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)
196 if (ext4_should_order_data(inode))
197 ext4_begin_ordered_truncate(inode, 0);
198 truncate_inode_pages(&inode->i_data, 0);
200 if (is_bad_inode(inode))
203 handle = ext4_journal_start(inode, blocks_for_truncate(inode)+3);
204 if (IS_ERR(handle)) {
205 ext4_std_error(inode->i_sb, PTR_ERR(handle));
207 * If we're going to skip the normal cleanup, we still need to
208 * make sure that the in-core orphan linked list is properly
211 ext4_orphan_del(NULL, inode);
218 err = ext4_mark_inode_dirty(handle, inode);
220 ext4_warning(inode->i_sb, __func__,
221 "couldn't mark inode dirty (err %d)", err);
225 ext4_truncate(inode);
228 * ext4_ext_truncate() doesn't reserve any slop when it
229 * restarts journal transactions; therefore there may not be
230 * enough credits left in the handle to remove the inode from
231 * the orphan list and set the dtime field.
233 if (handle->h_buffer_credits < 3) {
234 err = ext4_journal_extend(handle, 3);
236 err = ext4_journal_restart(handle, 3);
238 ext4_warning(inode->i_sb, __func__,
239 "couldn't extend journal (err %d)", err);
241 ext4_journal_stop(handle);
247 * Kill off the orphan record which ext4_truncate created.
248 * AKPM: I think this can be inside the above `if'.
249 * Note that ext4_orphan_del() has to be able to cope with the
250 * deletion of a non-existent orphan - this is because we don't
251 * know if ext4_truncate() actually created an orphan record.
252 * (Well, we could do this if we need to, but heck - it works)
254 ext4_orphan_del(handle, inode);
255 EXT4_I(inode)->i_dtime = get_seconds();
258 * One subtle ordering requirement: if anything has gone wrong
259 * (transaction abort, IO errors, whatever), then we can still
260 * do these next steps (the fs will already have been marked as
261 * having errors), but we can't free the inode if the mark_dirty
264 if (ext4_mark_inode_dirty(handle, inode))
265 /* If that failed, just do the required in-core inode clear. */
268 ext4_free_inode(handle, inode);
269 ext4_journal_stop(handle);
272 clear_inode(inode); /* We must guarantee clearing of inode... */
278 struct buffer_head *bh;
281 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
283 p->key = *(p->p = v);
288 * ext4_block_to_path - parse the block number into array of offsets
289 * @inode: inode in question (we are only interested in its superblock)
290 * @i_block: block number to be parsed
291 * @offsets: array to store the offsets in
292 * @boundary: set this non-zero if the referred-to block is likely to be
293 * followed (on disk) by an indirect block.
295 * To store the locations of file's data ext4 uses a data structure common
296 * for UNIX filesystems - tree of pointers anchored in the inode, with
297 * data blocks at leaves and indirect blocks in intermediate nodes.
298 * This function translates the block number into path in that tree -
299 * return value is the path length and @offsets[n] is the offset of
300 * pointer to (n+1)th node in the nth one. If @block is out of range
301 * (negative or too large) warning is printed and zero returned.
303 * Note: function doesn't find node addresses, so no IO is needed. All
304 * we need to know is the capacity of indirect blocks (taken from the
309 * Portability note: the last comparison (check that we fit into triple
310 * indirect block) is spelled differently, because otherwise on an
311 * architecture with 32-bit longs and 8Kb pages we might get into trouble
312 * if our filesystem had 8Kb blocks. We might use long long, but that would
313 * kill us on x86. Oh, well, at least the sign propagation does not matter -
314 * i_block would have to be negative in the very beginning, so we would not
318 static int ext4_block_to_path(struct inode *inode,
320 ext4_lblk_t offsets[4], int *boundary)
322 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
323 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
324 const long direct_blocks = EXT4_NDIR_BLOCKS,
325 indirect_blocks = ptrs,
326 double_blocks = (1 << (ptrs_bits * 2));
331 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
332 } else if (i_block < direct_blocks) {
333 offsets[n++] = i_block;
334 final = direct_blocks;
335 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
336 offsets[n++] = EXT4_IND_BLOCK;
337 offsets[n++] = i_block;
339 } else if ((i_block -= indirect_blocks) < double_blocks) {
340 offsets[n++] = EXT4_DIND_BLOCK;
341 offsets[n++] = i_block >> ptrs_bits;
342 offsets[n++] = i_block & (ptrs - 1);
344 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
345 offsets[n++] = EXT4_TIND_BLOCK;
346 offsets[n++] = i_block >> (ptrs_bits * 2);
347 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
348 offsets[n++] = i_block & (ptrs - 1);
351 ext4_warning(inode->i_sb, "ext4_block_to_path",
353 i_block + direct_blocks +
354 indirect_blocks + double_blocks);
357 *boundary = final - 1 - (i_block & (ptrs - 1));
362 * ext4_get_branch - read the chain of indirect blocks leading to data
363 * @inode: inode in question
364 * @depth: depth of the chain (1 - direct pointer, etc.)
365 * @offsets: offsets of pointers in inode/indirect blocks
366 * @chain: place to store the result
367 * @err: here we store the error value
369 * Function fills the array of triples <key, p, bh> and returns %NULL
370 * if everything went OK or the pointer to the last filled triple
371 * (incomplete one) otherwise. Upon the return chain[i].key contains
372 * the number of (i+1)-th block in the chain (as it is stored in memory,
373 * i.e. little-endian 32-bit), chain[i].p contains the address of that
374 * number (it points into struct inode for i==0 and into the bh->b_data
375 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
376 * block for i>0 and NULL for i==0. In other words, it holds the block
377 * numbers of the chain, addresses they were taken from (and where we can
378 * verify that chain did not change) and buffer_heads hosting these
381 * Function stops when it stumbles upon zero pointer (absent block)
382 * (pointer to last triple returned, *@err == 0)
383 * or when it gets an IO error reading an indirect block
384 * (ditto, *@err == -EIO)
385 * or when it reads all @depth-1 indirect blocks successfully and finds
386 * the whole chain, all way to the data (returns %NULL, *err == 0).
388 * Need to be called with
389 * down_read(&EXT4_I(inode)->i_data_sem)
391 static Indirect *ext4_get_branch(struct inode *inode, int depth,
392 ext4_lblk_t *offsets,
393 Indirect chain[4], int *err)
395 struct super_block *sb = inode->i_sb;
397 struct buffer_head *bh;
400 /* i_data is not going away, no lock needed */
401 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
405 bh = sb_bread(sb, le32_to_cpu(p->key));
408 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
422 * ext4_find_near - find a place for allocation with sufficient locality
424 * @ind: descriptor of indirect block.
426 * This function returns the preferred place for block allocation.
427 * It is used when heuristic for sequential allocation fails.
429 * + if there is a block to the left of our position - allocate near it.
430 * + if pointer will live in indirect block - allocate near that block.
431 * + if pointer will live in inode - allocate in the same
434 * In the latter case we colour the starting block by the callers PID to
435 * prevent it from clashing with concurrent allocations for a different inode
436 * in the same block group. The PID is used here so that functionally related
437 * files will be close-by on-disk.
439 * Caller must make sure that @ind is valid and will stay that way.
441 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
443 struct ext4_inode_info *ei = EXT4_I(inode);
444 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
446 ext4_fsblk_t bg_start;
447 ext4_fsblk_t last_block;
448 ext4_grpblk_t colour;
450 /* Try to find previous block */
451 for (p = ind->p - 1; p >= start; p--) {
453 return le32_to_cpu(*p);
456 /* No such thing, so let's try location of indirect block */
458 return ind->bh->b_blocknr;
461 * It is going to be referred to from the inode itself? OK, just put it
462 * into the same cylinder group then.
464 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
465 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
467 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
468 colour = (current->pid % 16) *
469 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
471 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
472 return bg_start + colour;
476 * ext4_find_goal - find a preferred place for allocation.
478 * @block: block we want
479 * @partial: pointer to the last triple within a chain
481 * Normally this function find the preferred place for block allocation,
484 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
487 struct ext4_block_alloc_info *block_i;
489 block_i = EXT4_I(inode)->i_block_alloc_info;
492 * try the heuristic for sequential allocation,
493 * failing that at least try to get decent locality.
495 if (block_i && (block == block_i->last_alloc_logical_block + 1)
496 && (block_i->last_alloc_physical_block != 0)) {
497 return block_i->last_alloc_physical_block + 1;
500 return ext4_find_near(inode, partial);
504 * ext4_blks_to_allocate: Look up the block map and count the number
505 * of direct blocks need to be allocated for the given branch.
507 * @branch: chain of indirect blocks
508 * @k: number of blocks need for indirect blocks
509 * @blks: number of data blocks to be mapped.
510 * @blocks_to_boundary: the offset in the indirect block
512 * return the total number of blocks to be allocate, including the
513 * direct and indirect blocks.
515 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
516 int blocks_to_boundary)
518 unsigned long count = 0;
521 * Simple case, [t,d]Indirect block(s) has not allocated yet
522 * then it's clear blocks on that path have not allocated
525 /* right now we don't handle cross boundary allocation */
526 if (blks < blocks_to_boundary + 1)
529 count += blocks_to_boundary + 1;
534 while (count < blks && count <= blocks_to_boundary &&
535 le32_to_cpu(*(branch[0].p + count)) == 0) {
542 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
543 * @indirect_blks: the number of blocks need to allocate for indirect
546 * @new_blocks: on return it will store the new block numbers for
547 * the indirect blocks(if needed) and the first direct block,
548 * @blks: on return it will store the total number of allocated
551 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
552 ext4_lblk_t iblock, ext4_fsblk_t goal,
553 int indirect_blks, int blks,
554 ext4_fsblk_t new_blocks[4], int *err)
557 unsigned long count = 0, blk_allocated = 0;
559 ext4_fsblk_t current_block = 0;
563 * Here we try to allocate the requested multiple blocks at once,
564 * on a best-effort basis.
565 * To build a branch, we should allocate blocks for
566 * the indirect blocks(if not allocated yet), and at least
567 * the first direct block of this branch. That's the
568 * minimum number of blocks need to allocate(required)
570 /* first we try to allocate the indirect blocks */
571 target = indirect_blks;
574 /* allocating blocks for indirect blocks and direct blocks */
575 current_block = ext4_new_meta_blocks(handle, inode,
581 /* allocate blocks for indirect blocks */
582 while (index < indirect_blks && count) {
583 new_blocks[index++] = current_block++;
588 * save the new block number
589 * for the first direct block
591 new_blocks[index] = current_block;
592 printk(KERN_INFO "%s returned more blocks than "
593 "requested\n", __func__);
599 target = blks - count ;
600 blk_allocated = count;
603 /* Now allocate data blocks */
605 /* allocating blocks for data blocks */
606 current_block = ext4_new_blocks(handle, inode, iblock,
608 if (*err && (target == blks)) {
610 * if the allocation failed and we didn't allocate
616 if (target == blks) {
618 * save the new block number
619 * for the first direct block
621 new_blocks[index] = current_block;
623 blk_allocated += count;
626 /* total number of blocks allocated for direct blocks */
631 for (i = 0; i <index; i++)
632 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
637 * ext4_alloc_branch - allocate and set up a chain of blocks.
639 * @indirect_blks: number of allocated indirect blocks
640 * @blks: number of allocated direct blocks
641 * @offsets: offsets (in the blocks) to store the pointers to next.
642 * @branch: place to store the chain in.
644 * This function allocates blocks, zeroes out all but the last one,
645 * links them into chain and (if we are synchronous) writes them to disk.
646 * In other words, it prepares a branch that can be spliced onto the
647 * inode. It stores the information about that chain in the branch[], in
648 * the same format as ext4_get_branch() would do. We are calling it after
649 * we had read the existing part of chain and partial points to the last
650 * triple of that (one with zero ->key). Upon the exit we have the same
651 * picture as after the successful ext4_get_block(), except that in one
652 * place chain is disconnected - *branch->p is still zero (we did not
653 * set the last link), but branch->key contains the number that should
654 * be placed into *branch->p to fill that gap.
656 * If allocation fails we free all blocks we've allocated (and forget
657 * their buffer_heads) and return the error value the from failed
658 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
659 * as described above and return 0.
661 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
662 ext4_lblk_t iblock, int indirect_blks,
663 int *blks, ext4_fsblk_t goal,
664 ext4_lblk_t *offsets, Indirect *branch)
666 int blocksize = inode->i_sb->s_blocksize;
669 struct buffer_head *bh;
671 ext4_fsblk_t new_blocks[4];
672 ext4_fsblk_t current_block;
674 num = ext4_alloc_blocks(handle, inode, iblock, goal, indirect_blks,
675 *blks, new_blocks, &err);
679 branch[0].key = cpu_to_le32(new_blocks[0]);
681 * metadata blocks and data blocks are allocated.
683 for (n = 1; n <= indirect_blks; n++) {
685 * Get buffer_head for parent block, zero it out
686 * and set the pointer to new one, then send
689 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
692 BUFFER_TRACE(bh, "call get_create_access");
693 err = ext4_journal_get_create_access(handle, bh);
700 memset(bh->b_data, 0, blocksize);
701 branch[n].p = (__le32 *) bh->b_data + offsets[n];
702 branch[n].key = cpu_to_le32(new_blocks[n]);
703 *branch[n].p = branch[n].key;
704 if ( n == indirect_blks) {
705 current_block = new_blocks[n];
707 * End of chain, update the last new metablock of
708 * the chain to point to the new allocated
709 * data blocks numbers
711 for (i=1; i < num; i++)
712 *(branch[n].p + i) = cpu_to_le32(++current_block);
714 BUFFER_TRACE(bh, "marking uptodate");
715 set_buffer_uptodate(bh);
718 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
719 err = ext4_journal_dirty_metadata(handle, bh);
726 /* Allocation failed, free what we already allocated */
727 for (i = 1; i <= n ; i++) {
728 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
729 ext4_journal_forget(handle, branch[i].bh);
731 for (i = 0; i <indirect_blks; i++)
732 ext4_free_blocks(handle, inode, new_blocks[i], 1, 0);
734 ext4_free_blocks(handle, inode, new_blocks[i], num, 0);
740 * ext4_splice_branch - splice the allocated branch onto inode.
742 * @block: (logical) number of block we are adding
743 * @chain: chain of indirect blocks (with a missing link - see
745 * @where: location of missing link
746 * @num: number of indirect blocks we are adding
747 * @blks: number of direct blocks we are adding
749 * This function fills the missing link and does all housekeeping needed in
750 * inode (->i_blocks, etc.). In case of success we end up with the full
751 * chain to new block and return 0.
753 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
754 ext4_lblk_t block, Indirect *where, int num, int blks)
758 struct ext4_block_alloc_info *block_i;
759 ext4_fsblk_t current_block;
761 block_i = EXT4_I(inode)->i_block_alloc_info;
763 * If we're splicing into a [td]indirect block (as opposed to the
764 * inode) then we need to get write access to the [td]indirect block
768 BUFFER_TRACE(where->bh, "get_write_access");
769 err = ext4_journal_get_write_access(handle, where->bh);
775 *where->p = where->key;
778 * Update the host buffer_head or inode to point to more just allocated
779 * direct blocks blocks
781 if (num == 0 && blks > 1) {
782 current_block = le32_to_cpu(where->key) + 1;
783 for (i = 1; i < blks; i++)
784 *(where->p + i ) = cpu_to_le32(current_block++);
788 * update the most recently allocated logical & physical block
789 * in i_block_alloc_info, to assist find the proper goal block for next
793 block_i->last_alloc_logical_block = block + blks - 1;
794 block_i->last_alloc_physical_block =
795 le32_to_cpu(where[num].key) + blks - 1;
798 /* We are done with atomic stuff, now do the rest of housekeeping */
800 inode->i_ctime = ext4_current_time(inode);
801 ext4_mark_inode_dirty(handle, inode);
803 /* had we spliced it onto indirect block? */
806 * If we spliced it onto an indirect block, we haven't
807 * altered the inode. Note however that if it is being spliced
808 * onto an indirect block at the very end of the file (the
809 * file is growing) then we *will* alter the inode to reflect
810 * the new i_size. But that is not done here - it is done in
811 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
813 jbd_debug(5, "splicing indirect only\n");
814 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
815 err = ext4_journal_dirty_metadata(handle, where->bh);
820 * OK, we spliced it into the inode itself on a direct block.
821 * Inode was dirtied above.
823 jbd_debug(5, "splicing direct\n");
828 for (i = 1; i <= num; i++) {
829 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
830 ext4_journal_forget(handle, where[i].bh);
831 ext4_free_blocks(handle, inode,
832 le32_to_cpu(where[i-1].key), 1, 0);
834 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks, 0);
840 * Allocation strategy is simple: if we have to allocate something, we will
841 * have to go the whole way to leaf. So let's do it before attaching anything
842 * to tree, set linkage between the newborn blocks, write them if sync is
843 * required, recheck the path, free and repeat if check fails, otherwise
844 * set the last missing link (that will protect us from any truncate-generated
845 * removals - all blocks on the path are immune now) and possibly force the
846 * write on the parent block.
847 * That has a nice additional property: no special recovery from the failed
848 * allocations is needed - we simply release blocks and do not touch anything
849 * reachable from inode.
851 * `handle' can be NULL if create == 0.
853 * return > 0, # of blocks mapped or allocated.
854 * return = 0, if plain lookup failed.
855 * return < 0, error case.
858 * Need to be called with
859 * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
860 * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
862 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
863 ext4_lblk_t iblock, unsigned long maxblocks,
864 struct buffer_head *bh_result,
865 int create, int extend_disksize)
868 ext4_lblk_t offsets[4];
873 int blocks_to_boundary = 0;
875 struct ext4_inode_info *ei = EXT4_I(inode);
877 ext4_fsblk_t first_block = 0;
881 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
882 J_ASSERT(handle != NULL || create == 0);
883 depth = ext4_block_to_path(inode, iblock, offsets,
884 &blocks_to_boundary);
889 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
891 /* Simplest case - block found, no allocation needed */
893 first_block = le32_to_cpu(chain[depth - 1].key);
894 clear_buffer_new(bh_result);
897 while (count < maxblocks && count <= blocks_to_boundary) {
900 blk = le32_to_cpu(*(chain[depth-1].p + count));
902 if (blk == first_block + count)
910 /* Next simple case - plain lookup or failed read of indirect block */
911 if (!create || err == -EIO)
915 * Okay, we need to do block allocation. Lazily initialize the block
916 * allocation info here if necessary
918 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
919 ext4_init_block_alloc_info(inode);
921 goal = ext4_find_goal(inode, iblock, partial);
923 /* the number of blocks need to allocate for [d,t]indirect blocks */
924 indirect_blks = (chain + depth) - partial - 1;
927 * Next look up the indirect map to count the totoal number of
928 * direct blocks to allocate for this branch.
930 count = ext4_blks_to_allocate(partial, indirect_blks,
931 maxblocks, blocks_to_boundary);
933 * Block out ext4_truncate while we alter the tree
935 err = ext4_alloc_branch(handle, inode, iblock, indirect_blks,
937 offsets + (partial - chain), partial);
940 * The ext4_splice_branch call will free and forget any buffers
941 * on the new chain if there is a failure, but that risks using
942 * up transaction credits, especially for bitmaps where the
943 * credits cannot be returned. Can we handle this somehow? We
944 * may need to return -EAGAIN upwards in the worst case. --sct
947 err = ext4_splice_branch(handle, inode, iblock,
948 partial, indirect_blks, count);
950 * i_disksize growing is protected by i_data_sem. Don't forget to
951 * protect it if you're about to implement concurrent
952 * ext4_get_block() -bzzz
954 if (!err && extend_disksize) {
955 disksize = ((loff_t) iblock + count) << inode->i_blkbits;
956 if (disksize > i_size_read(inode))
957 disksize = i_size_read(inode);
958 if (disksize > ei->i_disksize)
959 ei->i_disksize = disksize;
964 set_buffer_new(bh_result);
966 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
967 if (count > blocks_to_boundary)
968 set_buffer_boundary(bh_result);
970 /* Clean up and exit */
971 partial = chain + depth - 1; /* the whole chain */
973 while (partial > chain) {
974 BUFFER_TRACE(partial->bh, "call brelse");
978 BUFFER_TRACE(bh_result, "returned");
984 * Calculate the number of metadata blocks need to reserve
985 * to allocate @blocks for non extent file based file
987 static int ext4_indirect_calc_metadata_amount(struct inode *inode, int blocks)
989 int icap = EXT4_ADDR_PER_BLOCK(inode->i_sb);
990 int ind_blks, dind_blks, tind_blks;
992 /* number of new indirect blocks needed */
993 ind_blks = (blocks + icap - 1) / icap;
995 dind_blks = (ind_blks + icap - 1) / icap;
999 return ind_blks + dind_blks + tind_blks;
1003 * Calculate the number of metadata blocks need to reserve
1004 * to allocate given number of blocks
1006 static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
1008 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
1009 return ext4_ext_calc_metadata_amount(inode, blocks);
1011 return ext4_indirect_calc_metadata_amount(inode, blocks);
1014 static void ext4_da_update_reserve_space(struct inode *inode, int used)
1016 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1017 int total, mdb, mdb_free;
1019 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1020 /* recalculate the number of metablocks still need to be reserved */
1021 total = EXT4_I(inode)->i_reserved_data_blocks - used;
1022 mdb = ext4_calc_metadata_amount(inode, total);
1024 /* figure out how many metablocks to release */
1025 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1026 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1028 /* Account for allocated meta_blocks */
1029 mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
1031 /* update fs free blocks counter for truncate case */
1032 percpu_counter_add(&sbi->s_freeblocks_counter, mdb_free);
1034 /* update per-inode reservations */
1035 BUG_ON(used > EXT4_I(inode)->i_reserved_data_blocks);
1036 EXT4_I(inode)->i_reserved_data_blocks -= used;
1038 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1039 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1040 EXT4_I(inode)->i_allocated_meta_blocks = 0;
1041 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1044 /* Maximum number of blocks we map for direct IO at once. */
1045 #define DIO_MAX_BLOCKS 4096
1047 * Number of credits we need for writing DIO_MAX_BLOCKS:
1048 * We need sb + group descriptor + bitmap + inode -> 4
1049 * For B blocks with A block pointers per block we need:
1050 * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect).
1051 * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25.
1053 #define DIO_CREDITS 25
1057 * The ext4_get_blocks_wrap() function try to look up the requested blocks,
1058 * and returns if the blocks are already mapped.
1060 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
1061 * and store the allocated blocks in the result buffer head and mark it
1064 * If file type is extents based, it will call ext4_ext_get_blocks(),
1065 * Otherwise, call with ext4_get_blocks_handle() to handle indirect mapping
1068 * On success, it returns the number of blocks being mapped or allocate.
1069 * if create==0 and the blocks are pre-allocated and uninitialized block,
1070 * the result buffer head is unmapped. If the create ==1, it will make sure
1071 * the buffer head is mapped.
1073 * It returns 0 if plain look up failed (blocks have not been allocated), in
1074 * that casem, buffer head is unmapped
1076 * It returns the error in case of allocation failure.
1078 int ext4_get_blocks_wrap(handle_t *handle, struct inode *inode, sector_t block,
1079 unsigned long max_blocks, struct buffer_head *bh,
1080 int create, int extend_disksize, int flag)
1084 clear_buffer_mapped(bh);
1087 * Try to see if we can get the block without requesting
1088 * for new file system block.
1090 down_read((&EXT4_I(inode)->i_data_sem));
1091 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1092 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1095 retval = ext4_get_blocks_handle(handle,
1096 inode, block, max_blocks, bh, 0, 0);
1098 up_read((&EXT4_I(inode)->i_data_sem));
1100 /* If it is only a block(s) look up */
1105 * Returns if the blocks have already allocated
1107 * Note that if blocks have been preallocated
1108 * ext4_ext_get_block() returns th create = 0
1109 * with buffer head unmapped.
1111 if (retval > 0 && buffer_mapped(bh))
1115 * New blocks allocate and/or writing to uninitialized extent
1116 * will possibly result in updating i_data, so we take
1117 * the write lock of i_data_sem, and call get_blocks()
1118 * with create == 1 flag.
1120 down_write((&EXT4_I(inode)->i_data_sem));
1123 * if the caller is from delayed allocation writeout path
1124 * we have already reserved fs blocks for allocation
1125 * let the underlying get_block() function know to
1126 * avoid double accounting
1129 EXT4_I(inode)->i_delalloc_reserved_flag = 1;
1131 * We need to check for EXT4 here because migrate
1132 * could have changed the inode type in between
1134 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
1135 retval = ext4_ext_get_blocks(handle, inode, block, max_blocks,
1136 bh, create, extend_disksize);
1138 retval = ext4_get_blocks_handle(handle, inode, block,
1139 max_blocks, bh, create, extend_disksize);
1141 if (retval > 0 && buffer_new(bh)) {
1143 * We allocated new blocks which will result in
1144 * i_data's format changing. Force the migrate
1145 * to fail by clearing migrate flags
1147 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
1153 EXT4_I(inode)->i_delalloc_reserved_flag = 0;
1155 * Update reserved blocks/metadata blocks
1156 * after successful block allocation
1157 * which were deferred till now
1159 if ((retval > 0) && buffer_delay(bh))
1160 ext4_da_update_reserve_space(inode, retval);
1163 up_write((&EXT4_I(inode)->i_data_sem));
1167 static int ext4_get_block(struct inode *inode, sector_t iblock,
1168 struct buffer_head *bh_result, int create)
1170 handle_t *handle = ext4_journal_current_handle();
1171 int ret = 0, started = 0;
1172 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
1174 if (create && !handle) {
1175 /* Direct IO write... */
1176 if (max_blocks > DIO_MAX_BLOCKS)
1177 max_blocks = DIO_MAX_BLOCKS;
1178 handle = ext4_journal_start(inode, DIO_CREDITS +
1179 2 * EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb));
1180 if (IS_ERR(handle)) {
1181 ret = PTR_ERR(handle);
1187 ret = ext4_get_blocks_wrap(handle, inode, iblock,
1188 max_blocks, bh_result, create, 0, 0);
1190 bh_result->b_size = (ret << inode->i_blkbits);
1194 ext4_journal_stop(handle);
1200 * `handle' can be NULL if create is zero
1202 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1203 ext4_lblk_t block, int create, int *errp)
1205 struct buffer_head dummy;
1208 J_ASSERT(handle != NULL || create == 0);
1211 dummy.b_blocknr = -1000;
1212 buffer_trace_init(&dummy.b_history);
1213 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1214 &dummy, create, 1, 0);
1216 * ext4_get_blocks_handle() returns number of blocks
1217 * mapped. 0 in case of a HOLE.
1225 if (!err && buffer_mapped(&dummy)) {
1226 struct buffer_head *bh;
1227 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1232 if (buffer_new(&dummy)) {
1233 J_ASSERT(create != 0);
1234 J_ASSERT(handle != NULL);
1237 * Now that we do not always journal data, we should
1238 * keep in mind whether this should always journal the
1239 * new buffer as metadata. For now, regular file
1240 * writes use ext4_get_block instead, so it's not a
1244 BUFFER_TRACE(bh, "call get_create_access");
1245 fatal = ext4_journal_get_create_access(handle, bh);
1246 if (!fatal && !buffer_uptodate(bh)) {
1247 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1248 set_buffer_uptodate(bh);
1251 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1252 err = ext4_journal_dirty_metadata(handle, bh);
1256 BUFFER_TRACE(bh, "not a new buffer");
1269 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1270 ext4_lblk_t block, int create, int *err)
1272 struct buffer_head * bh;
1274 bh = ext4_getblk(handle, inode, block, create, err);
1277 if (buffer_uptodate(bh))
1279 ll_rw_block(READ_META, 1, &bh);
1281 if (buffer_uptodate(bh))
1288 static int walk_page_buffers( handle_t *handle,
1289 struct buffer_head *head,
1293 int (*fn)( handle_t *handle,
1294 struct buffer_head *bh))
1296 struct buffer_head *bh;
1297 unsigned block_start, block_end;
1298 unsigned blocksize = head->b_size;
1300 struct buffer_head *next;
1302 for ( bh = head, block_start = 0;
1303 ret == 0 && (bh != head || !block_start);
1304 block_start = block_end, bh = next)
1306 next = bh->b_this_page;
1307 block_end = block_start + blocksize;
1308 if (block_end <= from || block_start >= to) {
1309 if (partial && !buffer_uptodate(bh))
1313 err = (*fn)(handle, bh);
1321 * To preserve ordering, it is essential that the hole instantiation and
1322 * the data write be encapsulated in a single transaction. We cannot
1323 * close off a transaction and start a new one between the ext4_get_block()
1324 * and the commit_write(). So doing the jbd2_journal_start at the start of
1325 * prepare_write() is the right place.
1327 * Also, this function can nest inside ext4_writepage() ->
1328 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1329 * has generated enough buffer credits to do the whole page. So we won't
1330 * block on the journal in that case, which is good, because the caller may
1333 * By accident, ext4 can be reentered when a transaction is open via
1334 * quota file writes. If we were to commit the transaction while thus
1335 * reentered, there can be a deadlock - we would be holding a quota
1336 * lock, and the commit would never complete if another thread had a
1337 * transaction open and was blocking on the quota lock - a ranking
1340 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1341 * will _not_ run commit under these circumstances because handle->h_ref
1342 * is elevated. We'll still have enough credits for the tiny quotafile
1345 static int do_journal_get_write_access(handle_t *handle,
1346 struct buffer_head *bh)
1348 if (!buffer_mapped(bh) || buffer_freed(bh))
1350 return ext4_journal_get_write_access(handle, bh);
1353 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1354 loff_t pos, unsigned len, unsigned flags,
1355 struct page **pagep, void **fsdata)
1357 struct inode *inode = mapping->host;
1358 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1365 index = pos >> PAGE_CACHE_SHIFT;
1366 from = pos & (PAGE_CACHE_SIZE - 1);
1370 handle = ext4_journal_start(inode, needed_blocks);
1371 if (IS_ERR(handle)) {
1372 ret = PTR_ERR(handle);
1376 page = __grab_cache_page(mapping, index);
1378 ext4_journal_stop(handle);
1384 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1387 if (!ret && ext4_should_journal_data(inode)) {
1388 ret = walk_page_buffers(handle, page_buffers(page),
1389 from, to, NULL, do_journal_get_write_access);
1394 ext4_journal_stop(handle);
1395 page_cache_release(page);
1398 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1404 /* For write_end() in data=journal mode */
1405 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1407 if (!buffer_mapped(bh) || buffer_freed(bh))
1409 set_buffer_uptodate(bh);
1410 return ext4_journal_dirty_metadata(handle, bh);
1414 * We need to pick up the new inode size which generic_commit_write gave us
1415 * `file' can be NULL - eg, when called from page_symlink().
1417 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1418 * buffers are managed internally.
1420 static int ext4_ordered_write_end(struct file *file,
1421 struct address_space *mapping,
1422 loff_t pos, unsigned len, unsigned copied,
1423 struct page *page, void *fsdata)
1425 handle_t *handle = ext4_journal_current_handle();
1426 struct inode *inode = mapping->host;
1429 ret = ext4_jbd2_file_inode(handle, inode);
1433 * generic_write_end() will run mark_inode_dirty() if i_size
1434 * changes. So let's piggyback the i_disksize mark_inode_dirty
1439 new_i_size = pos + copied;
1440 if (new_i_size > EXT4_I(inode)->i_disksize)
1441 EXT4_I(inode)->i_disksize = new_i_size;
1442 ret2 = generic_write_end(file, mapping, pos, len, copied,
1448 ret2 = ext4_journal_stop(handle);
1452 return ret ? ret : copied;
1455 static int ext4_writeback_write_end(struct file *file,
1456 struct address_space *mapping,
1457 loff_t pos, unsigned len, unsigned copied,
1458 struct page *page, void *fsdata)
1460 handle_t *handle = ext4_journal_current_handle();
1461 struct inode *inode = mapping->host;
1465 new_i_size = pos + copied;
1466 if (new_i_size > EXT4_I(inode)->i_disksize)
1467 EXT4_I(inode)->i_disksize = new_i_size;
1469 ret2 = generic_write_end(file, mapping, pos, len, copied,
1475 ret2 = ext4_journal_stop(handle);
1479 return ret ? ret : copied;
1482 static int ext4_journalled_write_end(struct file *file,
1483 struct address_space *mapping,
1484 loff_t pos, unsigned len, unsigned copied,
1485 struct page *page, void *fsdata)
1487 handle_t *handle = ext4_journal_current_handle();
1488 struct inode *inode = mapping->host;
1493 from = pos & (PAGE_CACHE_SIZE - 1);
1497 if (!PageUptodate(page))
1499 page_zero_new_buffers(page, from+copied, to);
1502 ret = walk_page_buffers(handle, page_buffers(page), from,
1503 to, &partial, write_end_fn);
1505 SetPageUptodate(page);
1506 if (pos+copied > inode->i_size)
1507 i_size_write(inode, pos+copied);
1508 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1509 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1510 EXT4_I(inode)->i_disksize = inode->i_size;
1511 ret2 = ext4_mark_inode_dirty(handle, inode);
1517 ret2 = ext4_journal_stop(handle);
1520 page_cache_release(page);
1522 return ret ? ret : copied;
1525 static int ext4_da_reserve_space(struct inode *inode, int nrblocks)
1527 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1528 unsigned long md_needed, mdblocks, total = 0;
1531 * recalculate the amount of metadata blocks to reserve
1532 * in order to allocate nrblocks
1533 * worse case is one extent per block
1535 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1536 total = EXT4_I(inode)->i_reserved_data_blocks + nrblocks;
1537 mdblocks = ext4_calc_metadata_amount(inode, total);
1538 BUG_ON(mdblocks < EXT4_I(inode)->i_reserved_meta_blocks);
1540 md_needed = mdblocks - EXT4_I(inode)->i_reserved_meta_blocks;
1541 total = md_needed + nrblocks;
1543 if (ext4_has_free_blocks(sbi, total) < total) {
1544 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1547 /* reduce fs free blocks counter */
1548 percpu_counter_sub(&sbi->s_freeblocks_counter, total);
1550 EXT4_I(inode)->i_reserved_data_blocks += nrblocks;
1551 EXT4_I(inode)->i_reserved_meta_blocks = mdblocks;
1553 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1554 return 0; /* success */
1557 static void ext4_da_release_space(struct inode *inode, int to_free)
1559 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1560 int total, mdb, mdb_free, release;
1562 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1563 /* recalculate the number of metablocks still need to be reserved */
1564 total = EXT4_I(inode)->i_reserved_data_blocks - to_free;
1565 mdb = ext4_calc_metadata_amount(inode, total);
1567 /* figure out how many metablocks to release */
1568 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1569 mdb_free = EXT4_I(inode)->i_reserved_meta_blocks - mdb;
1571 release = to_free + mdb_free;
1573 /* update fs free blocks counter for truncate case */
1574 percpu_counter_add(&sbi->s_freeblocks_counter, release);
1576 /* update per-inode reservations */
1577 BUG_ON(to_free > EXT4_I(inode)->i_reserved_data_blocks);
1578 EXT4_I(inode)->i_reserved_data_blocks -= to_free;
1580 BUG_ON(mdb > EXT4_I(inode)->i_reserved_meta_blocks);
1581 EXT4_I(inode)->i_reserved_meta_blocks = mdb;
1582 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1585 static void ext4_da_page_release_reservation(struct page *page,
1586 unsigned long offset)
1589 struct buffer_head *head, *bh;
1590 unsigned int curr_off = 0;
1592 head = page_buffers(page);
1595 unsigned int next_off = curr_off + bh->b_size;
1597 if ((offset <= curr_off) && (buffer_delay(bh))) {
1599 clear_buffer_delay(bh);
1601 curr_off = next_off;
1602 } while ((bh = bh->b_this_page) != head);
1603 ext4_da_release_space(page->mapping->host, to_release);
1607 * Delayed allocation stuff
1610 struct mpage_da_data {
1611 struct inode *inode;
1612 struct buffer_head lbh; /* extent of blocks */
1613 unsigned long first_page, next_page; /* extent of pages */
1614 get_block_t *get_block;
1615 struct writeback_control *wbc;
1619 * mpage_da_submit_io - walks through extent of pages and try to write
1620 * them with __mpage_writepage()
1622 * @mpd->inode: inode
1623 * @mpd->first_page: first page of the extent
1624 * @mpd->next_page: page after the last page of the extent
1625 * @mpd->get_block: the filesystem's block mapper function
1627 * By the time mpage_da_submit_io() is called we expect all blocks
1628 * to be allocated. this may be wrong if allocation failed.
1630 * As pages are already locked by write_cache_pages(), we can't use it
1632 static int mpage_da_submit_io(struct mpage_da_data *mpd)
1634 struct address_space *mapping = mpd->inode->i_mapping;
1635 struct mpage_data mpd_pp = {
1637 .last_block_in_bio = 0,
1638 .get_block = mpd->get_block,
1641 int ret = 0, err, nr_pages, i;
1642 unsigned long index, end;
1643 struct pagevec pvec;
1645 BUG_ON(mpd->next_page <= mpd->first_page);
1647 pagevec_init(&pvec, 0);
1648 index = mpd->first_page;
1649 end = mpd->next_page - 1;
1651 while (index <= end) {
1652 /* XXX: optimize tail */
1653 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1656 for (i = 0; i < nr_pages; i++) {
1657 struct page *page = pvec.pages[i];
1659 index = page->index;
1664 err = __mpage_writepage(page, mpd->wbc, &mpd_pp);
1667 * In error case, we have to continue because
1668 * remaining pages are still locked
1669 * XXX: unlock and re-dirty them?
1674 pagevec_release(&pvec);
1677 mpage_bio_submit(WRITE, mpd_pp.bio);
1683 * mpage_put_bnr_to_bhs - walk blocks and assign them actual numbers
1685 * @mpd->inode - inode to walk through
1686 * @exbh->b_blocknr - first block on a disk
1687 * @exbh->b_size - amount of space in bytes
1688 * @logical - first logical block to start assignment with
1690 * the function goes through all passed space and put actual disk
1691 * block numbers into buffer heads, dropping BH_Delay
1693 static void mpage_put_bnr_to_bhs(struct mpage_da_data *mpd, sector_t logical,
1694 struct buffer_head *exbh)
1696 struct inode *inode = mpd->inode;
1697 struct address_space *mapping = inode->i_mapping;
1698 int blocks = exbh->b_size >> inode->i_blkbits;
1699 sector_t pblock = exbh->b_blocknr, cur_logical;
1700 struct buffer_head *head, *bh;
1701 unsigned long index, end;
1702 struct pagevec pvec;
1705 index = logical >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1706 end = (logical + blocks - 1) >> (PAGE_CACHE_SHIFT - inode->i_blkbits);
1707 cur_logical = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1709 pagevec_init(&pvec, 0);
1711 while (index <= end) {
1712 /* XXX: optimize tail */
1713 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1716 for (i = 0; i < nr_pages; i++) {
1717 struct page *page = pvec.pages[i];
1719 index = page->index;
1724 BUG_ON(!PageLocked(page));
1725 BUG_ON(PageWriteback(page));
1726 BUG_ON(!page_has_buffers(page));
1728 bh = page_buffers(page);
1731 /* skip blocks out of the range */
1733 if (cur_logical >= logical)
1736 } while ((bh = bh->b_this_page) != head);
1739 if (cur_logical >= logical + blocks)
1741 if (buffer_delay(bh)) {
1742 bh->b_blocknr = pblock;
1743 clear_buffer_delay(bh);
1744 bh->b_bdev = inode->i_sb->s_bdev;
1745 } else if (buffer_unwritten(bh)) {
1746 bh->b_blocknr = pblock;
1747 clear_buffer_unwritten(bh);
1748 set_buffer_mapped(bh);
1750 bh->b_bdev = inode->i_sb->s_bdev;
1751 } else if (buffer_mapped(bh))
1752 BUG_ON(bh->b_blocknr != pblock);
1756 } while ((bh = bh->b_this_page) != head);
1758 pagevec_release(&pvec);
1764 * __unmap_underlying_blocks - just a helper function to unmap
1765 * set of blocks described by @bh
1767 static inline void __unmap_underlying_blocks(struct inode *inode,
1768 struct buffer_head *bh)
1770 struct block_device *bdev = inode->i_sb->s_bdev;
1773 blocks = bh->b_size >> inode->i_blkbits;
1774 for (i = 0; i < blocks; i++)
1775 unmap_underlying_metadata(bdev, bh->b_blocknr + i);
1779 * mpage_da_map_blocks - go through given space
1781 * @mpd->lbh - bh describing space
1782 * @mpd->get_block - the filesystem's block mapper function
1784 * The function skips space we know is already mapped to disk blocks.
1786 * The function ignores errors ->get_block() returns, thus real
1787 * error handling is postponed to __mpage_writepage()
1789 static void mpage_da_map_blocks(struct mpage_da_data *mpd)
1791 struct buffer_head *lbh = &mpd->lbh;
1792 int err = 0, remain = lbh->b_size;
1793 sector_t next = lbh->b_blocknr;
1794 struct buffer_head new;
1797 * We consider only non-mapped and non-allocated blocks
1799 if (buffer_mapped(lbh) && !buffer_delay(lbh))
1803 new.b_state = lbh->b_state;
1805 new.b_size = remain;
1806 err = mpd->get_block(mpd->inode, next, &new, 1);
1809 * Rather than implement own error handling
1810 * here, we just leave remaining blocks
1811 * unallocated and try again with ->writepage()
1815 BUG_ON(new.b_size == 0);
1817 if (buffer_new(&new))
1818 __unmap_underlying_blocks(mpd->inode, &new);
1821 * If blocks are delayed marked, we need to
1822 * put actual blocknr and drop delayed bit
1824 if (buffer_delay(lbh) || buffer_unwritten(lbh))
1825 mpage_put_bnr_to_bhs(mpd, next, &new);
1827 /* go for the remaining blocks */
1828 next += new.b_size >> mpd->inode->i_blkbits;
1829 remain -= new.b_size;
1833 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1834 (1 << BH_Delay) | (1 << BH_Unwritten))
1837 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1839 * @mpd->lbh - extent of blocks
1840 * @logical - logical number of the block in the file
1841 * @bh - bh of the block (used to access block's state)
1843 * the function is used to collect contig. blocks in same state
1845 static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
1846 sector_t logical, struct buffer_head *bh)
1848 struct buffer_head *lbh = &mpd->lbh;
1851 next = lbh->b_blocknr + (lbh->b_size >> mpd->inode->i_blkbits);
1854 * First block in the extent
1856 if (lbh->b_size == 0) {
1857 lbh->b_blocknr = logical;
1858 lbh->b_size = bh->b_size;
1859 lbh->b_state = bh->b_state & BH_FLAGS;
1864 * Can we merge the block to our big extent?
1866 if (logical == next && (bh->b_state & BH_FLAGS) == lbh->b_state) {
1867 lbh->b_size += bh->b_size;
1872 * We couldn't merge the block to our extent, so we
1873 * need to flush current extent and start new one
1875 mpage_da_map_blocks(mpd);
1878 * Now start a new extent
1880 lbh->b_size = bh->b_size;
1881 lbh->b_state = bh->b_state & BH_FLAGS;
1882 lbh->b_blocknr = logical;
1886 * __mpage_da_writepage - finds extent of pages and blocks
1888 * @page: page to consider
1889 * @wbc: not used, we just follow rules
1892 * The function finds extents of pages and scan them for all blocks.
1894 static int __mpage_da_writepage(struct page *page,
1895 struct writeback_control *wbc, void *data)
1897 struct mpage_da_data *mpd = data;
1898 struct inode *inode = mpd->inode;
1899 struct buffer_head *bh, *head, fake;
1903 * Can we merge this page to current extent?
1905 if (mpd->next_page != page->index) {
1907 * Nope, we can't. So, we map non-allocated blocks
1908 * and start IO on them using __mpage_writepage()
1910 if (mpd->next_page != mpd->first_page) {
1911 mpage_da_map_blocks(mpd);
1912 mpage_da_submit_io(mpd);
1916 * Start next extent of pages ...
1918 mpd->first_page = page->index;
1923 mpd->lbh.b_size = 0;
1924 mpd->lbh.b_state = 0;
1925 mpd->lbh.b_blocknr = 0;
1928 mpd->next_page = page->index + 1;
1929 logical = (sector_t) page->index <<
1930 (PAGE_CACHE_SHIFT - inode->i_blkbits);
1932 if (!page_has_buffers(page)) {
1934 * There is no attached buffer heads yet (mmap?)
1935 * we treat the page asfull of dirty blocks
1938 bh->b_size = PAGE_CACHE_SIZE;
1940 set_buffer_dirty(bh);
1941 set_buffer_uptodate(bh);
1942 mpage_add_bh_to_extent(mpd, logical, bh);
1945 * Page with regular buffer heads, just add all dirty ones
1947 head = page_buffers(page);
1950 BUG_ON(buffer_locked(bh));
1951 if (buffer_dirty(bh))
1952 mpage_add_bh_to_extent(mpd, logical, bh);
1954 } while ((bh = bh->b_this_page) != head);
1961 * mpage_da_writepages - walk the list of dirty pages of the given
1962 * address space, allocates non-allocated blocks, maps newly-allocated
1963 * blocks to existing bhs and issue IO them
1965 * @mapping: address space structure to write
1966 * @wbc: subtract the number of written pages from *@wbc->nr_to_write
1967 * @get_block: the filesystem's block mapper function.
1969 * This is a library function, which implements the writepages()
1970 * address_space_operation.
1972 * In order to avoid duplication of logic that deals with partial pages,
1973 * multiple bio per page, etc, we find non-allocated blocks, allocate
1974 * them with minimal calls to ->get_block() and re-use __mpage_writepage()
1976 * It's important that we call __mpage_writepage() only once for each
1977 * involved page, otherwise we'd have to implement more complicated logic
1978 * to deal with pages w/o PG_lock or w/ PG_writeback and so on.
1980 * See comments to mpage_writepages()
1982 static int mpage_da_writepages(struct address_space *mapping,
1983 struct writeback_control *wbc,
1984 get_block_t get_block)
1986 struct mpage_da_data mpd;
1990 return generic_writepages(mapping, wbc);
1993 mpd.inode = mapping->host;
1995 mpd.lbh.b_state = 0;
1996 mpd.lbh.b_blocknr = 0;
1999 mpd.get_block = get_block;
2001 ret = write_cache_pages(mapping, wbc, __mpage_da_writepage, &mpd);
2004 * Handle last extent of pages
2006 if (mpd.next_page != mpd.first_page) {
2007 mpage_da_map_blocks(&mpd);
2008 mpage_da_submit_io(&mpd);
2015 * this is a special callback for ->write_begin() only
2016 * it's intention is to return mapped block or reserve space
2018 static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
2019 struct buffer_head *bh_result, int create)
2023 BUG_ON(create == 0);
2024 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
2027 * first, we need to know whether the block is allocated already
2028 * preallocated blocks are unmapped but should treated
2029 * the same as allocated blocks.
2031 ret = ext4_get_blocks_wrap(NULL, inode, iblock, 1, bh_result, 0, 0, 0);
2032 if ((ret == 0) && !buffer_delay(bh_result)) {
2033 /* the block isn't (pre)allocated yet, let's reserve space */
2035 * XXX: __block_prepare_write() unmaps passed block,
2038 ret = ext4_da_reserve_space(inode, 1);
2040 /* not enough space to reserve */
2043 map_bh(bh_result, inode->i_sb, 0);
2044 set_buffer_new(bh_result);
2045 set_buffer_delay(bh_result);
2046 } else if (ret > 0) {
2047 bh_result->b_size = (ret << inode->i_blkbits);
2053 #define EXT4_DELALLOC_RSVED 1
2054 static int ext4_da_get_block_write(struct inode *inode, sector_t iblock,
2055 struct buffer_head *bh_result, int create)
2058 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2059 loff_t disksize = EXT4_I(inode)->i_disksize;
2060 handle_t *handle = NULL;
2062 handle = ext4_journal_current_handle();
2064 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2065 bh_result, 0, 0, 0);
2068 ret = ext4_get_blocks_wrap(handle, inode, iblock, max_blocks,
2069 bh_result, create, 0, EXT4_DELALLOC_RSVED);
2073 bh_result->b_size = (ret << inode->i_blkbits);
2076 * Update on-disk size along with block allocation
2077 * we don't use 'extend_disksize' as size may change
2078 * within already allocated block -bzzz
2080 disksize = ((loff_t) iblock + ret) << inode->i_blkbits;
2081 if (disksize > i_size_read(inode))
2082 disksize = i_size_read(inode);
2083 if (disksize > EXT4_I(inode)->i_disksize) {
2085 * XXX: replace with spinlock if seen contended -bzzz
2087 down_write(&EXT4_I(inode)->i_data_sem);
2088 if (disksize > EXT4_I(inode)->i_disksize)
2089 EXT4_I(inode)->i_disksize = disksize;
2090 up_write(&EXT4_I(inode)->i_data_sem);
2092 if (EXT4_I(inode)->i_disksize == disksize) {
2093 ret = ext4_mark_inode_dirty(handle, inode);
2102 static int ext4_bh_unmapped_or_delay(handle_t *handle, struct buffer_head *bh)
2105 * unmapped buffer is possible for holes.
2106 * delay buffer is possible with delayed allocation
2108 return ((!buffer_mapped(bh) || buffer_delay(bh)) && buffer_dirty(bh));
2111 static int ext4_normal_get_block_write(struct inode *inode, sector_t iblock,
2112 struct buffer_head *bh_result, int create)
2115 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
2118 * we don't want to do block allocation in writepage
2119 * so call get_block_wrap with create = 0
2121 ret = ext4_get_blocks_wrap(NULL, inode, iblock, max_blocks,
2122 bh_result, 0, 0, 0);
2124 bh_result->b_size = (ret << inode->i_blkbits);
2131 * get called vi ext4_da_writepages after taking page lock (have journal handle)
2132 * get called via journal_submit_inode_data_buffers (no journal handle)
2133 * get called via shrink_page_list via pdflush (no journal handle)
2134 * or grab_page_cache when doing write_begin (have journal handle)
2136 static int ext4_da_writepage(struct page *page,
2137 struct writeback_control *wbc)
2142 struct buffer_head *page_bufs;
2143 struct inode *inode = page->mapping->host;
2145 size = i_size_read(inode);
2146 if (page->index == size >> PAGE_CACHE_SHIFT)
2147 len = size & ~PAGE_CACHE_MASK;
2149 len = PAGE_CACHE_SIZE;
2151 if (page_has_buffers(page)) {
2152 page_bufs = page_buffers(page);
2153 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2154 ext4_bh_unmapped_or_delay)) {
2156 * We don't want to do block allocation
2157 * So redirty the page and return
2158 * We may reach here when we do a journal commit
2159 * via journal_submit_inode_data_buffers.
2160 * If we don't have mapping block we just ignore
2161 * them. We can also reach here via shrink_page_list
2163 redirty_page_for_writepage(wbc, page);
2169 * The test for page_has_buffers() is subtle:
2170 * We know the page is dirty but it lost buffers. That means
2171 * that at some moment in time after write_begin()/write_end()
2172 * has been called all buffers have been clean and thus they
2173 * must have been written at least once. So they are all
2174 * mapped and we can happily proceed with mapping them
2175 * and writing the page.
2177 * Try to initialize the buffer_heads and check whether
2178 * all are mapped and non delay. We don't want to
2179 * do block allocation here.
2181 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2182 ext4_normal_get_block_write);
2184 page_bufs = page_buffers(page);
2185 /* check whether all are mapped and non delay */
2186 if (walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2187 ext4_bh_unmapped_or_delay)) {
2188 redirty_page_for_writepage(wbc, page);
2194 * We can't do block allocation here
2195 * so just redity the page and unlock
2198 redirty_page_for_writepage(wbc, page);
2204 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
2205 ret = nobh_writepage(page, ext4_normal_get_block_write, wbc);
2207 ret = block_write_full_page(page,
2208 ext4_normal_get_block_write,
2215 * For now just follow the DIO way to estimate the max credits
2216 * needed to write out EXT4_MAX_WRITEBACK_PAGES.
2217 * todo: need to calculate the max credits need for
2218 * extent based files, currently the DIO credits is based on
2219 * indirect-blocks mapping way.
2221 * Probably should have a generic way to calculate credits
2222 * for DIO, writepages, and truncate
2224 #define EXT4_MAX_WRITEBACK_PAGES DIO_MAX_BLOCKS
2225 #define EXT4_MAX_WRITEBACK_CREDITS DIO_CREDITS
2227 static int ext4_da_writepages(struct address_space *mapping,
2228 struct writeback_control *wbc)
2230 struct inode *inode = mapping->host;
2231 handle_t *handle = NULL;
2235 loff_t range_start = 0;
2238 * No pages to write? This is mainly a kludge to avoid starting
2239 * a transaction for special inodes like journal inode on last iput()
2240 * because that could violate lock ordering on umount
2242 if (!mapping->nrpages)
2246 * Estimate the worse case needed credits to write out
2247 * EXT4_MAX_BUF_BLOCKS pages
2249 needed_blocks = EXT4_MAX_WRITEBACK_CREDITS;
2251 to_write = wbc->nr_to_write;
2252 if (!wbc->range_cyclic) {
2254 * If range_cyclic is not set force range_cont
2255 * and save the old writeback_index
2257 wbc->range_cont = 1;
2258 range_start = wbc->range_start;
2261 while (!ret && to_write) {
2262 /* start a new transaction*/
2263 handle = ext4_journal_start(inode, needed_blocks);
2264 if (IS_ERR(handle)) {
2265 ret = PTR_ERR(handle);
2266 goto out_writepages;
2268 if (ext4_should_order_data(inode)) {
2270 * With ordered mode we need to add
2271 * the inode to the journal handle
2272 * when we do block allocation.
2274 ret = ext4_jbd2_file_inode(handle, inode);
2276 ext4_journal_stop(handle);
2277 goto out_writepages;
2282 * set the max dirty pages could be write at a time
2283 * to fit into the reserved transaction credits
2285 if (wbc->nr_to_write > EXT4_MAX_WRITEBACK_PAGES)
2286 wbc->nr_to_write = EXT4_MAX_WRITEBACK_PAGES;
2288 to_write -= wbc->nr_to_write;
2289 ret = mpage_da_writepages(mapping, wbc,
2290 ext4_da_get_block_write);
2291 ext4_journal_stop(handle);
2292 if (wbc->nr_to_write) {
2294 * There is no more writeout needed
2295 * or we requested for a noblocking writeout
2296 * and we found the device congested
2298 to_write += wbc->nr_to_write;
2301 wbc->nr_to_write = to_write;
2305 wbc->nr_to_write = to_write;
2307 wbc->range_start = range_start;
2311 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2312 loff_t pos, unsigned len, unsigned flags,
2313 struct page **pagep, void **fsdata)
2315 int ret, retries = 0;
2319 struct inode *inode = mapping->host;
2322 index = pos >> PAGE_CACHE_SHIFT;
2323 from = pos & (PAGE_CACHE_SIZE - 1);
2328 * With delayed allocation, we don't log the i_disksize update
2329 * if there is delayed block allocation. But we still need
2330 * to journalling the i_disksize update if writes to the end
2331 * of file which has an already mapped buffer.
2333 handle = ext4_journal_start(inode, 1);
2334 if (IS_ERR(handle)) {
2335 ret = PTR_ERR(handle);
2339 page = __grab_cache_page(mapping, index);
2341 ext4_journal_stop(handle);
2347 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2348 ext4_da_get_block_prep);
2351 ext4_journal_stop(handle);
2352 page_cache_release(page);
2355 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2362 * Check if we should update i_disksize
2363 * when write to the end of file but not require block allocation
2365 static int ext4_da_should_update_i_disksize(struct page *page,
2366 unsigned long offset)
2368 struct buffer_head *bh;
2369 struct inode *inode = page->mapping->host;
2373 bh = page_buffers(page);
2374 idx = offset >> inode->i_blkbits;
2376 for (i=0; i < idx; i++)
2377 bh = bh->b_this_page;
2379 if (!buffer_mapped(bh) || (buffer_delay(bh)))
2384 static int ext4_da_write_end(struct file *file,
2385 struct address_space *mapping,
2386 loff_t pos, unsigned len, unsigned copied,
2387 struct page *page, void *fsdata)
2389 struct inode *inode = mapping->host;
2391 handle_t *handle = ext4_journal_current_handle();
2393 unsigned long start, end;
2395 start = pos & (PAGE_CACHE_SIZE - 1);
2396 end = start + copied -1;
2399 * generic_write_end() will run mark_inode_dirty() if i_size
2400 * changes. So let's piggyback the i_disksize mark_inode_dirty
2404 new_i_size = pos + copied;
2405 if (new_i_size > EXT4_I(inode)->i_disksize) {
2406 if (ext4_da_should_update_i_disksize(page, end)) {
2407 down_write(&EXT4_I(inode)->i_data_sem);
2408 if (new_i_size > EXT4_I(inode)->i_disksize) {
2410 * Updating i_disksize when extending file
2411 * without needing block allocation
2413 if (ext4_should_order_data(inode))
2414 ret = ext4_jbd2_file_inode(handle,
2417 EXT4_I(inode)->i_disksize = new_i_size;
2419 up_write(&EXT4_I(inode)->i_data_sem);
2422 ret2 = generic_write_end(file, mapping, pos, len, copied,
2427 ret2 = ext4_journal_stop(handle);
2431 return ret ? ret : copied;
2434 static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2437 * Drop reserved blocks
2439 BUG_ON(!PageLocked(page));
2440 if (!page_has_buffers(page))
2443 ext4_da_page_release_reservation(page, offset);
2446 ext4_invalidatepage(page, offset);
2453 * bmap() is special. It gets used by applications such as lilo and by
2454 * the swapper to find the on-disk block of a specific piece of data.
2456 * Naturally, this is dangerous if the block concerned is still in the
2457 * journal. If somebody makes a swapfile on an ext4 data-journaling
2458 * filesystem and enables swap, then they may get a nasty shock when the
2459 * data getting swapped to that swapfile suddenly gets overwritten by
2460 * the original zero's written out previously to the journal and
2461 * awaiting writeback in the kernel's buffer cache.
2463 * So, if we see any bmap calls here on a modified, data-journaled file,
2464 * take extra steps to flush any blocks which might be in the cache.
2466 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
2468 struct inode *inode = mapping->host;
2472 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2473 test_opt(inode->i_sb, DELALLOC)) {
2475 * With delalloc we want to sync the file
2476 * so that we can make sure we allocate
2479 filemap_write_and_wait(mapping);
2482 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
2484 * This is a REALLY heavyweight approach, but the use of
2485 * bmap on dirty files is expected to be extremely rare:
2486 * only if we run lilo or swapon on a freshly made file
2487 * do we expect this to happen.
2489 * (bmap requires CAP_SYS_RAWIO so this does not
2490 * represent an unprivileged user DOS attack --- we'd be
2491 * in trouble if mortal users could trigger this path at
2494 * NB. EXT4_STATE_JDATA is not set on files other than
2495 * regular files. If somebody wants to bmap a directory
2496 * or symlink and gets confused because the buffer
2497 * hasn't yet been flushed to disk, they deserve
2498 * everything they get.
2501 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
2502 journal = EXT4_JOURNAL(inode);
2503 jbd2_journal_lock_updates(journal);
2504 err = jbd2_journal_flush(journal);
2505 jbd2_journal_unlock_updates(journal);
2511 return generic_block_bmap(mapping,block,ext4_get_block);
2514 static int bget_one(handle_t *handle, struct buffer_head *bh)
2520 static int bput_one(handle_t *handle, struct buffer_head *bh)
2527 * Note that we don't need to start a transaction unless we're journaling data
2528 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2529 * need to file the inode to the transaction's list in ordered mode because if
2530 * we are writing back data added by write(), the inode is already there and if
2531 * we are writing back data modified via mmap(), noone guarantees in which
2532 * transaction the data will hit the disk. In case we are journaling data, we
2533 * cannot start transaction directly because transaction start ranks above page
2534 * lock so we have to do some magic.
2536 * In all journaling modes block_write_full_page() will start the I/O.
2540 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2545 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
2547 * Same applies to ext4_get_block(). We will deadlock on various things like
2548 * lock_journal and i_data_sem
2550 * Setting PF_MEMALLOC here doesn't work - too many internal memory
2553 * 16May01: If we're reentered then journal_current_handle() will be
2554 * non-zero. We simply *return*.
2556 * 1 July 2001: @@@ FIXME:
2557 * In journalled data mode, a data buffer may be metadata against the
2558 * current transaction. But the same file is part of a shared mapping
2559 * and someone does a writepage() on it.
2561 * We will move the buffer onto the async_data list, but *after* it has
2562 * been dirtied. So there's a small window where we have dirty data on
2565 * Note that this only applies to the last partial page in the file. The
2566 * bit which block_write_full_page() uses prepare/commit for. (That's
2567 * broken code anyway: it's wrong for msync()).
2569 * It's a rare case: affects the final partial page, for journalled data
2570 * where the file is subject to bith write() and writepage() in the same
2571 * transction. To fix it we'll need a custom block_write_full_page().
2572 * We'll probably need that anyway for journalling writepage() output.
2574 * We don't honour synchronous mounts for writepage(). That would be
2575 * disastrous. Any write() or metadata operation will sync the fs for
2579 static int __ext4_normal_writepage(struct page *page,
2580 struct writeback_control *wbc)
2582 struct inode *inode = page->mapping->host;
2584 if (test_opt(inode->i_sb, NOBH))
2585 return nobh_writepage(page,
2586 ext4_normal_get_block_write, wbc);
2588 return block_write_full_page(page,
2589 ext4_normal_get_block_write,
2593 static int ext4_normal_writepage(struct page *page,
2594 struct writeback_control *wbc)
2596 struct inode *inode = page->mapping->host;
2597 loff_t size = i_size_read(inode);
2600 J_ASSERT(PageLocked(page));
2601 if (page->index == size >> PAGE_CACHE_SHIFT)
2602 len = size & ~PAGE_CACHE_MASK;
2604 len = PAGE_CACHE_SIZE;
2606 if (page_has_buffers(page)) {
2607 /* if page has buffers it should all be mapped
2608 * and allocated. If there are not buffers attached
2609 * to the page we know the page is dirty but it lost
2610 * buffers. That means that at some moment in time
2611 * after write_begin() / write_end() has been called
2612 * all buffers have been clean and thus they must have been
2613 * written at least once. So they are all mapped and we can
2614 * happily proceed with mapping them and writing the page.
2616 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2617 ext4_bh_unmapped_or_delay));
2620 if (!ext4_journal_current_handle())
2621 return __ext4_normal_writepage(page, wbc);
2623 redirty_page_for_writepage(wbc, page);
2628 static int __ext4_journalled_writepage(struct page *page,
2629 struct writeback_control *wbc)
2631 struct address_space *mapping = page->mapping;
2632 struct inode *inode = mapping->host;
2633 struct buffer_head *page_bufs;
2634 handle_t *handle = NULL;
2638 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
2639 ext4_normal_get_block_write);
2643 page_bufs = page_buffers(page);
2644 walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, NULL,
2646 /* As soon as we unlock the page, it can go away, but we have
2647 * references to buffers so we are safe */
2650 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
2651 if (IS_ERR(handle)) {
2652 ret = PTR_ERR(handle);
2656 ret = walk_page_buffers(handle, page_bufs, 0,
2657 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
2659 err = walk_page_buffers(handle, page_bufs, 0,
2660 PAGE_CACHE_SIZE, NULL, write_end_fn);
2663 err = ext4_journal_stop(handle);
2667 walk_page_buffers(handle, page_bufs, 0,
2668 PAGE_CACHE_SIZE, NULL, bput_one);
2669 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
2678 static int ext4_journalled_writepage(struct page *page,
2679 struct writeback_control *wbc)
2681 struct inode *inode = page->mapping->host;
2682 loff_t size = i_size_read(inode);
2685 J_ASSERT(PageLocked(page));
2686 if (page->index == size >> PAGE_CACHE_SHIFT)
2687 len = size & ~PAGE_CACHE_MASK;
2689 len = PAGE_CACHE_SIZE;
2691 if (page_has_buffers(page)) {
2692 /* if page has buffers it should all be mapped
2693 * and allocated. If there are not buffers attached
2694 * to the page we know the page is dirty but it lost
2695 * buffers. That means that at some moment in time
2696 * after write_begin() / write_end() has been called
2697 * all buffers have been clean and thus they must have been
2698 * written at least once. So they are all mapped and we can
2699 * happily proceed with mapping them and writing the page.
2701 BUG_ON(walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
2702 ext4_bh_unmapped_or_delay));
2705 if (ext4_journal_current_handle())
2708 if (PageChecked(page)) {
2710 * It's mmapped pagecache. Add buffers and journal it. There
2711 * doesn't seem much point in redirtying the page here.
2713 ClearPageChecked(page);
2714 return __ext4_journalled_writepage(page, wbc);
2717 * It may be a page full of checkpoint-mode buffers. We don't
2718 * really know unless we go poke around in the buffer_heads.
2719 * But block_write_full_page will do the right thing.
2721 return block_write_full_page(page,
2722 ext4_normal_get_block_write,
2726 redirty_page_for_writepage(wbc, page);
2731 static int ext4_readpage(struct file *file, struct page *page)
2733 return mpage_readpage(page, ext4_get_block);
2737 ext4_readpages(struct file *file, struct address_space *mapping,
2738 struct list_head *pages, unsigned nr_pages)
2740 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
2743 static void ext4_invalidatepage(struct page *page, unsigned long offset)
2745 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2748 * If it's a full truncate we just forget about the pending dirtying
2751 ClearPageChecked(page);
2753 jbd2_journal_invalidatepage(journal, page, offset);
2756 static int ext4_releasepage(struct page *page, gfp_t wait)
2758 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2760 WARN_ON(PageChecked(page));
2761 if (!page_has_buffers(page))
2763 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2767 * If the O_DIRECT write will extend the file then add this inode to the
2768 * orphan list. So recovery will truncate it back to the original size
2769 * if the machine crashes during the write.
2771 * If the O_DIRECT write is intantiating holes inside i_size and the machine
2772 * crashes then stale disk data _may_ be exposed inside the file. But current
2773 * VFS code falls back into buffered path in that case so we are safe.
2775 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
2776 const struct iovec *iov, loff_t offset,
2777 unsigned long nr_segs)
2779 struct file *file = iocb->ki_filp;
2780 struct inode *inode = file->f_mapping->host;
2781 struct ext4_inode_info *ei = EXT4_I(inode);
2785 size_t count = iov_length(iov, nr_segs);
2788 loff_t final_size = offset + count;
2790 if (final_size > inode->i_size) {
2791 /* Credits for sb + inode write */
2792 handle = ext4_journal_start(inode, 2);
2793 if (IS_ERR(handle)) {
2794 ret = PTR_ERR(handle);
2797 ret = ext4_orphan_add(handle, inode);
2799 ext4_journal_stop(handle);
2803 ei->i_disksize = inode->i_size;
2804 ext4_journal_stop(handle);
2808 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2810 ext4_get_block, NULL);
2815 /* Credits for sb + inode write */
2816 handle = ext4_journal_start(inode, 2);
2817 if (IS_ERR(handle)) {
2818 /* This is really bad luck. We've written the data
2819 * but cannot extend i_size. Bail out and pretend
2820 * the write failed... */
2821 ret = PTR_ERR(handle);
2825 ext4_orphan_del(handle, inode);
2827 loff_t end = offset + ret;
2828 if (end > inode->i_size) {
2829 ei->i_disksize = end;
2830 i_size_write(inode, end);
2832 * We're going to return a positive `ret'
2833 * here due to non-zero-length I/O, so there's
2834 * no way of reporting error returns from
2835 * ext4_mark_inode_dirty() to userspace. So
2838 ext4_mark_inode_dirty(handle, inode);
2841 err = ext4_journal_stop(handle);
2850 * Pages can be marked dirty completely asynchronously from ext4's journalling
2851 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
2852 * much here because ->set_page_dirty is called under VFS locks. The page is
2853 * not necessarily locked.
2855 * We cannot just dirty the page and leave attached buffers clean, because the
2856 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
2857 * or jbddirty because all the journalling code will explode.
2859 * So what we do is to mark the page "pending dirty" and next time writepage
2860 * is called, propagate that into the buffers appropriately.
2862 static int ext4_journalled_set_page_dirty(struct page *page)
2864 SetPageChecked(page);
2865 return __set_page_dirty_nobuffers(page);
2868 static const struct address_space_operations ext4_ordered_aops = {
2869 .readpage = ext4_readpage,
2870 .readpages = ext4_readpages,
2871 .writepage = ext4_normal_writepage,
2872 .sync_page = block_sync_page,
2873 .write_begin = ext4_write_begin,
2874 .write_end = ext4_ordered_write_end,
2876 .invalidatepage = ext4_invalidatepage,
2877 .releasepage = ext4_releasepage,
2878 .direct_IO = ext4_direct_IO,
2879 .migratepage = buffer_migrate_page,
2880 .is_partially_uptodate = block_is_partially_uptodate,
2883 static const struct address_space_operations ext4_writeback_aops = {
2884 .readpage = ext4_readpage,
2885 .readpages = ext4_readpages,
2886 .writepage = ext4_normal_writepage,
2887 .sync_page = block_sync_page,
2888 .write_begin = ext4_write_begin,
2889 .write_end = ext4_writeback_write_end,
2891 .invalidatepage = ext4_invalidatepage,
2892 .releasepage = ext4_releasepage,
2893 .direct_IO = ext4_direct_IO,
2894 .migratepage = buffer_migrate_page,
2895 .is_partially_uptodate = block_is_partially_uptodate,
2898 static const struct address_space_operations ext4_journalled_aops = {
2899 .readpage = ext4_readpage,
2900 .readpages = ext4_readpages,
2901 .writepage = ext4_journalled_writepage,
2902 .sync_page = block_sync_page,
2903 .write_begin = ext4_write_begin,
2904 .write_end = ext4_journalled_write_end,
2905 .set_page_dirty = ext4_journalled_set_page_dirty,
2907 .invalidatepage = ext4_invalidatepage,
2908 .releasepage = ext4_releasepage,
2909 .is_partially_uptodate = block_is_partially_uptodate,
2912 static const struct address_space_operations ext4_da_aops = {
2913 .readpage = ext4_readpage,
2914 .readpages = ext4_readpages,
2915 .writepage = ext4_da_writepage,
2916 .writepages = ext4_da_writepages,
2917 .sync_page = block_sync_page,
2918 .write_begin = ext4_da_write_begin,
2919 .write_end = ext4_da_write_end,
2921 .invalidatepage = ext4_da_invalidatepage,
2922 .releasepage = ext4_releasepage,
2923 .direct_IO = ext4_direct_IO,
2924 .migratepage = buffer_migrate_page,
2925 .is_partially_uptodate = block_is_partially_uptodate,
2928 void ext4_set_aops(struct inode *inode)
2930 if (ext4_should_order_data(inode) &&
2931 test_opt(inode->i_sb, DELALLOC))
2932 inode->i_mapping->a_ops = &ext4_da_aops;
2933 else if (ext4_should_order_data(inode))
2934 inode->i_mapping->a_ops = &ext4_ordered_aops;
2935 else if (ext4_should_writeback_data(inode) &&
2936 test_opt(inode->i_sb, DELALLOC))
2937 inode->i_mapping->a_ops = &ext4_da_aops;
2938 else if (ext4_should_writeback_data(inode))
2939 inode->i_mapping->a_ops = &ext4_writeback_aops;
2941 inode->i_mapping->a_ops = &ext4_journalled_aops;
2945 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
2946 * up to the end of the block which corresponds to `from'.
2947 * This required during truncate. We need to physically zero the tail end
2948 * of that block so it doesn't yield old data if the file is later grown.
2950 int ext4_block_truncate_page(handle_t *handle,
2951 struct address_space *mapping, loff_t from)
2953 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
2954 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2955 unsigned blocksize, length, pos;
2957 struct inode *inode = mapping->host;
2958 struct buffer_head *bh;
2962 page = grab_cache_page(mapping, from >> PAGE_CACHE_SHIFT);
2966 blocksize = inode->i_sb->s_blocksize;
2967 length = blocksize - (offset & (blocksize - 1));
2968 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2971 * For "nobh" option, we can only work if we don't need to
2972 * read-in the page - otherwise we create buffers to do the IO.
2974 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
2975 ext4_should_writeback_data(inode) && PageUptodate(page)) {
2976 zero_user(page, offset, length);
2977 set_page_dirty(page);
2981 if (!page_has_buffers(page))
2982 create_empty_buffers(page, blocksize, 0);
2984 /* Find the buffer that contains "offset" */
2985 bh = page_buffers(page);
2987 while (offset >= pos) {
2988 bh = bh->b_this_page;
2994 if (buffer_freed(bh)) {
2995 BUFFER_TRACE(bh, "freed: skip");
2999 if (!buffer_mapped(bh)) {
3000 BUFFER_TRACE(bh, "unmapped");
3001 ext4_get_block(inode, iblock, bh, 0);
3002 /* unmapped? It's a hole - nothing to do */
3003 if (!buffer_mapped(bh)) {
3004 BUFFER_TRACE(bh, "still unmapped");
3009 /* Ok, it's mapped. Make sure it's up-to-date */
3010 if (PageUptodate(page))
3011 set_buffer_uptodate(bh);
3013 if (!buffer_uptodate(bh)) {
3015 ll_rw_block(READ, 1, &bh);
3017 /* Uhhuh. Read error. Complain and punt. */
3018 if (!buffer_uptodate(bh))
3022 if (ext4_should_journal_data(inode)) {
3023 BUFFER_TRACE(bh, "get write access");
3024 err = ext4_journal_get_write_access(handle, bh);
3029 zero_user(page, offset, length);
3031 BUFFER_TRACE(bh, "zeroed end of block");
3034 if (ext4_should_journal_data(inode)) {
3035 err = ext4_journal_dirty_metadata(handle, bh);
3037 if (ext4_should_order_data(inode))
3038 err = ext4_jbd2_file_inode(handle, inode);
3039 mark_buffer_dirty(bh);
3044 page_cache_release(page);
3049 * Probably it should be a library function... search for first non-zero word
3050 * or memcmp with zero_page, whatever is better for particular architecture.
3053 static inline int all_zeroes(__le32 *p, __le32 *q)
3062 * ext4_find_shared - find the indirect blocks for partial truncation.
3063 * @inode: inode in question
3064 * @depth: depth of the affected branch
3065 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
3066 * @chain: place to store the pointers to partial indirect blocks
3067 * @top: place to the (detached) top of branch
3069 * This is a helper function used by ext4_truncate().
3071 * When we do truncate() we may have to clean the ends of several
3072 * indirect blocks but leave the blocks themselves alive. Block is
3073 * partially truncated if some data below the new i_size is refered
3074 * from it (and it is on the path to the first completely truncated
3075 * data block, indeed). We have to free the top of that path along
3076 * with everything to the right of the path. Since no allocation
3077 * past the truncation point is possible until ext4_truncate()
3078 * finishes, we may safely do the latter, but top of branch may
3079 * require special attention - pageout below the truncation point
3080 * might try to populate it.
3082 * We atomically detach the top of branch from the tree, store the
3083 * block number of its root in *@top, pointers to buffer_heads of
3084 * partially truncated blocks - in @chain[].bh and pointers to
3085 * their last elements that should not be removed - in
3086 * @chain[].p. Return value is the pointer to last filled element
3089 * The work left to caller to do the actual freeing of subtrees:
3090 * a) free the subtree starting from *@top
3091 * b) free the subtrees whose roots are stored in
3092 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
3093 * c) free the subtrees growing from the inode past the @chain[0].
3094 * (no partially truncated stuff there). */
3096 static Indirect *ext4_find_shared(struct inode *inode, int depth,
3097 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
3099 Indirect *partial, *p;
3103 /* Make k index the deepest non-null offest + 1 */
3104 for (k = depth; k > 1 && !offsets[k-1]; k--)
3106 partial = ext4_get_branch(inode, k, offsets, chain, &err);
3107 /* Writer: pointers */
3109 partial = chain + k-1;
3111 * If the branch acquired continuation since we've looked at it -
3112 * fine, it should all survive and (new) top doesn't belong to us.
3114 if (!partial->key && *partial->p)
3117 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
3120 * OK, we've found the last block that must survive. The rest of our
3121 * branch should be detached before unlocking. However, if that rest
3122 * of branch is all ours and does not grow immediately from the inode
3123 * it's easier to cheat and just decrement partial->p.
3125 if (p == chain + k - 1 && p > chain) {
3129 /* Nope, don't do this in ext4. Must leave the tree intact */
3136 while(partial > p) {
3137 brelse(partial->bh);
3145 * Zero a number of block pointers in either an inode or an indirect block.
3146 * If we restart the transaction we must again get write access to the
3147 * indirect block for further modification.
3149 * We release `count' blocks on disk, but (last - first) may be greater
3150 * than `count' because there can be holes in there.
3152 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
3153 struct buffer_head *bh, ext4_fsblk_t block_to_free,
3154 unsigned long count, __le32 *first, __le32 *last)
3157 if (try_to_extend_transaction(handle, inode)) {
3159 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
3160 ext4_journal_dirty_metadata(handle, bh);
3162 ext4_mark_inode_dirty(handle, inode);
3163 ext4_journal_test_restart(handle, inode);
3165 BUFFER_TRACE(bh, "retaking write access");
3166 ext4_journal_get_write_access(handle, bh);
3171 * Any buffers which are on the journal will be in memory. We find
3172 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
3173 * on them. We've already detached each block from the file, so
3174 * bforget() in jbd2_journal_forget() should be safe.
3176 * AKPM: turn on bforget in jbd2_journal_forget()!!!
3178 for (p = first; p < last; p++) {
3179 u32 nr = le32_to_cpu(*p);
3181 struct buffer_head *tbh;
3184 tbh = sb_find_get_block(inode->i_sb, nr);
3185 ext4_forget(handle, 0, inode, tbh, nr);
3189 ext4_free_blocks(handle, inode, block_to_free, count, 0);
3193 * ext4_free_data - free a list of data blocks
3194 * @handle: handle for this transaction
3195 * @inode: inode we are dealing with
3196 * @this_bh: indirect buffer_head which contains *@first and *@last
3197 * @first: array of block numbers
3198 * @last: points immediately past the end of array
3200 * We are freeing all blocks refered from that array (numbers are stored as
3201 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
3203 * We accumulate contiguous runs of blocks to free. Conveniently, if these
3204 * blocks are contiguous then releasing them at one time will only affect one
3205 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
3206 * actually use a lot of journal space.
3208 * @this_bh will be %NULL if @first and @last point into the inode's direct
3211 static void ext4_free_data(handle_t *handle, struct inode *inode,
3212 struct buffer_head *this_bh,
3213 __le32 *first, __le32 *last)
3215 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
3216 unsigned long count = 0; /* Number of blocks in the run */
3217 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
3220 ext4_fsblk_t nr; /* Current block # */
3221 __le32 *p; /* Pointer into inode/ind
3222 for current block */
3225 if (this_bh) { /* For indirect block */
3226 BUFFER_TRACE(this_bh, "get_write_access");
3227 err = ext4_journal_get_write_access(handle, this_bh);
3228 /* Important: if we can't update the indirect pointers
3229 * to the blocks, we can't free them. */
3234 for (p = first; p < last; p++) {
3235 nr = le32_to_cpu(*p);
3237 /* accumulate blocks to free if they're contiguous */
3240 block_to_free_p = p;
3242 } else if (nr == block_to_free + count) {
3245 ext4_clear_blocks(handle, inode, this_bh,
3247 count, block_to_free_p, p);
3249 block_to_free_p = p;
3256 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
3257 count, block_to_free_p, p);
3260 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
3263 * The buffer head should have an attached journal head at this
3264 * point. However, if the data is corrupted and an indirect
3265 * block pointed to itself, it would have been detached when
3266 * the block was cleared. Check for this instead of OOPSing.
3269 ext4_journal_dirty_metadata(handle, this_bh);
3271 ext4_error(inode->i_sb, __func__,
3272 "circular indirect block detected, "
3273 "inode=%lu, block=%llu",
3275 (unsigned long long) this_bh->b_blocknr);
3280 * ext4_free_branches - free an array of branches
3281 * @handle: JBD handle for this transaction
3282 * @inode: inode we are dealing with
3283 * @parent_bh: the buffer_head which contains *@first and *@last
3284 * @first: array of block numbers
3285 * @last: pointer immediately past the end of array
3286 * @depth: depth of the branches to free
3288 * We are freeing all blocks refered from these branches (numbers are
3289 * stored as little-endian 32-bit) and updating @inode->i_blocks
3292 static void ext4_free_branches(handle_t *handle, struct inode *inode,
3293 struct buffer_head *parent_bh,
3294 __le32 *first, __le32 *last, int depth)
3299 if (is_handle_aborted(handle))
3303 struct buffer_head *bh;
3304 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3306 while (--p >= first) {
3307 nr = le32_to_cpu(*p);
3309 continue; /* A hole */
3311 /* Go read the buffer for the next level down */
3312 bh = sb_bread(inode->i_sb, nr);
3315 * A read failure? Report error and clear slot
3319 ext4_error(inode->i_sb, "ext4_free_branches",
3320 "Read failure, inode=%lu, block=%llu",
3325 /* This zaps the entire block. Bottom up. */
3326 BUFFER_TRACE(bh, "free child branches");
3327 ext4_free_branches(handle, inode, bh,
3328 (__le32*)bh->b_data,
3329 (__le32*)bh->b_data + addr_per_block,
3333 * We've probably journalled the indirect block several
3334 * times during the truncate. But it's no longer
3335 * needed and we now drop it from the transaction via
3336 * jbd2_journal_revoke().
3338 * That's easy if it's exclusively part of this
3339 * transaction. But if it's part of the committing
3340 * transaction then jbd2_journal_forget() will simply
3341 * brelse() it. That means that if the underlying
3342 * block is reallocated in ext4_get_block(),
3343 * unmap_underlying_metadata() will find this block
3344 * and will try to get rid of it. damn, damn.
3346 * If this block has already been committed to the
3347 * journal, a revoke record will be written. And
3348 * revoke records must be emitted *before* clearing
3349 * this block's bit in the bitmaps.
3351 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
3354 * Everything below this this pointer has been
3355 * released. Now let this top-of-subtree go.
3357 * We want the freeing of this indirect block to be
3358 * atomic in the journal with the updating of the
3359 * bitmap block which owns it. So make some room in
3362 * We zero the parent pointer *after* freeing its
3363 * pointee in the bitmaps, so if extend_transaction()
3364 * for some reason fails to put the bitmap changes and
3365 * the release into the same transaction, recovery
3366 * will merely complain about releasing a free block,
3367 * rather than leaking blocks.
3369 if (is_handle_aborted(handle))
3371 if (try_to_extend_transaction(handle, inode)) {
3372 ext4_mark_inode_dirty(handle, inode);
3373 ext4_journal_test_restart(handle, inode);
3376 ext4_free_blocks(handle, inode, nr, 1, 1);
3380 * The block which we have just freed is
3381 * pointed to by an indirect block: journal it
3383 BUFFER_TRACE(parent_bh, "get_write_access");
3384 if (!ext4_journal_get_write_access(handle,
3387 BUFFER_TRACE(parent_bh,
3388 "call ext4_journal_dirty_metadata");
3389 ext4_journal_dirty_metadata(handle,
3395 /* We have reached the bottom of the tree. */
3396 BUFFER_TRACE(parent_bh, "free data blocks");
3397 ext4_free_data(handle, inode, parent_bh, first, last);
3401 int ext4_can_truncate(struct inode *inode)
3403 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3405 if (S_ISREG(inode->i_mode))
3407 if (S_ISDIR(inode->i_mode))
3409 if (S_ISLNK(inode->i_mode))
3410 return !ext4_inode_is_fast_symlink(inode);
3417 * We block out ext4_get_block() block instantiations across the entire
3418 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3419 * simultaneously on behalf of the same inode.
3421 * As we work through the truncate and commmit bits of it to the journal there
3422 * is one core, guiding principle: the file's tree must always be consistent on
3423 * disk. We must be able to restart the truncate after a crash.
3425 * The file's tree may be transiently inconsistent in memory (although it
3426 * probably isn't), but whenever we close off and commit a journal transaction,
3427 * the contents of (the filesystem + the journal) must be consistent and
3428 * restartable. It's pretty simple, really: bottom up, right to left (although
3429 * left-to-right works OK too).
3431 * Note that at recovery time, journal replay occurs *before* the restart of
3432 * truncate against the orphan inode list.
3434 * The committed inode has the new, desired i_size (which is the same as
3435 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3436 * that this inode's truncate did not complete and it will again call
3437 * ext4_truncate() to have another go. So there will be instantiated blocks
3438 * to the right of the truncation point in a crashed ext4 filesystem. But
3439 * that's fine - as long as they are linked from the inode, the post-crash
3440 * ext4_truncate() run will find them and release them.
3442 void ext4_truncate(struct inode *inode)
3445 struct ext4_inode_info *ei = EXT4_I(inode);
3446 __le32 *i_data = ei->i_data;
3447 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
3448 struct address_space *mapping = inode->i_mapping;
3449 ext4_lblk_t offsets[4];
3454 ext4_lblk_t last_block;
3455 unsigned blocksize = inode->i_sb->s_blocksize;
3457 if (!ext4_can_truncate(inode))
3460 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
3461 ext4_ext_truncate(inode);
3465 handle = start_transaction(inode);
3467 return; /* AKPM: return what? */
3469 last_block = (inode->i_size + blocksize-1)
3470 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
3472 if (inode->i_size & (blocksize - 1))
3473 if (ext4_block_truncate_page(handle, mapping, inode->i_size))
3476 n = ext4_block_to_path(inode, last_block, offsets, NULL);
3478 goto out_stop; /* error */
3481 * OK. This truncate is going to happen. We add the inode to the
3482 * orphan list, so that if this truncate spans multiple transactions,
3483 * and we crash, we will resume the truncate when the filesystem
3484 * recovers. It also marks the inode dirty, to catch the new size.
3486 * Implication: the file must always be in a sane, consistent
3487 * truncatable state while each transaction commits.
3489 if (ext4_orphan_add(handle, inode))
3493 * From here we block out all ext4_get_block() callers who want to
3494 * modify the block allocation tree.
3496 down_write(&ei->i_data_sem);
3498 ext4_discard_reservation(inode);
3501 * The orphan list entry will now protect us from any crash which
3502 * occurs before the truncate completes, so it is now safe to propagate
3503 * the new, shorter inode size (held for now in i_size) into the
3504 * on-disk inode. We do this via i_disksize, which is the value which
3505 * ext4 *really* writes onto the disk inode.
3507 ei->i_disksize = inode->i_size;
3509 if (n == 1) { /* direct blocks */
3510 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
3511 i_data + EXT4_NDIR_BLOCKS);
3515 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
3516 /* Kill the top of shared branch (not detached) */
3518 if (partial == chain) {
3519 /* Shared branch grows from the inode */
3520 ext4_free_branches(handle, inode, NULL,
3521 &nr, &nr+1, (chain+n-1) - partial);
3524 * We mark the inode dirty prior to restart,
3525 * and prior to stop. No need for it here.
3528 /* Shared branch grows from an indirect block */
3529 BUFFER_TRACE(partial->bh, "get_write_access");
3530 ext4_free_branches(handle, inode, partial->bh,
3532 partial->p+1, (chain+n-1) - partial);
3535 /* Clear the ends of indirect blocks on the shared branch */
3536 while (partial > chain) {
3537 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
3538 (__le32*)partial->bh->b_data+addr_per_block,
3539 (chain+n-1) - partial);
3540 BUFFER_TRACE(partial->bh, "call brelse");
3541 brelse (partial->bh);
3545 /* Kill the remaining (whole) subtrees */
3546 switch (offsets[0]) {
3548 nr = i_data[EXT4_IND_BLOCK];
3550 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
3551 i_data[EXT4_IND_BLOCK] = 0;
3553 case EXT4_IND_BLOCK:
3554 nr = i_data[EXT4_DIND_BLOCK];
3556 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
3557 i_data[EXT4_DIND_BLOCK] = 0;
3559 case EXT4_DIND_BLOCK:
3560 nr = i_data[EXT4_TIND_BLOCK];
3562 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
3563 i_data[EXT4_TIND_BLOCK] = 0;
3565 case EXT4_TIND_BLOCK:
3569 up_write(&ei->i_data_sem);
3570 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3571 ext4_mark_inode_dirty(handle, inode);
3574 * In a multi-transaction truncate, we only make the final transaction
3581 * If this was a simple ftruncate(), and the file will remain alive
3582 * then we need to clear up the orphan record which we created above.
3583 * However, if this was a real unlink then we were called by
3584 * ext4_delete_inode(), and we allow that function to clean up the
3585 * orphan info for us.
3588 ext4_orphan_del(handle, inode);
3590 ext4_journal_stop(handle);
3593 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
3594 unsigned long ino, struct ext4_iloc *iloc)
3596 ext4_group_t block_group;
3597 unsigned long offset;
3599 struct ext4_group_desc *gdp;
3601 if (!ext4_valid_inum(sb, ino)) {
3603 * This error is already checked for in namei.c unless we are
3604 * looking at an NFS filehandle, in which case no error
3610 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
3611 gdp = ext4_get_group_desc(sb, block_group, NULL);
3616 * Figure out the offset within the block group inode table
3618 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
3619 EXT4_INODE_SIZE(sb);
3620 block = ext4_inode_table(sb, gdp) +
3621 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
3623 iloc->block_group = block_group;
3624 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
3629 * ext4_get_inode_loc returns with an extra refcount against the inode's
3630 * underlying buffer_head on success. If 'in_mem' is true, we have all
3631 * data in memory that is needed to recreate the on-disk version of this
3634 static int __ext4_get_inode_loc(struct inode *inode,
3635 struct ext4_iloc *iloc, int in_mem)
3638 struct buffer_head *bh;
3640 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
3644 bh = sb_getblk(inode->i_sb, block);
3646 ext4_error (inode->i_sb, "ext4_get_inode_loc",
3647 "unable to read inode block - "
3648 "inode=%lu, block=%llu",
3649 inode->i_ino, block);
3652 if (!buffer_uptodate(bh)) {
3656 * If the buffer has the write error flag, we have failed
3657 * to write out another inode in the same block. In this
3658 * case, we don't have to read the block because we may
3659 * read the old inode data successfully.
3661 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3662 set_buffer_uptodate(bh);
3664 if (buffer_uptodate(bh)) {
3665 /* someone brought it uptodate while we waited */
3671 * If we have all information of the inode in memory and this
3672 * is the only valid inode in the block, we need not read the
3676 struct buffer_head *bitmap_bh;
3677 struct ext4_group_desc *desc;
3678 int inodes_per_buffer;
3679 int inode_offset, i;
3680 ext4_group_t block_group;
3683 block_group = (inode->i_ino - 1) /
3684 EXT4_INODES_PER_GROUP(inode->i_sb);
3685 inodes_per_buffer = bh->b_size /
3686 EXT4_INODE_SIZE(inode->i_sb);
3687 inode_offset = ((inode->i_ino - 1) %
3688 EXT4_INODES_PER_GROUP(inode->i_sb));
3689 start = inode_offset & ~(inodes_per_buffer - 1);
3691 /* Is the inode bitmap in cache? */
3692 desc = ext4_get_group_desc(inode->i_sb,
3697 bitmap_bh = sb_getblk(inode->i_sb,
3698 ext4_inode_bitmap(inode->i_sb, desc));
3703 * If the inode bitmap isn't in cache then the
3704 * optimisation may end up performing two reads instead
3705 * of one, so skip it.
3707 if (!buffer_uptodate(bitmap_bh)) {
3711 for (i = start; i < start + inodes_per_buffer; i++) {
3712 if (i == inode_offset)
3714 if (ext4_test_bit(i, bitmap_bh->b_data))
3718 if (i == start + inodes_per_buffer) {
3719 /* all other inodes are free, so skip I/O */
3720 memset(bh->b_data, 0, bh->b_size);
3721 set_buffer_uptodate(bh);
3729 * There are other valid inodes in the buffer, this inode
3730 * has in-inode xattrs, or we don't have this inode in memory.
3731 * Read the block from disk.
3734 bh->b_end_io = end_buffer_read_sync;
3735 submit_bh(READ_META, bh);
3737 if (!buffer_uptodate(bh)) {
3738 ext4_error(inode->i_sb, "ext4_get_inode_loc",
3739 "unable to read inode block - "
3740 "inode=%lu, block=%llu",
3741 inode->i_ino, block);
3751 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
3753 /* We have all inode data except xattrs in memory here. */
3754 return __ext4_get_inode_loc(inode, iloc,
3755 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
3758 void ext4_set_inode_flags(struct inode *inode)
3760 unsigned int flags = EXT4_I(inode)->i_flags;
3762 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
3763 if (flags & EXT4_SYNC_FL)
3764 inode->i_flags |= S_SYNC;
3765 if (flags & EXT4_APPEND_FL)
3766 inode->i_flags |= S_APPEND;
3767 if (flags & EXT4_IMMUTABLE_FL)
3768 inode->i_flags |= S_IMMUTABLE;
3769 if (flags & EXT4_NOATIME_FL)
3770 inode->i_flags |= S_NOATIME;
3771 if (flags & EXT4_DIRSYNC_FL)
3772 inode->i_flags |= S_DIRSYNC;
3775 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3776 void ext4_get_inode_flags(struct ext4_inode_info *ei)
3778 unsigned int flags = ei->vfs_inode.i_flags;
3780 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3781 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
3783 ei->i_flags |= EXT4_SYNC_FL;
3784 if (flags & S_APPEND)
3785 ei->i_flags |= EXT4_APPEND_FL;
3786 if (flags & S_IMMUTABLE)
3787 ei->i_flags |= EXT4_IMMUTABLE_FL;
3788 if (flags & S_NOATIME)
3789 ei->i_flags |= EXT4_NOATIME_FL;
3790 if (flags & S_DIRSYNC)
3791 ei->i_flags |= EXT4_DIRSYNC_FL;
3793 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
3794 struct ext4_inode_info *ei)
3797 struct inode *inode = &(ei->vfs_inode);
3798 struct super_block *sb = inode->i_sb;
3800 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3801 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3802 /* we are using combined 48 bit field */
3803 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3804 le32_to_cpu(raw_inode->i_blocks_lo);
3805 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
3806 /* i_blocks represent file system block size */
3807 return i_blocks << (inode->i_blkbits - 9);
3812 return le32_to_cpu(raw_inode->i_blocks_lo);
3816 struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
3818 struct ext4_iloc iloc;
3819 struct ext4_inode *raw_inode;
3820 struct ext4_inode_info *ei;
3821 struct buffer_head *bh;
3822 struct inode *inode;
3826 inode = iget_locked(sb, ino);
3828 return ERR_PTR(-ENOMEM);
3829 if (!(inode->i_state & I_NEW))
3833 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
3834 ei->i_acl = EXT4_ACL_NOT_CACHED;
3835 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
3837 ei->i_block_alloc_info = NULL;
3839 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3843 raw_inode = ext4_raw_inode(&iloc);
3844 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
3845 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3846 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
3847 if(!(test_opt (inode->i_sb, NO_UID32))) {
3848 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3849 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
3851 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
3854 ei->i_dir_start_lookup = 0;
3855 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3856 /* We now have enough fields to check if the inode was active or not.
3857 * This is needed because nfsd might try to access dead inodes
3858 * the test is that same one that e2fsck uses
3859 * NeilBrown 1999oct15
3861 if (inode->i_nlink == 0) {
3862 if (inode->i_mode == 0 ||
3863 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
3864 /* this inode is deleted */
3869 /* The only unlinked inodes we let through here have
3870 * valid i_mode and are being read by the orphan
3871 * recovery code: that's fine, we're about to complete
3872 * the process of deleting those. */
3874 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
3875 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
3876 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
3877 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
3878 cpu_to_le32(EXT4_OS_HURD)) {
3880 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
3882 inode->i_size = ext4_isize(raw_inode);
3883 ei->i_disksize = inode->i_size;
3884 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3885 ei->i_block_group = iloc.block_group;
3887 * NOTE! The in-memory inode i_data array is in little-endian order
3888 * even on big-endian machines: we do NOT byteswap the block numbers!
3890 for (block = 0; block < EXT4_N_BLOCKS; block++)
3891 ei->i_data[block] = raw_inode->i_block[block];
3892 INIT_LIST_HEAD(&ei->i_orphan);
3894 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3895 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3896 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3897 EXT4_INODE_SIZE(inode->i_sb)) {
3902 if (ei->i_extra_isize == 0) {
3903 /* The extra space is currently unused. Use it. */
3904 ei->i_extra_isize = sizeof(struct ext4_inode) -
3905 EXT4_GOOD_OLD_INODE_SIZE;
3907 __le32 *magic = (void *)raw_inode +
3908 EXT4_GOOD_OLD_INODE_SIZE +
3910 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
3911 ei->i_state |= EXT4_STATE_XATTR;
3914 ei->i_extra_isize = 0;
3916 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
3917 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
3918 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
3919 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
3921 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
3922 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3923 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
3925 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
3928 if (S_ISREG(inode->i_mode)) {
3929 inode->i_op = &ext4_file_inode_operations;
3930 inode->i_fop = &ext4_file_operations;
3931 ext4_set_aops(inode);
3932 } else if (S_ISDIR(inode->i_mode)) {
3933 inode->i_op = &ext4_dir_inode_operations;
3934 inode->i_fop = &ext4_dir_operations;
3935 } else if (S_ISLNK(inode->i_mode)) {
3936 if (ext4_inode_is_fast_symlink(inode))
3937 inode->i_op = &ext4_fast_symlink_inode_operations;
3939 inode->i_op = &ext4_symlink_inode_operations;
3940 ext4_set_aops(inode);
3943 inode->i_op = &ext4_special_inode_operations;
3944 if (raw_inode->i_block[0])
3945 init_special_inode(inode, inode->i_mode,
3946 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
3948 init_special_inode(inode, inode->i_mode,
3949 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
3952 ext4_set_inode_flags(inode);
3953 unlock_new_inode(inode);
3958 return ERR_PTR(ret);
3961 static int ext4_inode_blocks_set(handle_t *handle,
3962 struct ext4_inode *raw_inode,
3963 struct ext4_inode_info *ei)
3965 struct inode *inode = &(ei->vfs_inode);
3966 u64 i_blocks = inode->i_blocks;
3967 struct super_block *sb = inode->i_sb;
3970 if (i_blocks <= ~0U) {
3972 * i_blocks can be represnted in a 32 bit variable
3973 * as multiple of 512 bytes
3975 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3976 raw_inode->i_blocks_high = 0;
3977 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
3978 } else if (i_blocks <= 0xffffffffffffULL) {
3980 * i_blocks can be represented in a 48 bit variable
3981 * as multiple of 512 bytes
3983 err = ext4_update_rocompat_feature(handle, sb,
3984 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3987 /* i_block is stored in the split 48 bit fields */
3988 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
3989 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
3990 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
3993 * i_blocks should be represented in a 48 bit variable
3994 * as multiple of file system block size
3996 err = ext4_update_rocompat_feature(handle, sb,
3997 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
4000 ei->i_flags |= EXT4_HUGE_FILE_FL;
4001 /* i_block is stored in file system block size */
4002 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4003 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4004 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4011 * Post the struct inode info into an on-disk inode location in the
4012 * buffer-cache. This gobbles the caller's reference to the
4013 * buffer_head in the inode location struct.
4015 * The caller must have write access to iloc->bh.
4017 static int ext4_do_update_inode(handle_t *handle,
4018 struct inode *inode,
4019 struct ext4_iloc *iloc)
4021 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4022 struct ext4_inode_info *ei = EXT4_I(inode);
4023 struct buffer_head *bh = iloc->bh;
4024 int err = 0, rc, block;
4026 /* For fields not not tracking in the in-memory inode,
4027 * initialise them to zero for new inodes. */
4028 if (ei->i_state & EXT4_STATE_NEW)
4029 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4031 ext4_get_inode_flags(ei);
4032 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4033 if(!(test_opt(inode->i_sb, NO_UID32))) {
4034 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
4035 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
4037 * Fix up interoperability with old kernels. Otherwise, old inodes get
4038 * re-used with the upper 16 bits of the uid/gid intact
4041 raw_inode->i_uid_high =
4042 cpu_to_le16(high_16_bits(inode->i_uid));
4043 raw_inode->i_gid_high =
4044 cpu_to_le16(high_16_bits(inode->i_gid));
4046 raw_inode->i_uid_high = 0;
4047 raw_inode->i_gid_high = 0;
4050 raw_inode->i_uid_low =
4051 cpu_to_le16(fs_high2lowuid(inode->i_uid));
4052 raw_inode->i_gid_low =
4053 cpu_to_le16(fs_high2lowgid(inode->i_gid));
4054 raw_inode->i_uid_high = 0;
4055 raw_inode->i_gid_high = 0;
4057 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4059 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4060 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4061 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4062 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4064 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4066 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4067 /* clear the migrate flag in the raw_inode */
4068 raw_inode->i_flags = cpu_to_le32(ei->i_flags & ~EXT4_EXT_MIGRATE);
4069 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4070 cpu_to_le32(EXT4_OS_HURD))
4071 raw_inode->i_file_acl_high =
4072 cpu_to_le16(ei->i_file_acl >> 32);
4073 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4074 ext4_isize_set(raw_inode, ei->i_disksize);
4075 if (ei->i_disksize > 0x7fffffffULL) {
4076 struct super_block *sb = inode->i_sb;
4077 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4078 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4079 EXT4_SB(sb)->s_es->s_rev_level ==
4080 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4081 /* If this is the first large file
4082 * created, add a flag to the superblock.
4084 err = ext4_journal_get_write_access(handle,
4085 EXT4_SB(sb)->s_sbh);
4088 ext4_update_dynamic_rev(sb);
4089 EXT4_SET_RO_COMPAT_FEATURE(sb,
4090 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4093 err = ext4_journal_dirty_metadata(handle,
4094 EXT4_SB(sb)->s_sbh);
4097 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4098 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4099 if (old_valid_dev(inode->i_rdev)) {
4100 raw_inode->i_block[0] =
4101 cpu_to_le32(old_encode_dev(inode->i_rdev));
4102 raw_inode->i_block[1] = 0;
4104 raw_inode->i_block[0] = 0;
4105 raw_inode->i_block[1] =
4106 cpu_to_le32(new_encode_dev(inode->i_rdev));
4107 raw_inode->i_block[2] = 0;
4109 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
4110 raw_inode->i_block[block] = ei->i_data[block];
4112 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4113 if (ei->i_extra_isize) {
4114 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4115 raw_inode->i_version_hi =
4116 cpu_to_le32(inode->i_version >> 32);
4117 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
4121 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
4122 rc = ext4_journal_dirty_metadata(handle, bh);
4125 ei->i_state &= ~EXT4_STATE_NEW;
4129 ext4_std_error(inode->i_sb, err);
4134 * ext4_write_inode()
4136 * We are called from a few places:
4138 * - Within generic_file_write() for O_SYNC files.
4139 * Here, there will be no transaction running. We wait for any running
4140 * trasnaction to commit.
4142 * - Within sys_sync(), kupdate and such.
4143 * We wait on commit, if tol to.
4145 * - Within prune_icache() (PF_MEMALLOC == true)
4146 * Here we simply return. We can't afford to block kswapd on the
4149 * In all cases it is actually safe for us to return without doing anything,
4150 * because the inode has been copied into a raw inode buffer in
4151 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4154 * Note that we are absolutely dependent upon all inode dirtiers doing the
4155 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4156 * which we are interested.
4158 * It would be a bug for them to not do this. The code:
4160 * mark_inode_dirty(inode)
4162 * inode->i_size = expr;
4164 * is in error because a kswapd-driven write_inode() could occur while
4165 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4166 * will no longer be on the superblock's dirty inode list.
4168 int ext4_write_inode(struct inode *inode, int wait)
4170 if (current->flags & PF_MEMALLOC)
4173 if (ext4_journal_current_handle()) {
4174 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4182 return ext4_force_commit(inode->i_sb);
4188 * Called from notify_change.
4190 * We want to trap VFS attempts to truncate the file as soon as
4191 * possible. In particular, we want to make sure that when the VFS
4192 * shrinks i_size, we put the inode on the orphan list and modify
4193 * i_disksize immediately, so that during the subsequent flushing of
4194 * dirty pages and freeing of disk blocks, we can guarantee that any
4195 * commit will leave the blocks being flushed in an unused state on
4196 * disk. (On recovery, the inode will get truncated and the blocks will
4197 * be freed, so we have a strong guarantee that no future commit will
4198 * leave these blocks visible to the user.)
4200 * Another thing we have to assure is that if we are in ordered mode
4201 * and inode is still attached to the committing transaction, we must
4202 * we start writeout of all the dirty pages which are being truncated.
4203 * This way we are sure that all the data written in the previous
4204 * transaction are already on disk (truncate waits for pages under
4207 * Called with inode->i_mutex down.
4209 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
4211 struct inode *inode = dentry->d_inode;
4213 const unsigned int ia_valid = attr->ia_valid;
4215 error = inode_change_ok(inode, attr);
4219 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
4220 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
4223 /* (user+group)*(old+new) structure, inode write (sb,
4224 * inode block, ? - but truncate inode update has it) */
4225 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
4226 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
4227 if (IS_ERR(handle)) {
4228 error = PTR_ERR(handle);
4231 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
4233 ext4_journal_stop(handle);
4236 /* Update corresponding info in inode so that everything is in
4237 * one transaction */
4238 if (attr->ia_valid & ATTR_UID)
4239 inode->i_uid = attr->ia_uid;
4240 if (attr->ia_valid & ATTR_GID)
4241 inode->i_gid = attr->ia_gid;
4242 error = ext4_mark_inode_dirty(handle, inode);
4243 ext4_journal_stop(handle);
4246 if (attr->ia_valid & ATTR_SIZE) {
4247 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
4248 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4250 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
4257 if (S_ISREG(inode->i_mode) &&
4258 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
4261 handle = ext4_journal_start(inode, 3);
4262 if (IS_ERR(handle)) {
4263 error = PTR_ERR(handle);
4267 error = ext4_orphan_add(handle, inode);
4268 EXT4_I(inode)->i_disksize = attr->ia_size;
4269 rc = ext4_mark_inode_dirty(handle, inode);
4272 ext4_journal_stop(handle);
4274 if (ext4_should_order_data(inode)) {
4275 error = ext4_begin_ordered_truncate(inode,
4278 /* Do as much error cleanup as possible */
4279 handle = ext4_journal_start(inode, 3);
4280 if (IS_ERR(handle)) {
4281 ext4_orphan_del(NULL, inode);
4284 ext4_orphan_del(handle, inode);
4285 ext4_journal_stop(handle);
4291 rc = inode_setattr(inode, attr);
4293 /* If inode_setattr's call to ext4_truncate failed to get a
4294 * transaction handle at all, we need to clean up the in-core
4295 * orphan list manually. */
4297 ext4_orphan_del(NULL, inode);
4299 if (!rc && (ia_valid & ATTR_MODE))
4300 rc = ext4_acl_chmod(inode);
4303 ext4_std_error(inode->i_sb, error);
4309 int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4312 struct inode *inode;
4313 unsigned long delalloc_blocks;
4315 inode = dentry->d_inode;
4316 generic_fillattr(inode, stat);
4319 * We can't update i_blocks if the block allocation is delayed
4320 * otherwise in the case of system crash before the real block
4321 * allocation is done, we will have i_blocks inconsistent with
4322 * on-disk file blocks.
4323 * We always keep i_blocks updated together with real
4324 * allocation. But to not confuse with user, stat
4325 * will return the blocks that include the delayed allocation
4326 * blocks for this file.
4328 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
4329 delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks;
4330 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
4332 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4337 * How many blocks doth make a writepage()?
4339 * With N blocks per page, it may be:
4344 * N+5 bitmap blocks (from the above)
4345 * N+5 group descriptor summary blocks
4348 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
4350 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
4352 * With ordered or writeback data it's the same, less the N data blocks.
4354 * If the inode's direct blocks can hold an integral number of pages then a
4355 * page cannot straddle two indirect blocks, and we can only touch one indirect
4356 * and dindirect block, and the "5" above becomes "3".
4358 * This still overestimates under most circumstances. If we were to pass the
4359 * start and end offsets in here as well we could do block_to_path() on each
4360 * block and work out the exact number of indirects which are touched. Pah.
4363 int ext4_writepage_trans_blocks(struct inode *inode)
4365 int bpp = ext4_journal_blocks_per_page(inode);
4366 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
4369 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
4370 return ext4_ext_writepage_trans_blocks(inode, bpp);
4372 if (ext4_should_journal_data(inode))
4373 ret = 3 * (bpp + indirects) + 2;
4375 ret = 2 * (bpp + indirects) + 2;
4378 /* We know that structure was already allocated during DQUOT_INIT so
4379 * we will be updating only the data blocks + inodes */
4380 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
4387 * The caller must have previously called ext4_reserve_inode_write().
4388 * Give this, we know that the caller already has write access to iloc->bh.
4390 int ext4_mark_iloc_dirty(handle_t *handle,
4391 struct inode *inode, struct ext4_iloc *iloc)
4395 if (test_opt(inode->i_sb, I_VERSION))
4396 inode_inc_iversion(inode);
4398 /* the do_update_inode consumes one bh->b_count */
4401 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4402 err = ext4_do_update_inode(handle, inode, iloc);
4408 * On success, We end up with an outstanding reference count against
4409 * iloc->bh. This _must_ be cleaned up later.
4413 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4414 struct ext4_iloc *iloc)
4418 err = ext4_get_inode_loc(inode, iloc);
4420 BUFFER_TRACE(iloc->bh, "get_write_access");
4421 err = ext4_journal_get_write_access(handle, iloc->bh);
4428 ext4_std_error(inode->i_sb, err);
4433 * Expand an inode by new_extra_isize bytes.
4434 * Returns 0 on success or negative error number on failure.
4436 static int ext4_expand_extra_isize(struct inode *inode,
4437 unsigned int new_extra_isize,
4438 struct ext4_iloc iloc,
4441 struct ext4_inode *raw_inode;
4442 struct ext4_xattr_ibody_header *header;
4443 struct ext4_xattr_entry *entry;
4445 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4448 raw_inode = ext4_raw_inode(&iloc);
4450 header = IHDR(inode, raw_inode);
4451 entry = IFIRST(header);
4453 /* No extended attributes present */
4454 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
4455 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
4456 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4458 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4462 /* try to expand with EAs present */
4463 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4468 * What we do here is to mark the in-core inode as clean with respect to inode
4469 * dirtiness (it may still be data-dirty).
4470 * This means that the in-core inode may be reaped by prune_icache
4471 * without having to perform any I/O. This is a very good thing,
4472 * because *any* task may call prune_icache - even ones which
4473 * have a transaction open against a different journal.
4475 * Is this cheating? Not really. Sure, we haven't written the
4476 * inode out, but prune_icache isn't a user-visible syncing function.
4477 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4478 * we start and wait on commits.
4480 * Is this efficient/effective? Well, we're being nice to the system
4481 * by cleaning up our inodes proactively so they can be reaped
4482 * without I/O. But we are potentially leaving up to five seconds'
4483 * worth of inodes floating about which prune_icache wants us to
4484 * write out. One way to fix that would be to get prune_icache()
4485 * to do a write_super() to free up some memory. It has the desired
4488 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
4490 struct ext4_iloc iloc;
4491 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4492 static unsigned int mnt_count;
4496 err = ext4_reserve_inode_write(handle, inode, &iloc);
4497 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
4498 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
4500 * We need extra buffer credits since we may write into EA block
4501 * with this same handle. If journal_extend fails, then it will
4502 * only result in a minor loss of functionality for that inode.
4503 * If this is felt to be critical, then e2fsck should be run to
4504 * force a large enough s_min_extra_isize.
4506 if ((jbd2_journal_extend(handle,
4507 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4508 ret = ext4_expand_extra_isize(inode,
4509 sbi->s_want_extra_isize,
4512 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
4514 le16_to_cpu(sbi->s_es->s_mnt_count)) {
4515 ext4_warning(inode->i_sb, __func__,
4516 "Unable to expand inode %lu. Delete"
4517 " some EAs or run e2fsck.",
4520 le16_to_cpu(sbi->s_es->s_mnt_count);
4526 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
4531 * ext4_dirty_inode() is called from __mark_inode_dirty()
4533 * We're really interested in the case where a file is being extended.
4534 * i_size has been changed by generic_commit_write() and we thus need
4535 * to include the updated inode in the current transaction.
4537 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
4538 * are allocated to the file.
4540 * If the inode is marked synchronous, we don't honour that here - doing
4541 * so would cause a commit on atime updates, which we don't bother doing.
4542 * We handle synchronous inodes at the highest possible level.
4544 void ext4_dirty_inode(struct inode *inode)
4546 handle_t *current_handle = ext4_journal_current_handle();
4549 handle = ext4_journal_start(inode, 2);
4552 if (current_handle &&
4553 current_handle->h_transaction != handle->h_transaction) {
4554 /* This task has a transaction open against a different fs */
4555 printk(KERN_EMERG "%s: transactions do not match!\n",
4558 jbd_debug(5, "marking dirty. outer handle=%p\n",
4560 ext4_mark_inode_dirty(handle, inode);
4562 ext4_journal_stop(handle);
4569 * Bind an inode's backing buffer_head into this transaction, to prevent
4570 * it from being flushed to disk early. Unlike
4571 * ext4_reserve_inode_write, this leaves behind no bh reference and
4572 * returns no iloc structure, so the caller needs to repeat the iloc
4573 * lookup to mark the inode dirty later.
4575 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
4577 struct ext4_iloc iloc;
4581 err = ext4_get_inode_loc(inode, &iloc);
4583 BUFFER_TRACE(iloc.bh, "get_write_access");
4584 err = jbd2_journal_get_write_access(handle, iloc.bh);
4586 err = ext4_journal_dirty_metadata(handle,
4591 ext4_std_error(inode->i_sb, err);
4596 int ext4_change_inode_journal_flag(struct inode *inode, int val)
4603 * We have to be very careful here: changing a data block's
4604 * journaling status dynamically is dangerous. If we write a
4605 * data block to the journal, change the status and then delete
4606 * that block, we risk forgetting to revoke the old log record
4607 * from the journal and so a subsequent replay can corrupt data.
4608 * So, first we make sure that the journal is empty and that
4609 * nobody is changing anything.
4612 journal = EXT4_JOURNAL(inode);
4613 if (is_journal_aborted(journal))
4616 jbd2_journal_lock_updates(journal);
4617 jbd2_journal_flush(journal);
4620 * OK, there are no updates running now, and all cached data is
4621 * synced to disk. We are now in a completely consistent state
4622 * which doesn't have anything in the journal, and we know that
4623 * no filesystem updates are running, so it is safe to modify
4624 * the inode's in-core data-journaling state flag now.
4628 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
4630 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
4631 ext4_set_aops(inode);
4633 jbd2_journal_unlock_updates(journal);
4635 /* Finally we can mark the inode as dirty. */
4637 handle = ext4_journal_start(inode, 1);
4639 return PTR_ERR(handle);
4641 err = ext4_mark_inode_dirty(handle, inode);
4643 ext4_journal_stop(handle);
4644 ext4_std_error(inode->i_sb, err);
4649 static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4651 return !buffer_mapped(bh);
4654 int ext4_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4659 struct file *file = vma->vm_file;
4660 struct inode *inode = file->f_path.dentry->d_inode;
4661 struct address_space *mapping = inode->i_mapping;
4664 * Get i_alloc_sem to stop truncates messing with the inode. We cannot
4665 * get i_mutex because we are already holding mmap_sem.
4667 down_read(&inode->i_alloc_sem);
4668 size = i_size_read(inode);
4669 if (page->mapping != mapping || size <= page_offset(page)
4670 || !PageUptodate(page)) {
4671 /* page got truncated from under us? */
4675 if (PageMappedToDisk(page))
4678 if (page->index == size >> PAGE_CACHE_SHIFT)
4679 len = size & ~PAGE_CACHE_MASK;
4681 len = PAGE_CACHE_SIZE;
4683 if (page_has_buffers(page)) {
4684 /* return if we have all the buffers mapped */
4685 if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL,
4690 * OK, we need to fill the hole... Do write_begin write_end
4691 * to do block allocation/reservation.We are not holding
4692 * inode.i__mutex here. That allow * parallel write_begin,
4693 * write_end call. lock_page prevent this from happening
4694 * on the same page though
4696 ret = mapping->a_ops->write_begin(file, mapping, page_offset(page),
4697 len, AOP_FLAG_UNINTERRUPTIBLE, &page, NULL);
4700 ret = mapping->a_ops->write_end(file, mapping, page_offset(page),
4701 len, len, page, NULL);
4706 up_read(&inode->i_alloc_sem);