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/ext4_jbd2.h>
29 #include <linux/jbd2.h>
30 #include <linux/highuid.h>
31 #include <linux/pagemap.h>
32 #include <linux/quotaops.h>
33 #include <linux/string.h>
34 #include <linux/buffer_head.h>
35 #include <linux/writeback.h>
36 #include <linux/mpage.h>
37 #include <linux/uio.h>
38 #include <linux/bio.h>
43 * Test whether an inode is a fast symlink.
45 static int ext4_inode_is_fast_symlink(struct inode *inode)
47 int ea_blocks = EXT4_I(inode)->i_file_acl ?
48 (inode->i_sb->s_blocksize >> 9) : 0;
50 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
54 * The ext4 forget function must perform a revoke if we are freeing data
55 * which has been journaled. Metadata (eg. indirect blocks) must be
56 * revoked in all cases.
58 * "bh" may be NULL: a metadata block may have been freed from memory
59 * but there may still be a record of it in the journal, and that record
60 * still needs to be revoked.
62 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
63 struct buffer_head *bh, ext4_fsblk_t blocknr)
69 BUFFER_TRACE(bh, "enter");
71 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
73 bh, is_metadata, inode->i_mode,
74 test_opt(inode->i_sb, DATA_FLAGS));
76 /* Never use the revoke function if we are doing full data
77 * journaling: there is no need to, and a V1 superblock won't
78 * support it. Otherwise, only skip the revoke on un-journaled
81 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA ||
82 (!is_metadata && !ext4_should_journal_data(inode))) {
84 BUFFER_TRACE(bh, "call jbd2_journal_forget");
85 return ext4_journal_forget(handle, bh);
91 * data!=journal && (is_metadata || should_journal_data(inode))
93 BUFFER_TRACE(bh, "call ext4_journal_revoke");
94 err = ext4_journal_revoke(handle, blocknr, bh);
96 ext4_abort(inode->i_sb, __FUNCTION__,
97 "error %d when attempting revoke", err);
98 BUFFER_TRACE(bh, "exit");
103 * Work out how many blocks we need to proceed with the next chunk of a
104 * truncate transaction.
106 static unsigned long blocks_for_truncate(struct inode *inode)
110 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
112 /* Give ourselves just enough room to cope with inodes in which
113 * i_blocks is corrupt: we've seen disk corruptions in the past
114 * which resulted in random data in an inode which looked enough
115 * like a regular file for ext4 to try to delete it. Things
116 * will go a bit crazy if that happens, but at least we should
117 * try not to panic the whole kernel. */
121 /* But we need to bound the transaction so we don't overflow the
123 if (needed > EXT4_MAX_TRANS_DATA)
124 needed = EXT4_MAX_TRANS_DATA;
126 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
130 * Truncate transactions can be complex and absolutely huge. So we need to
131 * be able to restart the transaction at a conventient checkpoint to make
132 * sure we don't overflow the journal.
134 * start_transaction gets us a new handle for a truncate transaction,
135 * and extend_transaction tries to extend the existing one a bit. If
136 * extend fails, we need to propagate the failure up and restart the
137 * transaction in the top-level truncate loop. --sct
139 static handle_t *start_transaction(struct inode *inode)
143 result = ext4_journal_start(inode, blocks_for_truncate(inode));
147 ext4_std_error(inode->i_sb, PTR_ERR(result));
152 * Try to extend this transaction for the purposes of truncation.
154 * Returns 0 if we managed to create more room. If we can't create more
155 * room, and the transaction must be restarted we return 1.
157 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
159 if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
161 if (!ext4_journal_extend(handle, blocks_for_truncate(inode)))
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
171 static int ext4_journal_test_restart(handle_t *handle, struct inode *inode)
173 jbd_debug(2, "restarting handle %p\n", handle);
174 return ext4_journal_restart(handle, blocks_for_truncate(inode));
178 * Called at the last iput() if i_nlink is zero.
180 void ext4_delete_inode (struct inode * inode)
184 truncate_inode_pages(&inode->i_data, 0);
186 if (is_bad_inode(inode))
189 handle = start_transaction(inode);
190 if (IS_ERR(handle)) {
192 * If we're going to skip the normal cleanup, we still need to
193 * make sure that the in-core orphan linked list is properly
196 ext4_orphan_del(NULL, inode);
204 ext4_truncate(inode);
206 * Kill off the orphan record which ext4_truncate created.
207 * AKPM: I think this can be inside the above `if'.
208 * Note that ext4_orphan_del() has to be able to cope with the
209 * deletion of a non-existent orphan - this is because we don't
210 * know if ext4_truncate() actually created an orphan record.
211 * (Well, we could do this if we need to, but heck - it works)
213 ext4_orphan_del(handle, inode);
214 EXT4_I(inode)->i_dtime = get_seconds();
217 * One subtle ordering requirement: if anything has gone wrong
218 * (transaction abort, IO errors, whatever), then we can still
219 * do these next steps (the fs will already have been marked as
220 * having errors), but we can't free the inode if the mark_dirty
223 if (ext4_mark_inode_dirty(handle, inode))
224 /* If that failed, just do the required in-core inode clear. */
227 ext4_free_inode(handle, inode);
228 ext4_journal_stop(handle);
231 clear_inode(inode); /* We must guarantee clearing of inode... */
237 struct buffer_head *bh;
240 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
242 p->key = *(p->p = v);
246 static int verify_chain(Indirect *from, Indirect *to)
248 while (from <= to && from->key == *from->p)
254 * ext4_block_to_path - parse the block number into array of offsets
255 * @inode: inode in question (we are only interested in its superblock)
256 * @i_block: block number to be parsed
257 * @offsets: array to store the offsets in
258 * @boundary: set this non-zero if the referred-to block is likely to be
259 * followed (on disk) by an indirect block.
261 * To store the locations of file's data ext4 uses a data structure common
262 * for UNIX filesystems - tree of pointers anchored in the inode, with
263 * data blocks at leaves and indirect blocks in intermediate nodes.
264 * This function translates the block number into path in that tree -
265 * return value is the path length and @offsets[n] is the offset of
266 * pointer to (n+1)th node in the nth one. If @block is out of range
267 * (negative or too large) warning is printed and zero returned.
269 * Note: function doesn't find node addresses, so no IO is needed. All
270 * we need to know is the capacity of indirect blocks (taken from the
275 * Portability note: the last comparison (check that we fit into triple
276 * indirect block) is spelled differently, because otherwise on an
277 * architecture with 32-bit longs and 8Kb pages we might get into trouble
278 * if our filesystem had 8Kb blocks. We might use long long, but that would
279 * kill us on x86. Oh, well, at least the sign propagation does not matter -
280 * i_block would have to be negative in the very beginning, so we would not
284 static int ext4_block_to_path(struct inode *inode,
286 ext4_lblk_t offsets[4], int *boundary)
288 int ptrs = EXT4_ADDR_PER_BLOCK(inode->i_sb);
289 int ptrs_bits = EXT4_ADDR_PER_BLOCK_BITS(inode->i_sb);
290 const long direct_blocks = EXT4_NDIR_BLOCKS,
291 indirect_blocks = ptrs,
292 double_blocks = (1 << (ptrs_bits * 2));
297 ext4_warning (inode->i_sb, "ext4_block_to_path", "block < 0");
298 } else if (i_block < direct_blocks) {
299 offsets[n++] = i_block;
300 final = direct_blocks;
301 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
302 offsets[n++] = EXT4_IND_BLOCK;
303 offsets[n++] = i_block;
305 } else if ((i_block -= indirect_blocks) < double_blocks) {
306 offsets[n++] = EXT4_DIND_BLOCK;
307 offsets[n++] = i_block >> ptrs_bits;
308 offsets[n++] = i_block & (ptrs - 1);
310 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
311 offsets[n++] = EXT4_TIND_BLOCK;
312 offsets[n++] = i_block >> (ptrs_bits * 2);
313 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
314 offsets[n++] = i_block & (ptrs - 1);
317 ext4_warning(inode->i_sb, "ext4_block_to_path",
319 i_block + direct_blocks +
320 indirect_blocks + double_blocks);
323 *boundary = final - 1 - (i_block & (ptrs - 1));
328 * ext4_get_branch - read the chain of indirect blocks leading to data
329 * @inode: inode in question
330 * @depth: depth of the chain (1 - direct pointer, etc.)
331 * @offsets: offsets of pointers in inode/indirect blocks
332 * @chain: place to store the result
333 * @err: here we store the error value
335 * Function fills the array of triples <key, p, bh> and returns %NULL
336 * if everything went OK or the pointer to the last filled triple
337 * (incomplete one) otherwise. Upon the return chain[i].key contains
338 * the number of (i+1)-th block in the chain (as it is stored in memory,
339 * i.e. little-endian 32-bit), chain[i].p contains the address of that
340 * number (it points into struct inode for i==0 and into the bh->b_data
341 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
342 * block for i>0 and NULL for i==0. In other words, it holds the block
343 * numbers of the chain, addresses they were taken from (and where we can
344 * verify that chain did not change) and buffer_heads hosting these
347 * Function stops when it stumbles upon zero pointer (absent block)
348 * (pointer to last triple returned, *@err == 0)
349 * or when it gets an IO error reading an indirect block
350 * (ditto, *@err == -EIO)
351 * or when it notices that chain had been changed while it was reading
352 * (ditto, *@err == -EAGAIN)
353 * or when it reads all @depth-1 indirect blocks successfully and finds
354 * the whole chain, all way to the data (returns %NULL, *err == 0).
356 static Indirect *ext4_get_branch(struct inode *inode, int depth,
357 ext4_lblk_t *offsets,
358 Indirect chain[4], int *err)
360 struct super_block *sb = inode->i_sb;
362 struct buffer_head *bh;
365 /* i_data is not going away, no lock needed */
366 add_chain (chain, NULL, EXT4_I(inode)->i_data + *offsets);
370 bh = sb_bread(sb, le32_to_cpu(p->key));
373 /* Reader: pointers */
374 if (!verify_chain(chain, p))
376 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
394 * ext4_find_near - find a place for allocation with sufficient locality
396 * @ind: descriptor of indirect block.
398 * This function returns the prefered place for block allocation.
399 * It is used when heuristic for sequential allocation fails.
401 * + if there is a block to the left of our position - allocate near it.
402 * + if pointer will live in indirect block - allocate near that block.
403 * + if pointer will live in inode - allocate in the same
406 * In the latter case we colour the starting block by the callers PID to
407 * prevent it from clashing with concurrent allocations for a different inode
408 * in the same block group. The PID is used here so that functionally related
409 * files will be close-by on-disk.
411 * Caller must make sure that @ind is valid and will stay that way.
413 static ext4_fsblk_t ext4_find_near(struct inode *inode, Indirect *ind)
415 struct ext4_inode_info *ei = EXT4_I(inode);
416 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
418 ext4_fsblk_t bg_start;
419 ext4_grpblk_t colour;
421 /* Try to find previous block */
422 for (p = ind->p - 1; p >= start; p--) {
424 return le32_to_cpu(*p);
427 /* No such thing, so let's try location of indirect block */
429 return ind->bh->b_blocknr;
432 * It is going to be referred to from the inode itself? OK, just put it
433 * into the same cylinder group then.
435 bg_start = ext4_group_first_block_no(inode->i_sb, ei->i_block_group);
436 colour = (current->pid % 16) *
437 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
438 return bg_start + colour;
442 * ext4_find_goal - find a prefered place for allocation.
444 * @block: block we want
445 * @chain: chain of indirect blocks
446 * @partial: pointer to the last triple within a chain
447 * @goal: place to store the result.
449 * Normally this function find the prefered place for block allocation,
450 * stores it in *@goal and returns zero.
453 static ext4_fsblk_t ext4_find_goal(struct inode *inode, ext4_lblk_t block,
454 Indirect chain[4], Indirect *partial)
456 struct ext4_block_alloc_info *block_i;
458 block_i = EXT4_I(inode)->i_block_alloc_info;
461 * try the heuristic for sequential allocation,
462 * failing that at least try to get decent locality.
464 if (block_i && (block == block_i->last_alloc_logical_block + 1)
465 && (block_i->last_alloc_physical_block != 0)) {
466 return block_i->last_alloc_physical_block + 1;
469 return ext4_find_near(inode, partial);
473 * ext4_blks_to_allocate: Look up the block map and count the number
474 * of direct blocks need to be allocated for the given branch.
476 * @branch: chain of indirect blocks
477 * @k: number of blocks need for indirect blocks
478 * @blks: number of data blocks to be mapped.
479 * @blocks_to_boundary: the offset in the indirect block
481 * return the total number of blocks to be allocate, including the
482 * direct and indirect blocks.
484 static int ext4_blks_to_allocate(Indirect *branch, int k, unsigned long blks,
485 int blocks_to_boundary)
487 unsigned long count = 0;
490 * Simple case, [t,d]Indirect block(s) has not allocated yet
491 * then it's clear blocks on that path have not allocated
494 /* right now we don't handle cross boundary allocation */
495 if (blks < blocks_to_boundary + 1)
498 count += blocks_to_boundary + 1;
503 while (count < blks && count <= blocks_to_boundary &&
504 le32_to_cpu(*(branch[0].p + count)) == 0) {
511 * ext4_alloc_blocks: multiple allocate blocks needed for a branch
512 * @indirect_blks: the number of blocks need to allocate for indirect
515 * @new_blocks: on return it will store the new block numbers for
516 * the indirect blocks(if needed) and the first direct block,
517 * @blks: on return it will store the total number of allocated
520 static int ext4_alloc_blocks(handle_t *handle, struct inode *inode,
521 ext4_fsblk_t goal, int indirect_blks, int blks,
522 ext4_fsblk_t new_blocks[4], int *err)
525 unsigned long count = 0;
527 ext4_fsblk_t current_block = 0;
531 * Here we try to allocate the requested multiple blocks at once,
532 * on a best-effort basis.
533 * To build a branch, we should allocate blocks for
534 * the indirect blocks(if not allocated yet), and at least
535 * the first direct block of this branch. That's the
536 * minimum number of blocks need to allocate(required)
538 target = blks + indirect_blks;
542 /* allocating blocks for indirect blocks and direct blocks */
543 current_block = ext4_new_blocks(handle,inode,goal,&count,err);
548 /* allocate blocks for indirect blocks */
549 while (index < indirect_blks && count) {
550 new_blocks[index++] = current_block++;
558 /* save the new block number for the first direct block */
559 new_blocks[index] = current_block;
561 /* total number of blocks allocated for direct blocks */
566 for (i = 0; i <index; i++)
567 ext4_free_blocks(handle, inode, new_blocks[i], 1);
572 * ext4_alloc_branch - allocate and set up a chain of blocks.
574 * @indirect_blks: number of allocated indirect blocks
575 * @blks: number of allocated direct blocks
576 * @offsets: offsets (in the blocks) to store the pointers to next.
577 * @branch: place to store the chain in.
579 * This function allocates blocks, zeroes out all but the last one,
580 * links them into chain and (if we are synchronous) writes them to disk.
581 * In other words, it prepares a branch that can be spliced onto the
582 * inode. It stores the information about that chain in the branch[], in
583 * the same format as ext4_get_branch() would do. We are calling it after
584 * we had read the existing part of chain and partial points to the last
585 * triple of that (one with zero ->key). Upon the exit we have the same
586 * picture as after the successful ext4_get_block(), except that in one
587 * place chain is disconnected - *branch->p is still zero (we did not
588 * set the last link), but branch->key contains the number that should
589 * be placed into *branch->p to fill that gap.
591 * If allocation fails we free all blocks we've allocated (and forget
592 * their buffer_heads) and return the error value the from failed
593 * ext4_alloc_block() (normally -ENOSPC). Otherwise we set the chain
594 * as described above and return 0.
596 static int ext4_alloc_branch(handle_t *handle, struct inode *inode,
597 int indirect_blks, int *blks, ext4_fsblk_t goal,
598 ext4_lblk_t *offsets, Indirect *branch)
600 int blocksize = inode->i_sb->s_blocksize;
603 struct buffer_head *bh;
605 ext4_fsblk_t new_blocks[4];
606 ext4_fsblk_t current_block;
608 num = ext4_alloc_blocks(handle, inode, goal, indirect_blks,
609 *blks, new_blocks, &err);
613 branch[0].key = cpu_to_le32(new_blocks[0]);
615 * metadata blocks and data blocks are allocated.
617 for (n = 1; n <= indirect_blks; n++) {
619 * Get buffer_head for parent block, zero it out
620 * and set the pointer to new one, then send
623 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
626 BUFFER_TRACE(bh, "call get_create_access");
627 err = ext4_journal_get_create_access(handle, bh);
634 memset(bh->b_data, 0, blocksize);
635 branch[n].p = (__le32 *) bh->b_data + offsets[n];
636 branch[n].key = cpu_to_le32(new_blocks[n]);
637 *branch[n].p = branch[n].key;
638 if ( n == indirect_blks) {
639 current_block = new_blocks[n];
641 * End of chain, update the last new metablock of
642 * the chain to point to the new allocated
643 * data blocks numbers
645 for (i=1; i < num; i++)
646 *(branch[n].p + i) = cpu_to_le32(++current_block);
648 BUFFER_TRACE(bh, "marking uptodate");
649 set_buffer_uptodate(bh);
652 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
653 err = ext4_journal_dirty_metadata(handle, bh);
660 /* Allocation failed, free what we already allocated */
661 for (i = 1; i <= n ; i++) {
662 BUFFER_TRACE(branch[i].bh, "call jbd2_journal_forget");
663 ext4_journal_forget(handle, branch[i].bh);
665 for (i = 0; i <indirect_blks; i++)
666 ext4_free_blocks(handle, inode, new_blocks[i], 1);
668 ext4_free_blocks(handle, inode, new_blocks[i], num);
674 * ext4_splice_branch - splice the allocated branch onto inode.
676 * @block: (logical) number of block we are adding
677 * @chain: chain of indirect blocks (with a missing link - see
679 * @where: location of missing link
680 * @num: number of indirect blocks we are adding
681 * @blks: number of direct blocks we are adding
683 * This function fills the missing link and does all housekeeping needed in
684 * inode (->i_blocks, etc.). In case of success we end up with the full
685 * chain to new block and return 0.
687 static int ext4_splice_branch(handle_t *handle, struct inode *inode,
688 ext4_lblk_t block, Indirect *where, int num, int blks)
692 struct ext4_block_alloc_info *block_i;
693 ext4_fsblk_t current_block;
695 block_i = EXT4_I(inode)->i_block_alloc_info;
697 * If we're splicing into a [td]indirect block (as opposed to the
698 * inode) then we need to get write access to the [td]indirect block
702 BUFFER_TRACE(where->bh, "get_write_access");
703 err = ext4_journal_get_write_access(handle, where->bh);
709 *where->p = where->key;
712 * Update the host buffer_head or inode to point to more just allocated
713 * direct blocks blocks
715 if (num == 0 && blks > 1) {
716 current_block = le32_to_cpu(where->key) + 1;
717 for (i = 1; i < blks; i++)
718 *(where->p + i ) = cpu_to_le32(current_block++);
722 * update the most recently allocated logical & physical block
723 * in i_block_alloc_info, to assist find the proper goal block for next
727 block_i->last_alloc_logical_block = block + blks - 1;
728 block_i->last_alloc_physical_block =
729 le32_to_cpu(where[num].key) + blks - 1;
732 /* We are done with atomic stuff, now do the rest of housekeeping */
734 inode->i_ctime = ext4_current_time(inode);
735 ext4_mark_inode_dirty(handle, inode);
737 /* had we spliced it onto indirect block? */
740 * If we spliced it onto an indirect block, we haven't
741 * altered the inode. Note however that if it is being spliced
742 * onto an indirect block at the very end of the file (the
743 * file is growing) then we *will* alter the inode to reflect
744 * the new i_size. But that is not done here - it is done in
745 * generic_commit_write->__mark_inode_dirty->ext4_dirty_inode.
747 jbd_debug(5, "splicing indirect only\n");
748 BUFFER_TRACE(where->bh, "call ext4_journal_dirty_metadata");
749 err = ext4_journal_dirty_metadata(handle, where->bh);
754 * OK, we spliced it into the inode itself on a direct block.
755 * Inode was dirtied above.
757 jbd_debug(5, "splicing direct\n");
762 for (i = 1; i <= num; i++) {
763 BUFFER_TRACE(where[i].bh, "call jbd2_journal_forget");
764 ext4_journal_forget(handle, where[i].bh);
765 ext4_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1);
767 ext4_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks);
773 * Allocation strategy is simple: if we have to allocate something, we will
774 * have to go the whole way to leaf. So let's do it before attaching anything
775 * to tree, set linkage between the newborn blocks, write them if sync is
776 * required, recheck the path, free and repeat if check fails, otherwise
777 * set the last missing link (that will protect us from any truncate-generated
778 * removals - all blocks on the path are immune now) and possibly force the
779 * write on the parent block.
780 * That has a nice additional property: no special recovery from the failed
781 * allocations is needed - we simply release blocks and do not touch anything
782 * reachable from inode.
784 * `handle' can be NULL if create == 0.
786 * The BKL may not be held on entry here. Be sure to take it early.
787 * return > 0, # of blocks mapped or allocated.
788 * return = 0, if plain lookup failed.
789 * return < 0, error case.
791 int ext4_get_blocks_handle(handle_t *handle, struct inode *inode,
792 ext4_lblk_t iblock, unsigned long maxblocks,
793 struct buffer_head *bh_result,
794 int create, int extend_disksize)
797 ext4_lblk_t offsets[4];
802 int blocks_to_boundary = 0;
804 struct ext4_inode_info *ei = EXT4_I(inode);
806 ext4_fsblk_t first_block = 0;
809 J_ASSERT(!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL));
810 J_ASSERT(handle != NULL || create == 0);
811 depth = ext4_block_to_path(inode, iblock, offsets,
812 &blocks_to_boundary);
817 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
819 /* Simplest case - block found, no allocation needed */
821 first_block = le32_to_cpu(chain[depth - 1].key);
822 clear_buffer_new(bh_result);
825 while (count < maxblocks && count <= blocks_to_boundary) {
828 if (!verify_chain(chain, partial)) {
830 * Indirect block might be removed by
831 * truncate while we were reading it.
832 * Handling of that case: forget what we've
833 * got now. Flag the err as EAGAIN, so it
840 blk = le32_to_cpu(*(chain[depth-1].p + count));
842 if (blk == first_block + count)
851 /* Next simple case - plain lookup or failed read of indirect block */
852 if (!create || err == -EIO)
855 mutex_lock(&ei->truncate_mutex);
858 * If the indirect block is missing while we are reading
859 * the chain(ext4_get_branch() returns -EAGAIN err), or
860 * if the chain has been changed after we grab the semaphore,
861 * (either because another process truncated this branch, or
862 * another get_block allocated this branch) re-grab the chain to see if
863 * the request block has been allocated or not.
865 * Since we already block the truncate/other get_block
866 * at this point, we will have the current copy of the chain when we
867 * splice the branch into the tree.
869 if (err == -EAGAIN || !verify_chain(chain, partial)) {
870 while (partial > chain) {
874 partial = ext4_get_branch(inode, depth, offsets, chain, &err);
877 mutex_unlock(&ei->truncate_mutex);
880 clear_buffer_new(bh_result);
886 * Okay, we need to do block allocation. Lazily initialize the block
887 * allocation info here if necessary
889 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
890 ext4_init_block_alloc_info(inode);
892 goal = ext4_find_goal(inode, iblock, chain, partial);
894 /* the number of blocks need to allocate for [d,t]indirect blocks */
895 indirect_blks = (chain + depth) - partial - 1;
898 * Next look up the indirect map to count the totoal number of
899 * direct blocks to allocate for this branch.
901 count = ext4_blks_to_allocate(partial, indirect_blks,
902 maxblocks, blocks_to_boundary);
904 * Block out ext4_truncate while we alter the tree
906 err = ext4_alloc_branch(handle, inode, indirect_blks, &count, goal,
907 offsets + (partial - chain), partial);
910 * The ext4_splice_branch call will free and forget any buffers
911 * on the new chain if there is a failure, but that risks using
912 * up transaction credits, especially for bitmaps where the
913 * credits cannot be returned. Can we handle this somehow? We
914 * may need to return -EAGAIN upwards in the worst case. --sct
917 err = ext4_splice_branch(handle, inode, iblock,
918 partial, indirect_blks, count);
920 * i_disksize growing is protected by truncate_mutex. Don't forget to
921 * protect it if you're about to implement concurrent
922 * ext4_get_block() -bzzz
924 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
925 ei->i_disksize = inode->i_size;
926 mutex_unlock(&ei->truncate_mutex);
930 set_buffer_new(bh_result);
932 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
933 if (count > blocks_to_boundary)
934 set_buffer_boundary(bh_result);
936 /* Clean up and exit */
937 partial = chain + depth - 1; /* the whole chain */
939 while (partial > chain) {
940 BUFFER_TRACE(partial->bh, "call brelse");
944 BUFFER_TRACE(bh_result, "returned");
949 #define DIO_CREDITS (EXT4_RESERVE_TRANS_BLOCKS + 32)
951 static int ext4_get_block(struct inode *inode, sector_t iblock,
952 struct buffer_head *bh_result, int create)
954 handle_t *handle = ext4_journal_current_handle();
956 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
959 goto get_block; /* A read */
962 goto get_block; /* A single block get */
964 if (handle->h_transaction->t_state == T_LOCKED) {
966 * Huge direct-io writes can hold off commits for long
967 * periods of time. Let this commit run.
969 ext4_journal_stop(handle);
970 handle = ext4_journal_start(inode, DIO_CREDITS);
972 ret = PTR_ERR(handle);
976 if (handle->h_buffer_credits <= EXT4_RESERVE_TRANS_BLOCKS) {
978 * Getting low on buffer credits...
980 ret = ext4_journal_extend(handle, DIO_CREDITS);
983 * Couldn't extend the transaction. Start a new one.
985 ret = ext4_journal_restart(handle, DIO_CREDITS);
991 ret = ext4_get_blocks_wrap(handle, inode, iblock,
992 max_blocks, bh_result, create, 0);
994 bh_result->b_size = (ret << inode->i_blkbits);
1002 * `handle' can be NULL if create is zero
1004 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
1005 ext4_lblk_t block, int create, int *errp)
1007 struct buffer_head dummy;
1010 J_ASSERT(handle != NULL || create == 0);
1013 dummy.b_blocknr = -1000;
1014 buffer_trace_init(&dummy.b_history);
1015 err = ext4_get_blocks_wrap(handle, inode, block, 1,
1018 * ext4_get_blocks_handle() returns number of blocks
1019 * mapped. 0 in case of a HOLE.
1027 if (!err && buffer_mapped(&dummy)) {
1028 struct buffer_head *bh;
1029 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
1034 if (buffer_new(&dummy)) {
1035 J_ASSERT(create != 0);
1036 J_ASSERT(handle != NULL);
1039 * Now that we do not always journal data, we should
1040 * keep in mind whether this should always journal the
1041 * new buffer as metadata. For now, regular file
1042 * writes use ext4_get_block instead, so it's not a
1046 BUFFER_TRACE(bh, "call get_create_access");
1047 fatal = ext4_journal_get_create_access(handle, bh);
1048 if (!fatal && !buffer_uptodate(bh)) {
1049 memset(bh->b_data,0,inode->i_sb->s_blocksize);
1050 set_buffer_uptodate(bh);
1053 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1054 err = ext4_journal_dirty_metadata(handle, bh);
1058 BUFFER_TRACE(bh, "not a new buffer");
1071 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1072 ext4_lblk_t block, int create, int *err)
1074 struct buffer_head * bh;
1076 bh = ext4_getblk(handle, inode, block, create, err);
1079 if (buffer_uptodate(bh))
1081 ll_rw_block(READ_META, 1, &bh);
1083 if (buffer_uptodate(bh))
1090 static int walk_page_buffers( handle_t *handle,
1091 struct buffer_head *head,
1095 int (*fn)( handle_t *handle,
1096 struct buffer_head *bh))
1098 struct buffer_head *bh;
1099 unsigned block_start, block_end;
1100 unsigned blocksize = head->b_size;
1102 struct buffer_head *next;
1104 for ( bh = head, block_start = 0;
1105 ret == 0 && (bh != head || !block_start);
1106 block_start = block_end, bh = next)
1108 next = bh->b_this_page;
1109 block_end = block_start + blocksize;
1110 if (block_end <= from || block_start >= to) {
1111 if (partial && !buffer_uptodate(bh))
1115 err = (*fn)(handle, bh);
1123 * To preserve ordering, it is essential that the hole instantiation and
1124 * the data write be encapsulated in a single transaction. We cannot
1125 * close off a transaction and start a new one between the ext4_get_block()
1126 * and the commit_write(). So doing the jbd2_journal_start at the start of
1127 * prepare_write() is the right place.
1129 * Also, this function can nest inside ext4_writepage() ->
1130 * block_write_full_page(). In that case, we *know* that ext4_writepage()
1131 * has generated enough buffer credits to do the whole page. So we won't
1132 * block on the journal in that case, which is good, because the caller may
1135 * By accident, ext4 can be reentered when a transaction is open via
1136 * quota file writes. If we were to commit the transaction while thus
1137 * reentered, there can be a deadlock - we would be holding a quota
1138 * lock, and the commit would never complete if another thread had a
1139 * transaction open and was blocking on the quota lock - a ranking
1142 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1143 * will _not_ run commit under these circumstances because handle->h_ref
1144 * is elevated. We'll still have enough credits for the tiny quotafile
1147 static int do_journal_get_write_access(handle_t *handle,
1148 struct buffer_head *bh)
1150 if (!buffer_mapped(bh) || buffer_freed(bh))
1152 return ext4_journal_get_write_access(handle, bh);
1155 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1156 loff_t pos, unsigned len, unsigned flags,
1157 struct page **pagep, void **fsdata)
1159 struct inode *inode = mapping->host;
1160 int ret, needed_blocks = ext4_writepage_trans_blocks(inode);
1167 index = pos >> PAGE_CACHE_SHIFT;
1168 from = pos & (PAGE_CACHE_SIZE - 1);
1172 page = __grab_cache_page(mapping, index);
1177 handle = ext4_journal_start(inode, needed_blocks);
1178 if (IS_ERR(handle)) {
1180 page_cache_release(page);
1181 ret = PTR_ERR(handle);
1185 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1188 if (!ret && ext4_should_journal_data(inode)) {
1189 ret = walk_page_buffers(handle, page_buffers(page),
1190 from, to, NULL, do_journal_get_write_access);
1194 ext4_journal_stop(handle);
1196 page_cache_release(page);
1199 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
1205 int ext4_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1207 int err = jbd2_journal_dirty_data(handle, bh);
1209 ext4_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1214 /* For write_end() in data=journal mode */
1215 static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1217 if (!buffer_mapped(bh) || buffer_freed(bh))
1219 set_buffer_uptodate(bh);
1220 return ext4_journal_dirty_metadata(handle, bh);
1224 * Generic write_end handler for ordered and writeback ext4 journal modes.
1225 * We can't use generic_write_end, because that unlocks the page and we need to
1226 * unlock the page after ext4_journal_stop, but ext4_journal_stop must run
1227 * after block_write_end.
1229 static int ext4_generic_write_end(struct file *file,
1230 struct address_space *mapping,
1231 loff_t pos, unsigned len, unsigned copied,
1232 struct page *page, void *fsdata)
1234 struct inode *inode = file->f_mapping->host;
1236 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1238 if (pos+copied > inode->i_size) {
1239 i_size_write(inode, pos+copied);
1240 mark_inode_dirty(inode);
1247 * We need to pick up the new inode size which generic_commit_write gave us
1248 * `file' can be NULL - eg, when called from page_symlink().
1250 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1251 * buffers are managed internally.
1253 static int ext4_ordered_write_end(struct file *file,
1254 struct address_space *mapping,
1255 loff_t pos, unsigned len, unsigned copied,
1256 struct page *page, void *fsdata)
1258 handle_t *handle = ext4_journal_current_handle();
1259 struct inode *inode = file->f_mapping->host;
1263 from = pos & (PAGE_CACHE_SIZE - 1);
1266 ret = walk_page_buffers(handle, page_buffers(page),
1267 from, to, NULL, ext4_journal_dirty_data);
1271 * generic_write_end() will run mark_inode_dirty() if i_size
1272 * changes. So let's piggyback the i_disksize mark_inode_dirty
1277 new_i_size = pos + copied;
1278 if (new_i_size > EXT4_I(inode)->i_disksize)
1279 EXT4_I(inode)->i_disksize = new_i_size;
1280 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1285 ret2 = ext4_journal_stop(handle);
1289 page_cache_release(page);
1291 return ret ? ret : copied;
1294 static int ext4_writeback_write_end(struct file *file,
1295 struct address_space *mapping,
1296 loff_t pos, unsigned len, unsigned copied,
1297 struct page *page, void *fsdata)
1299 handle_t *handle = ext4_journal_current_handle();
1300 struct inode *inode = file->f_mapping->host;
1304 new_i_size = pos + copied;
1305 if (new_i_size > EXT4_I(inode)->i_disksize)
1306 EXT4_I(inode)->i_disksize = new_i_size;
1308 copied = ext4_generic_write_end(file, mapping, pos, len, copied,
1313 ret2 = ext4_journal_stop(handle);
1317 page_cache_release(page);
1319 return ret ? ret : copied;
1322 static int ext4_journalled_write_end(struct file *file,
1323 struct address_space *mapping,
1324 loff_t pos, unsigned len, unsigned copied,
1325 struct page *page, void *fsdata)
1327 handle_t *handle = ext4_journal_current_handle();
1328 struct inode *inode = mapping->host;
1333 from = pos & (PAGE_CACHE_SIZE - 1);
1337 if (!PageUptodate(page))
1339 page_zero_new_buffers(page, from+copied, to);
1342 ret = walk_page_buffers(handle, page_buffers(page), from,
1343 to, &partial, write_end_fn);
1345 SetPageUptodate(page);
1346 if (pos+copied > inode->i_size)
1347 i_size_write(inode, pos+copied);
1348 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1349 if (inode->i_size > EXT4_I(inode)->i_disksize) {
1350 EXT4_I(inode)->i_disksize = inode->i_size;
1351 ret2 = ext4_mark_inode_dirty(handle, inode);
1356 ret2 = ext4_journal_stop(handle);
1360 page_cache_release(page);
1362 return ret ? ret : copied;
1366 * bmap() is special. It gets used by applications such as lilo and by
1367 * the swapper to find the on-disk block of a specific piece of data.
1369 * Naturally, this is dangerous if the block concerned is still in the
1370 * journal. If somebody makes a swapfile on an ext4 data-journaling
1371 * filesystem and enables swap, then they may get a nasty shock when the
1372 * data getting swapped to that swapfile suddenly gets overwritten by
1373 * the original zero's written out previously to the journal and
1374 * awaiting writeback in the kernel's buffer cache.
1376 * So, if we see any bmap calls here on a modified, data-journaled file,
1377 * take extra steps to flush any blocks which might be in the cache.
1379 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
1381 struct inode *inode = mapping->host;
1385 if (EXT4_I(inode)->i_state & EXT4_STATE_JDATA) {
1387 * This is a REALLY heavyweight approach, but the use of
1388 * bmap on dirty files is expected to be extremely rare:
1389 * only if we run lilo or swapon on a freshly made file
1390 * do we expect this to happen.
1392 * (bmap requires CAP_SYS_RAWIO so this does not
1393 * represent an unprivileged user DOS attack --- we'd be
1394 * in trouble if mortal users could trigger this path at
1397 * NB. EXT4_STATE_JDATA is not set on files other than
1398 * regular files. If somebody wants to bmap a directory
1399 * or symlink and gets confused because the buffer
1400 * hasn't yet been flushed to disk, they deserve
1401 * everything they get.
1404 EXT4_I(inode)->i_state &= ~EXT4_STATE_JDATA;
1405 journal = EXT4_JOURNAL(inode);
1406 jbd2_journal_lock_updates(journal);
1407 err = jbd2_journal_flush(journal);
1408 jbd2_journal_unlock_updates(journal);
1414 return generic_block_bmap(mapping,block,ext4_get_block);
1417 static int bget_one(handle_t *handle, struct buffer_head *bh)
1423 static int bput_one(handle_t *handle, struct buffer_head *bh)
1429 static int jbd2_journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1431 if (buffer_mapped(bh))
1432 return ext4_journal_dirty_data(handle, bh);
1437 * Note that we always start a transaction even if we're not journalling
1438 * data. This is to preserve ordering: any hole instantiation within
1439 * __block_write_full_page -> ext4_get_block() should be journalled
1440 * along with the data so we don't crash and then get metadata which
1441 * refers to old data.
1443 * In all journalling modes block_write_full_page() will start the I/O.
1447 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1452 * ext4_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1454 * Same applies to ext4_get_block(). We will deadlock on various things like
1455 * lock_journal and i_truncate_mutex.
1457 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1460 * 16May01: If we're reentered then journal_current_handle() will be
1461 * non-zero. We simply *return*.
1463 * 1 July 2001: @@@ FIXME:
1464 * In journalled data mode, a data buffer may be metadata against the
1465 * current transaction. But the same file is part of a shared mapping
1466 * and someone does a writepage() on it.
1468 * We will move the buffer onto the async_data list, but *after* it has
1469 * been dirtied. So there's a small window where we have dirty data on
1472 * Note that this only applies to the last partial page in the file. The
1473 * bit which block_write_full_page() uses prepare/commit for. (That's
1474 * broken code anyway: it's wrong for msync()).
1476 * It's a rare case: affects the final partial page, for journalled data
1477 * where the file is subject to bith write() and writepage() in the same
1478 * transction. To fix it we'll need a custom block_write_full_page().
1479 * We'll probably need that anyway for journalling writepage() output.
1481 * We don't honour synchronous mounts for writepage(). That would be
1482 * disastrous. Any write() or metadata operation will sync the fs for
1485 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1486 * we don't need to open a transaction here.
1488 static int ext4_ordered_writepage(struct page *page,
1489 struct writeback_control *wbc)
1491 struct inode *inode = page->mapping->host;
1492 struct buffer_head *page_bufs;
1493 handle_t *handle = NULL;
1497 J_ASSERT(PageLocked(page));
1500 * We give up here if we're reentered, because it might be for a
1501 * different filesystem.
1503 if (ext4_journal_current_handle())
1506 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1508 if (IS_ERR(handle)) {
1509 ret = PTR_ERR(handle);
1513 if (!page_has_buffers(page)) {
1514 create_empty_buffers(page, inode->i_sb->s_blocksize,
1515 (1 << BH_Dirty)|(1 << BH_Uptodate));
1517 page_bufs = page_buffers(page);
1518 walk_page_buffers(handle, page_bufs, 0,
1519 PAGE_CACHE_SIZE, NULL, bget_one);
1521 ret = block_write_full_page(page, ext4_get_block, wbc);
1524 * The page can become unlocked at any point now, and
1525 * truncate can then come in and change things. So we
1526 * can't touch *page from now on. But *page_bufs is
1527 * safe due to elevated refcount.
1531 * And attach them to the current transaction. But only if
1532 * block_write_full_page() succeeded. Otherwise they are unmapped,
1533 * and generally junk.
1536 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1537 NULL, jbd2_journal_dirty_data_fn);
1541 walk_page_buffers(handle, page_bufs, 0,
1542 PAGE_CACHE_SIZE, NULL, bput_one);
1543 err = ext4_journal_stop(handle);
1549 redirty_page_for_writepage(wbc, page);
1554 static int ext4_writeback_writepage(struct page *page,
1555 struct writeback_control *wbc)
1557 struct inode *inode = page->mapping->host;
1558 handle_t *handle = NULL;
1562 if (ext4_journal_current_handle())
1565 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1566 if (IS_ERR(handle)) {
1567 ret = PTR_ERR(handle);
1571 if (test_opt(inode->i_sb, NOBH) && ext4_should_writeback_data(inode))
1572 ret = nobh_writepage(page, ext4_get_block, wbc);
1574 ret = block_write_full_page(page, ext4_get_block, wbc);
1576 err = ext4_journal_stop(handle);
1582 redirty_page_for_writepage(wbc, page);
1587 static int ext4_journalled_writepage(struct page *page,
1588 struct writeback_control *wbc)
1590 struct inode *inode = page->mapping->host;
1591 handle_t *handle = NULL;
1595 if (ext4_journal_current_handle())
1598 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1599 if (IS_ERR(handle)) {
1600 ret = PTR_ERR(handle);
1604 if (!page_has_buffers(page) || PageChecked(page)) {
1606 * It's mmapped pagecache. Add buffers and journal it. There
1607 * doesn't seem much point in redirtying the page here.
1609 ClearPageChecked(page);
1610 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1613 ext4_journal_stop(handle);
1616 ret = walk_page_buffers(handle, page_buffers(page), 0,
1617 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1619 err = walk_page_buffers(handle, page_buffers(page), 0,
1620 PAGE_CACHE_SIZE, NULL, write_end_fn);
1623 EXT4_I(inode)->i_state |= EXT4_STATE_JDATA;
1627 * It may be a page full of checkpoint-mode buffers. We don't
1628 * really know unless we go poke around in the buffer_heads.
1629 * But block_write_full_page will do the right thing.
1631 ret = block_write_full_page(page, ext4_get_block, wbc);
1633 err = ext4_journal_stop(handle);
1640 redirty_page_for_writepage(wbc, page);
1646 static int ext4_readpage(struct file *file, struct page *page)
1648 return mpage_readpage(page, ext4_get_block);
1652 ext4_readpages(struct file *file, struct address_space *mapping,
1653 struct list_head *pages, unsigned nr_pages)
1655 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
1658 static void ext4_invalidatepage(struct page *page, unsigned long offset)
1660 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1663 * If it's a full truncate we just forget about the pending dirtying
1666 ClearPageChecked(page);
1668 jbd2_journal_invalidatepage(journal, page, offset);
1671 static int ext4_releasepage(struct page *page, gfp_t wait)
1673 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
1675 WARN_ON(PageChecked(page));
1676 if (!page_has_buffers(page))
1678 return jbd2_journal_try_to_free_buffers(journal, page, wait);
1682 * If the O_DIRECT write will extend the file then add this inode to the
1683 * orphan list. So recovery will truncate it back to the original size
1684 * if the machine crashes during the write.
1686 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1687 * crashes then stale disk data _may_ be exposed inside the file.
1689 static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
1690 const struct iovec *iov, loff_t offset,
1691 unsigned long nr_segs)
1693 struct file *file = iocb->ki_filp;
1694 struct inode *inode = file->f_mapping->host;
1695 struct ext4_inode_info *ei = EXT4_I(inode);
1696 handle_t *handle = NULL;
1699 size_t count = iov_length(iov, nr_segs);
1702 loff_t final_size = offset + count;
1704 handle = ext4_journal_start(inode, DIO_CREDITS);
1705 if (IS_ERR(handle)) {
1706 ret = PTR_ERR(handle);
1709 if (final_size > inode->i_size) {
1710 ret = ext4_orphan_add(handle, inode);
1714 ei->i_disksize = inode->i_size;
1718 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1720 ext4_get_block, NULL);
1723 * Reacquire the handle: ext4_get_block() can restart the transaction
1725 handle = ext4_journal_current_handle();
1731 if (orphan && inode->i_nlink)
1732 ext4_orphan_del(handle, inode);
1733 if (orphan && ret > 0) {
1734 loff_t end = offset + ret;
1735 if (end > inode->i_size) {
1736 ei->i_disksize = end;
1737 i_size_write(inode, end);
1739 * We're going to return a positive `ret'
1740 * here due to non-zero-length I/O, so there's
1741 * no way of reporting error returns from
1742 * ext4_mark_inode_dirty() to userspace. So
1745 ext4_mark_inode_dirty(handle, inode);
1748 err = ext4_journal_stop(handle);
1757 * Pages can be marked dirty completely asynchronously from ext4's journalling
1758 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1759 * much here because ->set_page_dirty is called under VFS locks. The page is
1760 * not necessarily locked.
1762 * We cannot just dirty the page and leave attached buffers clean, because the
1763 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1764 * or jbddirty because all the journalling code will explode.
1766 * So what we do is to mark the page "pending dirty" and next time writepage
1767 * is called, propagate that into the buffers appropriately.
1769 static int ext4_journalled_set_page_dirty(struct page *page)
1771 SetPageChecked(page);
1772 return __set_page_dirty_nobuffers(page);
1775 static const struct address_space_operations ext4_ordered_aops = {
1776 .readpage = ext4_readpage,
1777 .readpages = ext4_readpages,
1778 .writepage = ext4_ordered_writepage,
1779 .sync_page = block_sync_page,
1780 .write_begin = ext4_write_begin,
1781 .write_end = ext4_ordered_write_end,
1783 .invalidatepage = ext4_invalidatepage,
1784 .releasepage = ext4_releasepage,
1785 .direct_IO = ext4_direct_IO,
1786 .migratepage = buffer_migrate_page,
1789 static const struct address_space_operations ext4_writeback_aops = {
1790 .readpage = ext4_readpage,
1791 .readpages = ext4_readpages,
1792 .writepage = ext4_writeback_writepage,
1793 .sync_page = block_sync_page,
1794 .write_begin = ext4_write_begin,
1795 .write_end = ext4_writeback_write_end,
1797 .invalidatepage = ext4_invalidatepage,
1798 .releasepage = ext4_releasepage,
1799 .direct_IO = ext4_direct_IO,
1800 .migratepage = buffer_migrate_page,
1803 static const struct address_space_operations ext4_journalled_aops = {
1804 .readpage = ext4_readpage,
1805 .readpages = ext4_readpages,
1806 .writepage = ext4_journalled_writepage,
1807 .sync_page = block_sync_page,
1808 .write_begin = ext4_write_begin,
1809 .write_end = ext4_journalled_write_end,
1810 .set_page_dirty = ext4_journalled_set_page_dirty,
1812 .invalidatepage = ext4_invalidatepage,
1813 .releasepage = ext4_releasepage,
1816 void ext4_set_aops(struct inode *inode)
1818 if (ext4_should_order_data(inode))
1819 inode->i_mapping->a_ops = &ext4_ordered_aops;
1820 else if (ext4_should_writeback_data(inode))
1821 inode->i_mapping->a_ops = &ext4_writeback_aops;
1823 inode->i_mapping->a_ops = &ext4_journalled_aops;
1827 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
1828 * up to the end of the block which corresponds to `from'.
1829 * This required during truncate. We need to physically zero the tail end
1830 * of that block so it doesn't yield old data if the file is later grown.
1832 int ext4_block_truncate_page(handle_t *handle, struct page *page,
1833 struct address_space *mapping, loff_t from)
1835 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
1836 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1837 unsigned blocksize, length, pos;
1839 struct inode *inode = mapping->host;
1840 struct buffer_head *bh;
1843 blocksize = inode->i_sb->s_blocksize;
1844 length = blocksize - (offset & (blocksize - 1));
1845 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1848 * For "nobh" option, we can only work if we don't need to
1849 * read-in the page - otherwise we create buffers to do the IO.
1851 if (!page_has_buffers(page) && test_opt(inode->i_sb, NOBH) &&
1852 ext4_should_writeback_data(inode) && PageUptodate(page)) {
1853 zero_user_page(page, offset, length, KM_USER0);
1854 set_page_dirty(page);
1858 if (!page_has_buffers(page))
1859 create_empty_buffers(page, blocksize, 0);
1861 /* Find the buffer that contains "offset" */
1862 bh = page_buffers(page);
1864 while (offset >= pos) {
1865 bh = bh->b_this_page;
1871 if (buffer_freed(bh)) {
1872 BUFFER_TRACE(bh, "freed: skip");
1876 if (!buffer_mapped(bh)) {
1877 BUFFER_TRACE(bh, "unmapped");
1878 ext4_get_block(inode, iblock, bh, 0);
1879 /* unmapped? It's a hole - nothing to do */
1880 if (!buffer_mapped(bh)) {
1881 BUFFER_TRACE(bh, "still unmapped");
1886 /* Ok, it's mapped. Make sure it's up-to-date */
1887 if (PageUptodate(page))
1888 set_buffer_uptodate(bh);
1890 if (!buffer_uptodate(bh)) {
1892 ll_rw_block(READ, 1, &bh);
1894 /* Uhhuh. Read error. Complain and punt. */
1895 if (!buffer_uptodate(bh))
1899 if (ext4_should_journal_data(inode)) {
1900 BUFFER_TRACE(bh, "get write access");
1901 err = ext4_journal_get_write_access(handle, bh);
1906 zero_user_page(page, offset, length, KM_USER0);
1908 BUFFER_TRACE(bh, "zeroed end of block");
1911 if (ext4_should_journal_data(inode)) {
1912 err = ext4_journal_dirty_metadata(handle, bh);
1914 if (ext4_should_order_data(inode))
1915 err = ext4_journal_dirty_data(handle, bh);
1916 mark_buffer_dirty(bh);
1921 page_cache_release(page);
1926 * Probably it should be a library function... search for first non-zero word
1927 * or memcmp with zero_page, whatever is better for particular architecture.
1930 static inline int all_zeroes(__le32 *p, __le32 *q)
1939 * ext4_find_shared - find the indirect blocks for partial truncation.
1940 * @inode: inode in question
1941 * @depth: depth of the affected branch
1942 * @offsets: offsets of pointers in that branch (see ext4_block_to_path)
1943 * @chain: place to store the pointers to partial indirect blocks
1944 * @top: place to the (detached) top of branch
1946 * This is a helper function used by ext4_truncate().
1948 * When we do truncate() we may have to clean the ends of several
1949 * indirect blocks but leave the blocks themselves alive. Block is
1950 * partially truncated if some data below the new i_size is refered
1951 * from it (and it is on the path to the first completely truncated
1952 * data block, indeed). We have to free the top of that path along
1953 * with everything to the right of the path. Since no allocation
1954 * past the truncation point is possible until ext4_truncate()
1955 * finishes, we may safely do the latter, but top of branch may
1956 * require special attention - pageout below the truncation point
1957 * might try to populate it.
1959 * We atomically detach the top of branch from the tree, store the
1960 * block number of its root in *@top, pointers to buffer_heads of
1961 * partially truncated blocks - in @chain[].bh and pointers to
1962 * their last elements that should not be removed - in
1963 * @chain[].p. Return value is the pointer to last filled element
1966 * The work left to caller to do the actual freeing of subtrees:
1967 * a) free the subtree starting from *@top
1968 * b) free the subtrees whose roots are stored in
1969 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1970 * c) free the subtrees growing from the inode past the @chain[0].
1971 * (no partially truncated stuff there). */
1973 static Indirect *ext4_find_shared(struct inode *inode, int depth,
1974 ext4_lblk_t offsets[4], Indirect chain[4], __le32 *top)
1976 Indirect *partial, *p;
1980 /* Make k index the deepest non-null offest + 1 */
1981 for (k = depth; k > 1 && !offsets[k-1]; k--)
1983 partial = ext4_get_branch(inode, k, offsets, chain, &err);
1984 /* Writer: pointers */
1986 partial = chain + k-1;
1988 * If the branch acquired continuation since we've looked at it -
1989 * fine, it should all survive and (new) top doesn't belong to us.
1991 if (!partial->key && *partial->p)
1994 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1997 * OK, we've found the last block that must survive. The rest of our
1998 * branch should be detached before unlocking. However, if that rest
1999 * of branch is all ours and does not grow immediately from the inode
2000 * it's easier to cheat and just decrement partial->p.
2002 if (p == chain + k - 1 && p > chain) {
2006 /* Nope, don't do this in ext4. Must leave the tree intact */
2013 while(partial > p) {
2014 brelse(partial->bh);
2022 * Zero a number of block pointers in either an inode or an indirect block.
2023 * If we restart the transaction we must again get write access to the
2024 * indirect block for further modification.
2026 * We release `count' blocks on disk, but (last - first) may be greater
2027 * than `count' because there can be holes in there.
2029 static void ext4_clear_blocks(handle_t *handle, struct inode *inode,
2030 struct buffer_head *bh, ext4_fsblk_t block_to_free,
2031 unsigned long count, __le32 *first, __le32 *last)
2034 if (try_to_extend_transaction(handle, inode)) {
2036 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2037 ext4_journal_dirty_metadata(handle, bh);
2039 ext4_mark_inode_dirty(handle, inode);
2040 ext4_journal_test_restart(handle, inode);
2042 BUFFER_TRACE(bh, "retaking write access");
2043 ext4_journal_get_write_access(handle, bh);
2048 * Any buffers which are on the journal will be in memory. We find
2049 * them on the hash table so jbd2_journal_revoke() will run jbd2_journal_forget()
2050 * on them. We've already detached each block from the file, so
2051 * bforget() in jbd2_journal_forget() should be safe.
2053 * AKPM: turn on bforget in jbd2_journal_forget()!!!
2055 for (p = first; p < last; p++) {
2056 u32 nr = le32_to_cpu(*p);
2058 struct buffer_head *tbh;
2061 tbh = sb_find_get_block(inode->i_sb, nr);
2062 ext4_forget(handle, 0, inode, tbh, nr);
2066 ext4_free_blocks(handle, inode, block_to_free, count);
2070 * ext4_free_data - free a list of data blocks
2071 * @handle: handle for this transaction
2072 * @inode: inode we are dealing with
2073 * @this_bh: indirect buffer_head which contains *@first and *@last
2074 * @first: array of block numbers
2075 * @last: points immediately past the end of array
2077 * We are freeing all blocks refered from that array (numbers are stored as
2078 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
2080 * We accumulate contiguous runs of blocks to free. Conveniently, if these
2081 * blocks are contiguous then releasing them at one time will only affect one
2082 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
2083 * actually use a lot of journal space.
2085 * @this_bh will be %NULL if @first and @last point into the inode's direct
2088 static void ext4_free_data(handle_t *handle, struct inode *inode,
2089 struct buffer_head *this_bh,
2090 __le32 *first, __le32 *last)
2092 ext4_fsblk_t block_to_free = 0; /* Starting block # of a run */
2093 unsigned long count = 0; /* Number of blocks in the run */
2094 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
2097 ext4_fsblk_t nr; /* Current block # */
2098 __le32 *p; /* Pointer into inode/ind
2099 for current block */
2102 if (this_bh) { /* For indirect block */
2103 BUFFER_TRACE(this_bh, "get_write_access");
2104 err = ext4_journal_get_write_access(handle, this_bh);
2105 /* Important: if we can't update the indirect pointers
2106 * to the blocks, we can't free them. */
2111 for (p = first; p < last; p++) {
2112 nr = le32_to_cpu(*p);
2114 /* accumulate blocks to free if they're contiguous */
2117 block_to_free_p = p;
2119 } else if (nr == block_to_free + count) {
2122 ext4_clear_blocks(handle, inode, this_bh,
2124 count, block_to_free_p, p);
2126 block_to_free_p = p;
2133 ext4_clear_blocks(handle, inode, this_bh, block_to_free,
2134 count, block_to_free_p, p);
2137 BUFFER_TRACE(this_bh, "call ext4_journal_dirty_metadata");
2138 ext4_journal_dirty_metadata(handle, this_bh);
2143 * ext4_free_branches - free an array of branches
2144 * @handle: JBD handle for this transaction
2145 * @inode: inode we are dealing with
2146 * @parent_bh: the buffer_head which contains *@first and *@last
2147 * @first: array of block numbers
2148 * @last: pointer immediately past the end of array
2149 * @depth: depth of the branches to free
2151 * We are freeing all blocks refered from these branches (numbers are
2152 * stored as little-endian 32-bit) and updating @inode->i_blocks
2155 static void ext4_free_branches(handle_t *handle, struct inode *inode,
2156 struct buffer_head *parent_bh,
2157 __le32 *first, __le32 *last, int depth)
2162 if (is_handle_aborted(handle))
2166 struct buffer_head *bh;
2167 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2169 while (--p >= first) {
2170 nr = le32_to_cpu(*p);
2172 continue; /* A hole */
2174 /* Go read the buffer for the next level down */
2175 bh = sb_bread(inode->i_sb, nr);
2178 * A read failure? Report error and clear slot
2182 ext4_error(inode->i_sb, "ext4_free_branches",
2183 "Read failure, inode=%lu, block=%llu",
2188 /* This zaps the entire block. Bottom up. */
2189 BUFFER_TRACE(bh, "free child branches");
2190 ext4_free_branches(handle, inode, bh,
2191 (__le32*)bh->b_data,
2192 (__le32*)bh->b_data + addr_per_block,
2196 * We've probably journalled the indirect block several
2197 * times during the truncate. But it's no longer
2198 * needed and we now drop it from the transaction via
2199 * jbd2_journal_revoke().
2201 * That's easy if it's exclusively part of this
2202 * transaction. But if it's part of the committing
2203 * transaction then jbd2_journal_forget() will simply
2204 * brelse() it. That means that if the underlying
2205 * block is reallocated in ext4_get_block(),
2206 * unmap_underlying_metadata() will find this block
2207 * and will try to get rid of it. damn, damn.
2209 * If this block has already been committed to the
2210 * journal, a revoke record will be written. And
2211 * revoke records must be emitted *before* clearing
2212 * this block's bit in the bitmaps.
2214 ext4_forget(handle, 1, inode, bh, bh->b_blocknr);
2217 * Everything below this this pointer has been
2218 * released. Now let this top-of-subtree go.
2220 * We want the freeing of this indirect block to be
2221 * atomic in the journal with the updating of the
2222 * bitmap block which owns it. So make some room in
2225 * We zero the parent pointer *after* freeing its
2226 * pointee in the bitmaps, so if extend_transaction()
2227 * for some reason fails to put the bitmap changes and
2228 * the release into the same transaction, recovery
2229 * will merely complain about releasing a free block,
2230 * rather than leaking blocks.
2232 if (is_handle_aborted(handle))
2234 if (try_to_extend_transaction(handle, inode)) {
2235 ext4_mark_inode_dirty(handle, inode);
2236 ext4_journal_test_restart(handle, inode);
2239 ext4_free_blocks(handle, inode, nr, 1);
2243 * The block which we have just freed is
2244 * pointed to by an indirect block: journal it
2246 BUFFER_TRACE(parent_bh, "get_write_access");
2247 if (!ext4_journal_get_write_access(handle,
2250 BUFFER_TRACE(parent_bh,
2251 "call ext4_journal_dirty_metadata");
2252 ext4_journal_dirty_metadata(handle,
2258 /* We have reached the bottom of the tree. */
2259 BUFFER_TRACE(parent_bh, "free data blocks");
2260 ext4_free_data(handle, inode, parent_bh, first, last);
2267 * We block out ext4_get_block() block instantiations across the entire
2268 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
2269 * simultaneously on behalf of the same inode.
2271 * As we work through the truncate and commmit bits of it to the journal there
2272 * is one core, guiding principle: the file's tree must always be consistent on
2273 * disk. We must be able to restart the truncate after a crash.
2275 * The file's tree may be transiently inconsistent in memory (although it
2276 * probably isn't), but whenever we close off and commit a journal transaction,
2277 * the contents of (the filesystem + the journal) must be consistent and
2278 * restartable. It's pretty simple, really: bottom up, right to left (although
2279 * left-to-right works OK too).
2281 * Note that at recovery time, journal replay occurs *before* the restart of
2282 * truncate against the orphan inode list.
2284 * The committed inode has the new, desired i_size (which is the same as
2285 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
2286 * that this inode's truncate did not complete and it will again call
2287 * ext4_truncate() to have another go. So there will be instantiated blocks
2288 * to the right of the truncation point in a crashed ext4 filesystem. But
2289 * that's fine - as long as they are linked from the inode, the post-crash
2290 * ext4_truncate() run will find them and release them.
2292 void ext4_truncate(struct inode *inode)
2295 struct ext4_inode_info *ei = EXT4_I(inode);
2296 __le32 *i_data = ei->i_data;
2297 int addr_per_block = EXT4_ADDR_PER_BLOCK(inode->i_sb);
2298 struct address_space *mapping = inode->i_mapping;
2299 ext4_lblk_t offsets[4];
2304 ext4_lblk_t last_block;
2305 unsigned blocksize = inode->i_sb->s_blocksize;
2308 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2309 S_ISLNK(inode->i_mode)))
2311 if (ext4_inode_is_fast_symlink(inode))
2313 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2317 * We have to lock the EOF page here, because lock_page() nests
2318 * outside jbd2_journal_start().
2320 if ((inode->i_size & (blocksize - 1)) == 0) {
2321 /* Block boundary? Nothing to do */
2324 page = grab_cache_page(mapping,
2325 inode->i_size >> PAGE_CACHE_SHIFT);
2330 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL) {
2331 ext4_ext_truncate(inode, page);
2335 handle = start_transaction(inode);
2336 if (IS_ERR(handle)) {
2338 clear_highpage(page);
2339 flush_dcache_page(page);
2341 page_cache_release(page);
2343 return; /* AKPM: return what? */
2346 last_block = (inode->i_size + blocksize-1)
2347 >> EXT4_BLOCK_SIZE_BITS(inode->i_sb);
2350 ext4_block_truncate_page(handle, page, mapping, inode->i_size);
2352 n = ext4_block_to_path(inode, last_block, offsets, NULL);
2354 goto out_stop; /* error */
2357 * OK. This truncate is going to happen. We add the inode to the
2358 * orphan list, so that if this truncate spans multiple transactions,
2359 * and we crash, we will resume the truncate when the filesystem
2360 * recovers. It also marks the inode dirty, to catch the new size.
2362 * Implication: the file must always be in a sane, consistent
2363 * truncatable state while each transaction commits.
2365 if (ext4_orphan_add(handle, inode))
2369 * The orphan list entry will now protect us from any crash which
2370 * occurs before the truncate completes, so it is now safe to propagate
2371 * the new, shorter inode size (held for now in i_size) into the
2372 * on-disk inode. We do this via i_disksize, which is the value which
2373 * ext4 *really* writes onto the disk inode.
2375 ei->i_disksize = inode->i_size;
2378 * From here we block out all ext4_get_block() callers who want to
2379 * modify the block allocation tree.
2381 mutex_lock(&ei->truncate_mutex);
2383 if (n == 1) { /* direct blocks */
2384 ext4_free_data(handle, inode, NULL, i_data+offsets[0],
2385 i_data + EXT4_NDIR_BLOCKS);
2389 partial = ext4_find_shared(inode, n, offsets, chain, &nr);
2390 /* Kill the top of shared branch (not detached) */
2392 if (partial == chain) {
2393 /* Shared branch grows from the inode */
2394 ext4_free_branches(handle, inode, NULL,
2395 &nr, &nr+1, (chain+n-1) - partial);
2398 * We mark the inode dirty prior to restart,
2399 * and prior to stop. No need for it here.
2402 /* Shared branch grows from an indirect block */
2403 BUFFER_TRACE(partial->bh, "get_write_access");
2404 ext4_free_branches(handle, inode, partial->bh,
2406 partial->p+1, (chain+n-1) - partial);
2409 /* Clear the ends of indirect blocks on the shared branch */
2410 while (partial > chain) {
2411 ext4_free_branches(handle, inode, partial->bh, partial->p + 1,
2412 (__le32*)partial->bh->b_data+addr_per_block,
2413 (chain+n-1) - partial);
2414 BUFFER_TRACE(partial->bh, "call brelse");
2415 brelse (partial->bh);
2419 /* Kill the remaining (whole) subtrees */
2420 switch (offsets[0]) {
2422 nr = i_data[EXT4_IND_BLOCK];
2424 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 1);
2425 i_data[EXT4_IND_BLOCK] = 0;
2427 case EXT4_IND_BLOCK:
2428 nr = i_data[EXT4_DIND_BLOCK];
2430 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 2);
2431 i_data[EXT4_DIND_BLOCK] = 0;
2433 case EXT4_DIND_BLOCK:
2434 nr = i_data[EXT4_TIND_BLOCK];
2436 ext4_free_branches(handle, inode, NULL, &nr, &nr+1, 3);
2437 i_data[EXT4_TIND_BLOCK] = 0;
2439 case EXT4_TIND_BLOCK:
2443 ext4_discard_reservation(inode);
2445 mutex_unlock(&ei->truncate_mutex);
2446 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
2447 ext4_mark_inode_dirty(handle, inode);
2450 * In a multi-transaction truncate, we only make the final transaction
2457 * If this was a simple ftruncate(), and the file will remain alive
2458 * then we need to clear up the orphan record which we created above.
2459 * However, if this was a real unlink then we were called by
2460 * ext4_delete_inode(), and we allow that function to clean up the
2461 * orphan info for us.
2464 ext4_orphan_del(handle, inode);
2466 ext4_journal_stop(handle);
2469 static ext4_fsblk_t ext4_get_inode_block(struct super_block *sb,
2470 unsigned long ino, struct ext4_iloc *iloc)
2472 unsigned long desc, group_desc;
2473 ext4_group_t block_group;
2474 unsigned long offset;
2476 struct buffer_head *bh;
2477 struct ext4_group_desc * gdp;
2479 if (!ext4_valid_inum(sb, ino)) {
2481 * This error is already checked for in namei.c unless we are
2482 * looking at an NFS filehandle, in which case no error
2488 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
2489 if (block_group >= EXT4_SB(sb)->s_groups_count) {
2490 ext4_error(sb,"ext4_get_inode_block","group >= groups count");
2494 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
2495 desc = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
2496 bh = EXT4_SB(sb)->s_group_desc[group_desc];
2498 ext4_error (sb, "ext4_get_inode_block",
2499 "Descriptor not loaded");
2503 gdp = (struct ext4_group_desc *)((__u8 *)bh->b_data +
2504 desc * EXT4_DESC_SIZE(sb));
2506 * Figure out the offset within the block group inode table
2508 offset = ((ino - 1) % EXT4_INODES_PER_GROUP(sb)) *
2509 EXT4_INODE_SIZE(sb);
2510 block = ext4_inode_table(sb, gdp) +
2511 (offset >> EXT4_BLOCK_SIZE_BITS(sb));
2513 iloc->block_group = block_group;
2514 iloc->offset = offset & (EXT4_BLOCK_SIZE(sb) - 1);
2519 * ext4_get_inode_loc returns with an extra refcount against the inode's
2520 * underlying buffer_head on success. If 'in_mem' is true, we have all
2521 * data in memory that is needed to recreate the on-disk version of this
2524 static int __ext4_get_inode_loc(struct inode *inode,
2525 struct ext4_iloc *iloc, int in_mem)
2528 struct buffer_head *bh;
2530 block = ext4_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2534 bh = sb_getblk(inode->i_sb, block);
2536 ext4_error (inode->i_sb, "ext4_get_inode_loc",
2537 "unable to read inode block - "
2538 "inode=%lu, block=%llu",
2539 inode->i_ino, block);
2542 if (!buffer_uptodate(bh)) {
2544 if (buffer_uptodate(bh)) {
2545 /* someone brought it uptodate while we waited */
2551 * If we have all information of the inode in memory and this
2552 * is the only valid inode in the block, we need not read the
2556 struct buffer_head *bitmap_bh;
2557 struct ext4_group_desc *desc;
2558 int inodes_per_buffer;
2559 int inode_offset, i;
2560 ext4_group_t block_group;
2563 block_group = (inode->i_ino - 1) /
2564 EXT4_INODES_PER_GROUP(inode->i_sb);
2565 inodes_per_buffer = bh->b_size /
2566 EXT4_INODE_SIZE(inode->i_sb);
2567 inode_offset = ((inode->i_ino - 1) %
2568 EXT4_INODES_PER_GROUP(inode->i_sb));
2569 start = inode_offset & ~(inodes_per_buffer - 1);
2571 /* Is the inode bitmap in cache? */
2572 desc = ext4_get_group_desc(inode->i_sb,
2577 bitmap_bh = sb_getblk(inode->i_sb,
2578 ext4_inode_bitmap(inode->i_sb, desc));
2583 * If the inode bitmap isn't in cache then the
2584 * optimisation may end up performing two reads instead
2585 * of one, so skip it.
2587 if (!buffer_uptodate(bitmap_bh)) {
2591 for (i = start; i < start + inodes_per_buffer; i++) {
2592 if (i == inode_offset)
2594 if (ext4_test_bit(i, bitmap_bh->b_data))
2598 if (i == start + inodes_per_buffer) {
2599 /* all other inodes are free, so skip I/O */
2600 memset(bh->b_data, 0, bh->b_size);
2601 set_buffer_uptodate(bh);
2609 * There are other valid inodes in the buffer, this inode
2610 * has in-inode xattrs, or we don't have this inode in memory.
2611 * Read the block from disk.
2614 bh->b_end_io = end_buffer_read_sync;
2615 submit_bh(READ_META, bh);
2617 if (!buffer_uptodate(bh)) {
2618 ext4_error(inode->i_sb, "ext4_get_inode_loc",
2619 "unable to read inode block - "
2620 "inode=%lu, block=%llu",
2621 inode->i_ino, block);
2631 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
2633 /* We have all inode data except xattrs in memory here. */
2634 return __ext4_get_inode_loc(inode, iloc,
2635 !(EXT4_I(inode)->i_state & EXT4_STATE_XATTR));
2638 void ext4_set_inode_flags(struct inode *inode)
2640 unsigned int flags = EXT4_I(inode)->i_flags;
2642 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
2643 if (flags & EXT4_SYNC_FL)
2644 inode->i_flags |= S_SYNC;
2645 if (flags & EXT4_APPEND_FL)
2646 inode->i_flags |= S_APPEND;
2647 if (flags & EXT4_IMMUTABLE_FL)
2648 inode->i_flags |= S_IMMUTABLE;
2649 if (flags & EXT4_NOATIME_FL)
2650 inode->i_flags |= S_NOATIME;
2651 if (flags & EXT4_DIRSYNC_FL)
2652 inode->i_flags |= S_DIRSYNC;
2655 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
2656 void ext4_get_inode_flags(struct ext4_inode_info *ei)
2658 unsigned int flags = ei->vfs_inode.i_flags;
2660 ei->i_flags &= ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
2661 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|EXT4_DIRSYNC_FL);
2663 ei->i_flags |= EXT4_SYNC_FL;
2664 if (flags & S_APPEND)
2665 ei->i_flags |= EXT4_APPEND_FL;
2666 if (flags & S_IMMUTABLE)
2667 ei->i_flags |= EXT4_IMMUTABLE_FL;
2668 if (flags & S_NOATIME)
2669 ei->i_flags |= EXT4_NOATIME_FL;
2670 if (flags & S_DIRSYNC)
2671 ei->i_flags |= EXT4_DIRSYNC_FL;
2673 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
2674 struct ext4_inode_info *ei)
2677 struct inode *inode = &(ei->vfs_inode);
2678 struct super_block *sb = inode->i_sb;
2680 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
2681 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2682 /* we are using combined 48 bit field */
2683 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
2684 le32_to_cpu(raw_inode->i_blocks_lo);
2685 if (ei->i_flags & EXT4_HUGE_FILE_FL) {
2686 /* i_blocks represent file system block size */
2687 return i_blocks << (inode->i_blkbits - 9);
2692 return le32_to_cpu(raw_inode->i_blocks_lo);
2696 void ext4_read_inode(struct inode * inode)
2698 struct ext4_iloc iloc;
2699 struct ext4_inode *raw_inode;
2700 struct ext4_inode_info *ei = EXT4_I(inode);
2701 struct buffer_head *bh;
2704 #ifdef CONFIG_EXT4DEV_FS_POSIX_ACL
2705 ei->i_acl = EXT4_ACL_NOT_CACHED;
2706 ei->i_default_acl = EXT4_ACL_NOT_CACHED;
2708 ei->i_block_alloc_info = NULL;
2710 if (__ext4_get_inode_loc(inode, &iloc, 0))
2713 raw_inode = ext4_raw_inode(&iloc);
2714 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2715 inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2716 inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2717 if(!(test_opt (inode->i_sb, NO_UID32))) {
2718 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2719 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2721 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2724 ei->i_dir_start_lookup = 0;
2725 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2726 /* We now have enough fields to check if the inode was active or not.
2727 * This is needed because nfsd might try to access dead inodes
2728 * the test is that same one that e2fsck uses
2729 * NeilBrown 1999oct15
2731 if (inode->i_nlink == 0) {
2732 if (inode->i_mode == 0 ||
2733 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
2734 /* this inode is deleted */
2738 /* The only unlinked inodes we let through here have
2739 * valid i_mode and are being read by the orphan
2740 * recovery code: that's fine, we're about to complete
2741 * the process of deleting those. */
2743 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2744 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
2745 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
2746 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2747 cpu_to_le32(EXT4_OS_HURD)) {
2749 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
2751 inode->i_size = ext4_isize(raw_inode);
2752 ei->i_disksize = inode->i_size;
2753 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2754 ei->i_block_group = iloc.block_group;
2756 * NOTE! The in-memory inode i_data array is in little-endian order
2757 * even on big-endian machines: we do NOT byteswap the block numbers!
2759 for (block = 0; block < EXT4_N_BLOCKS; block++)
2760 ei->i_data[block] = raw_inode->i_block[block];
2761 INIT_LIST_HEAD(&ei->i_orphan);
2763 if (inode->i_ino >= EXT4_FIRST_INO(inode->i_sb) + 1 &&
2764 EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
2766 * When mke2fs creates big inodes it does not zero out
2767 * the unused bytes above EXT4_GOOD_OLD_INODE_SIZE,
2768 * so ignore those first few inodes.
2770 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
2771 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
2772 EXT4_INODE_SIZE(inode->i_sb)) {
2776 if (ei->i_extra_isize == 0) {
2777 /* The extra space is currently unused. Use it. */
2778 ei->i_extra_isize = sizeof(struct ext4_inode) -
2779 EXT4_GOOD_OLD_INODE_SIZE;
2781 __le32 *magic = (void *)raw_inode +
2782 EXT4_GOOD_OLD_INODE_SIZE +
2784 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2785 ei->i_state |= EXT4_STATE_XATTR;
2788 ei->i_extra_isize = 0;
2790 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
2791 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
2792 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
2793 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
2795 if (S_ISREG(inode->i_mode)) {
2796 inode->i_op = &ext4_file_inode_operations;
2797 inode->i_fop = &ext4_file_operations;
2798 ext4_set_aops(inode);
2799 } else if (S_ISDIR(inode->i_mode)) {
2800 inode->i_op = &ext4_dir_inode_operations;
2801 inode->i_fop = &ext4_dir_operations;
2802 } else if (S_ISLNK(inode->i_mode)) {
2803 if (ext4_inode_is_fast_symlink(inode))
2804 inode->i_op = &ext4_fast_symlink_inode_operations;
2806 inode->i_op = &ext4_symlink_inode_operations;
2807 ext4_set_aops(inode);
2810 inode->i_op = &ext4_special_inode_operations;
2811 if (raw_inode->i_block[0])
2812 init_special_inode(inode, inode->i_mode,
2813 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2815 init_special_inode(inode, inode->i_mode,
2816 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2819 ext4_set_inode_flags(inode);
2823 make_bad_inode(inode);
2827 static int ext4_inode_blocks_set(handle_t *handle,
2828 struct ext4_inode *raw_inode,
2829 struct ext4_inode_info *ei)
2831 struct inode *inode = &(ei->vfs_inode);
2832 u64 i_blocks = inode->i_blocks;
2833 struct super_block *sb = inode->i_sb;
2836 if (i_blocks <= ~0U) {
2838 * i_blocks can be represnted in a 32 bit variable
2839 * as multiple of 512 bytes
2841 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
2842 raw_inode->i_blocks_high = 0;
2843 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
2844 } else if (i_blocks <= 0xffffffffffffULL) {
2846 * i_blocks can be represented in a 48 bit variable
2847 * as multiple of 512 bytes
2849 err = ext4_update_rocompat_feature(handle, sb,
2850 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2853 /* i_block is stored in the split 48 bit fields */
2854 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
2855 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
2856 ei->i_flags &= ~EXT4_HUGE_FILE_FL;
2859 * i_blocks should be represented in a 48 bit variable
2860 * as multiple of file system block size
2862 err = ext4_update_rocompat_feature(handle, sb,
2863 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
2866 ei->i_flags |= EXT4_HUGE_FILE_FL;
2867 /* i_block is stored in file system block size */
2868 i_blocks = i_blocks >> (inode->i_blkbits - 9);
2869 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
2870 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
2877 * Post the struct inode info into an on-disk inode location in the
2878 * buffer-cache. This gobbles the caller's reference to the
2879 * buffer_head in the inode location struct.
2881 * The caller must have write access to iloc->bh.
2883 static int ext4_do_update_inode(handle_t *handle,
2884 struct inode *inode,
2885 struct ext4_iloc *iloc)
2887 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
2888 struct ext4_inode_info *ei = EXT4_I(inode);
2889 struct buffer_head *bh = iloc->bh;
2890 int err = 0, rc, block;
2892 /* For fields not not tracking in the in-memory inode,
2893 * initialise them to zero for new inodes. */
2894 if (ei->i_state & EXT4_STATE_NEW)
2895 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2897 ext4_get_inode_flags(ei);
2898 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2899 if(!(test_opt(inode->i_sb, NO_UID32))) {
2900 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid));
2901 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid));
2903 * Fix up interoperability with old kernels. Otherwise, old inodes get
2904 * re-used with the upper 16 bits of the uid/gid intact
2907 raw_inode->i_uid_high =
2908 cpu_to_le16(high_16_bits(inode->i_uid));
2909 raw_inode->i_gid_high =
2910 cpu_to_le16(high_16_bits(inode->i_gid));
2912 raw_inode->i_uid_high = 0;
2913 raw_inode->i_gid_high = 0;
2916 raw_inode->i_uid_low =
2917 cpu_to_le16(fs_high2lowuid(inode->i_uid));
2918 raw_inode->i_gid_low =
2919 cpu_to_le16(fs_high2lowgid(inode->i_gid));
2920 raw_inode->i_uid_high = 0;
2921 raw_inode->i_gid_high = 0;
2923 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2925 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
2926 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
2927 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
2928 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
2930 if (ext4_inode_blocks_set(handle, raw_inode, ei))
2932 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2933 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2934 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
2935 cpu_to_le32(EXT4_OS_HURD))
2936 raw_inode->i_file_acl_high =
2937 cpu_to_le16(ei->i_file_acl >> 32);
2938 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
2939 ext4_isize_set(raw_inode, ei->i_disksize);
2940 if (ei->i_disksize > 0x7fffffffULL) {
2941 struct super_block *sb = inode->i_sb;
2942 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
2943 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
2944 EXT4_SB(sb)->s_es->s_rev_level ==
2945 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
2946 /* If this is the first large file
2947 * created, add a flag to the superblock.
2949 err = ext4_journal_get_write_access(handle,
2950 EXT4_SB(sb)->s_sbh);
2953 ext4_update_dynamic_rev(sb);
2954 EXT4_SET_RO_COMPAT_FEATURE(sb,
2955 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
2958 err = ext4_journal_dirty_metadata(handle,
2959 EXT4_SB(sb)->s_sbh);
2962 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2963 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2964 if (old_valid_dev(inode->i_rdev)) {
2965 raw_inode->i_block[0] =
2966 cpu_to_le32(old_encode_dev(inode->i_rdev));
2967 raw_inode->i_block[1] = 0;
2969 raw_inode->i_block[0] = 0;
2970 raw_inode->i_block[1] =
2971 cpu_to_le32(new_encode_dev(inode->i_rdev));
2972 raw_inode->i_block[2] = 0;
2974 } else for (block = 0; block < EXT4_N_BLOCKS; block++)
2975 raw_inode->i_block[block] = ei->i_data[block];
2977 if (ei->i_extra_isize)
2978 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
2980 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
2981 rc = ext4_journal_dirty_metadata(handle, bh);
2984 ei->i_state &= ~EXT4_STATE_NEW;
2988 ext4_std_error(inode->i_sb, err);
2993 * ext4_write_inode()
2995 * We are called from a few places:
2997 * - Within generic_file_write() for O_SYNC files.
2998 * Here, there will be no transaction running. We wait for any running
2999 * trasnaction to commit.
3001 * - Within sys_sync(), kupdate and such.
3002 * We wait on commit, if tol to.
3004 * - Within prune_icache() (PF_MEMALLOC == true)
3005 * Here we simply return. We can't afford to block kswapd on the
3008 * In all cases it is actually safe for us to return without doing anything,
3009 * because the inode has been copied into a raw inode buffer in
3010 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
3013 * Note that we are absolutely dependent upon all inode dirtiers doing the
3014 * right thing: they *must* call mark_inode_dirty() after dirtying info in
3015 * which we are interested.
3017 * It would be a bug for them to not do this. The code:
3019 * mark_inode_dirty(inode)
3021 * inode->i_size = expr;
3023 * is in error because a kswapd-driven write_inode() could occur while
3024 * `stuff()' is running, and the new i_size will be lost. Plus the inode
3025 * will no longer be on the superblock's dirty inode list.
3027 int ext4_write_inode(struct inode *inode, int wait)
3029 if (current->flags & PF_MEMALLOC)
3032 if (ext4_journal_current_handle()) {
3033 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
3041 return ext4_force_commit(inode->i_sb);
3047 * Called from notify_change.
3049 * We want to trap VFS attempts to truncate the file as soon as
3050 * possible. In particular, we want to make sure that when the VFS
3051 * shrinks i_size, we put the inode on the orphan list and modify
3052 * i_disksize immediately, so that during the subsequent flushing of
3053 * dirty pages and freeing of disk blocks, we can guarantee that any
3054 * commit will leave the blocks being flushed in an unused state on
3055 * disk. (On recovery, the inode will get truncated and the blocks will
3056 * be freed, so we have a strong guarantee that no future commit will
3057 * leave these blocks visible to the user.)
3059 * Called with inode->sem down.
3061 int ext4_setattr(struct dentry *dentry, struct iattr *attr)
3063 struct inode *inode = dentry->d_inode;
3065 const unsigned int ia_valid = attr->ia_valid;
3067 error = inode_change_ok(inode, attr);
3071 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3072 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3075 /* (user+group)*(old+new) structure, inode write (sb,
3076 * inode block, ? - but truncate inode update has it) */
3077 handle = ext4_journal_start(inode, 2*(EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)+
3078 EXT4_QUOTA_DEL_BLOCKS(inode->i_sb))+3);
3079 if (IS_ERR(handle)) {
3080 error = PTR_ERR(handle);
3083 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3085 ext4_journal_stop(handle);
3088 /* Update corresponding info in inode so that everything is in
3089 * one transaction */
3090 if (attr->ia_valid & ATTR_UID)
3091 inode->i_uid = attr->ia_uid;
3092 if (attr->ia_valid & ATTR_GID)
3093 inode->i_gid = attr->ia_gid;
3094 error = ext4_mark_inode_dirty(handle, inode);
3095 ext4_journal_stop(handle);
3098 if (attr->ia_valid & ATTR_SIZE) {
3099 if (!(EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)) {
3100 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3102 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
3109 if (S_ISREG(inode->i_mode) &&
3110 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
3113 handle = ext4_journal_start(inode, 3);
3114 if (IS_ERR(handle)) {
3115 error = PTR_ERR(handle);
3119 error = ext4_orphan_add(handle, inode);
3120 EXT4_I(inode)->i_disksize = attr->ia_size;
3121 rc = ext4_mark_inode_dirty(handle, inode);
3124 ext4_journal_stop(handle);
3127 rc = inode_setattr(inode, attr);
3129 /* If inode_setattr's call to ext4_truncate failed to get a
3130 * transaction handle at all, we need to clean up the in-core
3131 * orphan list manually. */
3133 ext4_orphan_del(NULL, inode);
3135 if (!rc && (ia_valid & ATTR_MODE))
3136 rc = ext4_acl_chmod(inode);
3139 ext4_std_error(inode->i_sb, error);
3147 * How many blocks doth make a writepage()?
3149 * With N blocks per page, it may be:
3154 * N+5 bitmap blocks (from the above)
3155 * N+5 group descriptor summary blocks
3158 * 2 * EXT4_SINGLEDATA_TRANS_BLOCKS for the quote files
3160 * 3 * (N + 5) + 2 + 2 * EXT4_SINGLEDATA_TRANS_BLOCKS
3162 * With ordered or writeback data it's the same, less the N data blocks.
3164 * If the inode's direct blocks can hold an integral number of pages then a
3165 * page cannot straddle two indirect blocks, and we can only touch one indirect
3166 * and dindirect block, and the "5" above becomes "3".
3168 * This still overestimates under most circumstances. If we were to pass the
3169 * start and end offsets in here as well we could do block_to_path() on each
3170 * block and work out the exact number of indirects which are touched. Pah.
3173 int ext4_writepage_trans_blocks(struct inode *inode)
3175 int bpp = ext4_journal_blocks_per_page(inode);
3176 int indirects = (EXT4_NDIR_BLOCKS % bpp) ? 5 : 3;
3179 if (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL)
3180 return ext4_ext_writepage_trans_blocks(inode, bpp);
3182 if (ext4_should_journal_data(inode))
3183 ret = 3 * (bpp + indirects) + 2;
3185 ret = 2 * (bpp + indirects) + 2;
3188 /* We know that structure was already allocated during DQUOT_INIT so
3189 * we will be updating only the data blocks + inodes */
3190 ret += 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
3197 * The caller must have previously called ext4_reserve_inode_write().
3198 * Give this, we know that the caller already has write access to iloc->bh.
3200 int ext4_mark_iloc_dirty(handle_t *handle,
3201 struct inode *inode, struct ext4_iloc *iloc)
3205 /* the do_update_inode consumes one bh->b_count */
3208 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
3209 err = ext4_do_update_inode(handle, inode, iloc);
3215 * On success, We end up with an outstanding reference count against
3216 * iloc->bh. This _must_ be cleaned up later.
3220 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
3221 struct ext4_iloc *iloc)
3225 err = ext4_get_inode_loc(inode, iloc);
3227 BUFFER_TRACE(iloc->bh, "get_write_access");
3228 err = ext4_journal_get_write_access(handle, iloc->bh);
3235 ext4_std_error(inode->i_sb, err);
3240 * Expand an inode by new_extra_isize bytes.
3241 * Returns 0 on success or negative error number on failure.
3243 static int ext4_expand_extra_isize(struct inode *inode,
3244 unsigned int new_extra_isize,
3245 struct ext4_iloc iloc,
3248 struct ext4_inode *raw_inode;
3249 struct ext4_xattr_ibody_header *header;
3250 struct ext4_xattr_entry *entry;
3252 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
3255 raw_inode = ext4_raw_inode(&iloc);
3257 header = IHDR(inode, raw_inode);
3258 entry = IFIRST(header);
3260 /* No extended attributes present */
3261 if (!(EXT4_I(inode)->i_state & EXT4_STATE_XATTR) ||
3262 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
3263 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
3265 EXT4_I(inode)->i_extra_isize = new_extra_isize;
3269 /* try to expand with EAs present */
3270 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
3275 * What we do here is to mark the in-core inode as clean with respect to inode
3276 * dirtiness (it may still be data-dirty).
3277 * This means that the in-core inode may be reaped by prune_icache
3278 * without having to perform any I/O. This is a very good thing,
3279 * because *any* task may call prune_icache - even ones which
3280 * have a transaction open against a different journal.
3282 * Is this cheating? Not really. Sure, we haven't written the
3283 * inode out, but prune_icache isn't a user-visible syncing function.
3284 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
3285 * we start and wait on commits.
3287 * Is this efficient/effective? Well, we're being nice to the system
3288 * by cleaning up our inodes proactively so they can be reaped
3289 * without I/O. But we are potentially leaving up to five seconds'
3290 * worth of inodes floating about which prune_icache wants us to
3291 * write out. One way to fix that would be to get prune_icache()
3292 * to do a write_super() to free up some memory. It has the desired
3295 int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
3297 struct ext4_iloc iloc;
3298 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3299 static unsigned int mnt_count;
3303 err = ext4_reserve_inode_write(handle, inode, &iloc);
3304 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
3305 !(EXT4_I(inode)->i_state & EXT4_STATE_NO_EXPAND)) {
3307 * We need extra buffer credits since we may write into EA block
3308 * with this same handle. If journal_extend fails, then it will
3309 * only result in a minor loss of functionality for that inode.
3310 * If this is felt to be critical, then e2fsck should be run to
3311 * force a large enough s_min_extra_isize.
3313 if ((jbd2_journal_extend(handle,
3314 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
3315 ret = ext4_expand_extra_isize(inode,
3316 sbi->s_want_extra_isize,
3319 EXT4_I(inode)->i_state |= EXT4_STATE_NO_EXPAND;
3321 le16_to_cpu(sbi->s_es->s_mnt_count)) {
3322 ext4_warning(inode->i_sb, __FUNCTION__,
3323 "Unable to expand inode %lu. Delete"
3324 " some EAs or run e2fsck.",
3327 le16_to_cpu(sbi->s_es->s_mnt_count);
3333 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3338 * ext4_dirty_inode() is called from __mark_inode_dirty()
3340 * We're really interested in the case where a file is being extended.
3341 * i_size has been changed by generic_commit_write() and we thus need
3342 * to include the updated inode in the current transaction.
3344 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
3345 * are allocated to the file.
3347 * If the inode is marked synchronous, we don't honour that here - doing
3348 * so would cause a commit on atime updates, which we don't bother doing.
3349 * We handle synchronous inodes at the highest possible level.
3351 void ext4_dirty_inode(struct inode *inode)
3353 handle_t *current_handle = ext4_journal_current_handle();
3356 handle = ext4_journal_start(inode, 2);
3359 if (current_handle &&
3360 current_handle->h_transaction != handle->h_transaction) {
3361 /* This task has a transaction open against a different fs */
3362 printk(KERN_EMERG "%s: transactions do not match!\n",
3365 jbd_debug(5, "marking dirty. outer handle=%p\n",
3367 ext4_mark_inode_dirty(handle, inode);
3369 ext4_journal_stop(handle);
3376 * Bind an inode's backing buffer_head into this transaction, to prevent
3377 * it from being flushed to disk early. Unlike
3378 * ext4_reserve_inode_write, this leaves behind no bh reference and
3379 * returns no iloc structure, so the caller needs to repeat the iloc
3380 * lookup to mark the inode dirty later.
3382 static int ext4_pin_inode(handle_t *handle, struct inode *inode)
3384 struct ext4_iloc iloc;
3388 err = ext4_get_inode_loc(inode, &iloc);
3390 BUFFER_TRACE(iloc.bh, "get_write_access");
3391 err = jbd2_journal_get_write_access(handle, iloc.bh);
3393 err = ext4_journal_dirty_metadata(handle,
3398 ext4_std_error(inode->i_sb, err);
3403 int ext4_change_inode_journal_flag(struct inode *inode, int val)
3410 * We have to be very careful here: changing a data block's
3411 * journaling status dynamically is dangerous. If we write a
3412 * data block to the journal, change the status and then delete
3413 * that block, we risk forgetting to revoke the old log record
3414 * from the journal and so a subsequent replay can corrupt data.
3415 * So, first we make sure that the journal is empty and that
3416 * nobody is changing anything.
3419 journal = EXT4_JOURNAL(inode);
3420 if (is_journal_aborted(journal))
3423 jbd2_journal_lock_updates(journal);
3424 jbd2_journal_flush(journal);
3427 * OK, there are no updates running now, and all cached data is
3428 * synced to disk. We are now in a completely consistent state
3429 * which doesn't have anything in the journal, and we know that
3430 * no filesystem updates are running, so it is safe to modify
3431 * the inode's in-core data-journaling state flag now.
3435 EXT4_I(inode)->i_flags |= EXT4_JOURNAL_DATA_FL;
3437 EXT4_I(inode)->i_flags &= ~EXT4_JOURNAL_DATA_FL;
3438 ext4_set_aops(inode);
3440 jbd2_journal_unlock_updates(journal);
3442 /* Finally we can mark the inode as dirty. */
3444 handle = ext4_journal_start(inode, 1);
3446 return PTR_ERR(handle);
3448 err = ext4_mark_inode_dirty(handle, inode);
3450 ext4_journal_stop(handle);
3451 ext4_std_error(inode->i_sb, err);