2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/jbd2.h>
29 #include <linux/highuid.h>
30 #include <linux/pagemap.h>
31 #include <linux/quotaops.h>
32 #include <linux/string.h>
33 #include <linux/buffer_head.h>
34 #include <linux/writeback.h>
35 #include <linux/pagevec.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
39 #include "ext4_jbd2.h"
42 #include "ext4_extents.h"
44 static inline int ext4_begin_ordered_truncate(struct inode *inode,
47 return jbd2_journal_begin_ordered_truncate(&EXT4_I(inode)->jinode,
51 static void ext4_invalidatepage(struct page *page, unsigned long offset);
54 * Test whether an inode is a fast symlink.
56 static int ext4_inode_is_fast_symlink(struct inode *inode)
58 int ea_blocks = EXT4_I(inode)->i_file_acl ?
59 (inode->i_sb->s_blocksize >> 9) : 0;
61 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
65 * The ext4 forget function must perform a revoke if we are freeing data
66 * which has been journaled. Metadata (eg. indirect blocks) must be
67 * revoked in all cases.
69 * "bh" may be NULL: a metadata block may have been freed from memory
70 * but there may still be a record of it in the journal, and that record
71 * still needs to be revoked.
73 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
74 struct buffer_head *bh, ext4_fsblk_t blocknr)
80 BUFFER_TRACE(bh, "enter");
82 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
84 bh, is_metadata, inode->i_mode,
85 test_opt(inode->i_sb, DATA_FLAGS));
87 /* Never use the revoke function if we are doing full data
88 * journaling: there is no need to, and a V1 superblock won't
89 * support it. Otherwise, only skip the revoke on un-journaled
92 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
93 (!is_metadata && !ext4_should_journal_data(inode))) {
95 BUFFER_TRACE(bh, "call jbd2_journal_forget");
96 return ext4_journal_forget(handle, bh);
102 * data!=journal && (is_metadata || should_journal_data(inode))
104 BUFFER_TRACE(bh, "call ext4_journal_revoke");
105 err = ext4_journal_revoke(handle, blocknr, bh);
107 ext4_abort(inode->i_sb, __func__,
108 "error %d when attempting revoke", err);
109 BUFFER_TRACE(bh, "exit");
114 * Work out how many blocks we need to proceed with the next chunk of a
115 * truncate transaction.
117 static unsigned long blocks_for_truncate(struct inode *inode)
121 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
123 /* Give ourselves just enough room to cope with inodes in which
124 * i_blocks is corrupt: we've seen disk corruptions in the past
125 * which resulted in random data in an inode which looked enough
126 * like a regular file for ext4 to try to delete it. Things
127 * will go a bit crazy if that happens, but at least we should
128 * try not to panic the whole kernel. */
132 /* But we need to bound the transaction so we don't overflow the
134 if (needed > EXT4_MAX_TRANS_DATA)
135 needed = EXT4_MAX_TRANS_DATA;
137 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
141 * Truncate transactions can be complex and absolutely huge. So we need to
142 * be able to restart the transaction at a conventient checkpoint to make
143 * sure we don't overflow the journal.
145 * start_transaction gets us a new handle for a truncate transaction,
146 * and extend_transaction tries to extend the existing one a bit. If
147 * extend fails, we need to propagate the failure up and restart the
148 * transaction in the top-level truncate loop. --sct
150 static handle_t *start_transaction(struct inode *inode)
154 result = ext4_journal_start(inode, blocks_for_truncate(inode));
158 ext4_std_error(inode->i_sb, PTR_ERR(result));
163 * Try to extend this transaction for the purposes of truncation.
165 * Returns 0 if we managed to create more room. If we can't create more
166 * room, and the transaction must be restarted we return 1.
168 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
170 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
172 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
178 * Restart the transaction associated with *handle. This does a commit,
179 * so before we call here everything must be consistently dirtied against
182 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
184 jbd_debug(2, "restarting handle %p\n", handle);
185 return ext4_journal_restart(handle, blocks_for_truncate(inode));
189 * Called at the last iput() if i_nlink is zero.
191 void ext4_delete_inode (struct inode * inode)
195 if (ext4_should_order_data(inode))
196 ext4_begin_ordered_truncate(inode, 0);
197 truncate_inode_pages(&inode->i_data, 0);
199 if (is_bad_inode(inode))
202 handle = start_transaction(inode);
203 if (IS_ERR(handle)) {
205 * If we're going to skip the normal cleanup, we still need to
206 * make sure that the in-core orphan linked list is properly
209 ext4_orphan_del(NULL, inode);
217 ext4_truncate(inode);
219 * Kill off the orphan record which ext4_truncate created.
220 * AKPM: I think this can be inside the above `if'.
221 * Note that ext4_orphan_del() has to be able to cope with the
222 * deletion of a non-existent orphan - this is because we don't
223 * know if ext4_truncate() actually created an orphan record.
224 * (Well, we could do this if we need to, but heck - it works)
226 ext4_orphan_del(handle, inode);
227 EXT4_I(inode)->i_dtime = get_seconds();
230 * One subtle ordering requirement: if anything has gone wrong
231 * (transaction abort, IO errors, whatever), then we can still
232 * do these next steps (the fs will already have been marked as
233 * having errors), but we can't free the inode if the mark_dirty
236 if (ext4_mark_inode_dirty(handle, inode))
237 /* If that failed, just do the required in-core inode clear. */
240 ext4_free_inode(handle, inode);
241 ext4_journal_stop(handle);
244 clear_inode(inode); /* We must guarantee clearing of inode... */
250 struct buffer_head *bh;
253 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
255 p->key = *(p->p = v);
260 * ext4_block_to_path - parse the block number into array of offsets
261 * @inode: inode in question (we are only interested in its superblock)
262 * @i_block: block number to be parsed
263 * @offsets: array to store the offsets in
264 * @boundary: set this non-zero if the referred-to block is likely to be
265 * followed (on disk) by an indirect block.
267 * To store the locations of file's data ext4 uses a data structure common
268 * for UNIX filesystems - tree of pointers anchored in the inode, with
269 * data blocks at leaves and indirect blocks in intermediate nodes.
270 * This function translates the block number into path in that tree -
271 * return value is the path length and @offsets[n] is the offset of
272 * pointer to (n+1)th node in the nth one. If @block is out of range
273 * (negative or too large) warning is printed and zero returned.
275 * Note: function doesn't find node addresses, so no IO is needed. All
276 * we need to know is the capacity of indirect blocks (taken from the
281 * Portability note: the last comparison (check that we fit into triple
282 * indirect block) is spelled differently, because otherwise on an
283 * architecture with 32-bit longs and 8Kb pages we might get into trouble
284 * if our filesystem had 8Kb blocks. We might use long long, but that would
285 * kill us on x86. Oh, well, at least the sign propagation does not matter -
286 * i_block would have to be negative in the very beginning, so we would not
290 static int ext4_block_to_path(struct inode *inode,
292 ext4_lblk_t offsets[4], int *boundary)
294 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
295 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
296 const long direct_blocks = EXT4_NDIR_BLOCKS,
297 indirect_blocks = ptrs,
298 double_blocks = (1 << (ptrs_bits * 2));
303 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
304 } else if (i_block < direct_blocks) {
305 offsets[n++] = i_block;
306 final = direct_blocks;
307 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
308 offsets[n++] = EXT4_IND_BLOCK;
309 offsets[n++] = i_block;
311 } else if ((i_block -= indirect_blocks) < double_blocks) {
312 offsets[n++] = EXT4_DIND_BLOCK;
313 offsets[n++] = i_block >> ptrs_bits;
314 offsets[n++] = i_block & (ptrs - 1);
316 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
317 offsets[n++] = EXT4_TIND_BLOCK;
318 offsets[n++] = i_block >> (ptrs_bits * 2);
319 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
320 offsets[n++] = i_block & (ptrs - 1);
323 ext4_warning(inode->i_sb, "ext4_block_to_path",
325 i_block + direct_blocks +
326 indirect_blocks + double_blocks);
329 *boundary = final - 1 - (i_block & (ptrs - 1));
334 * ext4_get_branch - read the chain of indirect blocks leading to data
335 * @inode: inode in question
336 * @depth: depth of the chain (1 - direct pointer, etc.)
337 * @offsets: offsets of pointers in inode/indirect blocks
338 * @chain: place to store the result
339 * @err: here we store the error value
341 * Function fills the array of triples <key, p, bh> and returns %NULL
342 * if everything went OK or the pointer to the last filled triple
343 * (incomplete one) otherwise. Upon the return chain[i].key contains
344 * the number of (i+1)-th block in the chain (as it is stored in memory,
345 * i.e. little-endian 32-bit), chain[i].p contains the address of that
346 * number (it points into struct inode for i==0 and into the bh->b_data
347 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
348 * block for i>0 and NULL for i==0. In other words, it holds the block
349 * numbers of the chain, addresses they were taken from (and where we can
350 * verify that chain did not change) and buffer_heads hosting these
353 * Function stops when it stumbles upon zero pointer (absent block)
354 * (pointer to last triple returned, *@err == 0)
355 * or when it gets an IO error reading an indirect block
356 * (ditto, *@err == -EIO)
357 * or when it reads all @depth-1 indirect blocks successfully and finds
358 * the whole chain, all way to the data (returns %NULL, *err == 0).
360 * Need to be called with
361 * down_read(&EXT4_I(inode)->i_data_sem)
363 static Indirect *ext4_get_branch(struct inode *inode, int depth,
364 ext4_lblk_t *offsets,
365 Indirect chain[4], int *err)
367 struct super_block *sb = inode->i_sb;
369 struct buffer_head *bh;
372 /* i_data is not going away, no lock needed */
373 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
377 bh = sb_bread(sb, le32_to_cpu(p->key));
380 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
394 * ext4_find_near - find a place for allocation with sufficient locality
396 * @ind: descriptor of indirect block.
398 * This function returns the preferred place for block allocation.
399 * It is used when heuristic for sequential allocation fails.
401 * + if there is a block to the left of our position - allocate near it.
402 * + if pointer will live in indirect block - allocate near that block.
403 * + if pointer will live in inode - allocate in the same
406 * In the latter case we colour the starting block by the callers PID to
407 * prevent it from clashing with concurrent allocations for a different inode
408 * in the same block group. The PID is used here so that functionally related
409 * files will be close-by on-disk.
411 * Caller must make sure that @ind is valid and will stay that way.
413 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
415 struct ext4_inode_info *ei = EXT4_I(inode);
416 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
418 ext4_fsblk_t bg_start;
419 ext4_fsblk_t last_block;
420 ext4_grpblk_t colour;
422 /* Try to find previous block */
423 for (p = ind->p - 1; p >= start; p--) {
425 return le32_to_cpu(*p);
428 /* No such thing, so let's try location of indirect block */
430 return ind->bh->b_blocknr;
433 * It is going to be referred to from the inode itself? OK, just put it
434 * into the same cylinder group then.
436 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
437 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
439 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
440 colour = (current->pid % 16) *
441 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
443 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
444 return bg_start + colour;
448 * ext4_find_goal - find a preferred place for allocation.
450 * @block: block we want
451 * @partial: pointer to the last triple within a chain
453 * Normally this function find the preferred place for block allocation,
456 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
459 struct ext4_block_alloc_info *block_i;
461 block_i = EXT4_I(inode)->i_block_alloc_info;
464 * try the heuristic for sequential allocation,
465 * failing that at least try to get decent locality.
467 if (block_i && (block == block_i->last_alloc_logical_block + 1)
468 && (block_i->last_alloc_physical_block != 0)) {
469 return block_i->last_alloc_physical_block + 1;
472 return ext4_find_near(inode, partial);
476 * ext4_blks_to_allocate: Look up the block map and count the number
477 * of direct blocks need to be allocated for the given branch.
479 * @branch: chain of indirect blocks
480 * @k: number of blocks need for indirect blocks
481 * @blks: number of data blocks to be mapped.
482 * @blocks_to_boundary: the offset in the indirect block
484 * return the total number of blocks to be allocate, including the
485 * direct and indirect blocks.
487 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
488 int blocks_to_boundary)
490 unsigned long count = 0;
493 * Simple case, [t,d]Indirect block(s) has not allocated yet
494 * then it's clear blocks on that path have not allocated
497 /* right now we don't handle cross boundary allocation */
498 if (blks < blocks_to_boundary + 1)
501 count += blocks_to_boundary + 1;
506 while (count < blks && count <= blocks_to_boundary &&
507 le32_to_cpu(*(branch[0].p + count)) == 0) {
514 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
515 * @indirect_blks: the number of blocks need to allocate for indirect
518 * @new_blocks: on return it will store the new block numbers for
519 * the indirect blocks(if needed) and the first direct block,
520 * @blks: on return it will store the total number of allocated
523 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
524 ext4_lblk_t iblock, ext4_fsblk_t goal,
525 int indirect_blks, int blks,
526 ext4_fsblk_t new_blocks[4], int *err)
529 unsigned long count = 0, blk_allocated = 0;
531 ext4_fsblk_t current_block = 0;
535 * Here we try to allocate the requested multiple blocks at once,
536 * on a best-effort basis.
537 * To build a branch, we should allocate blocks for
538 * the indirect blocks(if not allocated yet), and at least
539 * the first direct block of this branch. That's the
540 * minimum number of blocks need to allocate(required)
542 /* first we try to allocate the indirect blocks */
543 target = indirect_blks;
546 /* allocating blocks for indirect blocks and direct blocks */
547 current_block = ext4_new_meta_blocks(handle, inode,
553 /* allocate blocks for indirect blocks */
554 while (index < indirect_blks && count) {
555 new_blocks[index++] = current_block++;
560 * save the new block number
561 * for the first direct block
563 new_blocks[index] = current_block;
564 printk(KERN_INFO "%s returned more blocks than "
565 "requested\n", __func__);
571 target = blks - count ;
572 blk_allocated = count;
575 /* Now allocate data blocks */
577 /* allocating blocks for data blocks */
578 current_block = ext4_new_blocks(handle, inode, iblock,
580 if (*err && (target == blks)) {
582 * if the allocation failed and we didn't allocate
588 if (target == blks) {
590 * save the new block number
591 * for the first direct block
593 new_blocks[index] = current_block;
595 blk_allocated += count;
598 /* total number of blocks allocated for direct blocks */
603 for (i = 0; i <index; i++)
604 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
609 * ext4_alloc_branch - allocate and set up a chain of blocks.
611 * @indirect_blks: number of allocated indirect blocks
612 * @blks: number of allocated direct blocks
613 * @offsets: offsets (in the blocks) to store the pointers to next.
614 * @branch: place to store the chain in.
616 * This function allocates blocks, zeroes out all but the last one,
617 * links them into chain and (if we are synchronous) writes them to disk.
618 * In other words, it prepares a branch that can be spliced onto the
619 * inode. It stores the information about that chain in the branch[], in
620 * the same format as ext4_get_branch() would do. We are calling it after
621 * we had read the existing part of chain and partial points to the last
622 * triple of that (one with zero ->key). Upon the exit we have the same
623 * picture as after the successful ext4_get_block(), except that in one
624 * place chain is disconnected - *branch->p is still zero (we did not
625 * set the last link), but branch->key contains the number that should
626 * be placed into *branch->p to fill that gap.
628 * If allocation fails we free all blocks we've allocated (and forget
629 * their buffer_heads) and return the error value the from failed
630 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
631 * as described above and return 0.
633 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
634 ext4_lblk_t iblock, int indirect_blks,
635 int *blks, ext4_fsblk_t goal,
636 ext4_lblk_t *offsets, Indirect *branch)
638 int blocksize = inode->i_sb->s_blocksize;
641 struct buffer_head *bh;
643 ext4_fsblk_t new_blocks[4];
644 ext4_fsblk_t current_block;
646 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
647 *blks, new_blocks, &err);
651 branch[0].key = cpu_to_le32(new_blocks[0]);
653 * metadata blocks and data blocks are allocated.
655 for (n = 1; n <= indirect_blks; n++) {
657 * Get buffer_head for parent block, zero it out
658 * and set the pointer to new one, then send
661 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
664 BUFFER_TRACE(bh, "call get_create_access");
665 err = ext4_journal_get_create_access(handle, bh);
672 memset(bh->b_data, 0, blocksize);
673 branch[n].p = (__le32 *) bh->b_data + offsets[n];
674 branch[n].key = cpu_to_le32(new_blocks[n]);
675 *branch[n].p = branch[n].key;
676 if ( n == indirect_blks) {
677 current_block = new_blocks[n];
679 * End of chain, update the last new metablock of
680 * the chain to point to the new allocated
681 * data blocks numbers
683 for (i=1; i < num; i++)
684 *(branch[n].p + i) = cpu_to_le32(++current_block);
686 BUFFER_TRACE(bh, "marking uptodate");
687 set_buffer_uptodate(bh);
690 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
691 err = ext4_journal_dirty_metadata(handle, bh);
698 /* Allocation failed, free what we already allocated */
699 for (i = 1; i <= n ; i++) {
700 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
701 ext4_journal_forget(handle, branch[i].bh);
703 for (i = 0; i <indirect_blks; i++)
704 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
706 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
712 * ext4_splice_branch - splice the allocated branch onto inode.
714 * @block: (logical) number of block we are adding
715 * @chain: chain of indirect blocks (with a missing link - see
717 * @where: location of missing link
718 * @num: number of indirect blocks we are adding
719 * @blks: number of direct blocks we are adding
721 * This function fills the missing link and does all housekeeping needed in
722 * inode (->i_blocks, etc.). In case of success we end up with the full
723 * chain to new block and return 0.
725 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
726 ext4_lblk_t block, Indirect *where, int num, int blks)
730 struct ext4_block_alloc_info *block_i;
731 ext4_fsblk_t current_block;
733 block_i = EXT4_I(inode)->i_block_alloc_info;
735 * If we're splicing into a [td]indirect block (as opposed to the
736 * inode) then we need to get write access to the [td]indirect block
740 BUFFER_TRACE(where->bh, "get_write_access");
741 err = ext4_journal_get_write_access(handle, where->bh);
747 *where->p = where->key;
750 * Update the host buffer_head or inode to point to more just allocated
751 * direct blocks blocks
753 if (num == 0 && blks > 1) {
754 current_block = le32_to_cpu(where->key) + 1;
755 for (i = 1; i < blks; i++)
756 *(where->p + i ) = cpu_to_le32(current_block++);
760 * update the most recently allocated logical & physical block
761 * in i_block_alloc_info, to assist find the proper goal block for next
765 block_i->last_alloc_logical_block = block + blks - 1;
766 block_i->last_alloc_physical_block =
767 le32_to_cpu(where[num].key) + blks - 1;
770 /* We are done with atomic stuff, now do the rest of housekeeping */
772 inode->i_ctime = ext4_current_time(inode);
773 ext4_mark_inode_dirty(handle, inode);
775 /* had we spliced it onto indirect block? */
778 * If we spliced it onto an indirect block, we haven't
779 * altered the inode. Note however that if it is being spliced
780 * onto an indirect block at the very end of the file (the
781 * file is growing) then we *will* alter the inode to reflect
782 * the new i_size. But that is not done here - it is done in
783 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
785 jbd_debug(5, "splicing indirect only\n");
786 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
787 err = ext4_journal_dirty_metadata(handle, where->bh);
792 * OK, we spliced it into the inode itself on a direct block.
793 * Inode was dirtied above.
795 jbd_debug(5, "splicing direct\n");
800 for (i = 1; i <= num; i++) {
801 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
802 ext4_journal_forget(handle, where[i].bh);
803 ext4_free_blocks(handle, inode,
804 le32_to_cpu(where[i-1].key), 1, 0);
806 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
812 * Allocation strategy is simple: if we have to allocate something, we will
813 * have to go the whole way to leaf. So let's do it before attaching anything
814 * to tree, set linkage between the newborn blocks, write them if sync is
815 * required, recheck the path, free and repeat if check fails, otherwise
816 * set the last missing link (that will protect us from any truncate-generated
817 * removals - all blocks on the path are immune now) and possibly force the
818 * write on the parent block.
819 * That has a nice additional property: no special recovery from the failed
820 * allocations is needed - we simply release blocks and do not touch anything
821 * reachable from inode.
823 * `handle' can be NULL if create == 0.
825 * return > 0, # of blocks mapped or allocated.
826 * return = 0, if plain lookup failed.
827 * return < 0, error case.
830 * Need to be called with
831 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
832 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
834 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
835 ext4_lblk_t iblock, unsigned long maxblocks,
836 struct buffer_head *bh_result,
837 int create, int extend_disksize)
840 ext4_lblk_t offsets[4];
845 int blocks_to_boundary = 0;
847 struct ext4_inode_info *ei = EXT4_I(inode);
849 ext4_fsblk_t first_block = 0;
853 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
854 J_ASSERT(handle != NULL || create == 0);
855 depth = ext4_block_to_path(inode, iblock, offsets,
856 &blocks_to_boundary);
861 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
863 /* Simplest case - block found, no allocation needed */
865 first_block = le32_to_cpu(chain[depth - 1].key);
866 clear_buffer_new(bh_result);
869 while (count < maxblocks && count <= blocks_to_boundary) {
872 blk = le32_to_cpu(*(chain[depth-1].p + count));
874 if (blk == first_block + count)
882 /* Next simple case - plain lookup or failed read of indirect block */
883 if (!create || err == -EIO)
887 * Okay, we need to do block allocation. Lazily initialize the block
888 * allocation info here if necessary
890 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
891 ext4_init_block_alloc_info(inode);
893 goal = ext4_find_goal(inode, iblock, partial);
895 /* the number of blocks need to allocate for [d,t]indirect blocks */
896 indirect_blks = (chain + depth) - partial - 1;
899 * Next look up the indirect map to count the totoal number of
900 * direct blocks to allocate for this branch.
902 count = ext4_blks_to_allocate(partial, indirect_blks,
903 maxblocks, blocks_to_boundary);
905 * Block out ext4_truncate while we alter the tree
907 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
909 offsets + (partial - chain), partial);
912 * The ext4_splice_branch call will free and forget any buffers
913 * on the new chain if there is a failure, but that risks using
914 * up transaction credits, especially for bitmaps where the
915 * credits cannot be returned. Can we handle this somehow? We
916 * may need to return -EAGAIN upwards in the worst case. --sct
919 err = ext4_splice_branch(handle, inode, iblock,
920 partial, indirect_blks, count);
922 * i_disksize growing is protected by i_data_sem. Don't forget to
923 * protect it if you're about to implement concurrent
924 * ext4_get_block() -bzzz
926 if (!err && extend_disksize) {
927 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
928 if (disksize > i_size_read(inode))
929 disksize = i_size_read(inode);
930 if (disksize > ei->i_disksize)
931 ei->i_disksize = disksize;
936 set_buffer_new(bh_result);
938 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
939 if (count > blocks_to_boundary)
940 set_buffer_boundary(bh_result);
942 /* Clean up and exit */
943 partial = chain + depth - 1; /* the whole chain */
945 while (partial > chain) {
946 BUFFER_TRACE(partial->bh, "call brelse");
950 BUFFER_TRACE(bh_result, "returned");
955 /* Maximum number of blocks we map for direct IO at once. */
956 #define DIO_MAX_BLOCKS 4096
958 * Number of credits we need for writing DIO_MAX_BLOCKS:
959 * We need sb + group descriptor + bitmap + inode -> 4
960 * For B blocks with A block pointers per block we need:
961 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
962 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
964 #define DIO_CREDITS 25
970 * ext4_ext4 get_block() wrapper function
971 * It will do a look up first, and returns if the blocks already mapped.
972 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
973 * and store the allocated blocks in the result buffer head and mark it
976 * If file type is extents based, it will call ext4_ext_get_blocks(),
977 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
980 * On success, it returns the number of blocks being mapped or allocate.
981 * if create==0 and the blocks are pre-allocated and uninitialized block,
982 * the result buffer head is unmapped. If the create ==1, it will make sure
983 * the buffer head is mapped.
985 * It returns 0 if plain look up failed (blocks have not been allocated), in
986 * that casem, buffer head is unmapped
988 * It returns the error in case of allocation failure.
990 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
991 unsigned long max_blocks, struct buffer_head *bh,
992 int create, int extend_disksize, int flag)
996 clear_buffer_mapped(bh);
999 * Try to see if we can get the block without requesting
1000 * for new file system block.
1002 down_read((&EXT4_I(inode)->i_data_sem));
1003 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1004 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1007 retval = ext4_get_blocks_handle(handle,
1008 inode, block, max_blocks, bh, 0, 0);
1010 up_read((&EXT4_I(inode)->i_data_sem));
1012 /* If it is only a block(s) look up */
1017 * Returns if the blocks have already allocated
1019 * Note that if blocks have been preallocated
1020 * ext4_ext_get_block() returns th create = 0
1021 * with buffer head unmapped.
1023 if (retval > 0 && buffer_mapped(bh))
1027 * New blocks allocate and/or writing to uninitialized extent
1028 * will possibly result in updating i_data, so we take
1029 * the write lock of i_data_sem, and call get_blocks()
1030 * with create == 1 flag.
1032 down_write((&EXT4_I(inode)->i_data_sem));
1035 * if the caller is from delayed allocation writeout path
1036 * we have already reserved fs blocks for allocation
1037 * let the underlying get_block() function know to
1038 * avoid double accounting
1041 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1043 * We need to check for EXT4 here because migrate
1044 * could have changed the inode type in between
1046 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1047 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1048 bh, create, extend_disksize);
1050 retval = ext4_get_blocks_handle(handle, inode, block,
1051 max_blocks, bh, create, extend_disksize);
1053 if (retval > 0 && buffer_new(bh)) {
1055 * We allocated new blocks which will result in
1056 * i_data's format changing. Force the migrate
1057 * to fail by clearing migrate flags
1059 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1065 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1067 * Update reserved blocks/metadata blocks
1068 * after successful block allocation
1069 * which were deferred till now
1071 if ((retval > 0) && buffer_delay(bh))
1072 ext4_da_release_space(inode, retval, 0);
1075 up_write((&EXT4_I(inode)->i_data_sem));
1079 static int ext4_get_block(struct inode *inode, sector_t iblock,
1080 struct buffer_head *bh_result, int create)
1082 handle_t *handle = ext4_journal_current_handle();
1083 int ret = 0, started = 0;
1084 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1086 if (create && !handle) {
1087 /* Direct IO write... */
1088 if (max_blocks > DIO_MAX_BLOCKS)
1089 max_blocks = DIO_MAX_BLOCKS;
1090 handle = ext4_journal_start(inode, DIO_CREDITS +
1091 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb));
1092 if (IS_ERR(handle)) {
1093 ret = PTR_ERR(handle);
1099 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1100 max_blocks, bh_result, create, 0, 0);
1102 bh_result->b_size = (ret << inode->i_blkbits);
1106 ext4_journal_stop(handle);
1112 * `handle' can be NULL if create is zero
1114 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1115 ext4_lblk_t block, int create, int *errp)
1117 struct buffer_head dummy;
1120 J_ASSERT(handle != NULL || create == 0);
1123 dummy.b_blocknr = -1000;
1124 buffer_trace_init(&dummy.b_history);
1125 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1126 &dummy, create, 1, 0);
1128 * ext4_get_blocks_handle() returns number of blocks
1129 * mapped. 0 in case of a HOLE.
1137 if (!err && buffer_mapped(&dummy)) {
1138 struct buffer_head *bh;
1139 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1144 if (buffer_new(&dummy)) {
1145 J_ASSERT(create != 0);
1146 J_ASSERT(handle != NULL);
1149 * Now that we do not always journal data, we should
1150 * keep in mind whether this should always journal the
1151 * new buffer as metadata. For now, regular file
1152 * writes use ext4_get_block instead, so it's not a
1156 BUFFER_TRACE(bh, "call get_create_access");
1157 fatal = ext4_journal_get_create_access(handle, bh);
1158 if (!fatal && !buffer_uptodate(bh)) {
1159 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1160 set_buffer_uptodate(bh);
1163 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1164 err = ext4_journal_dirty_metadata(handle, bh);
1168 BUFFER_TRACE(bh, "not a new buffer");
1181 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1182 ext4_lblk_t block, int create, int *err)
1184 struct buffer_head * bh;
1186 bh = ext4_getblk(handle, inode, block, create, err);
1189 if (buffer_uptodate(bh))
1191 ll_rw_block(READ_META, 1, &bh);
1193 if (buffer_uptodate(bh))
1200 static int walk_page_buffers( handle_t *handle,
1201 struct buffer_head *head,
1205 int (*fn)( handle_t *handle,
1206 struct buffer_head *bh))
1208 struct buffer_head *bh;
1209 unsigned block_start, block_end;
1210 unsigned blocksize = head->b_size;
1212 struct buffer_head *next;
1214 for ( bh = head, block_start = 0;
1215 ret == 0 && (bh != head || !block_start);
1216 block_start = block_end, bh = next)
1218 next = bh->b_this_page;
1219 block_end = block_start + blocksize;
1220 if (block_end <= from || block_start >= to) {
1221 if (partial && !buffer_uptodate(bh))
1225 err = (*fn)(handle, bh);
1233 * To preserve ordering, it is essential that the hole instantiation and
1234 * the data write be encapsulated in a single transaction. We cannot
1235 * close off a transaction and start a new one between the ext4_get_block()
1236 * and the commit_write(). So doing the jbd2_journal_start at the start of
1237 * prepare_write() is the right place.
1239 * Also, this function can nest inside ext4_writepage() ->
1240 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1241 * has generated enough buffer credits to do the whole page. So we won't
1242 * block on the journal in that case, which is good, because the caller may
1245 * By accident, ext4 can be reentered when a transaction is open via
1246 * quota file writes. If we were to commit the transaction while thus
1247 * reentered, there can be a deadlock - we would be holding a quota
1248 * lock, and the commit would never complete if another thread had a
1249 * transaction open and was blocking on the quota lock - a ranking
1252 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1253 * will _not_ run commit under these circumstances because handle->h_ref
1254 * is elevated. We'll still have enough credits for the tiny quotafile
1257 static int do_journal_get_write_access(handle_t *handle,
1258 struct buffer_head *bh)
1260 if (!buffer_mapped(bh) || buffer_freed(bh))
1262 return ext4_journal_get_write_access(handle, bh);
1265 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1266 loff_t pos, unsigned len, unsigned flags,
1267 struct page **pagep, void **fsdata)
1269 struct inode *inode = mapping->host;
1270 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1277 index = pos >> PAGE_CACHE_SHIFT;
1278 from = pos & (PAGE_CACHE_SIZE - 1);
1282 handle = ext4_journal_start(inode, needed_blocks);
1283 if (IS_ERR(handle)) {
1284 ret = PTR_ERR(handle);
1288 page = __grab_cache_page(mapping, index);
1290 ext4_journal_stop(handle);
1296 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1299 if (!ret && ext4_should_journal_data(inode)) {
1300 ret = walk_page_buffers(handle, page_buffers(page),
1301 from, to, NULL, do_journal_get_write_access);
1306 ext4_journal_stop(handle);
1307 page_cache_release(page);
1310 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1316 /* For write_end() in data=journal mode */
1317 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1319 if (!buffer_mapped(bh) || buffer_freed(bh))
1321 set_buffer_uptodate(bh);
1322 return ext4_journal_dirty_metadata(handle, bh);
1326 * We need to pick up the new inode size which generic_commit_write gave us
1327 * `file' can be NULL - eg, when called from page_symlink().
1329 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1330 * buffers are managed internally.
1332 static int ext4_ordered_write_end(struct file *file,
1333 struct address_space *mapping,
1334 loff_t pos, unsigned len, unsigned copied,
1335 struct page *page, void *fsdata)
1337 handle_t *handle = ext4_journal_current_handle();
1338 struct inode *inode = mapping->host;
1342 from = pos & (PAGE_CACHE_SIZE - 1);
1345 ret = ext4_jbd2_file_inode(handle, inode);
1349 * generic_write_end() will run mark_inode_dirty() if i_size
1350 * changes. So let's piggyback the i_disksize mark_inode_dirty
1355 new_i_size = pos + copied;
1356 if (new_i_size > EXT4_I(inode)->i_disksize)
1357 EXT4_I(inode)->i_disksize = new_i_size;
1358 ret2 = generic_write_end(file, mapping, pos, len, copied,
1364 ret2 = ext4_journal_stop(handle);
1368 return ret ? ret : copied;
1371 static int ext4_writeback_write_end(struct file *file,
1372 struct address_space *mapping,
1373 loff_t pos, unsigned len, unsigned copied,
1374 struct page *page, void *fsdata)
1376 handle_t *handle = ext4_journal_current_handle();
1377 struct inode *inode = mapping->host;
1381 new_i_size = pos + copied;
1382 if (new_i_size > EXT4_I(inode)->i_disksize)
1383 EXT4_I(inode)->i_disksize = new_i_size;
1385 ret2 = generic_write_end(file, mapping, pos, len, copied,
1391 ret2 = ext4_journal_stop(handle);
1395 return ret ? ret : copied;
1398 static int ext4_journalled_write_end(struct file *file,
1399 struct address_space *mapping,
1400 loff_t pos, unsigned len, unsigned copied,
1401 struct page *page, void *fsdata)
1403 handle_t *handle = ext4_journal_current_handle();
1404 struct inode *inode = mapping->host;
1409 from = pos & (PAGE_CACHE_SIZE - 1);
1413 if (!PageUptodate(page))
1415 page_zero_new_buffers(page, from+copied, to);
1418 ret = walk_page_buffers(handle, page_buffers(page), from,
1419 to, &partial, write_end_fn);
1421 SetPageUptodate(page);
1422 if (pos+copied > inode->i_size)
1423 i_size_write(inode, pos+copied);
1424 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1425 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1426 EXT4_I(inode)->i_disksize = inode->i_size;
1427 ret2 = ext4_mark_inode_dirty(handle, inode);
1433 ret2 = ext4_journal_stop(handle);
1436 page_cache_release(page);
1438 return ret ? ret : copied;
1441 * Calculate the number of metadata blocks need to reserve
1442 * to allocate @blocks for non extent file based file
1444 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
1446 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
1447 int ind_blks, dind_blks, tind_blks;
1449 /* number of new indirect blocks needed */
1450 ind_blks = (blocks + icap - 1) / icap;
1452 dind_blks = (ind_blks + icap - 1) / icap;
1456 return ind_blks + dind_blks + tind_blks;
1460 * Calculate the number of metadata blocks need to reserve
1461 * to allocate given number of blocks
1463 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1465 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1466 return ext4_ext_calc_metadata_amount(inode, blocks);
1468 return ext4_indirect_calc_metadata_amount(inode, blocks);
1471 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1473 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1474 unsigned long md_needed, mdblocks, total = 0;
1477 * recalculate the amount of metadata blocks to reserve
1478 * in order to allocate nrblocks
1479 * worse case is one extent per block
1481 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1482 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1483 mdblocks = ext4_calc_metadata_amount(inode, total);
1484 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1486 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1487 total = md_needed + nrblocks;
1489 if (ext4_has_free_blocks(sbi, total) < total) {
1490 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1494 /* reduce fs free blocks counter */
1495 percpu_counter_sub(&sbi->s_freeblocks_counter, total);
1497 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1498 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1500 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1501 return 0; /* success */
1504 void ext4_da_release_space(struct inode *inode, int used, int to_free)
1506 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1507 int total, mdb, mdb_free, release;
1509 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1510 /* recalculate the number of metablocks still need to be reserved */
1511 total = EXT4_I(inode)->i_reserved_data_blocks - used - to_free;
1512 mdb = ext4_calc_metadata_amount(inode, total);
1514 /* figure out how many metablocks to release */
1515 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1516 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1518 /* Account for allocated meta_blocks */
1519 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1521 release = to_free + mdb_free;
1523 /* update fs free blocks counter for truncate case */
1524 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1526 /* update per-inode reservations */
1527 BUG_ON(used + to_free > EXT4_I(inode)->i_reserved_data_blocks);
1528 EXT4_I(inode)->i_reserved_data_blocks -= (used + to_free);
1530 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1531 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1532 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1533 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1536 static void ext4_da_page_release_reservation(struct page *page,
1537 unsigned long offset)
1540 struct buffer_head *head, *bh;
1541 unsigned int curr_off = 0;
1543 head = page_buffers(page);
1546 unsigned int next_off = curr_off + bh->b_size;
1548 if ((offset <= curr_off) && (buffer_delay(bh))) {
1550 clear_buffer_delay(bh);
1552 curr_off = next_off;
1553 } while ((bh = bh->b_this_page) != head);
1554 ext4_da_release_space(page->mapping->host, 0, to_release);
1558 * Delayed allocation stuff
1561 struct mpage_da_data {
1562 struct inode *inode;
1563 struct buffer_head lbh; /* extent of blocks */
1564 unsigned long first_page, next_page; /* extent of pages */
1565 get_block_t *get_block;
1566 struct writeback_control *wbc;
1570 * mpage_da_submit_io - walks through extent of pages and try to write
1571 * them with __mpage_writepage()
1573 * @mpd->inode: inode
1574 * @mpd->first_page: first page of the extent
1575 * @mpd->next_page: page after the last page of the extent
1576 * @mpd->get_block: the filesystem's block mapper function
1578 * By the time mpage_da_submit_io() is called we expect all blocks
1579 * to be allocated. this may be wrong if allocation failed.
1581 * As pages are already locked by write_cache_pages(), we can't use it
1583 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1585 struct address_space *mapping = mpd->inode->i_mapping;
1586 struct mpage_data mpd_pp = {
1588 .last_block_in_bio = 0,
1589 .get_block = mpd->get_block,
1592 int ret = 0, err, nr_pages, i;
1593 unsigned long index, end;
1594 struct pagevec pvec;
1596 BUG_ON(mpd->next_page <= mpd->first_page);
1598 pagevec_init(&pvec, 0);
1599 index = mpd->first_page;
1600 end = mpd->next_page - 1;
1602 while (index <= end) {
1603 /* XXX: optimize tail */
1604 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1607 for (i = 0; i < nr_pages; i++) {
1608 struct page *page = pvec.pages[i];
1610 index = page->index;
1615 err = __mpage_writepage(page, mpd->wbc, &mpd_pp);
1618 * In error case, we have to continue because
1619 * remaining pages are still locked
1620 * XXX: unlock and re-dirty them?
1625 pagevec_release(&pvec);
1628 mpage_bio_submit(WRITE, mpd_pp.bio);
1634 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1636 * @mpd->inode - inode to walk through
1637 * @exbh->b_blocknr - first block on a disk
1638 * @exbh->b_size - amount of space in bytes
1639 * @logical - first logical block to start assignment with
1641 * the function goes through all passed space and put actual disk
1642 * block numbers into buffer heads, dropping BH_Delay
1644 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1645 struct buffer_head *exbh)
1647 struct inode *inode = mpd->inode;
1648 struct address_space *mapping = inode->i_mapping;
1649 int blocks = exbh->b_size >> inode->i_blkbits;
1650 sector_t pblock = exbh->b_blocknr, cur_logical;
1651 struct buffer_head *head, *bh;
1652 unsigned long index, end;
1653 struct pagevec pvec;
1656 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1657 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1658 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1660 pagevec_init(&pvec, 0);
1662 while (index <= end) {
1663 /* XXX: optimize tail */
1664 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1667 for (i = 0; i < nr_pages; i++) {
1668 struct page *page = pvec.pages[i];
1670 index = page->index;
1675 BUG_ON(!PageLocked(page));
1676 BUG_ON(PageWriteback(page));
1677 BUG_ON(!page_has_buffers(page));
1679 bh = page_buffers(page);
1682 /* skip blocks out of the range */
1684 if (cur_logical >= logical)
1687 } while ((bh = bh->b_this_page) != head);
1690 if (cur_logical >= logical + blocks)
1692 if (buffer_delay(bh)) {
1693 bh->b_blocknr = pblock;
1694 clear_buffer_delay(bh);
1695 } else if (buffer_mapped(bh))
1696 BUG_ON(bh->b_blocknr != pblock);
1700 } while ((bh = bh->b_this_page) != head);
1702 pagevec_release(&pvec);
1708 * __unmap_underlying_blocks - just a helper function to unmap
1709 * set of blocks described by @bh
1711 static inline void __unmap_underlying_blocks(struct inode *inode,
1712 struct buffer_head *bh)
1714 struct block_device *bdev = inode->i_sb->s_bdev;
1717 blocks = bh->b_size >> inode->i_blkbits;
1718 for (i = 0; i < blocks; i++)
1719 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1723 * mpage_da_map_blocks - go through given space
1725 * @mpd->lbh - bh describing space
1726 * @mpd->get_block - the filesystem's block mapper function
1728 * The function skips space we know is already mapped to disk blocks.
1730 * The function ignores errors ->get_block() returns, thus real
1731 * error handling is postponed to __mpage_writepage()
1733 static void mpage_da_map_blocks(struct mpage_da_data *mpd)
1735 struct buffer_head *lbh = &mpd->lbh;
1736 int err = 0, remain = lbh->b_size;
1737 sector_t next = lbh->b_blocknr;
1738 struct buffer_head new;
1741 * We consider only non-mapped and non-allocated blocks
1743 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1747 new.b_state = lbh->b_state;
1749 new.b_size = remain;
1750 err = mpd->get_block(mpd->inode, next, &new, 1);
1753 * Rather than implement own error handling
1754 * here, we just leave remaining blocks
1755 * unallocated and try again with ->writepage()
1759 BUG_ON(new.b_size == 0);
1761 if (buffer_new(&new))
1762 __unmap_underlying_blocks(mpd->inode, &new);
1765 * If blocks are delayed marked, we need to
1766 * put actual blocknr and drop delayed bit
1768 if (buffer_delay(lbh))
1769 mpage_put_bnr_to_bhs(mpd, next, &new);
1771 /* go for the remaining blocks */
1772 next += new.b_size >> mpd->inode->i_blkbits;
1773 remain -= new.b_size;
1777 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | (1 << BH_Delay))
1780 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1782 * @mpd->lbh - extent of blocks
1783 * @logical - logical number of the block in the file
1784 * @bh - bh of the block (used to access block's state)
1786 * the function is used to collect contig. blocks in same state
1788 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1789 sector_t logical, struct buffer_head *bh)
1791 struct buffer_head *lbh = &mpd->lbh;
1794 next = lbh->b_blocknr + (lbh->b_size >> mpd->inode->i_blkbits);
1797 * First block in the extent
1799 if (lbh->b_size == 0) {
1800 lbh->b_blocknr = logical;
1801 lbh->b_size = bh->b_size;
1802 lbh->b_state = bh->b_state & BH_FLAGS;
1807 * Can we merge the block to our big extent?
1809 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1810 lbh->b_size += bh->b_size;
1815 * We couldn't merge the block to our extent, so we
1816 * need to flush current extent and start new one
1818 mpage_da_map_blocks(mpd);
1821 * Now start a new extent
1823 lbh->b_size = bh->b_size;
1824 lbh->b_state = bh->b_state & BH_FLAGS;
1825 lbh->b_blocknr = logical;
1829 * __mpage_da_writepage - finds extent of pages and blocks
1831 * @page: page to consider
1832 * @wbc: not used, we just follow rules
1835 * The function finds extents of pages and scan them for all blocks.
1837 static int __mpage_da_writepage(struct page *page,
1838 struct writeback_control *wbc, void *data)
1840 struct mpage_da_data *mpd = data;
1841 struct inode *inode = mpd->inode;
1842 struct buffer_head *bh, *head, fake;
1846 * Can we merge this page to current extent?
1848 if (mpd->next_page != page->index) {
1850 * Nope, we can't. So, we map non-allocated blocks
1851 * and start IO on them using __mpage_writepage()
1853 if (mpd->next_page != mpd->first_page) {
1854 mpage_da_map_blocks(mpd);
1855 mpage_da_submit_io(mpd);
1859 * Start next extent of pages ...
1861 mpd->first_page = page->index;
1866 mpd->lbh.b_size = 0;
1867 mpd->lbh.b_state = 0;
1868 mpd->lbh.b_blocknr = 0;
1871 mpd->next_page = page->index + 1;
1872 logical = (sector_t) page->index <<
1873 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1875 if (!page_has_buffers(page)) {
1877 * There is no attached buffer heads yet (mmap?)
1878 * we treat the page asfull of dirty blocks
1881 bh->b_size = PAGE_CACHE_SIZE;
1883 set_buffer_dirty(bh);
1884 set_buffer_uptodate(bh);
1885 mpage_add_bh_to_extent(mpd, logical, bh);
1888 * Page with regular buffer heads, just add all dirty ones
1890 head = page_buffers(page);
1893 BUG_ON(buffer_locked(bh));
1894 if (buffer_dirty(bh))
1895 mpage_add_bh_to_extent(mpd, logical, bh);
1897 } while ((bh = bh->b_this_page) != head);
1904 * mpage_da_writepages - walk the list of dirty pages of the given
1905 * address space, allocates non-allocated blocks, maps newly-allocated
1906 * blocks to existing bhs and issue IO them
1908 * @mapping: address space structure to write
1909 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1910 * @get_block: the filesystem's block mapper function.
1912 * This is a library function, which implements the writepages()
1913 * address_space_operation.
1915 * In order to avoid duplication of logic that deals with partial pages,
1916 * multiple bio per page, etc, we find non-allocated blocks, allocate
1917 * them with minimal calls to ->get_block() and re-use __mpage_writepage()
1919 * It's important that we call __mpage_writepage() only once for each
1920 * involved page, otherwise we'd have to implement more complicated logic
1921 * to deal with pages w/o PG_lock or w/ PG_writeback and so on.
1923 * See comments to mpage_writepages()
1925 static int mpage_da_writepages(struct address_space *mapping,
1926 struct writeback_control *wbc,
1927 get_block_t get_block)
1929 struct mpage_da_data mpd;
1933 return generic_writepages(mapping, wbc);
1936 mpd.inode = mapping->host;
1938 mpd.lbh.b_state = 0;
1939 mpd.lbh.b_blocknr = 0;
1942 mpd.get_block = get_block;
1944 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
1947 * Handle last extent of pages
1949 if (mpd.next_page != mpd.first_page) {
1950 mpage_da_map_blocks(&mpd);
1951 mpage_da_submit_io(&mpd);
1958 * this is a special callback for ->write_begin() only
1959 * it's intention is to return mapped block or reserve space
1961 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1962 struct buffer_head *bh_result, int create)
1966 BUG_ON(create == 0);
1967 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
1970 * first, we need to know whether the block is allocated already
1971 * preallocated blocks are unmapped but should treated
1972 * the same as allocated blocks.
1974 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
1975 if ((ret == 0) && !buffer_delay(bh_result)) {
1976 /* the block isn't (pre)allocated yet, let's reserve space */
1978 * XXX: __block_prepare_write() unmaps passed block,
1981 ret = ext4_da_reserve_space(inode, 1);
1983 /* not enough space to reserve */
1986 map_bh(bh_result, inode->i_sb, 0);
1987 set_buffer_new(bh_result);
1988 set_buffer_delay(bh_result);
1989 } else if (ret > 0) {
1990 bh_result->b_size = (ret << inode->i_blkbits);
1996 #define EXT4_DELALLOC_RSVED 1
1997 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
1998 struct buffer_head *bh_result, int create)
2001 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2002 loff_t disksize = EXT4_I(inode)->i_disksize;
2003 handle_t *handle = NULL;
2005 handle = ext4_journal_current_handle();
2006 BUG_ON(handle == NULL);
2007 BUG_ON(create == 0);
2009 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2010 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2012 bh_result->b_size = (ret << inode->i_blkbits);
2015 * Update on-disk size along with block allocation
2016 * we don't use 'extend_disksize' as size may change
2017 * within already allocated block -bzzz
2019 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2020 if (disksize > i_size_read(inode))
2021 disksize = i_size_read(inode);
2022 if (disksize > EXT4_I(inode)->i_disksize) {
2024 * XXX: replace with spinlock if seen contended -bzzz
2026 down_write(&EXT4_I(inode)->i_data_sem);
2027 if (disksize > EXT4_I(inode)->i_disksize)
2028 EXT4_I(inode)->i_disksize = disksize;
2029 up_write(&EXT4_I(inode)->i_data_sem);
2031 if (EXT4_I(inode)->i_disksize == disksize) {
2032 ret = ext4_mark_inode_dirty(handle, inode);
2041 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2043 return !buffer_mapped(bh) || buffer_delay(bh);
2046 /* FIXME!! only support data=writeback mode */
2048 * get called vi ext4_da_writepages after taking page lock
2049 * We may end up doing block allocation here in case
2050 * mpage_da_map_blocks failed to allocate blocks.
2052 static int ext4_da_writepage(struct page *page,
2053 struct writeback_control *wbc)
2058 handle_t *handle = NULL;
2059 struct buffer_head *page_bufs;
2060 struct inode *inode = page->mapping->host;
2062 handle = ext4_journal_current_handle();
2065 * This can happen when we aren't called via
2066 * ext4_da_writepages() but directly (shrink_page_list).
2067 * We cannot easily start a transaction here so we just skip
2068 * writing the page in case we would have to do so.
2070 size = i_size_read(inode);
2072 page_bufs = page_buffers(page);
2073 if (page->index == size >> PAGE_CACHE_SHIFT)
2074 len = size & ~PAGE_CACHE_MASK;
2076 len = PAGE_CACHE_SIZE;
2078 if (walk_page_buffers(NULL, page_bufs, 0,
2079 len, NULL, ext4_bh_unmapped_or_delay)) {
2081 * We can't do block allocation under
2082 * page lock without a handle . So redirty
2083 * the page and return
2085 BUG_ON(wbc->sync_mode != WB_SYNC_NONE);
2086 redirty_page_for_writepage(wbc, page);
2092 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2093 ret = nobh_writepage(page, ext4_da_get_block_write, wbc);
2095 ret = block_write_full_page(page, ext4_da_get_block_write, wbc);
2102 * For now just follow the DIO way to estimate the max credits
2103 * needed to write out EXT4_MAX_WRITEBACK_PAGES.
2104 * todo: need to calculate the max credits need for
2105 * extent based files, currently the DIO credits is based on
2106 * indirect-blocks mapping way.
2108 * Probably should have a generic way to calculate credits
2109 * for DIO, writepages, and truncate
2111 #define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS
2112 #define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS
2114 static int ext4_da_writepages(struct address_space *mapping,
2115 struct writeback_control *wbc)
2117 struct inode *inode = mapping->host;
2118 handle_t *handle = NULL;
2122 loff_t range_start = 0;
2125 * No pages to write? This is mainly a kludge to avoid starting
2126 * a transaction for special inodes like journal inode on last iput()
2127 * because that could violate lock ordering on umount
2129 if (!mapping->nrpages)
2133 * Estimate the worse case needed credits to write out
2134 * EXT4_MAX_BUF_BLOCKS pages
2136 needed_blocks = EXT4_MAX_WRITEBACK_CREDITS;
2138 to_write = wbc->nr_to_write;
2139 if (!wbc->range_cyclic) {
2141 * If range_cyclic is not set force range_cont
2142 * and save the old writeback_index
2144 wbc->range_cont = 1;
2145 range_start = wbc->range_start;
2148 while (!ret && to_write) {
2149 /* start a new transaction*/
2150 handle = ext4_journal_start(inode, needed_blocks);
2151 if (IS_ERR(handle)) {
2152 ret = PTR_ERR(handle);
2153 goto out_writepages;
2156 * set the max dirty pages could be write at a time
2157 * to fit into the reserved transaction credits
2159 if (wbc->nr_to_write > EXT4_MAX_WRITEBACK_PAGES)
2160 wbc->nr_to_write = EXT4_MAX_WRITEBACK_PAGES;
2162 to_write -= wbc->nr_to_write;
2163 ret = mpage_da_writepages(mapping, wbc,
2164 ext4_da_get_block_write);
2165 ext4_journal_stop(handle);
2166 if (wbc->nr_to_write) {
2168 * There is no more writeout needed
2169 * or we requested for a noblocking writeout
2170 * and we found the device congested
2172 to_write += wbc->nr_to_write;
2175 wbc->nr_to_write = to_write;
2179 wbc->nr_to_write = to_write;
2181 wbc->range_start = range_start;
2185 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2186 loff_t pos, unsigned len, unsigned flags,
2187 struct page **pagep, void **fsdata)
2189 int ret, retries = 0;
2193 struct inode *inode = mapping->host;
2196 index = pos >> PAGE_CACHE_SHIFT;
2197 from = pos & (PAGE_CACHE_SIZE - 1);
2202 * With delayed allocation, we don't log the i_disksize update
2203 * if there is delayed block allocation. But we still need
2204 * to journalling the i_disksize update if writes to the end
2205 * of file which has an already mapped buffer.
2207 handle = ext4_journal_start(inode, 1);
2208 if (IS_ERR(handle)) {
2209 ret = PTR_ERR(handle);
2213 page = __grab_cache_page(mapping, index);
2218 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2219 ext4_da_get_block_prep);
2222 ext4_journal_stop(handle);
2223 page_cache_release(page);
2226 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2232 static int ext4_da_write_end(struct file *file,
2233 struct address_space *mapping,
2234 loff_t pos, unsigned len, unsigned copied,
2235 struct page *page, void *fsdata)
2237 struct inode *inode = mapping->host;
2239 handle_t *handle = ext4_journal_current_handle();
2243 * generic_write_end() will run mark_inode_dirty() if i_size
2244 * changes. So let's piggyback the i_disksize mark_inode_dirty
2248 new_i_size = pos + copied;
2249 if (new_i_size > EXT4_I(inode)->i_disksize)
2250 if (!walk_page_buffers(NULL, page_buffers(page),
2251 0, len, NULL, ext4_bh_unmapped_or_delay)){
2253 * Updating i_disksize when extending file without
2254 * needing block allocation
2256 if (ext4_should_order_data(inode))
2257 ret = ext4_jbd2_file_inode(handle, inode);
2259 EXT4_I(inode)->i_disksize = new_i_size;
2261 ret2 = generic_write_end(file, mapping, pos, len, copied,
2266 ret2 = ext4_journal_stop(handle);
2270 return ret ? ret : copied;
2273 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2276 * Drop reserved blocks
2278 BUG_ON(!PageLocked(page));
2279 if (!page_has_buffers(page))
2282 ext4_da_page_release_reservation(page, offset);
2285 ext4_invalidatepage(page, offset);
2292 * bmap() is special. It gets used by applications such as lilo and by
2293 * the swapper to find the on-disk block of a specific piece of data.
2295 * Naturally, this is dangerous if the block concerned is still in the
2296 * journal. If somebody makes a swapfile on an ext4 data-journaling
2297 * filesystem and enables swap, then they may get a nasty shock when the
2298 * data getting swapped to that swapfile suddenly gets overwritten by
2299 * the original zero's written out previously to the journal and
2300 * awaiting writeback in the kernel's buffer cache.
2302 * So, if we see any bmap calls here on a modified, data-journaled file,
2303 * take extra steps to flush any blocks which might be in the cache.
2305 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2307 struct inode *inode = mapping->host;
2311 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2312 test_opt(inode->i_sb, DELALLOC)) {
2314 * With delalloc we want to sync the file
2315 * so that we can make sure we allocate
2318 filemap_write_and_wait(mapping);
2321 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2323 * This is a REALLY heavyweight approach, but the use of
2324 * bmap on dirty files is expected to be extremely rare:
2325 * only if we run lilo or swapon on a freshly made file
2326 * do we expect this to happen.
2328 * (bmap requires CAP_SYS_RAWIO so this does not
2329 * represent an unprivileged user DOS attack --- we'd be
2330 * in trouble if mortal users could trigger this path at
2333 * NB. EXT4_STATE_JDATA is not set on files other than
2334 * regular files. If somebody wants to bmap a directory
2335 * or symlink and gets confused because the buffer
2336 * hasn't yet been flushed to disk, they deserve
2337 * everything they get.
2340 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2341 journal = EXT4_JOURNAL(inode);
2342 jbd2_journal_lock_updates(journal);
2343 err = jbd2_journal_flush(journal);
2344 jbd2_journal_unlock_updates(journal);
2350 return generic_block_bmap(mapping,block,ext4_get_block);
2353 static int bget_one(handle_t *handle, struct buffer_head *bh)
2359 static int bput_one(handle_t *handle, struct buffer_head *bh)
2366 * Note that we don't need to start a transaction unless we're journaling data
2367 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2368 * need to file the inode to the transaction's list in ordered mode because if
2369 * we are writing back data added by write(), the inode is already there and if
2370 * we are writing back data modified via mmap(), noone guarantees in which
2371 * transaction the data will hit the disk. In case we are journaling data, we
2372 * cannot start transaction directly because transaction start ranks above page
2373 * lock so we have to do some magic.
2375 * In all journaling modes block_write_full_page() will start the I/O.
2379 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2384 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2386 * Same applies to ext4_get_block(). We will deadlock on various things like
2387 * lock_journal and i_data_sem
2389 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2392 * 16May01: If we're reentered then journal_current_handle() will be
2393 * non-zero. We simply *return*.
2395 * 1 July 2001: @@@ FIXME:
2396 * In journalled data mode, a data buffer may be metadata against the
2397 * current transaction. But the same file is part of a shared mapping
2398 * and someone does a writepage() on it.
2400 * We will move the buffer onto the async_data list, but *after* it has
2401 * been dirtied. So there's a small window where we have dirty data on
2404 * Note that this only applies to the last partial page in the file. The
2405 * bit which block_write_full_page() uses prepare/commit for. (That's
2406 * broken code anyway: it's wrong for msync()).
2408 * It's a rare case: affects the final partial page, for journalled data
2409 * where the file is subject to bith write() and writepage() in the same
2410 * transction. To fix it we'll need a custom block_write_full_page().
2411 * We'll probably need that anyway for journalling writepage() output.
2413 * We don't honour synchronous mounts for writepage(). That would be
2414 * disastrous. Any write() or metadata operation will sync the fs for
2418 static int __ext4_normal_writepage(struct page *page,
2419 struct writeback_control *wbc)
2421 struct inode *inode = page->mapping->host;
2423 if (test_opt(inode->i_sb, NOBH))
2424 return nobh_writepage(page, ext4_get_block, wbc);
2426 return block_write_full_page(page, ext4_get_block, wbc);
2430 static int ext4_normal_writepage(struct page *page,
2431 struct writeback_control *wbc)
2433 struct inode *inode = page->mapping->host;
2434 loff_t size = i_size_read(inode);
2437 J_ASSERT(PageLocked(page));
2438 J_ASSERT(page_has_buffers(page));
2439 if (page->index == size >> PAGE_CACHE_SHIFT)
2440 len = size & ~PAGE_CACHE_MASK;
2442 len = PAGE_CACHE_SIZE;
2443 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2444 ext4_bh_unmapped_or_delay));
2446 if (!ext4_journal_current_handle())
2447 return __ext4_normal_writepage(page, wbc);
2449 redirty_page_for_writepage(wbc, page);
2454 static int __ext4_journalled_writepage(struct page *page,
2455 struct writeback_control *wbc)
2457 struct address_space *mapping = page->mapping;
2458 struct inode *inode = mapping->host;
2459 struct buffer_head *page_bufs;
2460 handle_t *handle = NULL;
2464 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE, ext4_get_block);
2468 page_bufs = page_buffers(page);
2469 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2471 /* As soon as we unlock the page, it can go away, but we have
2472 * references to buffers so we are safe */
2475 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2476 if (IS_ERR(handle)) {
2477 ret = PTR_ERR(handle);
2481 ret = walk_page_buffers(handle, page_bufs, 0,
2482 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2484 err = walk_page_buffers(handle, page_bufs, 0,
2485 PAGE_CACHE_SIZE, NULL, write_end_fn);
2488 err = ext4_journal_stop(handle);
2492 walk_page_buffers(handle, page_bufs, 0,
2493 PAGE_CACHE_SIZE, NULL, bput_one);
2494 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2503 static int ext4_journalled_writepage(struct page *page,
2504 struct writeback_control *wbc)
2506 struct inode *inode = page->mapping->host;
2507 loff_t size = i_size_read(inode);
2510 J_ASSERT(PageLocked(page));
2511 J_ASSERT(page_has_buffers(page));
2512 if (page->index == size >> PAGE_CACHE_SHIFT)
2513 len = size & ~PAGE_CACHE_MASK;
2515 len = PAGE_CACHE_SIZE;
2516 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2517 ext4_bh_unmapped_or_delay));
2519 if (ext4_journal_current_handle())
2522 if (PageChecked(page)) {
2524 * It's mmapped pagecache. Add buffers and journal it. There
2525 * doesn't seem much point in redirtying the page here.
2527 ClearPageChecked(page);
2528 return __ext4_journalled_writepage(page, wbc);
2531 * It may be a page full of checkpoint-mode buffers. We don't
2532 * really know unless we go poke around in the buffer_heads.
2533 * But block_write_full_page will do the right thing.
2535 return block_write_full_page(page, ext4_get_block, wbc);
2538 redirty_page_for_writepage(wbc, page);
2543 static int ext4_readpage(struct file *file, struct page *page)
2545 return mpage_readpage(page, ext4_get_block);
2549 ext4_readpages(struct file *file, struct address_space *mapping,
2550 struct list_head *pages, unsigned nr_pages)
2552 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2555 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2557 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2560 * If it's a full truncate we just forget about the pending dirtying
2563 ClearPageChecked(page);
2565 jbd2_journal_invalidatepage(journal, page, offset);
2568 static int ext4_releasepage(struct page *page, gfp_t wait)
2570 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2572 WARN_ON(PageChecked(page));
2573 if (!page_has_buffers(page))
2575 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2579 * If the O_DIRECT write will extend the file then add this inode to the
2580 * orphan list. So recovery will truncate it back to the original size
2581 * if the machine crashes during the write.
2583 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2584 * crashes then stale disk data _may_ be exposed inside the file. But current
2585 * VFS code falls back into buffered path in that case so we are safe.
2587 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2588 const struct iovec *iov, loff_t offset,
2589 unsigned long nr_segs)
2591 struct file *file = iocb->ki_filp;
2592 struct inode *inode = file->f_mapping->host;
2593 struct ext4_inode_info *ei = EXT4_I(inode);
2597 size_t count = iov_length(iov, nr_segs);
2600 loff_t final_size = offset + count;
2602 if (final_size > inode->i_size) {
2603 /* Credits for sb + inode write */
2604 handle = ext4_journal_start(inode, 2);
2605 if (IS_ERR(handle)) {
2606 ret = PTR_ERR(handle);
2609 ret = ext4_orphan_add(handle, inode);
2611 ext4_journal_stop(handle);
2615 ei->i_disksize = inode->i_size;
2616 ext4_journal_stop(handle);
2620 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2622 ext4_get_block, NULL);
2627 /* Credits for sb + inode write */
2628 handle = ext4_journal_start(inode, 2);
2629 if (IS_ERR(handle)) {
2630 /* This is really bad luck. We've written the data
2631 * but cannot extend i_size. Bail out and pretend
2632 * the write failed... */
2633 ret = PTR_ERR(handle);
2637 ext4_orphan_del(handle, inode);
2639 loff_t end = offset + ret;
2640 if (end > inode->i_size) {
2641 ei->i_disksize = end;
2642 i_size_write(inode, end);
2644 * We're going to return a positive `ret'
2645 * here due to non-zero-length I/O, so there's
2646 * no way of reporting error returns from
2647 * ext4_mark_inode_dirty() to userspace. So
2650 ext4_mark_inode_dirty(handle, inode);
2653 err = ext4_journal_stop(handle);
2662 * Pages can be marked dirty completely asynchronously from ext4's journalling
2663 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2664 * much here because ->set_page_dirty is called under VFS locks. The page is
2665 * not necessarily locked.
2667 * We cannot just dirty the page and leave attached buffers clean, because the
2668 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2669 * or jbddirty because all the journalling code will explode.
2671 * So what we do is to mark the page "pending dirty" and next time writepage
2672 * is called, propagate that into the buffers appropriately.
2674 static int ext4_journalled_set_page_dirty(struct page *page)
2676 SetPageChecked(page);
2677 return __set_page_dirty_nobuffers(page);
2680 static const struct address_space_operations ext4_ordered_aops = {
2681 .readpage = ext4_readpage,
2682 .readpages = ext4_readpages,
2683 .writepage = ext4_normal_writepage,
2684 .sync_page = block_sync_page,
2685 .write_begin = ext4_write_begin,
2686 .write_end = ext4_ordered_write_end,
2688 .invalidatepage = ext4_invalidatepage,
2689 .releasepage = ext4_releasepage,
2690 .direct_IO = ext4_direct_IO,
2691 .migratepage = buffer_migrate_page,
2694 static const struct address_space_operations ext4_writeback_aops = {
2695 .readpage = ext4_readpage,
2696 .readpages = ext4_readpages,
2697 .writepage = ext4_normal_writepage,
2698 .sync_page = block_sync_page,
2699 .write_begin = ext4_write_begin,
2700 .write_end = ext4_writeback_write_end,
2702 .invalidatepage = ext4_invalidatepage,
2703 .releasepage = ext4_releasepage,
2704 .direct_IO = ext4_direct_IO,
2705 .migratepage = buffer_migrate_page,
2708 static const struct address_space_operations ext4_journalled_aops = {
2709 .readpage = ext4_readpage,
2710 .readpages = ext4_readpages,
2711 .writepage = ext4_journalled_writepage,
2712 .sync_page = block_sync_page,
2713 .write_begin = ext4_write_begin,
2714 .write_end = ext4_journalled_write_end,
2715 .set_page_dirty = ext4_journalled_set_page_dirty,
2717 .invalidatepage = ext4_invalidatepage,
2718 .releasepage = ext4_releasepage,
2721 static const struct address_space_operations ext4_da_aops = {
2722 .readpage = ext4_readpage,
2723 .readpages = ext4_readpages,
2724 .writepage = ext4_da_writepage,
2725 .writepages = ext4_da_writepages,
2726 .sync_page = block_sync_page,
2727 .write_begin = ext4_da_write_begin,
2728 .write_end = ext4_da_write_end,
2730 .invalidatepage = ext4_da_invalidatepage,
2731 .releasepage = ext4_releasepage,
2732 .direct_IO = ext4_direct_IO,
2733 .migratepage = buffer_migrate_page,
2736 void ext4_set_aops(struct inode *inode)
2738 if (ext4_should_order_data(inode))
2739 inode->i_mapping->a_ops = &ext4_ordered_aops;
2740 else if (ext4_should_writeback_data(inode) &&
2741 test_opt(inode->i_sb, DELALLOC))
2742 inode->i_mapping->a_ops = &ext4_da_aops;
2743 else if (ext4_should_writeback_data(inode))
2744 inode->i_mapping->a_ops = &ext4_writeback_aops;
2746 inode->i_mapping->a_ops = &ext4_journalled_aops;
2750 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2751 * up to the end of the block which corresponds to `from'.
2752 * This required during truncate. We need to physically zero the tail end
2753 * of that block so it doesn't yield old data if the file is later grown.
2755 int ext4_block_truncate_page(handle_t *handle,
2756 struct address_space *mapping, loff_t from)
2758 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2759 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2760 unsigned blocksize, length, pos;
2762 struct inode *inode = mapping->host;
2763 struct buffer_head *bh;
2767 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
2771 blocksize = inode->i_sb->s_blocksize;
2772 length = blocksize - (offset & (blocksize - 1));
2773 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2776 * For "nobh" option, we can only work if we don't need to
2777 * read-in the page - otherwise we create buffers to do the IO.
2779 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
2780 ext4_should_writeback_data(inode) && PageUptodate(page)) {
2781 zero_user(page, offset, length);
2782 set_page_dirty(page);
2786 if (!page_has_buffers(page))
2787 create_empty_buffers(page, blocksize, 0);
2789 /* Find the buffer that contains "offset" */
2790 bh = page_buffers(page);
2792 while (offset >= pos) {
2793 bh = bh->b_this_page;
2799 if (buffer_freed(bh)) {
2800 BUFFER_TRACE(bh, "freed: skip");
2804 if (!buffer_mapped(bh)) {
2805 BUFFER_TRACE(bh, "unmapped");
2806 ext4_get_block(inode, iblock, bh, 0);
2807 /* unmapped? It's a hole - nothing to do */
2808 if (!buffer_mapped(bh)) {
2809 BUFFER_TRACE(bh, "still unmapped");
2814 /* Ok, it's mapped. Make sure it's up-to-date */
2815 if (PageUptodate(page))
2816 set_buffer_uptodate(bh);
2818 if (!buffer_uptodate(bh)) {
2820 ll_rw_block(READ, 1, &bh);
2822 /* Uhhuh. Read error. Complain and punt. */
2823 if (!buffer_uptodate(bh))
2827 if (ext4_should_journal_data(inode)) {
2828 BUFFER_TRACE(bh, "get write access");
2829 err = ext4_journal_get_write_access(handle, bh);
2834 zero_user(page, offset, length);
2836 BUFFER_TRACE(bh, "zeroed end of block");
2839 if (ext4_should_journal_data(inode)) {
2840 err = ext4_journal_dirty_metadata(handle, bh);
2842 if (ext4_should_order_data(inode))
2843 err = ext4_jbd2_file_inode(handle, inode);
2844 mark_buffer_dirty(bh);
2849 page_cache_release(page);
2854 * Probably it should be a library function... search for first non-zero word
2855 * or memcmp with zero_page, whatever is better for particular architecture.
2858 static inline int all_zeroes(__le32 *p, __le32 *q)
2867 * ext4_find_shared - find the indirect blocks for partial truncation.
2868 * @inode: inode in question
2869 * @depth: depth of the affected branch
2870 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
2871 * @chain: place to store the pointers to partial indirect blocks
2872 * @top: place to the (detached) top of branch
2874 * This is a helper function used by ext4_truncate().
2876 * When we do truncate() we may have to clean the ends of several
2877 * indirect blocks but leave the blocks themselves alive. Block is
2878 * partially truncated if some data below the new i_size is refered
2879 * from it (and it is on the path to the first completely truncated
2880 * data block, indeed). We have to free the top of that path along
2881 * with everything to the right of the path. Since no allocation
2882 * past the truncation point is possible until ext4_truncate()
2883 * finishes, we may safely do the latter, but top of branch may
2884 * require special attention - pageout below the truncation point
2885 * might try to populate it.
2887 * We atomically detach the top of branch from the tree, store the
2888 * block number of its root in *@top, pointers to buffer_heads of
2889 * partially truncated blocks - in @chain[].bh and pointers to
2890 * their last elements that should not be removed - in
2891 * @chain[].p. Return value is the pointer to last filled element
2894 * The work left to caller to do the actual freeing of subtrees:
2895 * a) free the subtree starting from *@top
2896 * b) free the subtrees whose roots are stored in
2897 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
2898 * c) free the subtrees growing from the inode past the @chain[0].
2899 * (no partially truncated stuff there). */
2901 static Indirect *ext4_find_shared(struct inode *inode, int depth,
2902 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
2904 Indirect *partial, *p;
2908 /* Make k index the deepest non-null offest + 1 */
2909 for (k = depth; k > 1 && !offsets[k-1]; k--)
2911 partial = ext4_get_branch(inode, k, offsets, chain, &err);
2912 /* Writer: pointers */
2914 partial = chain + k-1;
2916 * If the branch acquired continuation since we've looked at it -
2917 * fine, it should all survive and (new) top doesn't belong to us.
2919 if (!partial->key && *partial->p)
2922 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
2925 * OK, we've found the last block that must survive. The rest of our
2926 * branch should be detached before unlocking. However, if that rest
2927 * of branch is all ours and does not grow immediately from the inode
2928 * it's easier to cheat and just decrement partial->p.
2930 if (p == chain + k - 1 && p > chain) {
2934 /* Nope, don't do this in ext4. Must leave the tree intact */
2941 while(partial > p) {
2942 brelse(partial->bh);
2950 * Zero a number of block pointers in either an inode or an indirect block.
2951 * If we restart the transaction we must again get write access to the
2952 * indirect block for further modification.
2954 * We release `count' blocks on disk, but (last - first) may be greater
2955 * than `count' because there can be holes in there.
2957 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
2958 struct buffer_head *bh, ext4_fsblk_t block_to_free,
2959 unsigned long count, __le32 *first, __le32 *last)
2962 if (try_to_extend_transaction(handle, inode)) {
2964 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2965 ext4_journal_dirty_metadata(handle, bh);
2967 ext4_mark_inode_dirty(handle, inode);
2968 ext4_journal_test_restart(handle, inode);
2970 BUFFER_TRACE(bh, "retaking write access");
2971 ext4_journal_get_write_access(handle, bh);
2976 * Any buffers which are on the journal will be in memory. We find
2977 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
2978 * on them. We've already detached each block from the file, so
2979 * bforget() in jbd2_journal_forget() should be safe.
2981 * AKPM: turn on bforget in jbd2_journal_forget()!!!
2983 for (p = first; p < last; p++) {
2984 u32 nr = le32_to_cpu(*p);
2986 struct buffer_head *tbh;
2989 tbh = sb_find_get_block(inode->i_sb, nr);
2990 ext4_forget(handle, 0, inode, tbh, nr);
2994 ext4_free_blocks(handle, inode, block_to_free, count, 0);
2998 * ext4_free_data - free a list of data blocks
2999 * @handle: handle for this transaction
3000 * @inode: inode we are dealing with
3001 * @this_bh: indirect buffer_head which contains *@first and *@last
3002 * @first: array of block numbers
3003 * @last: points immediately past the end of array
3005 * We are freeing all blocks refered from that array (numbers are stored as
3006 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3008 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3009 * blocks are contiguous then releasing them at one time will only affect one
3010 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3011 * actually use a lot of journal space.
3013 * @this_bh will be %NULL if @first and @last point into the inode's direct
3016 static void ext4_free_data(handle_t *handle, struct inode *inode,
3017 struct buffer_head *this_bh,
3018 __le32 *first, __le32 *last)
3020 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3021 unsigned long count = 0; /* Number of blocks in the run */
3022 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3025 ext4_fsblk_t nr; /* Current block # */
3026 __le32 *p; /* Pointer into inode/ind
3027 for current block */
3030 if (this_bh) { /* For indirect block */
3031 BUFFER_TRACE(this_bh, "get_write_access");
3032 err = ext4_journal_get_write_access(handle, this_bh);
3033 /* Important: if we can't update the indirect pointers
3034 * to the blocks, we can't free them. */
3039 for (p = first; p < last; p++) {
3040 nr = le32_to_cpu(*p);
3042 /* accumulate blocks to free if they're contiguous */
3045 block_to_free_p = p;
3047 } else if (nr == block_to_free + count) {
3050 ext4_clear_blocks(handle, inode, this_bh,
3052 count, block_to_free_p, p);
3054 block_to_free_p = p;
3061 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3062 count, block_to_free_p, p);
3065 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3068 * The buffer head should have an attached journal head at this
3069 * point. However, if the data is corrupted and an indirect
3070 * block pointed to itself, it would have been detached when
3071 * the block was cleared. Check for this instead of OOPSing.
3074 ext4_journal_dirty_metadata(handle, this_bh);
3076 ext4_error(inode->i_sb, __func__,
3077 "circular indirect block detected, "
3078 "inode=%lu, block=%llu",
3080 (unsigned long long) this_bh->b_blocknr);
3085 * ext4_free_branches - free an array of branches
3086 * @handle: JBD handle for this transaction
3087 * @inode: inode we are dealing with
3088 * @parent_bh: the buffer_head which contains *@first and *@last
3089 * @first: array of block numbers
3090 * @last: pointer immediately past the end of array
3091 * @depth: depth of the branches to free
3093 * We are freeing all blocks refered from these branches (numbers are
3094 * stored as little-endian 32-bit) and updating @inode->i_blocks
3097 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3098 struct buffer_head *parent_bh,
3099 __le32 *first, __le32 *last, int depth)
3104 if (is_handle_aborted(handle))
3108 struct buffer_head *bh;
3109 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3111 while (--p >= first) {
3112 nr = le32_to_cpu(*p);
3114 continue; /* A hole */
3116 /* Go read the buffer for the next level down */
3117 bh = sb_bread(inode->i_sb, nr);
3120 * A read failure? Report error and clear slot
3124 ext4_error(inode->i_sb, "ext4_free_branches",
3125 "Read failure, inode=%lu, block=%llu",
3130 /* This zaps the entire block. Bottom up. */
3131 BUFFER_TRACE(bh, "free child branches");
3132 ext4_free_branches(handle, inode, bh,
3133 (__le32*)bh->b_data,
3134 (__le32*)bh->b_data + addr_per_block,
3138 * We've probably journalled the indirect block several
3139 * times during the truncate. But it's no longer
3140 * needed and we now drop it from the transaction via
3141 * jbd2_journal_revoke().
3143 * That's easy if it's exclusively part of this
3144 * transaction. But if it's part of the committing
3145 * transaction then jbd2_journal_forget() will simply
3146 * brelse() it. That means that if the underlying
3147 * block is reallocated in ext4_get_block(),
3148 * unmap_underlying_metadata() will find this block
3149 * and will try to get rid of it. damn, damn.
3151 * If this block has already been committed to the
3152 * journal, a revoke record will be written. And
3153 * revoke records must be emitted *before* clearing
3154 * this block's bit in the bitmaps.
3156 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3159 * Everything below this this pointer has been
3160 * released. Now let this top-of-subtree go.
3162 * We want the freeing of this indirect block to be
3163 * atomic in the journal with the updating of the
3164 * bitmap block which owns it. So make some room in
3167 * We zero the parent pointer *after* freeing its
3168 * pointee in the bitmaps, so if extend_transaction()
3169 * for some reason fails to put the bitmap changes and
3170 * the release into the same transaction, recovery
3171 * will merely complain about releasing a free block,
3172 * rather than leaking blocks.
3174 if (is_handle_aborted(handle))
3176 if (try_to_extend_transaction(handle, inode)) {
3177 ext4_mark_inode_dirty(handle, inode);
3178 ext4_journal_test_restart(handle, inode);
3181 ext4_free_blocks(handle, inode, nr, 1, 1);
3185 * The block which we have just freed is
3186 * pointed to by an indirect block: journal it
3188 BUFFER_TRACE(parent_bh, "get_write_access");
3189 if (!ext4_journal_get_write_access(handle,
3192 BUFFER_TRACE(parent_bh,
3193 "call ext4_journal_dirty_metadata");
3194 ext4_journal_dirty_metadata(handle,
3200 /* We have reached the bottom of the tree. */
3201 BUFFER_TRACE(parent_bh, "free data blocks");
3202 ext4_free_data(handle, inode, parent_bh, first, last);
3206 int ext4_can_truncate(struct inode *inode)
3208 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3210 if (S_ISREG(inode->i_mode))
3212 if (S_ISDIR(inode->i_mode))
3214 if (S_ISLNK(inode->i_mode))
3215 return !ext4_inode_is_fast_symlink(inode);
3222 * We block out ext4_get_block() block instantiations across the entire
3223 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3224 * simultaneously on behalf of the same inode.
3226 * As we work through the truncate and commmit bits of it to the journal there
3227 * is one core, guiding principle: the file's tree must always be consistent on
3228 * disk. We must be able to restart the truncate after a crash.
3230 * The file's tree may be transiently inconsistent in memory (although it
3231 * probably isn't), but whenever we close off and commit a journal transaction,
3232 * the contents of (the filesystem + the journal) must be consistent and
3233 * restartable. It's pretty simple, really: bottom up, right to left (although
3234 * left-to-right works OK too).
3236 * Note that at recovery time, journal replay occurs *before* the restart of
3237 * truncate against the orphan inode list.
3239 * The committed inode has the new, desired i_size (which is the same as
3240 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3241 * that this inode's truncate did not complete and it will again call
3242 * ext4_truncate() to have another go. So there will be instantiated blocks
3243 * to the right of the truncation point in a crashed ext4 filesystem. But
3244 * that's fine - as long as they are linked from the inode, the post-crash
3245 * ext4_truncate() run will find them and release them.
3247 void ext4_truncate(struct inode *inode)
3250 struct ext4_inode_info *ei = EXT4_I(inode);
3251 __le32 *i_data = ei->i_data;
3252 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3253 struct address_space *mapping = inode->i_mapping;
3254 ext4_lblk_t offsets[4];
3259 ext4_lblk_t last_block;
3260 unsigned blocksize = inode->i_sb->s_blocksize;
3262 if (!ext4_can_truncate(inode))
3265 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3266 ext4_ext_truncate(inode);
3270 handle = start_transaction(inode);
3272 return; /* AKPM: return what? */
3274 last_block = (inode->i_size + blocksize-1)
3275 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3277 if (inode->i_size & (blocksize - 1))
3278 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3281 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3283 goto out_stop; /* error */
3286 * OK. This truncate is going to happen. We add the inode to the
3287 * orphan list, so that if this truncate spans multiple transactions,
3288 * and we crash, we will resume the truncate when the filesystem
3289 * recovers. It also marks the inode dirty, to catch the new size.
3291 * Implication: the file must always be in a sane, consistent
3292 * truncatable state while each transaction commits.
3294 if (ext4_orphan_add(handle, inode))
3298 * The orphan list entry will now protect us from any crash which
3299 * occurs before the truncate completes, so it is now safe to propagate
3300 * the new, shorter inode size (held for now in i_size) into the
3301 * on-disk inode. We do this via i_disksize, which is the value which
3302 * ext4 *really* writes onto the disk inode.
3304 ei->i_disksize = inode->i_size;
3307 * From here we block out all ext4_get_block() callers who want to
3308 * modify the block allocation tree.
3310 down_write(&ei->i_data_sem);
3312 if (n == 1) { /* direct blocks */
3313 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3314 i_data + EXT4_NDIR_BLOCKS);
3318 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3319 /* Kill the top of shared branch (not detached) */
3321 if (partial == chain) {
3322 /* Shared branch grows from the inode */
3323 ext4_free_branches(handle, inode, NULL,
3324 &nr, &nr+1, (chain+n-1) - partial);
3327 * We mark the inode dirty prior to restart,
3328 * and prior to stop. No need for it here.
3331 /* Shared branch grows from an indirect block */
3332 BUFFER_TRACE(partial->bh, "get_write_access");
3333 ext4_free_branches(handle, inode, partial->bh,
3335 partial->p+1, (chain+n-1) - partial);
3338 /* Clear the ends of indirect blocks on the shared branch */
3339 while (partial > chain) {
3340 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3341 (__le32*)partial->bh->b_data+addr_per_block,
3342 (chain+n-1) - partial);
3343 BUFFER_TRACE(partial->bh, "call brelse");
3344 brelse (partial->bh);
3348 /* Kill the remaining (whole) subtrees */
3349 switch (offsets[0]) {
3351 nr = i_data[EXT4_IND_BLOCK];
3353 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3354 i_data[EXT4_IND_BLOCK] = 0;
3356 case EXT4_IND_BLOCK:
3357 nr = i_data[EXT4_DIND_BLOCK];
3359 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3360 i_data[EXT4_DIND_BLOCK] = 0;
3362 case EXT4_DIND_BLOCK:
3363 nr = i_data[EXT4_TIND_BLOCK];
3365 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3366 i_data[EXT4_TIND_BLOCK] = 0;
3368 case EXT4_TIND_BLOCK:
3372 ext4_discard_reservation(inode);
3374 up_write(&ei->i_data_sem);
3375 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3376 ext4_mark_inode_dirty(handle, inode);
3379 * In a multi-transaction truncate, we only make the final transaction
3386 * If this was a simple ftruncate(), and the file will remain alive
3387 * then we need to clear up the orphan record which we created above.
3388 * However, if this was a real unlink then we were called by
3389 * ext4_delete_inode(), and we allow that function to clean up the
3390 * orphan info for us.
3393 ext4_orphan_del(handle, inode);
3395 ext4_journal_stop(handle);
3398 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3399 unsigned long ino, struct ext4_iloc *iloc)
3401 ext4_group_t block_group;
3402 unsigned long offset;
3404 struct ext4_group_desc *gdp;
3406 if (!ext4_valid_inum(sb, ino)) {
3408 * This error is already checked for in namei.c unless we are
3409 * looking at an NFS filehandle, in which case no error
3415 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3416 gdp = ext4_get_group_desc(sb, block_group, NULL);
3421 * Figure out the offset within the block group inode table
3423 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3424 EXT4_INODE_SIZE(sb);
3425 block = ext4_inode_table(sb, gdp) +
3426 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3428 iloc->block_group = block_group;
3429 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3434 * ext4_get_inode_loc returns with an extra refcount against the inode's
3435 * underlying buffer_head on success. If 'in_mem' is true, we have all
3436 * data in memory that is needed to recreate the on-disk version of this
3439 static int __ext4_get_inode_loc(struct inode *inode,
3440 struct ext4_iloc *iloc, int in_mem)
3443 struct buffer_head *bh;
3445 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3449 bh = sb_getblk(inode->i_sb, block);
3451 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3452 "unable to read inode block - "
3453 "inode=%lu, block=%llu",
3454 inode->i_ino, block);
3457 if (!buffer_uptodate(bh)) {
3459 if (buffer_uptodate(bh)) {
3460 /* someone brought it uptodate while we waited */
3466 * If we have all information of the inode in memory and this
3467 * is the only valid inode in the block, we need not read the
3471 struct buffer_head *bitmap_bh;
3472 struct ext4_group_desc *desc;
3473 int inodes_per_buffer;
3474 int inode_offset, i;
3475 ext4_group_t block_group;
3478 block_group = (inode->i_ino - 1) /
3479 EXT4_INODES_PER_GROUP(inode->i_sb);
3480 inodes_per_buffer = bh->b_size /
3481 EXT4_INODE_SIZE(inode->i_sb);
3482 inode_offset = ((inode->i_ino - 1) %
3483 EXT4_INODES_PER_GROUP(inode->i_sb));
3484 start = inode_offset & ~(inodes_per_buffer - 1);
3486 /* Is the inode bitmap in cache? */
3487 desc = ext4_get_group_desc(inode->i_sb,
3492 bitmap_bh = sb_getblk(inode->i_sb,
3493 ext4_inode_bitmap(inode->i_sb, desc));
3498 * If the inode bitmap isn't in cache then the
3499 * optimisation may end up performing two reads instead
3500 * of one, so skip it.
3502 if (!buffer_uptodate(bitmap_bh)) {
3506 for (i = start; i < start + inodes_per_buffer; i++) {
3507 if (i == inode_offset)
3509 if (ext4_test_bit(i, bitmap_bh->b_data))
3513 if (i == start + inodes_per_buffer) {
3514 /* all other inodes are free, so skip I/O */
3515 memset(bh->b_data, 0, bh->b_size);
3516 set_buffer_uptodate(bh);
3524 * There are other valid inodes in the buffer, this inode
3525 * has in-inode xattrs, or we don't have this inode in memory.
3526 * Read the block from disk.
3529 bh->b_end_io = end_buffer_read_sync;
3530 submit_bh(READ_META, bh);
3532 if (!buffer_uptodate(bh)) {
3533 ext4_error(inode->i_sb, "ext4_get_inode_loc",
3534 "unable to read inode block - "
3535 "inode=%lu, block=%llu",
3536 inode->i_ino, block);
3546 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3548 /* We have all inode data except xattrs in memory here. */
3549 return __ext4_get_inode_loc(inode, iloc,
3550 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3553 void ext4_set_inode_flags(struct inode *inode)
3555 unsigned int flags = EXT4_I(inode)->i_flags;
3557 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3558 if (flags & EXT4_SYNC_FL)
3559 inode->i_flags |= S_SYNC;
3560 if (flags & EXT4_APPEND_FL)
3561 inode->i_flags |= S_APPEND;
3562 if (flags & EXT4_IMMUTABLE_FL)
3563 inode->i_flags |= S_IMMUTABLE;
3564 if (flags & EXT4_NOATIME_FL)
3565 inode->i_flags |= S_NOATIME;
3566 if (flags & EXT4_DIRSYNC_FL)
3567 inode->i_flags |= S_DIRSYNC;
3570 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3571 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3573 unsigned int flags = ei->vfs_inode.i_flags;
3575 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3576 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
3578 ei->i_flags |= EXT4_SYNC_FL;
3579 if (flags & S_APPEND)
3580 ei->i_flags |= EXT4_APPEND_FL;
3581 if (flags & S_IMMUTABLE)
3582 ei->i_flags |= EXT4_IMMUTABLE_FL;
3583 if (flags & S_NOATIME)
3584 ei->i_flags |= EXT4_NOATIME_FL;
3585 if (flags & S_DIRSYNC)
3586 ei->i_flags |= EXT4_DIRSYNC_FL;
3588 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3589 struct ext4_inode_info *ei)
3592 struct inode *inode = &(ei->vfs_inode);
3593 struct super_block *sb = inode->i_sb;
3595 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3596 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3597 /* we are using combined 48 bit field */
3598 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3599 le32_to_cpu(raw_inode->i_blocks_lo);
3600 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
3601 /* i_blocks represent file system block size */
3602 return i_blocks << (inode->i_blkbits - 9);
3607 return le32_to_cpu(raw_inode->i_blocks_lo);
3611 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3613 struct ext4_iloc iloc;
3614 struct ext4_inode *raw_inode;
3615 struct ext4_inode_info *ei;
3616 struct buffer_head *bh;
3617 struct inode *inode;
3621 inode = iget_locked(sb, ino);
3623 return ERR_PTR(-ENOMEM);
3624 if (!(inode->i_state & I_NEW))
3628 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3629 ei->i_acl = EXT4_ACL_NOT_CACHED;
3630 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
3632 ei->i_block_alloc_info = NULL;
3634 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3638 raw_inode = ext4_raw_inode(&iloc);
3639 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3640 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3641 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3642 if(!(test_opt (inode->i_sb, NO_UID32))) {
3643 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3644 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3646 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3649 ei->i_dir_start_lookup = 0;
3650 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3651 /* We now have enough fields to check if the inode was active or not.
3652 * This is needed because nfsd might try to access dead inodes
3653 * the test is that same one that e2fsck uses
3654 * NeilBrown 1999oct15
3656 if (inode->i_nlink == 0) {
3657 if (inode->i_mode == 0 ||
3658 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3659 /* this inode is deleted */
3664 /* The only unlinked inodes we let through here have
3665 * valid i_mode and are being read by the orphan
3666 * recovery code: that's fine, we're about to complete
3667 * the process of deleting those. */
3669 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3670 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3671 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3672 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3673 cpu_to_le32(EXT4_OS_HURD)) {
3675 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3677 inode->i_size = ext4_isize(raw_inode);
3678 ei->i_disksize = inode->i_size;
3679 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3680 ei->i_block_group = iloc.block_group;
3682 * NOTE! The in-memory inode i_data array is in little-endian order
3683 * even on big-endian machines: we do NOT byteswap the block numbers!
3685 for (block = 0; block < EXT4_N_BLOCKS; block++)
3686 ei->i_data[block] = raw_inode->i_block[block];
3687 INIT_LIST_HEAD(&ei->i_orphan);
3689 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3690 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3691 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3692 EXT4_INODE_SIZE(inode->i_sb)) {
3697 if (ei->i_extra_isize == 0) {
3698 /* The extra space is currently unused. Use it. */
3699 ei->i_extra_isize = sizeof(struct ext4_inode) -
3700 EXT4_GOOD_OLD_INODE_SIZE;
3702 __le32 *magic = (void *)raw_inode +
3703 EXT4_GOOD_OLD_INODE_SIZE +
3705 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3706 ei->i_state |= EXT4_STATE_XATTR;
3709 ei->i_extra_isize = 0;
3711 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3712 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3713 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3714 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3716 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3717 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3718 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3720 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3723 if (S_ISREG(inode->i_mode)) {
3724 inode->i_op = &ext4_file_inode_operations;
3725 inode->i_fop = &ext4_file_operations;
3726 ext4_set_aops(inode);
3727 } else if (S_ISDIR(inode->i_mode)) {
3728 inode->i_op = &ext4_dir_inode_operations;
3729 inode->i_fop = &ext4_dir_operations;
3730 } else if (S_ISLNK(inode->i_mode)) {
3731 if (ext4_inode_is_fast_symlink(inode))
3732 inode->i_op = &ext4_fast_symlink_inode_operations;
3734 inode->i_op = &ext4_symlink_inode_operations;
3735 ext4_set_aops(inode);
3738 inode->i_op = &ext4_special_inode_operations;
3739 if (raw_inode->i_block[0])
3740 init_special_inode(inode, inode->i_mode,
3741 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3743 init_special_inode(inode, inode->i_mode,
3744 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3747 ext4_set_inode_flags(inode);
3748 unlock_new_inode(inode);
3753 return ERR_PTR(ret);
3756 static int ext4_inode_blocks_set(handle_t *handle,
3757 struct ext4_inode *raw_inode,
3758 struct ext4_inode_info *ei)
3760 struct inode *inode = &(ei->vfs_inode);
3761 u64 i_blocks = inode->i_blocks;
3762 struct super_block *sb = inode->i_sb;
3765 if (i_blocks <= ~0U) {
3767 * i_blocks can be represnted in a 32 bit variable
3768 * as multiple of 512 bytes
3770 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3771 raw_inode->i_blocks_high = 0;
3772 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
3773 } else if (i_blocks <= 0xffffffffffffULL) {
3775 * i_blocks can be represented in a 48 bit variable
3776 * as multiple of 512 bytes
3778 err = ext4_update_rocompat_feature(handle, sb,
3779 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3782 /* i_block is stored in the split 48 bit fields */
3783 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3784 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3785 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
3788 * i_blocks should be represented in a 48 bit variable
3789 * as multiple of file system block size
3791 err = ext4_update_rocompat_feature(handle, sb,
3792 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3795 ei->i_flags |= EXT4_HUGE_FILE_FL;
3796 /* i_block is stored in file system block size */
3797 i_blocks = i_blocks >> (inode->i_blkbits - 9);
3798 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3799 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3806 * Post the struct inode info into an on-disk inode location in the
3807 * buffer-cache. This gobbles the caller's reference to the
3808 * buffer_head in the inode location struct.
3810 * The caller must have write access to iloc->bh.
3812 static int ext4_do_update_inode(handle_t *handle,
3813 struct inode *inode,
3814 struct ext4_iloc *iloc)
3816 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
3817 struct ext4_inode_info *ei = EXT4_I(inode);
3818 struct buffer_head *bh = iloc->bh;
3819 int err = 0, rc, block;
3821 /* For fields not not tracking in the in-memory inode,
3822 * initialise them to zero for new inodes. */
3823 if (ei->i_state & EXT4_STATE_NEW)
3824 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
3826 ext4_get_inode_flags(ei);
3827 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
3828 if(!(test_opt(inode->i_sb, NO_UID32))) {
3829 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
3830 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
3832 * Fix up interoperability with old kernels. Otherwise, old inodes get
3833 * re-used with the upper 16 bits of the uid/gid intact
3836 raw_inode->i_uid_high =
3837 cpu_to_le16(high_16_bits(inode->i_uid));
3838 raw_inode->i_gid_high =
3839 cpu_to_le16(high_16_bits(inode->i_gid));
3841 raw_inode->i_uid_high = 0;
3842 raw_inode->i_gid_high = 0;
3845 raw_inode->i_uid_low =
3846 cpu_to_le16(fs_high2lowuid(inode->i_uid));
3847 raw_inode->i_gid_low =
3848 cpu_to_le16(fs_high2lowgid(inode->i_gid));
3849 raw_inode->i_uid_high = 0;
3850 raw_inode->i_gid_high = 0;
3852 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
3854 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
3855 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
3856 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
3857 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
3859 if (ext4_inode_blocks_set(handle, raw_inode, ei))
3861 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
3862 /* clear the migrate flag in the raw_inode */
3863 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
3864 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3865 cpu_to_le32(EXT4_OS_HURD))
3866 raw_inode->i_file_acl_high =
3867 cpu_to_le16(ei->i_file_acl >> 32);
3868 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
3869 ext4_isize_set(raw_inode, ei->i_disksize);
3870 if (ei->i_disksize > 0x7fffffffULL) {
3871 struct super_block *sb = inode->i_sb;
3872 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
3873 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
3874 EXT4_SB(sb)->s_es->s_rev_level ==
3875 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
3876 /* If this is the first large file
3877 * created, add a flag to the superblock.
3879 err = ext4_journal_get_write_access(handle,
3880 EXT4_SB(sb)->s_sbh);
3883 ext4_update_dynamic_rev(sb);
3884 EXT4_SET_RO_COMPAT_FEATURE(sb,
3885 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
3888 err = ext4_journal_dirty_metadata(handle,
3889 EXT4_SB(sb)->s_sbh);
3892 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
3893 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
3894 if (old_valid_dev(inode->i_rdev)) {
3895 raw_inode->i_block[0] =
3896 cpu_to_le32(old_encode_dev(inode->i_rdev));
3897 raw_inode->i_block[1] = 0;
3899 raw_inode->i_block[0] = 0;
3900 raw_inode->i_block[1] =
3901 cpu_to_le32(new_encode_dev(inode->i_rdev));
3902 raw_inode->i_block[2] = 0;
3904 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
3905 raw_inode->i_block[block] = ei->i_data[block];
3907 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
3908 if (ei->i_extra_isize) {
3909 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3910 raw_inode->i_version_hi =
3911 cpu_to_le32(inode->i_version >> 32);
3912 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
3916 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3917 rc = ext4_journal_dirty_metadata(handle, bh);
3920 ei->i_state &= ~EXT4_STATE_NEW;
3924 ext4_std_error(inode->i_sb, err);
3929 * ext4_write_inode()
3931 * We are called from a few places:
3933 * - Within generic_file_write() for O_SYNC files.
3934 * Here, there will be no transaction running. We wait for any running
3935 * trasnaction to commit.
3937 * - Within sys_sync(), kupdate and such.
3938 * We wait on commit, if tol to.
3940 * - Within prune_icache() (PF_MEMALLOC == true)
3941 * Here we simply return. We can't afford to block kswapd on the
3944 * In all cases it is actually safe for us to return without doing anything,
3945 * because the inode has been copied into a raw inode buffer in
3946 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3949 * Note that we are absolutely dependent upon all inode dirtiers doing the
3950 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3951 * which we are interested.
3953 * It would be a bug for them to not do this. The code:
3955 * mark_inode_dirty(inode)
3957 * inode->i_size = expr;
3959 * is in error because a kswapd-driven write_inode() could occur while
3960 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3961 * will no longer be on the superblock's dirty inode list.
3963 int ext4_write_inode(struct inode *inode, int wait)
3965 if (current->flags & PF_MEMALLOC)
3968 if (ext4_journal_current_handle()) {
3969 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3977 return ext4_force_commit(inode->i_sb);
3983 * Called from notify_change.
3985 * We want to trap VFS attempts to truncate the file as soon as
3986 * possible. In particular, we want to make sure that when the VFS
3987 * shrinks i_size, we put the inode on the orphan list and modify
3988 * i_disksize immediately, so that during the subsequent flushing of
3989 * dirty pages and freeing of disk blocks, we can guarantee that any
3990 * commit will leave the blocks being flushed in an unused state on
3991 * disk. (On recovery, the inode will get truncated and the blocks will
3992 * be freed, so we have a strong guarantee that no future commit will
3993 * leave these blocks visible to the user.)
3995 * Another thing we have to assure is that if we are in ordered mode
3996 * and inode is still attached to the committing transaction, we must
3997 * we start writeout of all the dirty pages which are being truncated.
3998 * This way we are sure that all the data written in the previous
3999 * transaction are already on disk (truncate waits for pages under
4002 * Called with inode->i_mutex down.
4004 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4006 struct inode *inode = dentry->d_inode;
4008 const unsigned int ia_valid = attr->ia_valid;
4010 error = inode_change_ok(inode, attr);
4014 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4015 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4018 /* (user+group)*(old+new) structure, inode write (sb,
4019 * inode block, ? - but truncate inode update has it) */
4020 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4021 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4022 if (IS_ERR(handle)) {
4023 error = PTR_ERR(handle);
4026 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4028 ext4_journal_stop(handle);
4031 /* Update corresponding info in inode so that everything is in
4032 * one transaction */
4033 if (attr->ia_valid & ATTR_UID)
4034 inode->i_uid = attr->ia_uid;
4035 if (attr->ia_valid & ATTR_GID)
4036 inode->i_gid = attr->ia_gid;
4037 error = ext4_mark_inode_dirty(handle, inode);
4038 ext4_journal_stop(handle);
4041 if (attr->ia_valid & ATTR_SIZE) {
4042 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4043 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4045 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4052 if (S_ISREG(inode->i_mode) &&
4053 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4056 handle = ext4_journal_start(inode, 3);
4057 if (IS_ERR(handle)) {
4058 error = PTR_ERR(handle);
4062 error = ext4_orphan_add(handle, inode);
4063 EXT4_I(inode)->i_disksize = attr->ia_size;
4064 rc = ext4_mark_inode_dirty(handle, inode);
4067 ext4_journal_stop(handle);
4069 if (ext4_should_order_data(inode)) {
4070 error = ext4_begin_ordered_truncate(inode,
4073 /* Do as much error cleanup as possible */
4074 handle = ext4_journal_start(inode, 3);
4075 if (IS_ERR(handle)) {
4076 ext4_orphan_del(NULL, inode);
4079 ext4_orphan_del(handle, inode);
4080 ext4_journal_stop(handle);
4086 rc = inode_setattr(inode, attr);
4088 /* If inode_setattr's call to ext4_truncate failed to get a
4089 * transaction handle at all, we need to clean up the in-core
4090 * orphan list manually. */
4092 ext4_orphan_del(NULL, inode);
4094 if (!rc && (ia_valid & ATTR_MODE))
4095 rc = ext4_acl_chmod(inode);
4098 ext4_std_error(inode->i_sb, error);
4106 * How many blocks doth make a writepage()?
4108 * With N blocks per page, it may be:
4113 * N+5 bitmap blocks (from the above)
4114 * N+5 group descriptor summary blocks
4117 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
4119 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
4121 * With ordered or writeback data it's the same, less the N data blocks.
4123 * If the inode's direct blocks can hold an integral number of pages then a
4124 * page cannot straddle two indirect blocks, and we can only touch one indirect
4125 * and dindirect block, and the "5" above becomes "3".
4127 * This still overestimates under most circumstances. If we were to pass the
4128 * start and end offsets in here as well we could do block_to_path() on each
4129 * block and work out the exact number of indirects which are touched. Pah.
4132 int ext4_writepage_trans_blocks(struct inode *inode)
4134 int bpp = ext4_journal_blocks_per_page(inode);
4135 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
4138 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
4139 return ext4_ext_writepage_trans_blocks(inode, bpp);
4141 if (ext4_should_journal_data(inode))
4142 ret = 3 * (bpp + indirects) + 2;
4144 ret = 2 * (bpp + indirects) + 2;
4147 /* We know that structure was already allocated during DQUOT_INIT so
4148 * we will be updating only the data blocks + inodes */
4149 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
4156 * The caller must have previously called ext4_reserve_inode_write().
4157 * Give this, we know that the caller already has write access to iloc->bh.
4159 int ext4_mark_iloc_dirty(handle_t *handle,
4160 struct inode *inode, struct ext4_iloc *iloc)
4164 if (test_opt(inode->i_sb, I_VERSION))
4165 inode_inc_iversion(inode);
4167 /* the do_update_inode consumes one bh->b_count */
4170 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4171 err = ext4_do_update_inode(handle, inode, iloc);
4177 * On success, We end up with an outstanding reference count against
4178 * iloc->bh. This _must_ be cleaned up later.
4182 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4183 struct ext4_iloc *iloc)
4187 err = ext4_get_inode_loc(inode, iloc);
4189 BUFFER_TRACE(iloc->bh, "get_write_access");
4190 err = ext4_journal_get_write_access(handle, iloc->bh);
4197 ext4_std_error(inode->i_sb, err);
4202 * Expand an inode by new_extra_isize bytes.
4203 * Returns 0 on success or negative error number on failure.
4205 static int ext4_expand_extra_isize(struct inode *inode,
4206 unsigned int new_extra_isize,
4207 struct ext4_iloc iloc,
4210 struct ext4_inode *raw_inode;
4211 struct ext4_xattr_ibody_header *header;
4212 struct ext4_xattr_entry *entry;
4214 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4217 raw_inode = ext4_raw_inode(&iloc);
4219 header = IHDR(inode, raw_inode);
4220 entry = IFIRST(header);
4222 /* No extended attributes present */
4223 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4224 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4225 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4227 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4231 /* try to expand with EAs present */
4232 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4237 * What we do here is to mark the in-core inode as clean with respect to inode
4238 * dirtiness (it may still be data-dirty).
4239 * This means that the in-core inode may be reaped by prune_icache
4240 * without having to perform any I/O. This is a very good thing,
4241 * because *any* task may call prune_icache - even ones which
4242 * have a transaction open against a different journal.
4244 * Is this cheating? Not really. Sure, we haven't written the
4245 * inode out, but prune_icache isn't a user-visible syncing function.
4246 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4247 * we start and wait on commits.
4249 * Is this efficient/effective? Well, we're being nice to the system
4250 * by cleaning up our inodes proactively so they can be reaped
4251 * without I/O. But we are potentially leaving up to five seconds'
4252 * worth of inodes floating about which prune_icache wants us to
4253 * write out. One way to fix that would be to get prune_icache()
4254 * to do a write_super() to free up some memory. It has the desired
4257 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4259 struct ext4_iloc iloc;
4260 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4261 static unsigned int mnt_count;
4265 err = ext4_reserve_inode_write(handle, inode, &iloc);
4266 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4267 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4269 * We need extra buffer credits since we may write into EA block
4270 * with this same handle. If journal_extend fails, then it will
4271 * only result in a minor loss of functionality for that inode.
4272 * If this is felt to be critical, then e2fsck should be run to
4273 * force a large enough s_min_extra_isize.
4275 if ((jbd2_journal_extend(handle,
4276 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4277 ret = ext4_expand_extra_isize(inode,
4278 sbi->s_want_extra_isize,
4281 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4283 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4284 ext4_warning(inode->i_sb, __func__,
4285 "Unable to expand inode %lu. Delete"
4286 " some EAs or run e2fsck.",
4289 le16_to_cpu(sbi->s_es->s_mnt_count);
4295 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4300 * ext4_dirty_inode() is called from __mark_inode_dirty()
4302 * We're really interested in the case where a file is being extended.
4303 * i_size has been changed by generic_commit_write() and we thus need
4304 * to include the updated inode in the current transaction.
4306 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4307 * are allocated to the file.
4309 * If the inode is marked synchronous, we don't honour that here - doing
4310 * so would cause a commit on atime updates, which we don't bother doing.
4311 * We handle synchronous inodes at the highest possible level.
4313 void ext4_dirty_inode(struct inode *inode)
4315 handle_t *current_handle = ext4_journal_current_handle();
4318 handle = ext4_journal_start(inode, 2);
4321 if (current_handle &&
4322 current_handle->h_transaction != handle->h_transaction) {
4323 /* This task has a transaction open against a different fs */
4324 printk(KERN_EMERG "%s: transactions do not match!\n",
4327 jbd_debug(5, "marking dirty. outer handle=%p\n",
4329 ext4_mark_inode_dirty(handle, inode);
4331 ext4_journal_stop(handle);
4338 * Bind an inode's backing buffer_head into this transaction, to prevent
4339 * it from being flushed to disk early. Unlike
4340 * ext4_reserve_inode_write, this leaves behind no bh reference and
4341 * returns no iloc structure, so the caller needs to repeat the iloc
4342 * lookup to mark the inode dirty later.
4344 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4346 struct ext4_iloc iloc;
4350 err = ext4_get_inode_loc(inode, &iloc);
4352 BUFFER_TRACE(iloc.bh, "get_write_access");
4353 err = jbd2_journal_get_write_access(handle, iloc.bh);
4355 err = ext4_journal_dirty_metadata(handle,
4360 ext4_std_error(inode->i_sb, err);
4365 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4372 * We have to be very careful here: changing a data block's
4373 * journaling status dynamically is dangerous. If we write a
4374 * data block to the journal, change the status and then delete
4375 * that block, we risk forgetting to revoke the old log record
4376 * from the journal and so a subsequent replay can corrupt data.
4377 * So, first we make sure that the journal is empty and that
4378 * nobody is changing anything.
4381 journal = EXT4_JOURNAL(inode);
4382 if (is_journal_aborted(journal))
4385 jbd2_journal_lock_updates(journal);
4386 jbd2_journal_flush(journal);
4389 * OK, there are no updates running now, and all cached data is
4390 * synced to disk. We are now in a completely consistent state
4391 * which doesn't have anything in the journal, and we know that
4392 * no filesystem updates are running, so it is safe to modify
4393 * the inode's in-core data-journaling state flag now.
4397 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4399 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4400 ext4_set_aops(inode);
4402 jbd2_journal_unlock_updates(journal);
4404 /* Finally we can mark the inode as dirty. */
4406 handle = ext4_journal_start(inode, 1);
4408 return PTR_ERR(handle);
4410 err = ext4_mark_inode_dirty(handle, inode);
4412 ext4_journal_stop(handle);
4413 ext4_std_error(inode->i_sb, err);
4418 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4420 return !buffer_mapped(bh);
4423 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4428 struct file *file = vma->vm_file;
4429 struct inode *inode = file->f_path.dentry->d_inode;
4430 struct address_space *mapping = inode->i_mapping;
4433 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4434 * get i_mutex because we are already holding mmap_sem.
4436 down_read(&inode->i_alloc_sem);
4437 size = i_size_read(inode);
4438 if (page->mapping != mapping || size <= page_offset(page)
4439 || !PageUptodate(page)) {
4440 /* page got truncated from under us? */
4444 if (PageMappedToDisk(page))
4447 if (page->index == size >> PAGE_CACHE_SHIFT)
4448 len = size & ~PAGE_CACHE_MASK;
4450 len = PAGE_CACHE_SIZE;
4452 if (page_has_buffers(page)) {
4453 /* return if we have all the buffers mapped */
4454 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4459 * OK, we need to fill the hole... Do write_begin write_end
4460 * to do block allocation/reservation.We are not holding
4461 * inode.i__mutex here. That allow * parallel write_begin,
4462 * write_end call. lock_page prevent this from happening
4463 * on the same page though
4465 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4466 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4469 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4470 len, len, page, NULL);
4475 up_read(&inode->i_alloc_sem);