 BUFFER_TRACE(parent_bh, "free data blocks");
3644 ext4_free_data(handle, inode, parent_bh, first, last);
3648 int ext4_can_truncate(struct inode *inode)
3650 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3652 if (S_ISREG(inode->i_mode))
3654 if (S_ISDIR(inode->i_mode))
3656 if (S_ISLNK(inode->i_mode))
3657 return !ext4_inode_is_fast_symlink(inode);
3664 * We block out ext4_get_block() block instantiations across the entire
3665 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3666 * simultaneously on behalf of the same inode.
3668 * As we work through the truncate and commmit bits of it to the journal there
3669 * is one core, guiding principle: the file's tree must always be consistent on
3670 * disk. We must be able to restart the truncate after a crash.
3672 * The file's tree may be transiently inconsistent in memory (although it
3673 * probably isn't), but whenever we close off and commit a journal transaction,
3674 * the contents of (the filesystem + the journal) must be consistent and
3675 * restartable. It's pretty simple, really: bottom up, right to left (although
3676 * left-to-right works OK too).
3678 * Note that at recovery time, journal replay occurs *before* the restart of
3679 * truncate against the orphan inode list.
3681 * The committed inode has the new, desired i_size (which is the same as
3682 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3683 * that this inode's truncate did not complete and it will again call
3684 * ext4_truncate() to have another go. So there will be instantiated blocks
3685 * to the right of the truncation point in a crashed ext4 filesystem. But
3686 * that's fine - as long as they are linked from the inode, the post-crash
3687 * ext4_truncate() run will find them and release them.
3689 void ext4_truncate(struct inode *inode)
3692 struct ext4_inode_info *ei = EXT4_I(inode);
3693 __le32 *i_data = ei->i_data;
3694 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3695 struct address_space *mapping = inode->i_mapping;
3696 ext4_lblk_t offsets[4];
3701 ext4_lblk_t last_block;
3702 unsigned blocksize = inode->i_sb->s_blocksize;
3704 if (!ext4_can_truncate(inode))
3707 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3708 ext4_ext_truncate(inode);
3712 handle = start_transaction(inode);
3714 return; /* AKPM: return what? */
3716 last_block = (inode->i_size + blocksize-1)
3717 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3719 if (inode->i_size & (blocksize - 1))
3720 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3723 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3725 goto out_stop; /* error */
3728 * OK. This truncate is going to happen. We add the inode to the
3729 * orphan list, so that if this truncate spans multiple transactions,
3730 * and we crash, we will resume the truncate when the filesystem
3731 * recovers. It also marks the inode dirty, to catch the new size.
3733 * Implication: the file must always be in a sane, consistent
3734 * truncatable state while each transaction commits.
3736 if (ext4_orphan_add(handle, inode))
3740 * From here we block out all ext4_get_block() callers who want to
3741 * modify the block allocation tree.
3743 down_write(&ei->i_data_sem);
3745 ext4_discard_preallocations(inode);
3748 * The orphan list entry will now protect us from any crash which
3749 * occurs before the truncate completes, so it is now safe to propagate
3750 * the new, shorter inode size (held for now in i_size) into the
3751 * on-disk inode. We do this via i_disksize, which is the value which
3752 * ext4 *really* writes onto the disk inode.
3754 ei->i_disksize = inode->i_size;
3756 if (n == 1) { /* direct blocks */
3757 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3758 i_data + EXT4_NDIR_BLOCKS);
3762 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3763 /* Kill the top of shared branch (not detached) */
3765 if (partial == chain) {
3766 /* Shared branch grows from the inode */
3767 ext4_free_branches(handle, inode, NULL,
3768 &nr, &nr+1, (chain+n-1) - partial);
3771 * We mark the inode dirty prior to restart,
3772 * and prior to stop. No need for it here.
3775 /* Shared branch grows from an indirect block */
3776 BUFFER_TRACE(partial->bh, "get_write_access");
3777 ext4_free_branches(handle, inode, partial->bh,
3779 partial->p+1, (chain+n-1) - partial);
3782 /* Clear the ends of indirect blocks on the shared branch */
3783 while (partial > chain) {
3784 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3785 (__le32*)partial->bh->b_data+addr_per_block,
3786 (chain+n-1) - partial);
3787 BUFFER_TRACE(partial->bh, "call brelse");
3788 brelse (partial->bh);
3792 /* Kill the remaining (whole) subtrees */
3793 switch (offsets[0]) {
3795 nr = i_data[EXT4_IND_BLOCK];
3797 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3798 i_data[EXT4_IND_BLOCK] = 0;
3800 case EXT4_IND_BLOCK:
3801 nr = i_data[EXT4_DIND_BLOCK];
3803 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3804 i_data[EXT4_DIND_BLOCK] = 0;
3806 case EXT4_DIND_BLOCK:
3807 nr = i_data[EXT4_TIND_BLOCK];
3809 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3810 i_data[EXT4_TIND_BLOCK] = 0;
3812 case EXT4_TIND_BLOCK:
3816 up_write(&ei->i_data_sem);
3817 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3818 ext4_mark_inode_dirty(handle, inode);
3821 * In a multi-transaction truncate, we only make the final transaction
3828 * If this was a simple ftruncate(), and the file will remain alive
3829 * then we need to clear up the orphan record which we created above.
3830 * However, if this was a real unlink then we were called by
3831 * ext4_delete_inode(), and we allow that function to clean up the
3832 * orphan info for us.
3835 ext4_orphan_del(handle, inode);
3837 ext4_journal_stop(handle);
3841 * ext4_get_inode_loc returns with an extra refcount against the inode's
3842 * underlying buffer_head on success. If 'in_mem' is true, we have all
3843 * data in memory that is needed to recreate the on-disk version of this
3846 static int __ext4_get_inode_loc(struct inode *inode,
3847 struct ext4_iloc *iloc, int in_mem)
3849 struct ext4_group_desc *gdp;
3850 struct buffer_head *bh;
3851 struct super_block *sb = inode->i_sb;
3853 int inodes_per_block, inode_offset;
3856 if (!ext4_valid_inum(sb, inode->i_ino))
3859 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3860 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3865 * Figure out the offset within the block group inode table
3867 inodes_per_block = (EXT4_BLOCK_SIZE(sb) / EXT4_INODE_SIZE(sb));
3868 inode_offset = ((inode->i_ino - 1) %
3869 EXT4_INODES_PER_GROUP(sb));
3870 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3871 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3873 bh = sb_getblk(sb, block);
3875 ext4_error(sb, "ext4_get_inode_loc", "unable to read "
3876 "inode block - inode=%lu, block=%llu",
3877 inode->i_ino, block);
3880 if (!buffer_uptodate(bh)) {
3884 * If the buffer has the write error flag, we have failed
3885 * to write out another inode in the same block. In this
3886 * case, we don't have to read the block because we may
3887 * read the old inode data successfully.
3889 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3890 set_buffer_uptodate(bh);
3892 if (buffer_uptodate(bh)) {
3893 /* someone brought it uptodate while we waited */
3899 * If we have all information of the inode in memory and this
3900 * is the only valid inode in the block, we need not read the
3904 struct buffer_head *bitmap_bh;
3907 start = inode_offset & ~(inodes_per_block - 1);
3909 /* Is the inode bitmap in cache? */
3910 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
3915 * If the inode bitmap isn't in cache then the
3916 * optimisation may end up performing two reads instead
3917 * of one, so skip it.
3919 if (!buffer_uptodate(bitmap_bh)) {
3923 for (i = start; i < start + inodes_per_block; i++) {
3924 if (i == inode_offset)
3926 if (ext4_test_bit(i, bitmap_bh->b_data))
3930 if (i == start + inodes_per_block) {
3931 /* all other inodes are free, so skip I/O */
3932 memset(bh->b_data, 0, bh->b_size);
3933 set_buffer_uptodate(bh);
3941 * If we need to do any I/O, try to pre-readahead extra
3942 * blocks from the inode table.
3944 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3945 ext4_fsblk_t b, end, table;
3948 table = ext4_inode_table(sb, gdp);
3949 /* Make sure s_inode_readahead_blks is a power of 2 */
3950 while (EXT4_SB(sb)->s_inode_readahead_blks &
3951 (EXT4_SB(sb)->s_inode_readahead_blks-1))
3952 EXT4_SB(sb)->s_inode_readahead_blks =
3953 (EXT4_SB(sb)->s_inode_readahead_blks &
3954 (EXT4_SB(sb)->s_inode_readahead_blks-1));
3955 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3958 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3959 num = EXT4_INODES_PER_GROUP(sb);
3960 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3961 EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3962 num -= le16_to_cpu(gdp->bg_itable_unused);
3963 table += num / inodes_per_block;
3967 sb_breadahead(sb, b++);
3971 * There are other valid inodes in the buffer, this inode
3972 * has in-inode xattrs, or we don't have this inode in memory.
3973 * Read the block from disk.
3976 bh->b_end_io = end_buffer_read_sync;
3977 submit_bh(READ_META, bh);
3979 if (!buffer_uptodate(bh)) {
3980 ext4_error(sb, __func__,
3981 "unable to read inode block - inode=%lu, "
3982 "block=%llu", inode->i_ino, block);
3992 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3994 /* We have all inode data except xattrs in memory here. */
3995 return __ext4_get_inode_loc(inode, iloc,
3996 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3999 void ext4_set_inode_flags(struct inode *inode)
4001 unsigned int flags = EXT4_I(inode)->i_flags;
4003 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
4004 if (flags & EXT4_SYNC_FL)
4005 inode->i_flags |= S_SYNC;
4006 if (flags & EXT4_APPEND_FL)
4007 inode->i_flags |= S_APPEND;
4008 if (flags & EXT4_IMMUTABLE_FL)
4009 inode->i_flags |= S_IMMUTABLE;
4010 if (flags & EXT4_NOATIME_FL)
4011 inode->i_flags |= S_NOATIME;
4012 if (flags & EXT4_DIRSYNC_FL)
4013 inode->i_flags |= S_DIRSYNC;
4016 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4017 void ext4_get_inode_flags(struct ext4_inode_info *ei)
4019 unsigned int flags = ei->vfs_inode.i_flags;
4021 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
4022 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
4024 ei->i_flags |= EXT4_SYNC_FL;
4025 if (flags & S_APPEND)
4026 ei->i_flags |= EXT4_APPEND_FL;
4027 if (flags & S_IMMUTABLE)
4028 ei->i_flags |= EXT4_IMMUTABLE_FL;
4029 if (flags & S_NOATIME)
4030 ei->i_flags |= EXT4_NOATIME_FL;
4031 if (flags & S_DIRSYNC)
4032 ei->i_flags |= EXT4_DIRSYNC_FL;
4034 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4035 struct ext4_inode_info *ei)
4038 struct inode *inode = &(ei->vfs_inode);
4039 struct super_block *sb = inode->i_sb;
4041 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4042 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
4043 /* we are using combined 48 bit field */
4044 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4045 le32_to_cpu(raw_inode->i_blocks_lo);
4046 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
4047 /* i_blocks represent file system block size */
4048 return i_blocks << (inode->i_blkbits - 9);
4053 return le32_to_cpu(raw_inode->i_blocks_lo);
4057 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
4059 struct ext4_iloc iloc;
4060 struct ext4_inode *raw_inode;
4061 struct ext4_inode_info *ei;
4062 struct buffer_head *bh;
4063 struct inode *inode;
4067 inode = iget_locked(sb, ino);
4069 return ERR_PTR(-ENOMEM);
4070 if (!(inode->i_state & I_NEW))
4074 #ifdef CONFIG_EXT4_FS_POSIX_ACL
4075 ei->i_acl = EXT4_ACL_NOT_CACHED;
4076 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
4079 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4083 raw_inode = ext4_raw_inode(&iloc);
4084 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4085 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4086 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4087 if (!(test_opt(inode->i_sb, NO_UID32))) {
4088 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4089 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4091 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
4094 ei->i_dir_start_lookup = 0;
4095 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4096 /* We now have enough fields to check if the inode was active or not.
4097 * This is needed because nfsd might try to access dead inodes
4098 * the test is that same one that e2fsck uses
4099 * NeilBrown 1999oct15
4101 if (inode->i_nlink == 0) {
4102 if (inode->i_mode == 0 ||
4103 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
4104 /* this inode is deleted */
4109 /* The only unlinked inodes we let through here have
4110 * valid i_mode and are being read by the orphan
4111 * recovery code: that's fine, we're about to complete
4112 * the process of deleting those. */
4114 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4115 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4116 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4117 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4118 cpu_to_le32(EXT4_OS_HURD)) {
4120 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4122 inode->i_size = ext4_isize(raw_inode);
4123 ei->i_disksize = inode->i_size;
4124 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4125 ei->i_block_group = iloc.block_group;
4127 * NOTE! The in-memory inode i_data array is in little-endian order
4128 * even on big-endian machines: we do NOT byteswap the block numbers!
4130 for (block = 0; block < EXT4_N_BLOCKS; block++)
4131 ei->i_data[block] = raw_inode->i_block[block];
4132 INIT_LIST_HEAD(&ei->i_orphan);
4134 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4135 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4136 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4137 EXT4_INODE_SIZE(inode->i_sb)) {
4142 if (ei->i_extra_isize == 0) {
4143 /* The extra space is currently unused. Use it. */
4144 ei->i_extra_isize = sizeof(struct ext4_inode) -
4145 EXT4_GOOD_OLD_INODE_SIZE;
4147 __le32 *magic = (void *)raw_inode +
4148 EXT4_GOOD_OLD_INODE_SIZE +
4150 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
4151 ei->i_state |= EXT4_STATE_XATTR;
4154 ei->i_extra_isize = 0;
4156 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4157 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4158 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4159 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4161 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4162 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4163 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4165 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4168 if (S_ISREG(inode->i_mode)) {
4169 inode->i_op = &ext4_file_inode_operations;
4170 inode->i_fop = &ext4_file_operations;
4171 ext4_set_aops(inode);
4172 } else if (S_ISDIR(inode->i_mode)) {
4173 inode->i_op = &ext4_dir_inode_operations;
4174 inode->i_fop = &ext4_dir_operations;
4175 } else if (S_ISLNK(inode->i_mode)) {
4176 if (ext4_inode_is_fast_symlink(inode)) {
4177 inode->i_op = &ext4_fast_symlink_inode_operations;
4178 nd_terminate_link(ei->i_data, inode->i_size,
4179 sizeof(ei->i_data) - 1);
4181 inode->i_op = &ext4_symlink_inode_operations;
4182 ext4_set_aops(inode);
4185 inode->i_op = &ext4_special_inode_operations;
4186 if (raw_inode->i_block[0])
4187 init_special_inode(inode, inode->i_mode,
4188 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4190 init_special_inode(inode, inode->i_mode,
4191 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4194 ext4_set_inode_flags(inode);
4195 unlock_new_inode(inode);
4200 return ERR_PTR(ret);
4203 static int ext4_inode_blocks_set(handle_t *handle,
4204 struct ext4_inode *raw_inode,
4205 struct ext4_inode_info *ei)
4207 struct inode *inode = &(ei->vfs_inode);
4208 u64 i_blocks = inode->i_blocks;
4209 struct super_block *sb = inode->i_sb;
4211 if (i_blocks <= ~0U) {
4213 * i_blocks can be represnted in a 32 bit variable
4214 * as multiple of 512 bytes
4216 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4217 raw_inode->i_blocks_high = 0;
4218 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4221 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4224 if (i_blocks <= 0xffffffffffffULL) {
4226 * i_blocks can be represented in a 48 bit variable
4227 * as multiple of 512 bytes
4229 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4230 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4231 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
4233 ei->i_flags |= EXT4_HUGE_FILE_FL;
4234 /* i_block is stored in file system block size */
4235 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4236 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4237 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4243 * Post the struct inode info into an on-disk inode location in the
4244 * buffer-cache. This gobbles the caller's reference to the
4245 * buffer_head in the inode location struct.
4247 * The caller must have write access to iloc->bh.
4249 static int ext4_do_update_inode(handle_t *handle,
4250 struct inode *inode,
4251 struct ext4_iloc *iloc)
4253 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4254 struct ext4_inode_info *ei = EXT4_I(inode);
4255 struct buffer_head *bh = iloc->bh;
4256 int err = 0, rc, block;
4258 /* For fields not not tracking in the in-memory inode,
4259 * initialise them to zero for new inodes. */
4260 if (ei->i_state & EXT4_STATE_NEW)
4261 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4263 ext4_get_inode_flags(ei);
4264 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4265 if (!(test_opt(inode->i_sb, NO_UID32))) {
4266 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4267 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4269 * Fix up interoperability with old kernels. Otherwise, old inodes get
4270 * re-used with the upper 16 bits of the uid/gid intact
4273 raw_inode->i_uid_high =
4274 cpu_to_le16(high_16_bits(inode->i_uid));
4275 raw_inode->i_gid_high =
4276 cpu_to_le16(high_16_bits(inode->i_gid));
4278 raw_inode->i_uid_high = 0;
4279 raw_inode->i_gid_high = 0;
4282 raw_inode->i_uid_low =
4283 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4284 raw_inode->i_gid_low =
4285 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4286 raw_inode->i_uid_high = 0;
4287 raw_inode->i_gid_high = 0;
4289 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4291 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4292 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4293 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4294 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4296 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4298 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4299 /* clear the migrate flag in the raw_inode */
4300 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4301 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4302 cpu_to_le32(EXT4_OS_HURD))
4303 raw_inode->i_file_acl_high =
4304 cpu_to_le16(ei->i_file_acl >> 32);
4305 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4306 ext4_isize_set(raw_inode, ei->i_disksize);
4307 if (ei->i_disksize > 0x7fffffffULL) {
4308 struct super_block *sb = inode->i_sb;
4309 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4310 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4311 EXT4_SB(sb)->s_es->s_rev_level ==
4312 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4313 /* If this is the first large file
4314 * created, add a flag to the superblock.
4316 err = ext4_journal_get_write_access(handle,
4317 EXT4_SB(sb)->s_sbh);
4320 ext4_update_dynamic_rev(sb);
4321 EXT4_SET_RO_COMPAT_FEATURE(sb,
4322 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4325 err = ext4_journal_dirty_metadata(handle,
4326 EXT4_SB(sb)->s_sbh);
4329 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4330 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4331 if (old_valid_dev(inode->i_rdev)) {
4332 raw_inode->i_block[0] =
4333 cpu_to_le32(old_encode_dev(inode->i_rdev));
4334 raw_inode->i_block[1] = 0;
4336 raw_inode->i_block[0] = 0;
4337 raw_inode->i_block[1] =
4338 cpu_to_le32(new_encode_dev(inode->i_rdev));
4339 raw_inode->i_block[2] = 0;
4341 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4342 raw_inode->i_block[block] = ei->i_data[block];
4344 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4345 if (ei->i_extra_isize) {
4346 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4347 raw_inode->i_version_hi =
4348 cpu_to_le32(inode->i_version >> 32);
4349 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4353 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4354 rc = ext4_journal_dirty_metadata(handle, bh);
4357 ei->i_state &= ~EXT4_STATE_NEW;
4361 ext4_std_error(inode->i_sb, err);
4366 * ext4_write_inode()
4368 * We are called from a few places:
4370 * - Within generic_file_write() for O_SYNC files.
4371 * Here, there will be no transaction running. We wait for any running
4372 * trasnaction to commit.
4374 * - Within sys_sync(), kupdate and such.
4375 * We wait on commit, if tol to.
4377 * - Within prune_icache() (PF_MEMALLOC == true)
4378 * Here we simply return. We can't afford to block kswapd on the
4381 * In all cases it is actually safe for us to return without doing anything,
4382 * because the inode has been copied into a raw inode buffer in
4383 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4386 * Note that we are absolutely dependent upon all inode dirtiers doing the
4387 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4388 * which we are interested.
4390 * It would be a bug for them to not do this. The code:
4392 * mark_inode_dirty(inode)
4394 * inode->i_size = expr;
4396 * is in error because a kswapd-driven write_inode() could occur while
4397 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4398 * will no longer be on the superblock's dirty inode list.
4400 int ext4_write_inode(struct inode *inode, int wait)
4402 if (current->flags & PF_MEMALLOC)
4405 if (ext4_journal_current_handle()) {
4406 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4414 return ext4_force_commit(inode->i_sb);
4420 * Called from notify_change.
4422 * We want to trap VFS attempts to truncate the file as soon as
4423 * possible. In particular, we want to make sure that when the VFS
4424 * shrinks i_size, we put the inode on the orphan list and modify
4425 * i_disksize immediately, so that during the subsequent flushing of
4426 * dirty pages and freeing of disk blocks, we can guarantee that any
4427 * commit will leave the blocks being flushed in an unused state on
4428 * disk. (On recovery, the inode will get truncated and the blocks will
4429 * be freed, so we have a strong guarantee that no future commit will
4430 * leave these blocks visible to the user.)
4432 * Another thing we have to assure is that if we are in ordered mode
4433 * and inode is still attached to the committing transaction, we must
4434 * we start writeout of all the dirty pages which are being truncated.
4435 * This way we are sure that all the data written in the previous
4436 * transaction are already on disk (truncate waits for pages under
4439 * Called with inode->i_mutex down.
4441 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4443 struct inode *inode = dentry->d_inode;
4445 const unsigned int ia_valid = attr->ia_valid;
4447 error = inode_change_ok(inode, attr);
4451 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4452 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4455 /* (user+group)*(old+new) structure, inode write (sb,
4456 * inode block, ? - but truncate inode update has it) */
4457 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4458 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4459 if (IS_ERR(handle)) {
4460 error = PTR_ERR(handle);
4463 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4465 ext4_journal_stop(handle);
4468 /* Update corresponding info in inode so that everything is in
4469 * one transaction */
4470 if (attr->ia_valid & ATTR_UID)
4471 inode->i_uid = attr->ia_uid;
4472 if (attr->ia_valid & ATTR_GID)
4473 inode->i_gid = attr->ia_gid;
4474 error = ext4_mark_inode_dirty(handle, inode);
4475 ext4_journal_stop(handle);
4478 if (attr->ia_valid & ATTR_SIZE) {
4479 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4480 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4482 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4489 if (S_ISREG(inode->i_mode) &&
4490 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4493 handle = ext4_journal_start(inode, 3);
4494 if (IS_ERR(handle)) {
4495 error = PTR_ERR(handle);
4499 error = ext4_orphan_add(handle, inode);
4500 EXT4_I(inode)->i_disksize = attr->ia_size;
4501 rc = ext4_mark_inode_dirty(handle, inode);
4504 ext4_journal_stop(handle);
4506 if (ext4_should_order_data(inode)) {
4507 error = ext4_begin_ordered_truncate(inode,
4510 /* Do as much error cleanup as possible */
4511 handle = ext4_journal_start(inode, 3);
4512 if (IS_ERR(handle)) {
4513 ext4_orphan_del(NULL, inode);
4516 ext4_orphan_del(handle, inode);
4517 ext4_journal_stop(handle);
4523 rc = inode_setattr(inode, attr);
4525 /* If inode_setattr's call to ext4_truncate failed to get a
4526 * transaction handle at all, we need to clean up the in-core
4527 * orphan list manually. */
4529 ext4_orphan_del(NULL, inode);
4531 if (!rc && (ia_valid & ATTR_MODE))
4532 rc = ext4_acl_chmod(inode);
4535 ext4_std_error(inode->i_sb, error);
4541 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4544 struct inode *inode;
4545 unsigned long delalloc_blocks;
4547 inode = dentry->d_inode;
4548 generic_fillattr(inode, stat);
4551 * We can't update i_blocks if the block allocation is delayed
4552 * otherwise in the case of system crash before the real block
4553 * allocation is done, we will have i_blocks inconsistent with
4554 * on-disk file blocks.
4555 * We always keep i_blocks updated together with real
4556 * allocation. But to not confuse with user, stat
4557 * will return the blocks that include the delayed allocation
4558 * blocks for this file.
4560 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4561 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4562 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4564 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4568 static int ext4_indirect_trans_blocks(struct inode *inode, int nrblocks,
4573 /* if nrblocks are contiguous */
4576 * With N contiguous data blocks, it need at most
4577 * N/EXT4_ADDR_PER_BLOCK(inode->i_sb) indirect blocks
4578 * 2 dindirect blocks
4581 indirects = nrblocks / EXT4_ADDR_PER_BLOCK(inode->i_sb);
4582 return indirects + 3;
4585 * if nrblocks are not contiguous, worse case, each block touch
4586 * a indirect block, and each indirect block touch a double indirect
4587 * block, plus a triple indirect block
4589 indirects = nrblocks * 2 + 1;
4593 static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4595 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
4596 return ext4_indirect_trans_blocks(inode, nrblocks, chunk);
4597 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
4601 * Account for index blocks, block groups bitmaps and block group
4602 * descriptor blocks if modify datablocks and index blocks
4603 * worse case, the indexs blocks spread over different block groups
4605 * If datablocks are discontiguous, they are possible to spread over
4606 * different block groups too. If they are contiugous, with flexbg,
4607 * they could still across block group boundary.
4609 * Also account for superblock, inode, quota and xattr blocks
4611 int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4613 int groups, gdpblocks;
4618 * How many index blocks need to touch to modify nrblocks?
4619 * The "Chunk" flag indicating whether the nrblocks is
4620 * physically contiguous on disk
4622 * For Direct IO and fallocate, they calls get_block to allocate
4623 * one single extent at a time, so they could set the "Chunk" flag
4625 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4630 * Now let's see how many group bitmaps and group descriptors need
4640 if (groups > EXT4_SB(inode->i_sb)->s_groups_count)
4641 groups = EXT4_SB(inode->i_sb)->s_groups_count;
4642 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4643 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4645 /* bitmaps and block group descriptor blocks */
4646 ret += groups + gdpblocks;
4648 /* Blocks for super block, inode, quota and xattr blocks */
4649 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4655 * Calulate the total number of credits to reserve to fit
4656 * the modification of a single pages into a single transaction,
4657 * which may include multiple chunks of block allocations.
4659 * This could be called via ext4_write_begin()
4661 * We need to consider the worse case, when
4662 * one new block per extent.
4664 int ext4_writepage_trans_blocks(struct inode *inode)
4666 int bpp = ext4_journal_blocks_per_page(inode);
4669 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4671 /* Account for data blocks for journalled mode */
4672 if (ext4_should_journal_data(inode))
4678 * Calculate the journal credits for a chunk of data modification.
4680 * This is called from DIO, fallocate or whoever calling
4681 * ext4_get_blocks_wrap() to map/allocate a chunk of contigous disk blocks.
4683 * journal buffers for data blocks are not included here, as DIO
4684 * and fallocate do no need to journal data buffers.
4686 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4688 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4692 * The caller must have previously called ext4_reserve_inode_write().
4693 * Give this, we know that the caller already has write access to iloc->bh.
4695 int ext4_mark_iloc_dirty(handle_t *handle,
4696 struct inode *inode, struct ext4_iloc *iloc)
4700 if (test_opt(inode->i_sb, I_VERSION))
4701 inode_inc_iversion(inode);
4703 /* the do_update_inode consumes one bh->b_count */
4706 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4707 err = ext4_do_update_inode(handle, inode, iloc);
4713 * On success, We end up with an outstanding reference count against
4714 * iloc->bh. This _must_ be cleaned up later.
4718 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4719 struct ext4_iloc *iloc)
4723 err = ext4_get_inode_loc(inode, iloc);
4725 BUFFER_TRACE(iloc->bh, "get_write_access");
4726 err = ext4_journal_get_write_access(handle, iloc->bh);
4733 ext4_std_error(inode->i_sb, err);
4738 * Expand an inode by new_extra_isize bytes.
4739 * Returns 0 on success or negative error number on failure.
4741 static int ext4_expand_extra_isize(struct inode *inode,
4742 unsigned int new_extra_isize,
4743 struct ext4_iloc iloc,
4746 struct ext4_inode *raw_inode;
4747 struct ext4_xattr_ibody_header *header;
4748 struct ext4_xattr_entry *entry;
4750 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4753 raw_inode = ext4_raw_inode(&iloc);
4755 header = IHDR(inode, raw_inode);
4756 entry = IFIRST(header);
4758 /* No extended attributes present */
4759 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4760 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4761 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4763 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4767 /* try to expand with EAs present */
4768 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4773 * What we do here is to mark the in-core inode as clean with respect to inode
4774 * dirtiness (it may still be data-dirty).
4775 * This means that the in-core inode may be reaped by prune_icache
4776 * without having to perform any I/O. This is a very good thing,
4777 * because *any* task may call prune_icache - even ones which
4778 * have a transaction open against a different journal.
4780 * Is this cheating? Not really. Sure, we haven't written the
4781 * inode out, but prune_icache isn't a user-visible syncing function.
4782 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4783 * we start and wait on commits.
4785 * Is this efficient/effective? Well, we're being nice to the system
4786 * by cleaning up our inodes proactively so they can be reaped
4787 * without I/O. But we are potentially leaving up to five seconds'
4788 * worth of inodes floating about which prune_icache wants us to
4789 * write out. One way to fix that would be to get prune_icache()
4790 * to do a write_super() to free up some memory. It has the desired
4793 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4795 struct ext4_iloc iloc;
4796 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4797 static unsigned int mnt_count;
4801 err = ext4_reserve_inode_write(handle, inode, &iloc);
4802 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4803 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4805 * We need extra buffer credits since we may write into EA block
4806 * with this same handle. If journal_extend fails, then it will
4807 * only result in a minor loss of functionality for that inode.
4808 * If this is felt to be critical, then e2fsck should be run to
4809 * force a large enough s_min_extra_isize.
4811 if ((jbd2_journal_extend(handle,
4812 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4813 ret = ext4_expand_extra_isize(inode,
4814 sbi->s_want_extra_isize,
4817 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4819 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4820 ext4_warning(inode->i_sb, __func__,
4821 "Unable to expand inode %lu. Delete"
4822 " some EAs or run e2fsck.",
4825 le16_to_cpu(sbi->s_es->s_mnt_count);
4831 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4836 * ext4_dirty_inode() is called from __mark_inode_dirty()
4838 * We're really interested in the case where a file is being extended.
4839 * i_size has been changed by generic_commit_write() and we thus need
4840 * to include the updated inode in the current transaction.
4842 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4843 * are allocated to the file.
4845 * If the inode is marked synchronous, we don't honour that here - doing
4846 * so would cause a commit on atime updates, which we don't bother doing.
4847 * We handle synchronous inodes at the highest possible level.
4849 void ext4_dirty_inode(struct inode *inode)
4851 handle_t *current_handle = ext4_journal_current_handle();
4854 handle = ext4_journal_start(inode, 2);
4857 if (current_handle &&
4858 current_handle->h_transaction != handle->h_transaction) {
4859 /* This task has a transaction open against a different fs */
4860 printk(KERN_EMERG "%s: transactions do not match!\n",
4863 jbd_debug(5, "marking dirty. outer handle=%p\n",
4865 ext4_mark_inode_dirty(handle, inode);
4867 ext4_journal_stop(handle);
4874 * Bind an inode's backing buffer_head into this transaction, to prevent
4875 * it from being flushed to disk early. Unlike
4876 * ext4_reserve_inode_write, this leaves behind no bh reference and
4877 * returns no iloc structure, so the caller needs to repeat the iloc
4878 * lookup to mark the inode dirty later.
4880 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4882 struct ext4_iloc iloc;
4886 err = ext4_get_inode_loc(inode, &iloc);
4888 BUFFER_TRACE(iloc.bh, "get_write_access");
4889 err = jbd2_journal_get_write_access(handle, iloc.bh);
4891 err = ext4_journal_dirty_metadata(handle,
4896 ext4_std_error(inode->i_sb, err);
4901 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4908 * We have to be very careful here: changing a data block's
4909 * journaling status dynamically is dangerous. If we write a
4910 * data block to the journal, change the status and then delete
4911 * that block, we risk forgetting to revoke the old log record
4912 * from the journal and so a subsequent replay can corrupt data.
4913 * So, first we make sure that the journal is empty and that
4914 * nobody is changing anything.
4917 journal = EXT4_JOURNAL(inode);
4918 if (is_journal_aborted(journal))
4921 jbd2_journal_lock_updates(journal);
4922 jbd2_journal_flush(journal);
4925 * OK, there are no updates running now, and all cached data is
4926 * synced to disk. We are now in a completely consistent state
4927 * which doesn't have anything in the journal, and we know that
4928 * no filesystem updates are running, so it is safe to modify
4929 * the inode's in-core data-journaling state flag now.
4933 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4935 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4936 ext4_set_aops(inode);
4938 jbd2_journal_unlock_updates(journal);
4940 /* Finally we can mark the inode as dirty. */
4942 handle = ext4_journal_start(inode, 1);
4944 return PTR_ERR(handle);
4946 err = ext4_mark_inode_dirty(handle, inode);
4948 ext4_journal_stop(handle);
4949 ext4_std_error(inode->i_sb, err);
4954 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4956 return !buffer_mapped(bh);
4959 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4965 struct file *file = vma->vm_file;
4966 struct inode *inode = file->f_path.dentry->d_inode;
4967 struct address_space *mapping = inode->i_mapping;
4970 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4971 * get i_mutex because we are already holding mmap_sem.
4973 down_read(&inode->i_alloc_sem);
4974 size = i_size_read(inode);
4975 if (page->mapping != mapping || size <= page_offset(page)
4976 || !PageUptodate(page)) {
4977 /* page got truncated from under us? */
4981 if (PageMappedToDisk(page))
4984 if (page->index == size >> PAGE_CACHE_SHIFT)
4985 len = size & ~PAGE_CACHE_MASK;
4987 len = PAGE_CACHE_SIZE;
4989 if (page_has_buffers(page)) {
4990 /* return if we have all the buffers mapped */
4991 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4996 * OK, we need to fill the hole... Do write_begin write_end
4997 * to do block allocation/reservation.We are not holding
4998 * inode.i__mutex here. That allow * parallel write_begin,
4999 * write_end call. lock_page prevent this from happening
5000 * on the same page though
5002 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
5003 len, AOP_FLAG_UNINTERRUPTIBLE, &page, &fsdata);
5006 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
5007 len, len, page, fsdata);
5012 up_read(&inode->i_alloc_sem);